Types of studies

For longer-term studies, the GDG decided that the washout period for this review should be at least two weeks. Trials with no washout were not included in the analysis and trials with one week washout were considered if there was no other information. Crossover studies are not appropriate for short-term (rescue) medication.

Types of participants

For this review, participants with IBS were included, but the review was extended to include people with simple constipation as well because the same drugs are used. Studies in these participants were regarded as indirect as far as population was concerned.

Types of intervention

Studies were to include the following interventions:

• Bulk-forming laxatives, which, as a class, include some dietary fibres (see fibres review), but here we consider only non-dietary bulk forming agents:
  • Ispaghula husk (trade names: Fibrelief®; Fybogel®; Isogel®; Ispagel Orange®; Regulan®)
  • Methylcellulose (trade name: Celevac®)
  • Sterculia (trade names: Normacol®; Normacol Plus®).
• Stimulant laxatives:
  • Biascodyl (trade name: ^*Dulco-lax®), given as either oral tablets or rectal suppositories
  • Docusate sodium (synonym: dioctyl sodium sulphosuccinate; trade names: Dioctyl® (oral); Docusol® (oral); Norgalax Micro-enema® (rectal))
  • Glycerol (synonym: glycerine), given as rectal suppositories
  • Senna (non proprietary tablets; trade names: Senokot® granules; Manevac® granules (senna fruit 12.4%, ispaghula 54.2%)), given as an oral preparation
  • Sodium picosulphate (Trade names: Laxoberal® (oral elixir); ^*Dulco-lax® Liquid (oral elixir); ^*Dulco-lax Perles® (oral capsules)), given as an oral preparation.
• Faecal softeners:
  • Arachis oil (Trade name: Fletchers’Arachis Oil Retention Enema®), given as a rectal preparation
  • Liquid paraffin (Liquid Paraffin Oral Emulsion, BP), given as oral preparation.
• Osmotic laxatives:
  • Lactulose (trade names include: Duphalac®; Lactugal®; Regulose®), given as oral solution
  • Macrogols (synonyms: polyethylene glycols, PEG; trade names: Idrolax® (oral powder, PEG 4000); Movicol® (oral powder, PEG 3350); Movicol®-Half (oral powder, PEG 3350))
  • Magnesium salts (Magnesium Hydroxide Mixture, BP, oral aqueous suspension; Liquid Paraffin and Magnesium Hydroxide Oral Emulsion, BP (oral aqueous suspension), Magnesium Sulphate (Epsom Salts); trade name: Milpar®)
  • Sodium phosphate – this is not in the BNF but is in routine use and so was included by the GDG.

The following comparisons were included:

• Laxative versus placebo (or nothing)
• Laxative type 1 versus type 2
• Laxative dose 1 versus dose 2
• Laxative + another intervention versus the other intervention alone
• Laxative route of delivery 1 versus route 2
• Duration of treatment 1 versus duration 2.

NB: In spite of the large placebo effect associated with IBS, comparisons with no treatment are included.

The laxatives review was concerned with both longer-term maintenance treatment and short-term symptom relief.

The GDG decided that there should be a minimum duration of treatment of four weeks for maintenance in this review. Maintenance studies of shorter durations were not included in the analysis.

Subgroup analyses

We carried out subgroup analyses by type of laxative (bulk forming laxatives; stimulant laxatives; faecal softeners, and; osmotic laxatives); dose; route of delivery (oral, rectal), and; duration of intervention.

Study Design

There were four crossover studies (Cleveland 2001; Connolly 1975; Quah 2006; Sobhani 1996) in which participants were allocated to receive both the intervention and control treatments during the course of the study, in a random order. Two of these studies (Cleveland 2001; Sobhani 1996) had either no washout period or it was not reported, in which case this was assumed to be none. The other two studies had a washout period of one week. The GDG had specified a washout period of two weeks minimum, but would consider the two one-week studies if there was no other data. No crossover studies reported first-period results only or individual patient data.

The remaining 15 studies had a parallel design (Attar 1999; Bouhnik 2004; Chaussade 2003; Corazziari 1996; Corazziari 2000; Dettmar 1998; DiPalma 2000; Hamilton 1988; Kienzle-Horn 2006; Kienzle-Horn 2007; Marlett 1987; Medoff 2004; Rouse 1991; Wang 2004; Wulkow 2007).

The GDG had specified a minimum treatment period of four weeks for each intervention in the maintenance studies. Two studies had a treatment duration of one week (Connolly 1975; Marlett 1987); one had ten days (Hamilton 1988); three had two weeks (Cleveland 2001; DiPalma 2000; Wang 2004); one had three weeks (Sobhani 1996). All these studies were transferred to the excluded studies table.

The remaining studies had durations of four weeks (Attar 1999; Bouhnik 2004; Chaussade 2003; Dettmar 1998; Kienzle-Horn 2007; Medoff 2004; Quah 2006 (1-week washout crossover); Rouse 1991); eight weeks (Corazziari 1996), and; 20 weeks (Corazziari 2000).

Two studies investigated the use of laxatives for acute constipation (Kienzle-Horn 2006; Wulkow 2007).

Twelve studies were therefore included in the analysis (11 parallel and one 1-week washout crossover trial: Attar 1999; Bouhnik 2004; Chaussade 2003; Corazziari 1996; Corazziari 2000; Dettmar 1998; Kienzle-Horn 2006; Kienzle-Horn 2007; Medoff 2004; Quah 2006; Rouse 1991; Wulkow 2007).

One of the remaining studies had more than two arms: Chaussade (2003) compared four PEG interventions; there were thus 14 comparisons in the laxatives review. The rest of the description of studies will focus on these studies/comparisons.

Two were conducted in the UK (Dettmar 1998 and Rouse 1991); one in sites in the UK and France (Attar 1999); seven in the rest of Europe, one in each of the USA and China.

Setting: Seven studies took place in primary care (Bouhnik 2004; Chaussade 2003; Dettmar 1998; Kienzle-Horn 2006; Medoff 2004; Rouse 1991; Wulkow 2007); and five were in secondary care (Attar 1999 (with 31% from geriatric institutions); Corazziari 1996 Corazziari 2000; Kienzle-Horn 2007; Quah 2006).

The majority of studies (7/12) had fewer than 100 patients, with two having 25 or fewer in the intervention arm (Corazziari 1996; Medoff 2004). Two studies had more than 200 patients in total (Chaussade 2003; Dettmar 1998).

Funding: Six studies had some industry sponsorship: Bouhnik (2004) was sponsored by Solvay Pharmaceuticals, manufacturers of lactulose; Chaussade (2003) was supported by a grant from by Hoffmann La Roche, the manufacturers of PEG 3350, and; the Dettmar (1998) authors were from Reckitt & Colman, manufacturers of Fybogel. Kienzle-Horn (2006), Kienzle-Horn (2007) and; Wulkow (2007) were all funded by Boehringer-Ingelheim GmbH, manufacturers of Bisacodyl, and Sodium Picosulphate.

Population

• Only one study (Medoff 2004) definitely included patients with IBS: 7/43 patients had a diagnosis of constipation predominant IBS, but separate results were not reported. All the other studies stated they had patients with simple constipation, usually defined as 3 or less, or 2 or less bowel movements per week. Most studies defined a minimum period of constipation symptoms, ranging from 3 weeks (Rouse 1991) to 12 months (Corazziari 1996; Corazziari 2000). The GDG suspected, however, that many of these studies may have had the classification of ‘simple constipation’ because this was the primary symptom being treated by laxatives, rather than the only symptom. Indeed, they thought it was quite likely that the patients had IBS. This was further investigated. Communication with the authors of one study (Kienzle-Horn 2007) revealed that abdominal pain was not excluded and the author believed that some patients entering the studies would have had IBS if they had been checked for it by their physician. Seven of these studies (Attar 1999; Bouhnik 2004; Corazziari 1996; Corazziari 2000; Chaussade 2003; Dettmar 1998; Rouse 1991) reported that some patients had pain and/or bloating. However, Dettmar (1998) and Rouse (1991) did not report sufficient duration to be defined as IBS. Two studies did not mention the incidence of pain or bloating before treatment (Kienzle-Horn 2006; Quah 2006) but the patients had had constipation for at least 3 months.
• In Attar (1999), 20% and 35% of the control group had pain and bloating respectively during the trial. The patients were stated to have had chronic idiopathic constipation for at least 3 months.
• In Bouhnik (2004), in the washout period, 45 and 53% had bloating at washout and 30 and 45% had pain for lactulose and PEG respectively. The patients were stated to have had chronic idiopathic constipation for at least 6 months.
• In Chaussade (2003), at baseline, the bloating score was ~ 3 points on a scale of 1 to 4 (considerable) and pain was 2.6. There was an implied use of the Rome II criteria for chronic idiopathic constipation, which the patients were to have had for at least 3 months
• In Corazziari (1996), in the run-in period 52–60% pts had pain and 84–91% had bloating. The patients were stated to have chronic non-organic constipation and had had this for at least 12 months.
• In Corazziari (2000), the pain and bloating scores were non zero, even after the patients had received PEG for 4 weeks. Chronic constipation was defined using Rome criteria and the patients had had chronic constipation for at least 12 months.
• Dettmar (1998) reported that the majority of the patients experienced abdominal symptoms, including pain, distension or flatulence. The patients were said to have ‘simple constipation’ and there were no details about the duration of constipation.
• In Rouse (1991) 53–54% patients in both groups had abdominal pain after seven days. Bloating was not mentioned. The patients were treated for chronic constipation, which they had had for at least 3 weeks.

The GDG concluded that all of these studies, with the exception of Dettmar (1998) and Rouse (1991) were likely to have some, if not all, patients with IBS. It was unclear if Kienzle-Horn (2006) and Quah (2006) included patients with IBS. Further details are given in the included studies table (Appendix C).

One study (Corazziari 2000) gave PEG electrolyte to patients for 4 weeks and then randomised only the responders (at least two bowel movements per week with no other defaecatory disturbances or more than three bowel movements per week) to PEG electrolyte or placebo.

The age range of participants across the studies was 18 to 89 years, with the mean age (where given) ranging from 42 to 58 years. All the studies had more women than men.

Interventions

The studies varied in the type of laxatives used:

• Two used stimulant laxatives:
  • One bisacodyl (Kienzle-Horn 2006)
  • One sodium picosulphate (Wulkow 2007).
  • Eight had osmotic laxatives:
  • Four lactulose (Attar 1999; Bouhnik 2004; Quah 2006; Rouse 1991)
  • Five polyethylene glycol (Attar 1999; Chaussade 2003; Corazziari 1996; Corazziari 2000; Bouhnik 2004)
  • One sodium phosphate (Medoff 2004).
• One study (Dettmar 1998) allowed the patient any laxative (which was mainly lactulose) and also reported the lactulose patients as a subgroup.

Comparisons

The included studies covered the following comparisons:

• Four comparisons of laxatives versus placebo:
  • Two gave stimulant laxatives (Kienzle-Horn 2006; Wulkow 2007 treatment for acute episodes)
  • Two gave osmotic laxatives (Corazziari 1996; Corazziari 2000).
• Three studies compared a laxative with fibre:
  • Two compared an osmotic laxative (lactulose) with fibre (ispaghula), (Quah 2006; Rouse 1991)
  • One compared usual laxatives (mainly lactulose) with fibre (ispaghula) (Dettmar 1998).
• Seven comparisons of different types of laxative in the same class:
  • Two studies (Attar 1999; Bouhnik 2004) compared lactulose with PEG electrolyte (Osmotic Laxatives)
  • Four comparisons of PEG 3350 (Transipeg) plus electrolytes versus PEG 4000 without electrolytes (Forlax) (Chaussade 2003 x4) (Osmotic Laxatives)
  • One comparison of Bisacodyl versus Sodium Picosulphate (Kienzle–Horn 2007).
• Three comparisons of different doses of an osmotic laxative:
  • One comparison of 11.8g versus 5.9g PEG 3350 (Transipeg) plus electrolytes (Chaussade 2003)
  • One comparison of 20g versus 10g PEG 4000 (Forlax) (Chaussade 2003)
  • One comparison of 10.56g (mean) versus 6.84g (mean) sodium phosphate.

I. Treatment for acute episodes of constipation

A. Laxatives versus placebo

Two studies (Kienzle-Horn 2006; Wulkow 2007) in 112 patients compared laxatives with placebo for the treatment of acute episodes of constipation. It was unclear if the patients had IBS. Stimulant laxatives 10mg bisacodyl (Kienzle-Horn 2006) or 7mg sodium picosulphate (Wulkow 2007) or placebo was given once-a-day for three days.

1. Global symptoms

Global symptoms (pain, bloating and bowel habit) were not reported.

2. Individual symptoms

a. Bowel habits

i. Number of patients with improvement in bowel habit assessed by investigators

The investigators assessed the improvement in bowel habit, based on diary recordings of the patients. Overall, the relative risk was 1.34 (95%CI 1.02, 1.76) (Figure 1), i.e. statistically significant difference between laxative and placebo (p=0.04). This corresponded to a number needed to treat (NNT) of 6 (95%CI 3, 50) for a control group risk of 52 to 61%. There was no heterogeneity.

Figure 1

ii. Stool score - consistency

The consistency of stool was measured on the scale of 1 to 5, where 5=hard, 4=moderately hard, 3=well-formed, 2=soft, 1=liquid (Kienzle-Horn 2006) and on a 4 point scale where 4=hard, 3=well-formed, 2=pasty,1=liquid (Wulkow 2007). The Wulkow study reported the number of patients with soft and/or well formed stools. The relative risk was 1.51 (95%CI 1.06, 2.15) i.e. statistically significant difference between laxative and placebo (p=0.02) favouring laxative (Figure 1).

Kienzle-Horn (2006) reported baseline mean scores which were 5.0 for each group, so any decrease in score constituted an improvement. The study did not report the standard deviation for the placebo group, but gave the difference in change score between the two groups and the 95%CI (Figure 2). This was −1.4 (95%CI −2.0, −0.76), for a placebo group score of 4.2, i.e. a statistically significant difference between groups, such that the bisacodyl group had a value between soft and well-formed.

Figure 2

iii. Stool frequency

The stool frequency per day was statistically significantly higher for the bisacodyl group: mean difference 0.85 (95%CI 0.24, 1.46) for a placebo group mean of 0.95 stools/day (Figure 3).

Figure 3

3. Adverse effects

The confidence interval was too wide to decide if there was a difference in the number of patients reporting adverse effects that could have been drug related (Figure 4).

Figure 4

II. Laxatives for maintenance treatment

A. Laxatives versus Placebo

There were two studies included in the analysis that compared laxatives with placebo in patients with constipation (Corazziari 1996 and Corazziari 2000). Both studies gave the patients an isosmotic PEG electrolyte balanced solution (PMF-100) containing 14.6g PEG 4000, twice a day. However, in Corazziari (2000), all patients received 4 weeks of PMF-100 initially, with responders (more than 3 bowel movements per week) then randomised to PEG or placebo for a further 20 weeks. Thus the populations were different in the two trials, and Corazziari (2000) was regarded as an investigation of the effects of stopping the laxative. Their results are therefore reported separately. Both studies had patients who were outpatients in secondary care. The GDG considered it likely that both studies had at least some patients with IBS.

In both trials, patients were allowed other laxatives when they had no bowel movements for at least 5 consecutive days, and they were allowed to adjust the intervention dose downwards (but not upwards above 2 sachets per day).

Where outcomes were measured at different times during the study, we took the end-study results unless there were significant numbers of withdrawals or problems with compliance. Therefore, for the Corazziari (2000) study we took the values at eight weeks (i.e. from the start of randomisation).

1. Global symptoms

Neither study reported global symptoms.

2. Number of patients using additional laxatives / not using additional laxatives as rescue medication

The GDG considered this to be an important outcome for this review and gave it the status of primary outcome measure. We gave both the number of patients using additional laxatives (as reported in the papers), and the number not using additional laxatives (calculated).

The comparison of PEG and placebo over 8 weeks in 48 patients (Corazziari 1996) showed a statistically significant decrease in the number of patients using other laxatives as rescue medication; Figure 5; RR 0.33 (95%CI 0.12, 0.90), although the confidence interval was wide. This corresponded to a number needed to treat of 4 (95%CI 2, 15) for a placebo group risk of 48%.

Figure 5

Number of patients taking rescue laxatives.

For the outcome measure, the number of patients not using rescue medication was calculated for Corazziari (1996) (Figure 6). There was a statistically significant difference between PEG and placebo, favouring the former; RR 1.61 (95%CI 1.05, 2.47), which gave an NNT of 4 (95%CI 2, 15) for a control group risk of 52%.

Figure 6

Number of patients not taking rescue laxatives.

Corazziari (2000) (withdrawal of laxative following four weeks PEG electrolyte solution, in responders) reported that the use of other oral laxatives, rectal evacuants, suppositories and enemas was more frequent in the placebo group compared to the PEG group. At eight weeks the difference in number of other laxatives used per four weeks was statistically significant (Figure 6), but the confidence interval was fairly wide.

At the same time, statistically significantly more sachets of the intervention were used in the placebo group (even though the more severely constipated patients dropped out), compared with the PEG group (Figure 7).

Figure 7

In the PEG group, the authors reported that the use of other laxatives progressively decreased in the PEG group but increased in the placebo group.

3. Individual symptoms

a. Pain

There was no significant difference between PEG and placebo for this outcome (Corazziari 1996) in 48 patients. The confidence intervals were fairly wide so there was some uncertainty over the results for the difference between groups in the number of patients with abdominal pain (Figure 8).

Figure 8

Corazziari (2000) (withdrawal of laxative following 4 weeks PEG electrolyte solution, in responders) reported that the abdominal pain score progressively decreased in the PEG group and increased in the placebo group. No data were given.

b. Bloating

The comparison of PEG and placebo (Corazziari 1996) in 48 patients showed no statistically significant difference between groups in the number of patients with bloating (Figure 9).

Figure 9

Corazziari (2000) (withdrawal of laxative following 4 weeks PEG electrolyte solution, in responders) reported that bloating was less severe in the PEG group compared to the placebo group throughout the study. At 8 weeks the difference was statistically significant: p<0.001. No other statistics were given.

c. Bowel habits

i. Stool frequency

The comparison of PEG and placebo over 8 weeks in 48 patients (Corazziari 1996) showed a statistically significant increase in stool frequency per week (figure 10) of 2.00 (95%CI 0.89, 3.11), for a placebo group value of 2.8 stools per week.

Figure 10

Corazziari (2000) (withdrawal of laxative following 4 weeks PEG electrolyte solution, in responders) found a statistically significant increase in stool frequency per week for the PEG group compared to the placebo group throughout the study. At 8 weeks (Figure 11) the difference was 3.13 (95%CI 1.35, 4.91) for a placebo group value of 4.39 stools per week.

Figure 11

ii. Improvement in bowel habit

Corazziari (2000) (withdrawal of laxative following 4 weeks PEG electrolyte solution, in responders) found a statistically significantly greater number of patients with complete remission of constipation symptoms (more than three bowel movements per week, no use of other laxatives, no straining at defecation, no feeling of incomplete evacuation, no hard/pellety stools) for the PEG group compared to the placebo group throughout the study. At 8 weeks (Figure 12) the RR was 3.95 (95%CI 1.86, 8.42); this corresponds to an NNT of 2 (95%CI 2, 3) for a control group rate of 18%.

Figure 12

Figure 13

iii. Number of patients withdrawing from study

Corazziari (2000) (withdrawal of laxative following 4 weeks PEG electrolyte solution, in responders) reported statistically significantly more patients withdrew from the study by 20 weeks because of non-response to treatment in the placebo group compared to the PEG group. At the end of the study the RR was 0.13 (95%CI 0.03, 0.53), i.e. statistically significantly in favour of the PEG group, although the confidence interval was very wide. This corresponded to an NNT of 3 (95%CI 2, 5) for a placebo group rate of 46%.

Figure 14

4. Adverse effects

In the comparison of PEG and placebo over 8 weeks in 48 patients (Corazziari 1996), there was too much uncertainty to determine if there was a difference in the number of patients reporting anorexia, headache or asthenia (Figure 15).

Figure 15

Corazziari (2000) reported that there were no significant differences between groups in the incidence of adverse effects.

B. Osmotic Laxative type 1 versus Osmotic laxative type 2

Three studies compared different types of osmotic laxatives: two (Attar 1999; Bouhnik 2004) compared lactulose with PEG, and one compared different types of PEG: PEG 3350 plus electrolytes versus PEG 4000 without electrolytes (Chaussade 2003). We noted that the Bouhnik (2004) was sponsored by Solvay Pharmaceuticals, manufacturers of lactulose, and Chaussade (2003) was supported by a grant from by Hoffmann La Roche, the manufacturers of PEG 3350. The Attar (1999) study was in secondary care (of which 31% were in geriatric institutions), and the Chaussade (2003) and Bouhnik (2004) studies were in primary care. The GDG thought it likely that all of these studies had some patients with IBS.

B1. Lactulose versus PEG

Attar (1999) compared PEG 3350 plus electrolytes (Movicol) versus lactulose, and Bouhnik (2004) compared PEG 4000 plus electrolytes (Forlax) versus lactulose. Both studies had a duration of four weeks. The doses of PEG differed in the two studies: the patients in Attar (1999) started with 26.24 g (2 sachets) for the first two weeks, but could change to 1 or 3 sachets for the second two weeks. Patients in Bouhnik (2004) started at a dose of 20g (2 sachets) for the first week and then this could be varied to 10 or 30g. The lactulose dose in both studies was 20g which could also be varied as above.

In Attar (1999) patients could take suppositories or microenemas for relief of constipation, apparently without restriction. However, in Bouhnik (2004) patients were asked to stop enema/suppositories 48 hours before the first stool collection.

1. Global outcomes

a. Global improvement in symptoms score

One study (Attar 1999) in 99 patients recorded a global improvement score at four weeks on a VAS of 0 to 10 (0=no change, 10=excellent); comprising pain, bloating and bowel habit. The global improvement score was statistically significantly in favour of PEG electrolyte; mean difference 2.20 (95%CI 1.05, 3.35) for a control group value of 5.20.

Figure 16

2. Use of microenemas as rescue medication

Attar (1999) also reported the number of patients using microenemas as rescue medication after four weeks, and we also calculated the number of patients not using rescue medication. The study found that statistically significantly more patients used microenemas in the lactulose group than in the PEG group; RR 0.48 (95%CI 0.25, 0.95), which corresponded to an NNT of 6 (95%CI 3, 50) for a lactulose group risk of 35% (Figure 17a). The confidence interval was fairly wide.

Figure 17a

Number of patients using microenemas.

There was a statistically significant difference, favouring PEG, for the number of patients not using microenemas; RR 1.27 (95%CI 1.02, 1.59). This corresponded to an NNT of 6 (95%CI 3, 50) for a lactulose risk of 65% (Figure 17b).

Figure 17b

Number of patients not using microenemas.

3. Number of sachets of intervention used

In Attar (1999) the number of sachets of laxative used over four weeks was statistically significantly lower for the PEG group (Figure 18), but there was no difference between groups in Bouhnik (2004). This led to significant heterogeneity between studies (I²=83%, p=0.02). It is unclear if this was an effect of dose; type of PEG; use of other laxatives, or; any other reason.

Figure 18

4. Individual symptoms

a. Pain

Both studies reported the number of patients with abdominal pain at four weeks. Meta-analysis of 180 patients gave a wide confidence interval.

Figure 19

One study (Attar 1999) recorded pain on a scale of 0 to 3 (severe). The difference was not statistically significant.

Figure 20

b. Bloating

Both studies reported the number of patients with bloating at four weeks. Meta-analysis of 180 patients gave a fairly wide confidence interval and some heterogeneity (I²=50%; p=0.16). This difference may be an effect of dose or type of PEG. It is also noted that Bouhnik (2004) was sponsored by the manufacturers of lactulose and Attar (1999) allowed the patients to use other laxatives ad libitum.

Figure 21

c. Bowel habits

i. Stool frequency

Both studies reported the stool frequency per day at four weeks. Meta-analysis of 180 patients gave a statistically significant difference of 0.27 (95%CI 0.09, 0.45) stools/day, favouring PEG, but there was some heterogeneity (I²=50%; p=0.16).

5. Adverse effects

Both studies reported the number of patients with adverse effects at four weeks. Meta-analysis of 180 patients gave a wide confidence interval and no heterogeneity (Figure 22). One study (Attar 1999) reported the number of patients with liquid stools, but this also had a wide confidence interval (Figure 23).

Figure 22

Figure 23

Figure 24

B2. Comparison of different PEG laxatives

One study (Chaussade 2003) compared two doses of each of two types of PEG solution for a duration of four weeks. The PEG species were PEG 4000 (Forlax) without electrolytes and PEG 3350 plus electrolytes. Doses used were the maximum and standard recommended by the manufacturers.

1. Global improvement of symptoms

The study measured the patients’ global impression of efficacy on a VAS, but the results are not reported. The authors state that there was no significant difference between groups.

2. Individual symptoms

a. Abdominal pain

Pain scores at four weeks were recorded on a scale of 1 (none) to 4 (considerable). There was no significant difference between the two types of PEG for this outcome at either dose, and there was no heterogeneity (I²=0%).

Figure 25

b. Bloating

Bloating scores at four weeks were recorded on a scale of 1 (none) to 4 (considerable). There was no significant difference between the two types of PEG for this outcome at either dose, and no heterogeneity (I²=0%).

Figure 26

c. Bowel habits

i. Stool frequency per week

There was no significant difference between the two types of PEG for this outcome at four weeks at either dose, and no heterogeneity (I²=0%).

Figure 27

ii. Stool consistency

Consistency of stools at four weeks were recorded on a scale of 1 (liquid) to 6 (very hard). Meta-analysis of the two comparisons revealed heterogeneity (I²=65%, p=0.09). For the standard dose, there was a statistically significant difference between the two PEG solutions, favouring PEG 4000; mean difference 0.30 (95%CI 0.01, 0.59). This was a fairly small change. There was no significant difference for the maximum dose.

Figure 28

iii. Number of patients with normal stools

Meta-analysis showed no significant difference at four weeks between types of PEG.

3. Quality of life

The study measured quality of life on a 100mm VAS at four weeks. Meta-analysis showed no significant difference between types of PEG and no heterogeneity (I²=0%, p=0.93).

Figure 30

4. Adverse effects

For overall adverse effects at four weeks, the majority of which were gastrointestinal, meta-analysis showed no significant difference between types of PEG and no heterogeneity (I²=0%, p=0.55).

Figure 31

For the specific adverse effect of diarrhoea, there was no significant difference between types of PEG and no heterogeneity (I²=0%, p=0.71).

Figure 32

B3. Comparison of different doses of PEG laxatives

One study (Chaussade 2003) compared two doses of each of two types of PEG solution for a duration of four weeks. The PEG species were PEG 4000 (Forlax) without electrolytes and PEG 3350 plus electrolytes. Doses used were the maximum and standard recommended by the manufacturers.

1. Global improvement of symptoms

The study measured the patients’ global impression of efficacy at four weeks on a VAS, but the results are not reported. The authors state that there was no significant difference between groups.

2. Individual symptoms

a. Abdominal pain

Pain scores at four weeks were recorded on a scale of 1 (none) to 4 (considerable). Meta-analysis showed no significant difference between the two doses for this outcome, and no heterogeneity (I²=0%).

Figure 33

b. Bloating

Bloating scores at four weeks were recorded on a scale of 1 (none) to 4 (considerable). There was no significant difference between the two doses for this outcome, and no heterogeneity (I²=0%).

Figure 34

c. Bowel habits

i. Stool frequency

There was no significant difference between the two doses for this outcome at four weeks, and no heterogeneity (I²=0%).

Figure 35

ii. Stool consistency

Consistency of stools at four weeks was recorded on a scale of 1 (liquid) to 6 (very hard). Meta-analysis of the two comparisons revealed some heterogeneity (I²=65%, p=0.09). For the PEG 3350 electrolyte dose, there was a statistically significant difference between the two doses, favouring the maximum dose; mean difference 0.60 (95%CI 0.29, 0.91). There was no significant difference for the PEG 4000.

Figure 36

iii. Number of patients with normal stools

Meta-analysis of the two comparisons reveals statistically significantly more patients with normal stools at four weeks for the standard dose groups. The RR was 1.68 (95%CI 1.14, 2.48), which corresponded to an NNH of 7 (95%CI 4, 25) for the higher dose rate of 19% or 25%.

Figure 37

3. Quality of life

The study measured quality of life at four weeks on a 100mm VAS. Meta-analysis showed no significant difference between doses and no heterogeneity (I²=0%, p=0.93).

Figure 38

4. Adverse effects

For overall adverse effects at four weeks, the majority of which were gastrointestinal, meta-analysis showed no significant difference between doses and no heterogeneity (I²=0%, p=0.55).

Figure 39

For the specific adverse effect of diarrhoea, there was a statistically significant difference, favouring the standard dose, and no heterogeneity (I²=0%, p=0.68). The RR was 0.41 (95%CI 0.24, 0.70); this corresponded to an NNT of 6 (95%CI 4, 13) for the higher dose rate of 30%.

B4. Stimulant Laxative Type 1 versus Stimulant Laxative Type 2

One study compared two stimulant laxatives, bisacodyl versus sodium picosulphate (Kienzle-Horn 2007). Patients were treated daily for 4 weeks with 5 to10mg of either bisacodyl or sodium picosulphate. The primary outcome was the change in bowel habit recorded as the mean number of bowel movements per day and stool consistency measured on a 5 point scale where 5=hard, 4= moderately hard,3= well formed,2=soft, 1=liquid. Secondary outcomes included straining scored on a 4 point scale with 4 = severe and 0=absent. There was no statistically significant difference between the two laxatives for the number of bowel movements per day, WMD: −0,05 (95%CI −0.18, 0.08), and similarly for the stool consistency and straining score. Both were equally effective in treating constipation.

Figure 41

C. Laxative versus fibre

Three studies compared a laxative with fibre: two compared an osmotic laxative (lactulose) with fibre (ispaghula), (Quah 2006; Rouse 1991) and one compared usual laxatives (mainly lactulose) with fibre (ispaghula) (Dettmar 1998). Quah (2006) had a crossover design, with a washout period of 1 week, and this study was treated separately. Quah (2006) and Rouse (1991) compared respectively: 20 ml lactulose with 3.5g ispaghula husk, and 30 ml lactulose with 7g ispaghula husk. Dettmar (1998) did not record the dose of lactulose (other laxatives), but 7g ispaghula husk was given. Dettmar (1998) also reported results for the lactulose subgroup of ‘other laxatives’. Rouse (1991) and Dettmar (1998) were in primary care and Quah (2006) in secondary care. The authors of Dettmar (1998) were from Reckitt and Colman, manufacturers of Fybogel. The GDG considered it unlikely that the patients in Dettmar (1998) and Rouse (1991) had IBS. It was unclear if the patients in Quah (2006) had IBS.

1. Global symptoms

Two studies reported an outcome of global symptoms (Rouse 1991; Dettmar 1998).

a. Global improvement of symptoms

One study (Rouse 1991) in 112 patients showed little difference in global improvement of symptoms at four weeks between patients given 30ml lactulose and 7g ispaghula husk (figure 40).

Figure 40

Figure 42

a. Global effectiveness

Dettmar (1998), in 315 patients, asked the patients to rate the effectiveness at four weeks of treatment with ispaghula 7g and other laxatives, mainly lactulose. Statistically significantly more patients given ispaghula rated the effectiveness as excellent, good or satisfactory. RR (all other laxatives versus fibre) was 0.87 (95%CI 0.80, 0.96) and for the subgroup with lactulose the RR was 0.90 (95%CI 0.85, 0.96). It was noted that authors of Dettmar (1998) were from the manufacturers of ispaghula (Reckitt and Colman).

Figure 43

When these two studies were combined in a meta-analysis, the RR was 0.92 (0.85, 1.00) and there was a significant heterogeneity I² = 73.5%, p=0.05. It is unclear what caused this, but the overall effect was small.

Figure 44

2. Individual symptoms

a. Pain

Two studies (Rouse 1991; Quah 2006) in 93 and 78 patients respectively, recorded the number of patients with abdominal pain at four weeks. We did not combine these studies because one was a crossover study and the other parallel. We did not draw conclusions for the crossover study because the confidence interval was too wide; there was also only 1 week washout for this study. The confidence interval was fairly wide for Rouse (1991), but there was no significant difference between lactulose and ispaghula.

b. Bloating

Two studies recorded the number of patients with bloating at four weeks: Quah (2006) in 76 patients, and; Dettmar (1998) in 394 patients. There was little difference in the numbers with bloating, although the confidence intervals were fairly wide.

Figure 46

c. Bowel habits

i. Improvement in bowel score

One crossover study (Quah 2006) with 78 patients recorded improvement in bowel score at four weeks compared with baseline on a scale of 0 (no effect) to 10 (excellent) (Figure 45). There was a statistically significantly greater improvement with lactulose compared to ispaghula; mean difference 1.40 (95%CI 0.19, 2.61). It was noted that this crossover study had a washout period of only 1 week, so the results were treated with caution.

Figure 45

Figure 47

ii. Stool frequency

One study reported the stool frequency at four weeks (Quah 2006). There was a non-significant difference between lactulose and ispaghula, favouring the former; WMD 1.80 (95%CI −0.12, 3.72). It was noted that this crossover study had a washout period of only 1 week, so the results were treated with caution.

iii. Stool consistency

One study reported the stool consistency at four weeks on a scale of 0 (no bowel movement) to 3 (comfortable and solid) to 5 (loose) (Quah 2006). There was a borderline significant difference of 0.50 (95%CI 0.00, 1.00; p=0.05) between lactulose and ispaghula, favouring the former. However, since the normal rating is 3 and the fibre group is closer to this value (2.9) it could be argued that fibre is more favourable. It was noted that this crossover study had a washout period of only one week, so the results were treated with caution.

Figure 49

3. Adverse effects

Two studies reported the number of patients with adverse effects at four weeks. In all cases there were wide confidence intervals (Figure 48).

Figure 48

Figure 50

4. Patient preference

The crossover study, Quah (2006), recorded patient preference at four weeks between lactulose and ispaghula. Statistically significantly more patients preferred lactulose; RR 1.71 (95%CI 1.05, 2.79). This gave an NNT of 4 (95%CI 3, 25). It was noted that this crossover study had a washout period of only one week, so the results were treated with caution.

Figure 51

Adverse Effects

An adverse effects review has been carried out and is reported in section 8.5.1. The review included six RCTs (Quah 2006; Ferguson and Attar 1999; Bouhnik 2004; Corazziari 1996; Corazziari 2000; Chaussade 2003), and their results are reported in this effectiveness review. The RCTs were primarily aimed at assessing and reporting on the efficacy of the drug treatments. Evaluation of safety and reporting of adverse effects data was often cursory or nonexistent. Even in instances where the methods sections had explicitly stated the intention of monitoring for adverse effects, trial reports did not follow a structured format (e.g. by WHO system organ class) of reporting adverse effects. The interventions and comparators were extremely varied, as was the reporting of adverse effects.

One non-randomised study in the US reported a series of adverse effects of laxatives, but did not distinguish laxative class, and used doses higher than in the UK.

Many of the adverse outcomes of interest are very similar to the symptoms of the IBS itself. For instance, laxatives are associated with flatulence, cramps and abdominal pain – all of which are commonly seen in untreated IBS patients and also form part of the efficacy assessment. It is not always possible to determine whether deterioration in these symptoms is due to lack of efficacy, or the natural history of the disease, or the adverse effect of the drug. Generally, though, the RCT data on lactulose was consistent with the findings of the non-randomised data with regards to increased risk of abdominal symptoms. The GDG’s clinical experience of lactulose was that it caused bloating and other side effects.

Modelled response rates

In the basecase scenario the response rate of 45% in the no treatment arm is taken from the Mearin (2004) cohort study. This represents the group of patients whose symptoms improve without any specific intervention. The RR of response for PEG versus placebo is 1.61; therefore the response rate in the PEG arm is 72% (=45% x 1.61), giving an absolute difference in response between the intervention and no treatment arms of 27% (=72%–45%) during the first month for PEG. The RR of bisacodyl and sodium picosulfate is 1.34 compared to placebo, so if these interventions are used first line then we would expect an absolute difference in response between intervention and no treatment of 15% (=1.34*45%–45%). The first line use of bisacodyl and sodium picosulfate has not been modelled due to a lack of longer-term data on their effectiveness compared to placebo.

In the basecase scenario the response rate for the subsequent interventions is assumed to be equal to the response rate to the first intervention. If bisacodyl and sodium picosulfate are used second line in patients who do not respond to PEG, we would expect an additional 4.2% (=15% x 28%) of the original cohort to respond to the second laxative, and an additional 3.6% (=15% x 24%) to respond to the third laxative, giving an overall response rate of 80% for laxatives. The response rate over time for the basecase is given in Figure 52. It is assumed that bisacodyl is tried first after a failure to respond to PEG as it has a lower cost than sodium picosulfate.

Figure 52

Modelled response rates for laxatives (PEG followed by two switches to other laxatives for non responders) and no treatment.

We have also considered an alternative scenario in which no patient in the comparator arm achieves an improvement in symptoms, but the absolute gain in response rates is maintained from the basecase (e.g. for first line PEG, we modelled a zero response to no treatment but a 27% response to PEG for this scenario).

Table 2 gives the incremental cost-effectiveness for several laxative treatment pathways in order of the benefits they achieve. It shows that whilst PEG provides additional benefit for a cost per QALY of £7,779, compared to no treatment, further benefit can be achieved by allowing non responders to PEG to switch to bisacodyl and if that is not effective to switch to sodium picosulfate. Each of these additional switches for non responders has a low cost per QALY compared to no further treatment for non-responders (£4,488 and £6,561 respectively). However, it should be noted that the cost-effectiveness of these second line laxatives is based on clinical effectiveness evidence for bisacodyl and sodium picosulfate from short term trials lasting only 3 days.

Table 2

Incremental cost-effectiveness of allowing subsequent switches in laxative therapy.

These results are an estimate of the cost-effectiveness over the first 6 months after the initiation of laxative therapy. The cost per QALY for continuing laxative therapy beyond 6 months is lower than the cost per QALY during the initial 6 months provided that treatment is reviewed every 6 months and discontinued in patients who no longer experience a therapeutic benefit, either due to lack of effectiveness or a change in their symptom profile.

The probabilistic sensitivity analysis provides an estimate of the uncertainty in the cost per QALY estimate due to uncertainty in the efficacy estimate, the probability of response in the no treatment arm and the utility gain. The CEAC in Figure 53 shows the uncertainty surrounding the cost-effectiveness of PEG, compared to no treatment and the incremental cost-effectiveness of allowing non-responders to switch to other laxatives. It shows that there is an 83% likelihood that the cost per QALY for PEG compared to no treatment is under £20K, suggesting that PEG provides health benefit at an acceptable cost in patients with IBS-C. Allowing non responders to PEG one treatment switch has an 81% probability of a being cost-effective when a £20K threshold is applied compared to no further treatment for non responders. Similarly, allowing a second treatment switch for non-responders has a 74% probability of being cost-effective.

Figure 53

CEAC for PEG with up to two switches for non responders compared to no treatment (NT).

However, it should be noted that these estimates only consider the uncertainty in cost-effectiveness due to the accuracy of several input parameters and they do not reflect general uncertainty around the assumptions made in the model. The uncertainty from these assumptions was explored in the univariate sensitivity analysis.

Univariate sensitivity results for laxatives

The results of the univariate sensitivity analysis for PEG compared to no treatment are given in Table 3. Maintaining the 27% difference in response between the two arms but reducing the response rate in the no treatment arm from 45% to zero decreased the cost per QALY to £4,896. The use of higher cost formulations increased the cost per QALY to £9,980, whilst assuming that 50% of prescriptions were over the counter reduced the cost per QALY to £4,814.

Table 3

Sensitivity results for PEG compared to no treatment for 100 patients with IBS-C.

The cost per QALY is less favourable for patients who only use the medication on 25% of days as the upfront costs of initiating therapy and establishing response are constant despite lower benefit from less frequent use. PEG compared to no treatment has a cost per QALY of £13,325 when used on 25% of days.

We carried out a threshold analysis to determine whether laxative therapy would still be cost-effective for lower gains in health related quality of life. In the basecase it was assumed that patients who respond to therapy accumulate 0.071 QALYs more per annum than patients who do not respond. For comparison, a gain of 0.135 QALYs would represent a complete remission of IBS symptoms. If the QALY gain associated with a response to therapy was reduced to 0.027 QALYs, then the cost per QALY of providing PEG compared to no treatment would be above £20,000 per QALY. The methods used in this analysis vary from the methods used for other pharmacological interventions as we were unable to estimate the response rate in terms of an improvement in global symptoms from the trial data available. Instead we assumed that patients who did not use another laxative during the trial had responded to the trial intervention. This outcome was considered a less reliable indicator of whether there had been an overall improvement in HRQoL than an improvement in global symptoms. However, the threshold analysis shows that laxatives are cost-effective even if the utility gain associated with a therapeutic response is small.

We carried out a similar univariate sensitivity analysis on the incremental cost-effectiveness of allowing non responders to switch to an alternative laxative. The incremental cost per QALY for the first and second switches was increased to £8,784 and £12,624 when assuming that patients who demonstrate no response to PEG would be half as likely to respond to another laxative. The incremental cost per QALY for each subsequent treatment switch in non-responders was higher for patients using treatments intermittently with a cost per QALY of £8,468 for the first switch and a cost per QALY of £11,536 for the second switch in patients who use laxatives on only 50% of days. When laxatives are used on only 25% of days, the incremental cost per QALY estimates for the first and second switches are £16,428 and £21,485 respectively.

If a patient also takes another medication (an antispasmodic), then this medication can be reviewed at the same time, so it may be cost-effective to provide both therapies. For example, if laxatives are prescribed with the antispasmodic and both used on 25% of days then allowing up to 2 switches of both treatments was estimated to be cost-effective with a cost per QALY of £10,107 compared to no treatment a cost per QALY of £17,393 compared to 1 switch.

Types of participants

For this review, participants were required to have IBS and not to have inflammatory bowel disease or diarrhoea subsequent to surgery. This inclusion criterion was adhered to for the longer term maintenance studies, but, for short term relief of symptoms investigations, there were insufficient data for IBS patients. Therefore, for this section of the review only, the GDG extended the population, post-hoc, to include studies in patients with acute diarrhoea of any cause (including those with diarrhoea caused by infection or virus). Such studies were regarded as indirect as far as the population was concerned.

Types of studies

The GDG decided that the washout period for this review should be at least one week. Trials with shorter washout periods were not included in the analysis.

Types of intervention

Studies included the following interventions:

• Codeine phosphate
• Co-phenotrope (diphenoxylate and atropine mixture; Trade name: Lomotil®)
• Loperamide
  • Single drug: loperamide hydrochloride (Trade names: Norimode®, Imodium®)
  • Compound preparation: loperamide hydrochloride and simeticone (Trade name: Imodium® Plus)
• Morphine
  • Kaolin and Morphine mixture BP
  • Morphine preparations on sale to the public.

The following comparisons were included:

• Antimotility agent versus placebo (or nothing)
• Antimotility agent type 1 versus type 2
• Antimotility agent dose 1 versus dose 2
• Antimotility agent + another intervention versus the other intervention alone
• Antimotility agent delivery mode 1 versus delivery mode 2
• Duration of treatment 1 versus duration 2.

NB: In spite of the large placebo effect associated with IBS, comparisons with no treatment are included.

The antimotility agents review was concerned with both longer term maintenance treatment and short-term symptom relief.

For maintenance studies, the GDG had decided that there should be a minimum duration of treatment of four weeks, but on further reflection agreed to include studies of two weeks or more. Studies of shorter durations were excluded. Short-term/symptom relief studies had duration of less than one week.

Subgroup analyses

We planned to carry out subgroup analyses by type of antimotility agent, dose, mode of delivery (modified release/conventional) and duration of intervention.

Study Design

There were seven crossover studies (Allison 1988; Cann 1984; Corbett 1980; Harford 1980; Palmer 1980; Pelemans and Vantrappen 1976; Verhaegen 1974) in which participants were allocated to receive both the intervention and control treatments during the course of the study, in a random order. Four of these studies (Allison 1988; Cann 1984; Corbett 1980; Palmer 1980) had either no washout period or it was not reported, in which case this was assumed to be none. One acute study had a washout period of 12 to 24 hours (Harford 1980). One longer term study had a washout period of 2 to 7 days, and the drugs were discontinued until severe diarrhoea returned (Verhaegen 1974). The other longer term study (Pelemans and Vantrappen 1976) had a washout period of 3 to 20 days (median 7 days). As the GDG had specified a washout period of one week minimum for maintenance studies, the latter three studies were included on the basis of washout period, but those with no washout were excluded from the analysis and transferred to the excluded studies table. The remaining studies had a parallel design.

The GDG had specified a minimum treatment period of four weeks for each intervention in the maintenance studies. One study had a treatment duration of one week (Barbezat 1979); one had a duration of at least 10 days or until relapse (Tijtgat 1975); one had at least 12 days (Verhaegen 1974), one had from 14 to 49 days, median 25 days (Pelemans and Vantrappen 1976); one had three weeks (Hovdenak 1987). The GDG decided to exclude Barbezat (1979); Tijtgat (1975), and; Verhaegen (1974) on this basis, and to accept Hovdenak (1987), and; Pelemans and Vantrappen (1976). The remaining studies had durations of seven weeks (Efskind 1995), 2 months (Taneja 2004) and 13 weeks (Lavö 1987).

Thus, 15 studies were included in the analysis: thirteen parallel (Amery 1975; Cornett 1977; Dettmer 1994; Dom 1974; Dreverman 1995; Efskind 1995; Ericsson 1990; Hovdenak 1987; Jaffe 1977; Lavö 1987; Lee 1968; Lustman 1987; Taneja 2004), and two crossover (Harford 1980; Pelemans and Vantrappen 1976).

Ten studies investigated the use of anti-motility agents for acute diarrhoea (Amery 1975; Cornett 1977; Dettmer 1994; Dom 1974; Dreverman 1995; Ericsson 1990; Harford 1980; Jaffe 1977; Lee 1968; Lustman 1987). Only one of these had some patients with IBS (Harford 1980). Five studies investigated the effectiveness of anti-motility agents for the treatment chronic diarrhoea in IBS.

Three studies had more than two arms. Amery (1975) compared diphenoxylate with loperamide and placebo. Dettmer (1994) and Dreverman (1995) compared two dose of loperamide with placebo, giving a total of 21 comparisons.

Setting: seven studies were in primary care (Amery 1975; Dom 1974; Dreverman 1995; Efskind 1995; Jaffe 1977; Lee 1968; Lustman 1987); three were in secondary care (Dettmer 1994; Lavö 1987; Taneja 2004) and the others did not report the setting.

Funding: four studies (Amery 1975; Cornett 1977; Dettmer 1994; Dreverman 1995) were from Janssen Pharmaceutica (manufacturers of Imodium, i.e., loperamide) and Efskind (1995) stated that Janssen Pharmaceutica provided the drug, monitored the study and gave statistical support. Lustman (1987) did not specify funding, but the corresponding author was employed by Gold Cross Pharmaceuticals, a division of GD Searle & Co Ltd (manufacturers of Lomotil). Three studies were funded by non-industry sources (Ericsson 1990; Harford 1980; Taneja 2004) and the others did not state their funding.

Population

Four studies were in patients with IBS (Efskind 1995; Hovdenak 1987; Lavö 1987; Taneja 2004), and two studies had some patients with IBS. The acute study, Harford (1980), had 4/15 patients with IBS; and the maintenance study, Pelemans and Vantrappen (1976), had 4/23; 18 of the remaining patients in the latter study had inflammatory bowel disease. In both studies, individual patient data were reported for the IBS subgroup, but these were not stratified before randomisation and the small numbers give uncertainty and likely potential for bias. The definition of IBS varied between studies: one included patients meeting the Rome II criteria (Taneja 2004), one met criteria defined by the authors that were similar to the Rome criteria (Efskind 1995), and in the other studies, the authors stated that the patients had IBS, with no further explanation.

Most studies included patients with diarrhoea predominance, but one study (Hovednak 1987) had a mixture of types: IBS-D (16); IBS-A with pain (21); IBS-A without pain (12); IBS-C (9). None of the studies stated that any participants had IBS as result of gastrointestinal infection. The majority of studies did not state the number of participants with bloating. One study had some patients with bloating (Efskind 1995).

Most of the studies did not describe symptom severity. Two studies stated that participants had symptoms of mixed severity (Efskind 1995; Hovednak 1987), one of which excluded patients with mild symptoms (Hovednak 1987).

The remaining studies were in patients who did not have IBS and these were treated as indirect evidence. Further details are given in the included studies table.

The age range of participants across the IBS studies was 18 to 70 years, with the mean age, where given, ranging from 31 to 43 years. No study particularly identified elderly participants. The indirect studies included patients aged 9 to 95 years, with four of the studies definitely including children: Amery (1975) included patients aged 9 to 82 years, with a median age of 31 years; Cornett (1977) had an age range of 11 to 84 years, with 8% patients in the age group 10 to 19 years; Dom (1974) had a range of 14 to 95 years with a median of 35 years; and Drevermann (1995) had patients aged 16 to 75 years.

Five studies had more women than men (Efskind 1995; Harford 1980; Lavö 1987; Lee 1968; Lustman 1987); two studies had about the same number of men and women (Dettmer 1994; Pelemans and Vantrappen 1976); one study examined only men (Taneja 2004) and the other indirect studies had more men than women. One study did not report the numbers of men and women (Ericsson 1990).

Interventions

The studies varied in the type of antimotility agent used:

• No studies examined codeine phosphate
• Seven acute studies gave the patients co-phenotrope (Amery 1975; Cornett 1977; Dom 1974; Harford 1980; Jaffe 1977; Lee 1968; Lustman 1987). The Amery study stated that they gave the patients 2.5mg diphenoxylate and that the contents of the capsules were identical to Lomotil (co-phenotrope)
• Twelve studies (seven acute) gave the patients loperamide (Amery 1975; Cornett 1977; Dettmer 1994; Dom 1974; Dreverman 1995; Efskind 1995; Ericsson 1990; Hovdenak 1987; Jaffe 1977; Lavö 1987; Pelemans and Vantrappen 1976; Taneja 2004)
• One acute study examined morphine (Lee 1968 used a kaolin and morphine mixture).

Ten of the included studies used antimotility agents for acute diarrhoea (Amery 1975; Cornett 1977; Dettmer 1994; Dom 1974; Dreverman 1995; Ericsson 1990; Harford 1980; Jaffe 1977; Lee 1968; Lustman 1987).

In the maintenance studies, no study allowed rescue medication, but patients were allowed to vary the dose of study drug in three studies (Efskind 1995; Lavö 1987; Pelemans and Vantrappen 1976). A fixed dose was used in the remaining studies.

Comparisons

The included studies covered the following comparisons:

• Ten comparisons of antimotility agent versus placebo:
  • Three gave co-phenotrope for acute episodes (Amery 1975; Harford 1980; Lustman 1987)
  • Four gave loperamide for acute episodes (Amery 1975; Dettmer 1994, in a dose of 1mg or 2mg; Dreverman 1995 in a dose of 1mg or 0.5mg; Ericsson 1990 2mg)
  • Three gave loperamide for maintenance treatment (Efskind 1995; Hovdenak 1987; Lavö 1987).
• Three comparisons of different types of antimotility agent:
  • Four studies compared loperamide and diphenoxylate for acute episodes (Amery 1975; Cornett 1977; Dom 1974; Jaffe 1977)
  • One study compared loperamide and co-phenotrope for maintenance treatment (Pelemans and Vantrappen 1976)
  • One study compared co-phenotrope with kaolin-and-morphine for acute episodes (Lee 1968).
• Two acute studies compared two doses of loperamide (Dettmer 1994: 1mg versus 2mg; Dreverman 1995 1mg versus 0.5mg)
• One maintenance study compared loperamide with yoga (Taneja 2004).

I. TREATMENT FOR ACUTE EPISODES OF DIARRHOEA

For the treatment of acute diarrhoea the GDG simply wished to know whether anti-motility agents were effective in stopping diarrhoea, regardless of its cause. Therefore only outcomes relating to bowel habit are reported.

A. Anti-motility agent versus placebo

Three studies gave co-phenotrope for acute episodes (Amery 1975; Harford 1980; Lustman 1987).

• Amery (1975) gave 5 mg diphenoxylate per day (1 tablet twice a day) for 24 hours
• Harford (1980) gave 10 or 20 mg per day (1 or 2 tablets 4 times a day) for 3 days.
• Lustmann (1987) gave 10 mg initially (4 tablets) then 20mg per day (2 tablets four times per day) for 3 days.

The recommended initial dose for co-phenotrope is initially 10 mg (4 tablets) then 20mg per day (2 tablets four times a day), thus the Amery (1975) study can be considered to have a low dose of diphenoxylate.

Three studies gave loperamide for acute episodes compared with placebo (Amery 1975; Dettmer 1994, 2 doses; Dreverman 1995, 2 doses).

• Amery (1975) gave 4 mg loperamide per day (1 tablet twice a day) for 24 hours
• Dreverman (1995a) gave 1mg loperamide (2 tablets) initially then up to 3.5mg (7 tablets) per day for 3 days
• Dreverman (1995b) gave 2mg loperamide (2 tablets) initially then up to 7mg (7 tablets) per day for 3 days
• Dettmer (1994a) gave loperamide in a slow release formulation, 2 mg initially (2 tablets) then up to 8 mg (8 tablets) per day for 3 days
• Dettmer (1994b) gave loperamide in a slow release formulation, 4 mg initially (2 tablets) then up to 16 mg (8 tablets) per day for 3 days.

The recommended dose is 4 mg initially (2 tablets) followed by 2 mg (1 tablet) after each loose stool for up to 5 days.

We noted that all of the loperamide studies were funded by Janssen Pharmaceutica, manufacturers of Imodium (i.e. loperamide).

A1. Co-phenotrope versus placebo

i. Number of patients with improvement in bowel habit

One study (Amery 1975) reported the number of patients without recurrence (unformed stools) after 1, 2, 4 and 24 hours. There was no significant difference between diphenoxylate and placebo at any duration.

Figure 1Acute diarrhoea – co-phenotrope

ii. Stool frequency

Harford (1980) recorded stool frequency averaged over 3 days of co-phenotrope or placebo in a crossover study in 15 patients; individual patient data were given. Results are reported for all patients and for the IBS subgroup separately. For all patients, there was a statistically significant difference in stool frequency of −2.29 stools per day (95%CI −4.47, −0.11), favouring co-phenotrope. For the IBS subgroup, the effect was not statistically significant, but the confidence interval was wide.

Figure 2Stool frequency

Lustman (1987) reported the change in the median number of bowel actions (24 hours prior to start of therapy minus first 24 hours of treatment). This fell from 5 in each group to 3 on diphenoxylate, compared to 4 on placebo, which was a statistically significant difference (p=0.046).

iii. Time to recurrence of unformed stools

Amery (1975) also reported the median time to first recurrence of unformed stools. This was 2 hours for both diphenoxylate and placebo groups, i.e. no significant difference (p=0.48).

Lustman (1987) reported the median time to the last loose or watery stool, which was 25 hours on diphenoxylate compared with 30 hours for placebo (not statistically significant).

Lustman (1987) also reported the median time (hours) to the start of an interval of at least 12 hours between bowel actions: 14 hours for diphenoxylate vs. 24 hours for placebo, p=0.025.

iv. Stool weight

Harford (1980) recorded stool weight averaged over 3 days of co-phenotrope or placebo in a crossover study in 15 patients. The results are reported for all patients and for the IBS subgroup separately. The confidence interval was too wide to determine if there was a difference for all patients and there was no statistically significant difference for the IBS patients.

Figure 3Stool weight

A2. Loperamide versus placebo

i. Number of patients with improvement in bowel habit

Amery (1975) reported the number of patients without recurrence (unformed stools) after 1, 2, 4 and 24 hours. Dettmer (1994) also recorded the number of patients without diarrhoea at 3 days for the two loperamide doses combined.

At 1 and 2 hours Amery (1975) showed no significant difference between loperamide and placebo. After 4 hours there were significantly fewer patients with unformed stools in the loperamide group compared to placebo, with a number needed to treat of 5 (95%CI 3, 17), for a control group risk of 36%. After 24 hours, loperamide showed fewer patients with unformed stools, with borderline significance, but the confidence interval was fairly wide. At 72 hours, Dettmar (1994) showed a small statistically significant difference between loperamide and placebo, with an NNT of 7 (95%CI 4, 34), for a control group risk of 73%.

Figure 4Acute diarrhoea – loperamide

Dreverman (1995) reported the number of patients achieving first relief (the start of a 24 hour period during which no more than 1 pasty stool and no watery or loose stools were passed). There were statistically significantly more patients achieving relief for each dose of loperamide compared with placebo, but the confidence intervals are wide.

Figure 5

ii. Time taken to first relief of symptoms

Two studies reported the median time to first relief of symptoms, but different doses were used.

• 0.5 mg versus placebo
  Dreverman (1995) in 156 patients: 20 hours 15 minutes for loperamide 0.5mg versus 24 hours 50 minutes for placebo (p=0.012), i.e. statistically significantly in favour of loperamide
• 1 mg versus placebo (2 studies)
  • Dettmer (1994): 22.4 hours for loperamide 1mg versus 30 hours for placebo
  • Dreverman (1995) in 158 patients: 15 hours 30 minutes for loperamide 1mg versus 24 hours 50 minutes for placebo (p=0.003), i.e. highly statistically significant, in favour of loperamide
• 2 mg versus placebo
  • Dettmer (1994): 22.1 hours for loperamide 1mg versus 30 hours for placebo.

In Dreverman (1995), the median time to complete relief of symptoms was 26 hours 30 minutes for loperamide 1mg; 25 hours 40 minutes for loperamide 0.5mg; and 34 hours 15 minutes for placebo (p=0.041 for 0.5 mg and 0.044 for 1mg).

iii. Time to first recurrence

Amery (1975) also reported the median time to first recurrence of unformed stools. For the loperamide and placebo groups, this was 24, and 2 hours respectively. The p value for the difference between loperamide and placebo was p=0.016, i.e. statistically significant.

B. Anti-motility agent dose 1 versus dose 2

i. Number of patients with improvement in bowel habit

Dreverman (1995) reported the number of patients achieving first relief. There was no significant difference between the doses and the confidence interval was fairly wide.

Figure 6

ii. Median time to first relief of symptoms

• 0.5 mg versus 1 mg
  Dreverman (1995): 20 hours 15 minutes for loperamide 0.5mg versus 15 hours 30 minutes for loperamide 1mg
• Dettmer (1994): 22.4 hours for loperamide 1mg versus 22.1 hours for loperamide

In Dettmar (1994), the median time to complete relief of symptoms was 27.55 hours for loperamide 1mg and 25.00 hours for loperamide 2mg, with no significant difference between dose groups.

C. Anti-motility agent type 1 versus type 2

C1. Co-phenotrope versus loperamide

Four studies compared co-phenotrope and loperamide for acute episodes (indirect population) (Amery 1975; Cornett 1977; Dom 1974; Jaffe 1977). Although different doses were compared across studies, the proportions of drugs were the same.

• Amery (1975): 5 mg diphenoxylate per day (1 tablet twice a day) versus loperamide 4 mg (2 mg twice a day) for 24 hours.
• Cornett (1977): initially 5mg (2 capsules) co-phenotrope then up to 20 mg per day versus 4 mg (2 capsules) loperamide initially, then up to 16 mg per day
• Dom (1974): initially 5 mg (2 capsules) co-phenotrope then up to 25 mg per day versus 4 mg loperamide intitially (2 capsules) then up to 20 mg per day
• Jaffe (1977): initially 10 mg (4 capsules) co-phenotrope then 20 mg per day (2 capsules × 4) versus 4 mg (2 capsules) loperamide initially, then up to 16 mg per day.

We noted that Amery (1975) and Cornett (1977) were funded by Janssen Pharmaceutica, manufacturers of Imodium (i.e. loperamide).

i. Number of patients with improvement in bowel habit

Three studies reported the number of patients with no unformed stools at different durations: Amery (1975) reported the number of patients without recurrence (no unformed stools) after 1, 2, 4 and 24 hours; Dom (1974) recorded the same outcome after 24 hours and 2 and 3 days; Cornett (1977) recorded the number of patients with unformed stools, from which we calculated the number with no unformed stools after 24 hours and 2 and 3 days. Jaffe (1977) reported the number of patients not reaching a ‘cure’ after 4 days (figure 10). At 1 and 2 hours there were statistically significantly more patients who were diarrhoea free for the loperamide group compared to the diphenoxylate. NNTs are 4 (95%CI 3, 12) and 5 (9%CI 3, 34). There was no significant difference at 4 hours. At 24 hours, meta-analysis of three studies showed statistically significantly more patients had no unformed stools for the loperamide group; RR 0.78 (95%CI 0.62, 0.98); with some heterogeneity (I²=47%, p=0.15). The NNT was 20 (95%CI 10, 100) for a control group rate of 21 to 41%. At 2 and 3 days there was still a statistically significant effect favouring loperamide, but there was significant heterogeneity for the 3 day meta-analysis. We note, however, that some of these trials were industry sponsored.

Figure 10

Figure 7Number of patients with no unformed stools

ii. Time taken to first stools

Jaffe (1977) reported the mean time to the 1^st, 2^nd, and 3^rd motion. In all cases the confidence intervals were wide.

Figure 8

Amery (1975) reported the median time to first recurrence of unformed stools. For the loperamide and diphenoxylate groups, this was 24 and 2 hours respectively. This was a statistically significant difference (p=0.024), favouring loperamide. We note, however, that this was an industry funded study.

iii. Stool frequency

Jaffe (1977) reported the stool frequency in the first 6 hours; first 12 hours; first 24 hours; and first 48 hours following the first dose of the medication. In all cases the confidence intervals were wide, so conclusions were not drawn.

Figure 9

Cornett (1977) also reported the frequency of unformed stools over 72 hours. From a mean of 9.14 unformed stools before treatment, patients on diphenoxylate reduced to 5.47 (a fall of 3.67) while those on loperamide fell from 8.75 to 4.38 (a fall of 4.37). No standard deviations were given.

Dom (1974) reported the change in the mean number of unformed stools in a 72 hour period (from 8.06 before treatment with diphenoxylate and 8.02 in the loperamide group to 3.68 and 2.65 respectively, i.e. a fall of 4.38 and 5.37 respectively, this was statistically significantly in favour of loperamide (p=0.011).

iv. Time to cure

Jaffe (1977) reported the mean time to cure. There was a wide confidence interval so conclusions were not drawn.

C2. Co-phenotrope versus morphine

i. Number of patients with normal stool frequency

One study (Lee 1968) compared cophenotrope-with-neomycin to kaolin-and-morphine. This study reported the number of patients with abnormal stool frequency after 12 hours, 24 hours, 2 days, 3 days and 4 days. We have reported the number of patients with normal stool frequency. There was a statistically significant difference favouring co-phenotrope at durations up to 48 hours. At 12 hours the NNT was 4 (95%CI 3, 6), for a control group risk of 14%.

Figure 11

ii. Number of patients with normal stool consistency

One study (Lee 1968) compared cophenotrope-with-neomycin to kaolin-and-morphine. This study reported the number of patients with abnormal stool consistency after 12 hours, 24 hours, 2 days, 3 days and 4 days. We have reported this as the number of patients with normal stool consistency. There was a statistically significant effect, favouring co-phenotrope at 12 hours (RR 3.49 (95%CI 1.60, 7.60); NNT 5 (95%CI 4, 10), for a control group risk of 9%. Otherwise there was no significant difference between interventions.

Figure 12

II. MAINTENANCE TREATMENT FOR CHRONIC DIARRHOEA

A. Antimotility agent versus placebo

Three studies gave loperamide for maintenance treatment (Efskind 1995; Hovdenak 1987; Lavö 1987). All were in patients with IBS, although Hovdenak (1987) stated that the patients had different types of IBS, including IBS-C. Subgroup analyses were presented, but these did not constitute stratification before randomisation. This study also had a duration of 3 weeks, but the GDG agreed that this was acceptable. We noted that Efskind (1995) was industry supported (by Janssen Pharmaceutica, the manufacturers of loperamide).

The dose in Hovdenak (1987) was fixed at 4 mg at night, whilst the patients were able to adjust the dose in the other two studies. Both started the patients on 2 mg at night, which was increased to 6 mg (Efskind 1995) or 8 mg (Lavö 1987) as required.

1. Global symptoms

a. Number of patients with improvement in global symptoms

One study (Hovdenak 1987) reported the number of patients with improvement in global symptoms after 3 weeks. The study reported results for three of the different IBS subgroups (IBS-C results were not reported). The exclusion of the IBS-C results breaks the randomisation, so these are post-hoc subgroups, but still gives some information. We have grouped together the results for IBS-D, IBS-A (with pain), IBS-A (without pain), and the separate subgroups. There was a statistically significant improvement in symptom score for the 3-type IBS group and for the IBD-D group alone, although the confidence interval for the latter was wide. The relative risks were 2.00 (95%CI 1.15, 3.48) and 4.00 (1.20, 13.28) respectively. These corresponded to a number needed to treat of 3 (95%CI 2, 8) and 2 (95%CI 1, 3), for control group rates of 39% and 25% respectively.

Figure 13

b. Global symptom improvement score

One study (Lavö 1987) reported a subjective overall response for the whole 13 week study period, which was said to statistically significantly favour loperamide (p<0.03).

2. Individual symptoms

a. Pain

i. Number of patients with less pain

Two studies reported the number of patients with less pain, one (Hovdenak 1987) after 3 weeks in a range of different IBS subgroups; and the other (Lavö 1987) after 13 weeks in patients with IBS-D only. For the former study we grouped together the results for IBS-D, IBS-A (with pain), IBS-A (without pain) and also reported these separately.

The confidence intervals were wide for each study. Meta-analysis of the 3-type IBS group with the Lavö study, in 70 patients gave a statistically significant difference between interventions, favouring loperamide; RR 2.60 (95%CI 1.02, 6.61) but there was some heterogeneity (I²=48%, p=0.17). The NNT was 5 (95%CI 3, 25), for a control group rate of 8 to 17%.

ii. Number of patients with increased pain

Two studies reported the number of patients with more pain, one (Hovdenak 1987) after 3 weeks in a range of different IBS subgroups; and the other (Lavö 1987) after 13 weeks in patients with IBS-D only. For the former study we grouped together the results for IBS-D, IBS-A (with pain), IBS-A (without pain) and also reported these separately.

The confidence intervals were wide for each study. Meta-analysis of the 3-type IBS group with the Lavö (1987) study, in 70 patients, gave a statistically significant difference between interventions, favouring loperamide; RR 0.38 (95%CI 0.15, 0.96) and there was no heterogeneity (I²=0%), although the confidence interval was wide. The NNT was 5 (95%CI 3, 25), for a control group risk of 22 to 67%. For the meta-analysis of 40 patients with IBS-D, the difference was still statistically significant with no heterogeneity; RR 0.36 (95%CI 0.14, 0.96), i.e. 3 times more risk of pain for the placebo group (Figure 14). The NNT was 3 (95%CI 2, 13) for a control group risk of 38 to 67%, but the confidence interval was wide.

Figure 14

Number of patients with less pain.

iii. Number of days with pain

One study (Hovdenak 1987) recorded the number of days with pain over 3 weeks. In the subgroup of IBS-A with pain, there was a statistically significant mean difference of 6.1 days between loperamide and placebo (p<0.01), favouring the former.

iv. Pain score

One study (Efskind 1996) in 69 patients reported no significant difference between interventions in the change from baseline in abdominal pain measured on a visual analogue scale. This was the case for both their 2 week dose adjustment period and for 5 weeks of fixed dose loperamide or placebo. However, the study found significantly more pain at night for the loperamide group (p<0.05 for both periods). In contrast, another study in 21 patients (Lavö 1987) found a statistically significant difference in pain score, favouring loperamide (p<0.05).

b. Bowel habit

i. Number of patients with improved bowel habit (frequency)

Two studies reported the number of patients with improved stool frequency, one (Hovdenak 1987) after 3 weeks in different IBS subgroups; and the other (Lavö 1987) after 13 weeks in patients with IBS-D only. For the former study we grouped together the results for IBS-D, IBS-A (with pain), IBS-A (without pain) and also reported these separately.

Meta-analysis of the 3-type IBS group with the Lavö study, in 70 patients gave a statistically significant difference between interventions, favouring loperamide; RR 2.38 (95%CI 1.53, 3.70) and there was no heterogeneity (I²=0%). This corresponded to an NNT of 2 (95%CI 2, 4) for a control group risk of 25–43%. For the meta-analysis of 40 patients with IBS-D, the difference was still statistically significant with no heterogeneity; RR 2.83 (95%CI 1.43, 5.63). This corresponded to a number needed to treat of 2 (95%CI 2, 4) for a control group risk of 25–38% (figure 15). Statistically significant improvements were also achieved for the IBS-A group, which had an NNT of 2 (95%CI 2, 4), for a control group risk of 45%. Confidence intervals for these meta-analyses were fairly wide.

Figure 15

Number of patients with more pain.

Figure 16Improvement in stool frequency

ii. Number of patients with improved bowel habit (consistency)

The same two studies reported the number of patients with improved stool consistency (Hovdenak 1987; Lavö 1987). Meta-analysis of the 3-type IBS group with the Lavö (1987) study, in 70 patients gave a statistically significant difference between interventions, favouring loperamide; but there was significant heterogeneity (I²=69%; p=0.07). This heterogeneity may possibly be explained by differences in the study duration or the intervention dose, both of which were higher for the Lavö (1987) study. For the meta-analysis of 40 patients with IBS-D, the difference was still statistically significant with some heterogeneity (I²=59%; p=0.12). Random effects analysis showed wide confidence intervals and neither analysis was statistically significant. Statistically significant improvements were also achieved for the combined IBS-A group; RR: 2.10 (95%CI 1.23, 3.58), which corresponded to an NNT of 3 (95%CI 2, 5) for a control group risk of 44%.

Figure 17Improvement in stool consistency

iii. Stool scores

Two studies (Efskind 1996 (n=69) and Lavö 1987 (n=21)) reported a stool score measure for both consistency and frequency. Each found a statistically significant improvement in stool consistency (p<0.002 and p<0.001) respectively, favouring loperamide. Efskind (1996) also found a statistically significant difference over 7 weeks for frequency (p<0.05) but Lavö (1987) found no significant difference. Lavö (1987) found a highly significant difference in the number of formed stools, favouring the loperamide group.

Figure 18Number of formed stools

iv. Number of patients with improvement in urgency

One study (Lavö 1987) in 21 patients reported the number of patients with less urgency and found this to be significantly greater for the loperamide group, although the confidence interval was wide; RR 3.00 (95%CI 1.07, 8.43). This corresponds to an NNT of 2 (95%CI 2, 7), for a control group rate of 25%.

Figure 19Number of patients with less urgency

3. Adverse effects

Only one small study in 24 patients reported the number of patients with adverse effects (Lavö 1987). In all cases there are very wide confidence intervals and conclusions cannot be drawn.

Figure 20Adverse effects

C. Antimotility agent versus yoga

1. Bowel symptom score

Taneja (2004) reported bowel symptom scores 1 and 2 months after starting either loperamide or yoga. Patients were asked the following questions: Is the frequency of stools >3 times per day? Is the frequency of stools <3 times per week? Are the stools lumpy or hard? Is there straining? Are the stools loose or watery? Is there urgency to pass the stools? Are the stools mucoid? For each symptom, if present >25% of the time, a positive answer scored 1 and a negative answer 0, so the range is 0–7. Both groups reduced their scores from baseline (loperamide 4.08 (SD 0.9); yoga 3.77 (SD 1.2)), but there was no significant difference between the groups.

Figure 21Bowel symptom score

Adverse Effects

Two RCTs with adverse effects data were identified (Lavo 1987; Cann 1984). One was a crossover trial which should be treated with caution (Cann 1984). No clear trend could be identified from these two studies but abdominal cramps, dizziness, drowsiness, and skin reactions including urticaria; paralytic ileus and abdominal bloating have been reported (BNF 2007).

Modelled response rates

In the basecase scenario the response rate of 45% in the no treatment arm is taken from the Mearin (2004) cohort study. This represents the group of patients for whom we would expect an improvement in symptoms without any specific intervention. The RR of response for antimotility vs placebo is 2.00; therefore the response rate in the intervention arm is 90% (=45% × 2.00) giving an absolute difference in response between the intervention and no treatment arms of 45% (=90%-45%) during the 1^st month.

We have also considered an alternative scenario in which no patient in the comparator arm achieves an improvement in symptoms but there is still a 45% chance of response for the intervention arm.

In the basecase analysis, treatment with loperamide for 100 patients with IBS-D or IBS-A results in an additional 1.60 QALYs but costs an additional £3,055 over a 6 month time frame, compared to no treatment, giving a cost per QALY of £1,914. The basecase analysis assumes that the intervention is used on a daily basis, meaning this estimate may be more relevant to patients with IBS-D. The cost-effectiveness of using loperamide on 25% to 75% of days is considered in the sensitivity analysis, and this estimate may be more relevant to patients with IBS-A whose primary symptom may vary from constipation to diarrhoea.

These results are an estimate of the cost-effectiveness over the first 6 months after initiating treatment with loperamide. The cost per QALY for continuing treatment with loperamide beyond 6 months is lower than the cost per QALY during the initial 6 months provided that treatment is reviewed every 6 months and discontinued in patients who no longer experience a therapeutic benefit, either due to lack of effectiveness or a change in their symptom profile.

The probabilistic sensitivity analysis considered the uncertainty in the basecase estimate of cost-effectiveness due to uncertainty in the parameters used to estimate the cost-effectiveness. The CEAC presented in Figure 22 shows the probability that the cost per QALY for loperamide compared to no treatment is under various cost per QALY thresholds. For example, it shows that there is an 83% probability that loperamide has a cost per QALY of under £5,000 compared to no treatment and a 98% probability that it has a cost per QALY under £20,000. However, it should be noted that these estimates only consider the uncertainty in cost-effectiveness due to the accuracy of several input parameters and they do not reflect general uncertainty around the assumptions made in the model. The uncertainty from these assumptions was explored in the univariate sensitivity analysis.

Figure 22

Cost-effectiveness acceptability curve for loperamide compared to no treatment.

Univariate sensitivity analysis

Maintaining the 45% difference in response between the two arms but reducing the response rate in the no treatment arm from 45% to zero marginally decreased the cost per QALY to £1,593.

We carried out a threshold analysis to determine whether antimotility therapy is still cost-effective for lower gains in health related quality of life. In the basecase it is assumed that patients who respond to therapy accumulate 0.071 QALYs more per annum than patients who do not respond. For comparison, a gain of 0.135 QALYs would represent a complete remission of IBS symptoms. If the QALY gain associated with a response to therapy was reduced to 0.006 QALYs, then the cost per QALY of providing antimotility therapy would be above £20,000 per QALY.

The basecase analysis assumes that the lowest cost preparation is prescribed. If, the highest cost preparation (Imodium syrup) is prescribed at the same dose, then the cost per QALY is increased to £3,173. If the dose is increased to the maximum dose of 16mg per day, which is double the highest dose used in the maintenance therapy trials, and the highest cost preparation is prescribed, then the monthly cost of loperamide rises to £23.85, but the cost per QALY is still well below £20,000 per QALY at £9,311.

If a patient takes a therapy on an as needed basis, it is assumed that they take the therapy on days when their quality of life is significantly affected by their IBS symptoms but not on days when their symptoms are less severe. They only accrue QALY benefits and drug costs on the days they take the therapy. However, it is still necessary to assess all patients for response after 1 month of initiating therapy. This means that it is less cost-effective to initiate therapy in patients who use the therapy on fewer days as the monitoring costs are just as high but the benefits are lower. This is shown by the cost per QALY of £5,297 for patients who take the therapy on 25% on days.

Types of studies

The GDG decided that the washout period for this review should be at least one week. Trials with shorter washout periods were not included in the analysis.

Types of intervention

Studies were to include the following interventions:

• Anti-muscarinic agents:
  • Atropine (synonyms: hyoscyamine)
  • Dicycloverine (synonyms: dicyclomine, diethylaminocarbethoxybicyclohexyl hydrochloride; trade name: Merbentyl®)
  • Hyoscine (synonyms: scopolamine; trade name: Buscopan®)
  • Propantheline (Trade name: Pro-Banthine®).
• Direct-action smooth muscle relaxants:
  • Alverine (Trade name: Spasmonal®)
  • Mebeverine: includes modified release and conventional drug delivery (Trade name: Colofac®)
  • Peppermint oil: includes modified release and enteric coated (Trade names: Colpermin®; Mintec®).

Studies with interventions not available in the UK were included in the review. These studies were listed, but not included in the sections on analysis or characteristics of studies.

The following comparisons were included

• Antispasmodic versus placebo (or nothing)
• Antispasmodic type 1 versus type 2
• Antispasmodic dose 1 versus dose 2
• Antispasmodic + another intervention versus the other intervention alone
• Antispasmodic modified release versus conventional delivery
• Duration of treatment 1 versus duration 2.

NB: In spite of the large placebo effect associated with IBS, comparisons with no treatment were included.

The antispasmodics review was concerned with both longer term maintenance treatment and short term symptom relief.

The GDG decided that there should be a minimum duration of treatment of four weeks for maintenance in this review. Studies of shorter durations were included but dealt with in sensitivity analyses.

Subgroup analyses

We planned to carry out subgroup analyses by type of antispasmodic (antimuscarinic and direct-acting smooth muscle relaxants); dose; mode of delivery (modified release/conventional), and; duration of intervention.

Search strategy for identification of studies

The initial search identified a Cochrane Review (Quartero 2005, Bulking agents, antispasmodic and antidepressant medication for the treatment of irritable bowel syndrome). Searches were partly based on the terms in this review. Searches were performed on the following core databases: MEDLINE, EMBASE, CINAHL and The Cochrane Library (1966 to current day with guidance from the GDG). Additional databases were not searched for this review. The search strategies are listed in Appendix B.

The titles and abstracts identified by the search strategy were assessed. One-hundred and three were identified as being potentially relevant to the review and these papers were retrieved in full. Twenty-three studies met the inclusion criteria for the review. The reference lists for each of the retrieved studies were inspected for further potential papers, but none were identified. The 80 excluded studies are listed in Appendix E, along with the reasons for exclusion

Study Design

Setting: The majority of studies took place in secondary care; two studies were in primary care (Gilbody 2000; GP research group); five studies did not report the setting or it was not possible to determine it because the report was not in English (Berthelot 1981; Czalbert 1990; Evans 1998; Flavell Matts 1967; Schafer 1990).

There were ten crossover studies (Dew 1984; Evans 1998; Flavell Matts 1967; GP research group; Hennessy 1975; Lawson 1988; Nash 1986; Prout 1983; Tasman-Jones 1973; van Outryve 1995) in which participants were allocated to receive both the intervention and control treatments during the course of the study, in a random order. All of these studies had either no ‘washout period’ or it was not reported, in which case this was assumed to be none. As the GDG had specified a washout period of one week minimum, all these studies were eliminated from the analysis. One additional crossover study (Carling 1989) reported first period results for the global improvement of symptoms outcome, so these data were combined with those from parallel design studies.

The remaining 12 studies had a parallel design (Berthelot 1981; Czalbert 1990; Gilbody 2000; Inauen 1994; Kruis 1986; Lech 1988; Liu 1997; Mitchell 2002; Nigam 1984; Page 1981; Richie 1979; Schafer 1990).

Thirteen studies were therefore included in the analysis (12 parallel and one crossover trial, first period only).

Two studies had more than two arms: Schafer (1990) compared hyoscine plus paracetamol and hyoscine alone with placebo; Carling (1989) compared peppermint oil with hyoscamine and placebo; Ritchie (1979) compared hyoscine with placebo, and hyoscine+lorazepam with lorazepam alone. There were thus 17 comparisons in the antispasmodics review.

The rest of the description of studies will focus on these studies/comparisons.

Population

The definition of IBS varied between studies: one used the Rome I criteria (Mitchell 2002); one used the Rome II criteria (Gilbody 2000); three met criteria defined by the authors that were similar to the above (Inauen 1994; Nigam 1984; Page 1981). In seven comparisons, the authors stated that the patients had IBS, with no further explanation (Berthelot 1981; Czalbert 1990; Liu 1997; 2 x Ritchie 1979; 2 x Schafer 1990). The remaining five comparisons (3 x Carling 1989; Kruis 1986; Lech 1988) did not use a formal definition but described a range of symptoms consistent with IBS.

Most studies included a combination of IBS types, one study specified constipation (Page 1981); one study (Carling 1989) included only patients with IBS-C and IBS-A; and eight were unclear (Berthelot 1981; Czalbert 1990; Gilbody 2000; Lech 1988; Liu 1997; Nigam 1984; 2 x Ritchie 1979).

None of the studies stated that any participants had IBS as result of gastrointestinal infection.

The majority of studies did not state the number of participants with bloating. Four studies had some patients with bloating (Czalbert 1990; Inauen 1994; Kruis 1986; Liu 1997). Three comparisons (3 x Carling 1989) stated that all patients had bloating.

Most of the studies did not describe symptom severity. Two studies stated that participants had symptoms of mixed severity (Inauen 1994; Mitchell 2002).

The age range of participants across studies was 16 to 79 years, with the mean age (where given) ranging from 34.5 to 48 years (Czalbert 1990). None of the studies particularly identified elderly participants.

Most studies had more women than men; two studies had more men than women (Liu 1997; Nigam 1984).

Interventions

The studies varied in the type of antispasmodic used:

• Eight comparisons used anti-muscarinics:
  • One dicyclomine bromide (Page 1981)
  • Two hyoscamine (2 x Carling 1989)
  • Three hyoscine (Nigam 1984; Schafer 1990; Ritchie 1979)
  • One hyoscine plus paracetamol (Schafer 1990)
  • One hyoscine plus diazepam (Ritchie 1979).
• Nine had direct-action smooth muscle relaxants:
  • One alverine (Mitchell 2002)
  • Four mebeverine (Berthelot 1981; Kruis 1986; Gilbody 2000; Inauen 1994 (the last two were also modified release)
  • Four peppermint oil (Carling 1989; Czalbert 1990; Lech 1988; Liu 1997).

None of the included studies used antispasmodics for short-term symptom relief. The duration ranged from two weeks to 16 weeks (Kruis 1986). The most common durations were 12 weeks (four studies), four weeks (four studies) and two weeks (four studies). Studies of less than 4 weeks duration were considered in sensitivity analyses because the GDG preferred a minimum of four weeks duration. This meant the following: two weeks – four studies (3 x Carling 1989; Page 1981); three weeks – one study (Inauen 1994).

Comparisons

The included studies covered the following comparisons (we have indicated with an asterisk the studies with less than 4 weeks duration):

• 13 comparisons of antispasmodics versus placebo:
  • Five gave anti-muscarinics (Carling 1989*; Nigam 1984; Page 1981*; Richie 1979; Schafer 1990)
  • Seven gave direct-action smooth muscle relaxants (Berthelot 1981; Carling 1989*; Czalbert 1990; Kruis 1986; Lech 1988; Liu 1997; Mitchell 2002)
• One study (Ritchie 1979) compared antispasmodic plus diazepam versus diazepam alone
• One study (Schafer 1990) compared antispasmodic plus paracetamol versus paracetamol alone
• One study compared different classes of antispasmodic:
  • One compared an anti-muscarinic with a smooth muscle relaxant (Carling 1989*)
• Two studies compared different types of antispasmodic in the same class (smooth muscle relaxant):
  • Two studies (Gilbody 2000; Inauen 1994*) compared mebeverine modified release (200 g bid) with mebeverine conventional (135g tid).

Methodological Quality

The results of the quality assessment for included trials are shown in Appendix D.

The method of randomisation was reported in none of the studies. Allocation concealment was reported in one study (Ritchie 1979), which reported an adequate method, in which the drugs were issued in random order by the pharmacist. All but two of the studies reported that the patients were blinded to the interventions. One stated that the patients were not blinded (Inauen 1994). One study (Czalbert 1990) was unclear about patient blinding. Only one study (Mitchell 2002) described an a-priori power calculation. Several studies included in the review demonstrated baseline comparability of the groups, but 11 did not give baseline characteristics or were in non-English languages (Berthelot 1981; Lech 1988; Nigam 1984; Ritchie 1979; Schafer 1990). One study was not comparable at baseline (Liu 1997) for the severity of stool frequency (more severe for Colpermin). Three studies reported no withdrawals (Czalbert 1990; Nigam 1984; Ritchie 1979). One study (Page 1981) reported that more than 20% of patients in at least one arm (or overall) were not analysed or were lost to follow-up (attrition bias): 33% in the intervention group and 39% in the placebo group.

The GDG preferred a minimum intervention period of four weeks as this was felt to be clinically significant relative to potential effect. This meant the following were treated with caution: two weeks: Carling (1989); Inauen (1994); Page (1981).

The risk of bias was assessed for each included study and no studies were excluded from the analysis. Four studies were assessed as being at higher risk of bias: Page (1981) (attrition bias and duration less than four weeks); Inauen (1994) (patients not blinded and duration less than four weeks); Liu 1997 (lack of baseline comparability), and; Carling (1989) (duration less than 4 weeks). All these studies were treated with caution and sensitivity analyses were carried out. This was also done for Czalbert as discussed earlier.

A. Antispasmodics versus Placebo

There were 12 studies included in the analysis that compared antispasmodics with placebo, and a further two studies that compared antispasmodics + another intervention versus that other intervention alone. The GDG decided that it was inappropriate to combine the two types of comparison, even though each could be considered to be a comparison of antispasmodics versus placebo. The GDG’s view was that the drugs could interact and their effects would not simply be additive.

One of the studies was in patients with constipation-predominant IBS (Page 1981); eight did not specify the type of IBS and the remainder had patients of mixed IBS-type. Therefore the studies were not stratified by IBS type. Similarly, there was too little information to separate by severity, post-infective cause or bloating status. Only one study (Mitchell 2002) reported using established criteria to diagnose IBS (Rome I).

Where outcomes were measured at different times during the study, we took the end-study results unless there were significant numbers of withdrawals or problems with compliance. Therefore, for the Kruis (1986) study we took the values at four weeks.

1. Global symptoms

a. Number of patients with improvement in global symptoms

Eight studies with 731 patients reported this outcome for the comparison antispasmodics versus placebo. As described in the general section, patient assessments of improvement were combined with symptom score related methods (unlike the Cochrane review). Overall the relative risk was 1.32 (95%CI 1.18, 1.48), i.e. statistically significant, favouring antispasmodics. There was some heterogeneity, but it was not significant (p=0.09; I²=43%). The sensitivity analysis without Page (1981) made little difference to the summary statistics.

The full meta-analysis corresponded to a number needed to treat (NNT) of 6 (95%CI 5, 10), for a placebo group rate of 0 to 71%. This meant that clinicians needed to treat 6 patients to get additional benefit in relief of symptoms for one patient. Typically this is viewed as a low NNT.

Figure 1

Subgroup analysis into anti-muscarinic agents and direct-acting smooth muscle relaxants, for all antispasmodics-placebo comparisons (Figure 2) suggested there was little difference between classes of antispasmodics, although there was some heterogeneity (I2=57%, p=0.08) for the anti-muscarinics group. The RR for the random effects model was 1.51 (95%CI 1.00, 2.28), i.e. this result was sensitive to the method of analysis.

Figure 2

We noted that three of these studies were in hyoscine and the remaining one (Page 1981) was in dicyclomine bromide. Page (1981) had over 30% missing data and was only two weeks in duration. In its absence there was little difference to the fixed effects result, but a substantial difference to the random effects model.

b. Global symptom score (mean)

This outcome was recorded by one study (Carling 1989), which compared the change in symptom score (before and after intervention) for peppermint oil and hyoscamine (atropine) compared with placebo. This was a crossover study that reported first period results. The study reported p values for the difference:

• Peppermint oil versus placebo: mean difference for change score was −11.8 (on a scale of 105 for a week); p=0.063 (i.e. statistically significant).
• Atropine versus placebo: mean difference for change score was −1.0 (on a scale of 105 for a week); p=0.46 (i.e. not statistically significant).

2. Individual symptoms

a. Pain

The following studies measured pain:

1. Number of patients with no pain: one study (Liu 1997)
2. Number of patients with less pain: four studies (Lech 1988; Liu 1997; Mitchell 2002; Page 1981 (physician assessed)). The Kruis (1986) study was not included because of poor compliance – 4 week results were not reported for this outcome.
3. Pain score (change): one study (Berthelot 1981).

Figure 3 shows the number of patients with less pain. There was a statistically significant increase in the number with less pain, favouring antispasmodics. However, there was also significant heterogeneity within the smooth muscle relaxants group (p=0.07, I²=63%). This was possibly a drug specific effect: Lech (1988) and Liu (1997) gave peppermint oil and Mitchell (2002) gave alverine citrate. However, the duration of Mitchell (2002) was also longer (12 weeks versus 4 weeks).

Figure 3

A sensitivity analysis without Mitchell (2002) shows that this was responsible for the heterogeneity, but it was not clear why. It was interesting to note that this study was the only one that carried out a power calculation.

Figure 4Sensitivity analysis

The one study (Berthelot 1981) recording a pain score showed a statistically significant decrease in pain score, favouring the antispasmodic (figure 5). The mean difference was −0.56 (95%CI −1.03, −0.09) on a scale of 1 to 4.

Figure 5

b. Bloating

Only two studies (Liu 1997 and Mitchell 2002) reported the number of patients with less bloating (Figure 6). Meta-analysis showed a high degree of heterogeneity between studies (p=0.0002, I²=93%). It was not clear why these studies were so different, but see the discussion above regarding the pain outcome.

Figure 6

c. Bowel habits

i. Number of patients with improved bowel habits

One study (Page 1981), with 71 patients and attrition bias, recorded the number of patients with improved bowel habits (figure 7). There was a statistically significant improvement with the antispasmodic.

Figure 7

ii. Stool score

One study reported a stool score (Berthelot 1981) on a scale of 1–4. There was a statistically significant difference between antispasmodic and placebo, favouring the former.

Figure 8

3. Adverse effects

Three studies (four comparisons) reported the number of patients with adverse effects. These were grouped by antispasmodic. In all cases there were wide confidence intervals, but it appeared that there were significantly more side effects for both atropine and dicyclomine bromide than placebo (Figure 9). In particular, the statistically significant effect of atropine was manifested as dry mouth and blurred vision, but the confidence intervals were very wide, as demonstrated in figure 10.

Figure 9

Figure 10

Specific side effects.

B. Antispasmodic type 1 versus Antispasmodic type 2

One study compared different types of antispasmodic: Carling (1989) compared hyoscamine (atropine) with peppermint oil (smooth muscle relaxant).

2. Adverse effects

The study showed a statistically significant increase in side effects for atropine compared with peppermint oil, although the confidence interval was wide.

Figure 11

C. Comparison of antispasmodics in the same class

Two studies (Gilbody 2000; Inauen 1994) compared modified release mebeverine twice daily (total 400 mg) with conventional mebeverine three times daily (total 405 mg). The two studies had different durations: Gilbody (2000) had duration of eight weeks, and; Inauen (1994) had duration of three weeks. Gilbody (2000) was also carried out in general practice.

1. Global improvement of symptoms

a. Number of patients with improved symptoms

There appeared to be little difference between the two formulations (figure 12).

Figure 12

2. Adverse effects

a. Pain

One study in general practice (Gilbody 2000) showed little difference between the interventions for the number of patients with adverse effects, although it was noted that the number of adverse effects was high in both groups and the study said this included symptoms associated with IBS (pain and diarrhoea).

Figure 13

Adverse Effects

An adverse effects review has been carried out and is reported in section 8.5.1. There were six included RCTs in the review of adverse effects of antispasmodics (Grillage 1990; Gilbody 2000; Schaffstein 1990; Mitchell 2002; Liu 1997; Van Outryve 1995). The interventions and comparators were extremely varied, as was the reporting of adverse effects outcomes. In view of this, no meta-analysis was performed. Dry mouth, drowsiness, dizziness and constipation were the most common complaints reported amongst people taking antispasmodics.

One of the limitations of the adverse effects review was that many of the adverse outcomes of interest were very similar to the symptoms of the IBS itself. For instance, antispasmodics were associated with drowsiness and constipation, both of which are commonly seen in people with untreated IBS. This made it difficult for the investigators to differentiate between the progress of the condition and the harmful effects of the drug.

Modelled response rates

In the basecase scenario the response rate of 45% in the no treatment arm is taken from the Mearin (2004) cohort study. This represents the group of patients for whom we would expect an improvement in symptoms without any specific intervention. The RR of response for antispasmodics versus placebo is 1.32; therefore the response rate in the intervention arm is 59% (=45% × 1.32), giving an absolute difference in response between the intervention and no treatment arms of 14% (=59%–45%) during the 1^st month. In the basecase scenario the response rate for the subsequent interventions is assumed to be equal to the response rate for the first intervention. Therefore an additional 14% of those who do not respond to the first intervention achieve a therapeutic response to the second intervention, increasing the overall response rate to 65% (=59%+14%*[1–59%]) after the second month. The response rate over time for the basecase is given in Figure 14.

Figure 14

Modelled response rate for antispasmodic therapy, when allowing up to 4 switches of antispasmodic therapy, and no treatment.

We have also considered an alternative scenario in which no patient in the comparator arm achieves an improvement in symptoms but there is still a 14% chance of response for the intervention arm.

Results

Table 2 gives the incremental cost-effectiveness for several treatment pathways with different numbers of therapy switches included. These results demonstrate that although a treatment pathway which allows up to 4 switches had a cost per QALY under £20,000 compared to no treatment, the incremental cost-effectiveness compared to a pathway which allows up to 3 switches was greater than £20,000. Therefore, at a willingness to pay threshold of £20,000 per QALY, it would only be cost-effective to allow up to 3 switches as the additional switch would not provide sufficient additional benefit.

Table 2

Incremental cost-effectiveness of allowing subsequent switches in antispasmodic therapy.

These results are an estimate of the cost-effectiveness over the first 6 months after the initiation of antispasmodic therapy. The cost per QALY for continuing antispasmodic therapy beyond 6 months is lower than the cost per QALY during the initial 6 months provided that treatment is reviewed every 6 months and discontinued in patients who no longer experience a therapeutic benefit, either due to lack of effectiveness or a change in their symptom profile.

The probabilistic sensitivity analysis provided an estimate of the uncertainty in the cost per QALY estimate due to uncertainty in the efficacy estimate, the probability of response in the no treatment arm and the utility gain. The CEAC in Figure 15 shows the uncertainty surrounding the relative cost-effectiveness of allowing each additional switch in therapy. For example, the first curve on the left of Figure 15 shows that it is highly likely that using a first antispasmodic therapy would be cost-effective compared to no treatment as the probability of the cost per QALY being under a £20K threshold is over 90%. The second curve from the left is very close to the first and it shows that it is highly likely that allowing patients to switch once to an alternative antispasmodic therapy would be cost-effective compared to no further antispasmodic therapy for those who do not respond to the first. Similarly, the second switch is also highly likely to be cost-effective as it has a 91% probability of being under £20K. However, once the third switch of therapy is considered, it is only fairly likely to be cost-effective as there is a 53% likelihood that the true cost per QALY is under £20K and a 79% likelihood that it is under £30K. Providing four switches is fairly unlikely to be cost-effective compared to 3 switches as the cost per QALY has a 23% probability of being under £20K and a 51% probability of being under £30K. However, it should be noted that these estimates only consider the uncertainty in cost-effectiveness due to the accuracy of several input parameters and they do not reflect general uncertainty around the assumptions made in the model. The uncertainty from these assumptions was explored in the univariate sensitivity analysis.

Figure 15

Cost-effectiveness acceptability curves for antispasmodics (AS) compared to no treatment (NT) and for each additional switch of antispasmodic for non responders (AS).

Univariate sensitivity results for up to three switches

The results in Table 3 show that initiating treatment with an antispasmodic and allowing up to three switches in therapy for non-responders had a cost per QALY of £7,952 compared to no treatment. When the response rate in the no treatment arm was taken from the average response rate in the RCTs of pharmacological interventions (47%), rather than from the cohort study (45%), the cost per QALY was very similar at £7,539. Maintaining the 14% difference in response between the two arms but reducing the response rate in the no treatment arm from 45% to zero marginally decreased the cost per QALY to £7,772. Reducing the response rate for subsequent antispasmodics, in patients who have not responded to initial therapy, by 50% significantly increased the cost per QALY for each subsequent switch of therapy, such that only 2 rather than 3 switches of therapy could be provided for a cost per QALY under £20,000. However, the cost per QALY for 3 switches compared to no treatment was only moderately increased to £10,003 per QALY and even when the response rate for subsequent therapy was set to zero, the cost per QALY for antispasmodic therapy with up to 3 switches remained under £20,000 when compared to no treatment.

Table 3

Sensitivity results for antispasmodic therapy with up to 3 switches compared to no treatment for 100 patients with IBS (all subtypes).

We carried out a threshold analysis to determine whether antispasmodic therapy would still be cost-effective for lower gains in health related quality of life. In the basecase it was assumed that patients who respond to therapy accumulate 0.071 QALYs more per annum than patients who do not respond. For comparison, a gain of 0.135 QALYs would represent a complete remission of IBS symptoms. When the QALY gain associated with a response to therapy was reduced to 0.028 QALYs, the cost per QALY of providing antispasmodic therapy with up to 3 switches was estimated to be above £20,000 per QALY compared to no treatment.

If a patient takes a therapy on an as needed basis, it would be reasonable to assume that they take the therapy on days when their quality of life is significantly affected by their IBS symptoms but not on days when their symptoms are less severe. It has therefore been assumed in the model that they only accrue QALY benefits and drug costs on the days they take the therapy. However, it is still necessary to assess all patients for response after 1 month of initiating therapy. This means that it would be less cost-effective to initiate therapy in patients who use the therapy on fewer days, as the monitoring costs are just as high but the benefits are lower. This is shown by the estimated cost per QALY of £20,578 for patients who take the therapy on 25% on days. However, a more detailed look at the results for these patients (data not tabled) shows that up to 1 switch of therapy could be provided for a cost per QALY of £19,414 with a 45% likelihood of being under £20,000 per QALY and a 73% likelihood of being under £30K per QALY. So, cost-effective treatment strategies may be available for patients who do not experience severe symptoms as frequently.

If a patient also takes another medication (such as a laxative or anti-motility agent), then these medications can be reviewed at the same time, so it may be cost-effective to provide both therapies. For example, if laxatives are prescribed with the antispasmodic and both used on 25% of days then allowing up to 2 switches of both treatments was estimated to be cost-effective with a cost per QALY of £10,107 compared to no treatment a cost per QALY of £17,393 compared to 1 switch.

Types of patients

For this review, patients were required to have IBS and not to have inflammatory bowel disease or major psychiatric disorders.

Types of studies

Studies that investigated drugs not listed in the BNF were excluded. These included: Amineptine; Amoxapine; Desipramine, and; Pirenzepine.

The GDG decided that crossover studies were acceptable and that the washout period for this review should be at least one week. Trials with shorter washout periods were not included in the analysis.

Types of intervention

Studies included the following interventions:

Tricyclics and related antidepressants:

• Amitriptyline (Triptafen®, Triptafen-M® (with perphenazine)
• Clomipramine (Anafranil®)
• Dosulephin (Prothiaden®)
• Doxepin (Sinepin®)
• Imipramine
• Lofepramine (Feprapax®, Lomont®, Gamanil®)
• Nortriptyline (Allegron®)
• Trimipramine(Surmontil®)
• Mianserin
• Trazodone (Molipaxin®).

Selective serotonin re-uptake inhibitors (SSRIs):

• Citalopram (Cipramil®)
• Escitalopram (Cipralex®)
• Fluoxetine (Prozac®)
• Fluvoxamine (Faverin®)
• Paroxetine (Seroxat®)
• Sertraline (Lustral®).

Monoamine oxidase inhibitors (MAOIs):

• Phenelzine (Nardil®)
• Isocarboxazid
• Tranylcypromine.

Reversible MAOIs:

• Moclobemide (Manerix®).

Others:

• Duloxetine (Cymbalta®)
• Flupentixol (Fluanxol®)
• Mirtazapine (Zispin Soltab®)
• Reboxetine (Edronax®)
• L-Tryptophan (Optimax®).

The following comparisons were included:

• Antidepressant versus placebo (or nothing)
• Antidepressant type 1 versus Antidepressant type 2
• Antidepressant dose 1 versus Antidepressant dose 2
• Antidepressant versus other interventions.

NB: In spite of the large placebo effect associated with IBS, comparisons with no treatment were included.

The antidepressants review was concerned with medium and longer term symptom relief. Medium term treatment was defined as three months and long term as between six months and one year.

Subgroup analyses

We planned to carry out subgroup analyses by type of antidepressant; dose; mode of delivery, and; duration of intervention.

Search strategy for identification of studies

Searches were performed on the following core databases: MEDLINE; EMBASE; CINAHL, and; The Cochrane Library (1966 to current day with guidance from the GDG). Additional databases were not searched for this review. The search strategies are listed in Appendix B.

The search strategy identified 1458 studies. The titles and abstracts of these studies were assessed. Thirty were identified to be potentially relevant to the review and these papers were retrieved in full. The reference lists for each of the retrieved studies were inspected for further potential papers, but none were identified. The 17 excluded studies are listed in the Appendix, along with reasons for exclusion. The remaining 13 studies were included (Boerner 1988; Creed 2003; Kuiken 2003; Myren 1982; Myren 1984; Rajagopalan 1998; Schrivastava 1984; Steinhart 1982; Tabas 2004; Tanum and Malt 1996; Tripathi 1983; Vij 1991; Quartero 2007). One of these studies was a Cochrane review (Quartero 2007). The Myren (1982) study did not state that it was randomised although it was double blind. Since this study was included in the Cochrane review (under the name of Block (1983), which is an identical paper in Norwegian) we included it too, but treated it with caution.

Study Design

Setting: The majority of studies took place in secondary care, one of which treated inpatients (Tripathi 1983); one was stated to take place in primary care in a later paper (Myren 1982); three included patients from both primary and secondary care settings (Boerner 1988; Myren 1984; Tabas 2004) and one study (Steinhart 1982) did not report the setting.

Funding: three studies received some funding from industry: Kuiken (2003) was sponsored by Eli Lilly (manufacturers of fluoxetine); Tripathi (1983) received the study medication and placebo from May & Baker India Ltd. (manufacturers of trimipramine); Steinhart (1982) received the study medication and placebo from Merck Sharp & Dohme. Six studies (Myren 1982; Myren 1984; Rajagopalan 1998; Vij 1991; Boerner 1988) did not state source of funding. The remaining studies were funded by non-industry sources (Tabas 2004; Tanum and Malt 1996; Creed 2003).

Population

The age range of patients across the IBS studies was 13 to 75 years, with the mean age, where given, ranging from 35 to 41 years. The study with a lower age of 13 was Tripathi (1983); in this study the range was 13 to 60 years, with a mean of 37. The GDG did not consider this level of children to be important. No study particularly identified elderly patients.

Five studies had more women than men (Creed 2003; Myren 1982; Steinhart 1982; Tabas 2004; Tanum and Malt 1996); three studies had about the same number of men and women (Kuiken 2003; Rajagopalan 1998; Myren 1984) and two studies did not give the proportions of patients by gender (Boerner 1988; Tripathi 1983).

Six studies identified the type of IBS as mixed (Boerner 1988; Creed 2003; Kuiken 2003; Steinhart 1982; Tabas 2004; Vij 1991). The remaining studies gave no information regarding type of IBS. In Tanum and Malt (1996) only 60% of the patients had IBS; the rest had non ulcer dyspepsia (NUD). The authors reported that there was no significant difference in response between IBS and NUD patients, although there was a trend towards a slightly better response in the NUD patients.

Two studies stated that the patients had severe IBS (Creed 2003; Steinhart 1981); one study had mixed severity patients (Vij 1991). Creed (2003) was stratified by pain level before randomisation. Tabas (2004) also selected patients who were non-responders to placebo.

Four studies reported the inclusion of refractory IBS patients: Creed (2003) included patients that had failed to respond to usual treatment and had a median duration of IBS of eight years; Kuiken (2003) stated that the patients had all been treated unsuccessfully previously. Steinhart (1981) reported a mean duration of IBS of 5 years and stated that all patients had received antispasmodics previously; Tabas (2004) included patients who had failed to respond to a high fibre diet.

Three studies reported a long duration of IBS: Tanum and Malt (1996) had a mean duration of symptoms of about 8 years; Rajagopalan (1998) had a mean duration of about 4 years; Kuiken (2003) had a mean duration of symptoms of 5.9 years.

One study implied that some of the patients did not have refractory IBS: in Myren (1984), 45 to 61% were not taking other drugs before the study started, and the other studies did not report previous treatments.

Five studies stated that the patients had some depression: Tabas (2004) reported that 27/81 (33%) had a score greater than 10 on the Beck Depression Inventory (although major psychiatric illnesses were excluded); Creed (2003) reported that 47% had anxiety or depression; Steinhart (1981) stated that 57% had depression and 79% anxiety; Vij (1991) had 57% with psychiatric co-morbidities; Boerner (1988) reported that some patients had depression (mid point on the Hamilton depression scale). Kuiken (2003) excluded patients with depression.

One study stated that all the patients had anxiety (Tripathi 1983). Three studies reported that the patients did not have psychiatric disorders: Rajagopalan (1998) stated that patients had no major medical or psychiatric illnesses; Tanum and Malt (1986) excluded patients with schizophrenia, anxiety or depression, and; Myren (1982) and Myren (1984) both showed low scores on depression and anxiety scales.

Interventions

The interventions included:

• Tricyclic Amitriptyline up to 75mg for 3 months (Rajagopalan 1998) and 50mg for 1 month (Steinhart 1982) – c.f. BNF levels for depression treatment: initially 75mg, increasing to 150 to 200mg
• Tricyclic Trimipramine 30 to 50mg for 4 to 6 weeks in three studies (Myren 1982; Myren 1984; Tripathi 1983) – c.f. BNF levels for depression: initially 50–75mg, then 150 to 300mg
• Tricyclic Doxepin 50mg for 8 weeks (Boerner 1988) and 75mg for 6 weeks (Vij 1991) – c.f. BNF levels for depression: initially 75mg, then 30 to 300mg
• Tricyclic-related Mianserin 30 mg initially, then up to 120mg for weeks 2 to 7, then tapered in week 8 – c.f. BNF levels for depression: 30 to 40mg initially; usual dose 30 to 90mg
• SSRI Paroxetine 20mg per day for 3 months (Creed 2003) and up to 40mg for 3 months; 23% 10mg; 43% 20mg; 33% 40mg (Tabas 2004) – c.f. BNF levels for depression: initially 20mg, then up to 50mg
• SSRI Fluoxetine 20mg per day for 6 weeks – c.f. BNF levels for major depression: 20mg once daily increased after 3 weeks if necessary, usual dose 20 to 60mg.

Comparisons

The majority of comparisons were of antidepressants versus placebo. One study (Creed 2003) compared antidepressants with usual care. One study compared different doses of antidepressants (Myren 1984).

Antidepressant versus placebo

• Tricyclics versus placebo:
  • Trimepramine (Myren 1982; Myren 1984; Tripathi 1983)
  • Amitriptyline (Rajagopalan 1998; Steinhart 1982)
  • Doxepin (Boerner 1988; Vij 1991)
  • Mianserin (Tanum and Malt 1996).
• SSRI versus placebo:
  • Paroxetine (Tabas 2004) included high fibre diet (>25g daily) in both the treatment and placebo groups (NB. these patients had already been identified as non-responders to high fibre)
  • Fluoxetine (Kuiken 2003).

Antidepressant versus usual care

• SSRI versus usual care:
  • Paroxetine (Creed 2003) versus ‘routine care’ by gastroenterologist and GP including antispasmodics, laxatives, antidiarrhoeal medication or additional analgesics.

Antidepressant versus psychotherapy (this is covered in the psychotherapy review)

• Paroxetine versus psychotherapy (Creed 2003).

It was decided to combine the SSRI studies with comparators of placebo and usual care using subgroup analyses.

A. Antidepressants versus placebo or usual care

1. Global symptoms

a. Number of patients with global improvement of symptoms (pain, bloating and bowel habit)

Six studies, in 434 patients, reported the number of patients with improvement in global symptoms (Boerner 1988; Creed 2003; Kuiken 2003; Tabas 2004; Myren 1982; Vij 1991). The studies were combined in a meta-analysis, but as separate subgroups by type of antidepressant. The controlled trial (Myren 1982) was included in the tricyclics subgroup and examined in a sensitivity analysis. The comparisons of paroxetine with placebo and with usual care were also considered in sensitivity analyses. ‘Usual care’ was defined as patients receiving IBS treatment that was deemed appropriate by either their gastroenterologist consultant or general practitioner.

The difference between the antidepressants and placebo was statistically significant overall and for each subgroup. Within subgroups there was no heterogeneity, but between groups there was some (I²=42%, p=0.12). The overall relative risk (RR) for the meta-analysis of 434 patients was 1.55 (95% CI 1.30, 1.84), which corresponded to a number needed to treat of 5 (95% CI 4, 7), for a control group rate of 22 to 68%.

For the tricyclic subgroup (n=180) the number with global improvement of symptoms was statistically significantly higher for the antidepressant group; RR 1.31 (95% CI 1.04, 1.64), which gave an NNT of 6 (95% CI 4, 34), for a control group rate of 22–68%. In the absence of the Myren (1982) study, which was not stated to be randomised, the RR for this group was 1.37 (95% CI 0.99 1.91), i.e. no longer significant and with some heterogeneity (I²=62%, p=0.11).

For the meta-analysis of the three SSRI studies (n=254), there was a significant difference favouring antidepressant, RR 1.80 (95% CI 1.38, 2.34), with no heterogeneity (I²=0%, p=0.48). This corresponded to an NNT of 4 (95% CI 3, 7), for a control group rate of 28–41%. In the absence of the Creed (2003) study the effect was slightly bigger (RR 1.85 (95% CI 1.17, 2.91). We decided to use the results in Figure 1 in the health economic modelling.

Figure 1

b. Global symptom score

One study with 28 patients reported the global symptom score, but no details of the scale were given. There was no significant difference between interventions.

Figure 2

2. Individual symptoms

a. Pain

i. Number of patients with less pain

Three studies reported this outcome. Two were tricyclics versus placebo (Vij 1991; Tanum and Malt 1996) and the other was an SSRI versus placebo (Tabas 2004). These were included as subgroups in a meta-analysis in 150 patients. Overall there was significant heterogeneity (I²=84%, p=0.004), which is attributed to the type of antidepressant.

There was a large significant difference in favour of tricyclics compared to placebo in the number of patients with reduced pain; RR 3.91 (95% CI 1.93, 7.93), with no heterogeneity (I²=0%, p=0.81), although the confidence interval was fairly wide. This corresponded to an NNT of 2 (95% CI 2, 4), for a placebo group rate of 16–18%. We noted that this analysis included Tanum and Malt (1996) which had only 60% of patients with IBS, and; Vij (1991) which had 24% missing data in each arm.

There was no significant difference between the SSRI and placebo in one study in 66 patients (Tabas 2004); RR 0.88 (95% CI 0.54, 1.45).

Figure 3

ii. Number of patients with pain

One study in 34 non-depressed patients (Kuiken 2003) reported the number of patients with ‘significant pain’. There was no statistically significant difference between interventions, although the confidence interval was fairly wide.

Figure 4

iii. Pain score

Four studies (Creed 2003; Tanum and Malt 1996; Myren 1982; Myren 1984) reported a type of pain score, all appeared to use a VAS of 100 mm or 10 cm. Myren (1982) recorded an abdominal obstruction discomfort measurement, which the GDG decided was a different outcome to pain. Myren (1984) gave means and p values only. We decided to combine the first two studies in a meta-analysis by subgroup (Creed 2003 compared an SSRI with usual care). We also noted that Tanum and Malt (1996) had only 60% of patients with IBS.

Figure 5

At the end of treatment, meta-analysis showed significant heterogeneity (I²=77%, p=0.04), with different effect sizes being found for the two studies. This may be an effect of type of antidepressant, type of comparator, severity of IBS, or it may be that Tanum and Malt (1996) overestimated the effect because there were only 60% of patients with IBS. Individually, there was a statistically significant difference for the Tanum and Malt (1996) study (tricyclic versus placebo in 60% IBS): mean difference −25.90 (95% CI −38.82, −12.98) and for the Creed (2003) study: mean difference −9.20 (95% CI −18.35, −0.05).

Boerner (1988) compared a tricyclic with placebo, and recorded the median improvement in pain on a scale of 0 to 4. There was a difference in median change score of 0.3 units, in favour of the tricyclic antidepressant, which was reported to be statistically significant at the level of p<0.05.

b. Bowel habit

i. Number of patients with improvement in bowel habit

Meta-analysis of two studies in 110 patients showed a statistically significant increase in the number of patients with improvement in bowel habit; RR 1.92 (95% CI 1.19, 3.07), with no heterogeneity between the tricyclics and SSRI subgroups (I²=0; p=0.51) corresponding to an NNT of 4 (95% CI 3, 13). The confidence interval was fairly wide. The results for individual classes of antidepressant were not statistically significant, although the antidepressant was favoured; Vij (1991) (tricyclics) had a wide confidence interval and Tabas (2004) (SSRI) was fairly wide.

Figure 6

c. Bloating

i. Number of patients with less bloating

Two studies recorded the number of patients with less bloating, one a tricyclic (Boerner 1998) and the other using an SSRI (Tabas 2004), both compared with placebo. Meta-analysis could not be carried out because Boerner (1998) reported the median.

Boerner (1988) recorded the median improvement in the feeling of fullness on a scale of 0 to 4. There was a difference in median change score of 0.23 units, in favour of the tricyclic, but this was reported to be not statistically significant.

Tabas (2004), comparing SSRI with placebo, did not demonstrate a significant effect on the number of patients with less bloating, but the confidence interval was fairly wide so this conclusion was uncertain.

Figure 7

ii. Number of patients with bloating

One study (Kuiken 2003) reported the number of patients with bloating in 34 non-depressed patients. There was no significant difference between the SSRI and placebo, although the confidence interval was fairly wide. We note that the study was sponsored by Eli Lilly (manufacturers of fluoxetine).

Figure 8

3. Mental health

a. Psychological Distress

This was measured by the SCL-90 global severity index (90 item, 5 point rating scale; range 90 to 450, high = bad). There was a small, statistically significant difference between SSRI (paroxetine) and usual care at 3 months, WMD: −0.28 (95% CI −0.43, −0.09), favouring the antidepressant.

Figure 9

b. Depression score

One study (Myren 1982, CCT) recorded depression on a VAS of 0–10 cm for tricyclics versus placebo, in which a high score indicated increased depression. There was statistically significantly less depression for the patients taking antidepressants. We noted that this study, at baseline, showed low scores on depression and anxiety scales.

Figure 10

c. Anxiety

Two studies reported anxiety; one (Myren 1982, CCT) compared tricyclics versus placebo on a VAS of 0 to 10 and the other (Tabas 2004) reported the number of patients with anxiety for the comparison SSRI with placebo. The former study, at baseline, showed low scores on depression and anxiety scales, and the latter reported that 27/81 (33%) had a score greater than 10 on the Beck Depression Inventory.

i. Anxiety score

There was no significant difference in the degree of anxiety.

Figure 11

ii. Number of patients with anxiety

The confidence interval was wide and the result was not statistically significant.

Figure 12

4. Quality of life

Two studies reported quality of life measures, both for SSRIs (Creed 2003; Tabas 2004).

a. IBS-QoL

One study (Tabas 2004) comparing SSRI and placebo examined the change compared with baseline in three components of IBS-QoL: food avoidance; work function; social function. Tabas (2004) found the following differences for SSRI versus placebo:

• Food avoidance subscale – improvement of 12.7%; p value = 0.03 (statistically significant)
• Work function subscale – improvement of 2.1%; p value = 0.08 (not significant)
• Social function subscale – improvement of 13.4%; p value = 0.76 (not significant).

b. Change in SF36 Mental component score at 3 months

There was no significant difference in the mental health quality of life SF-36 score (scale 0 to 100, high=good) at 3 months between SSRI (paroxetine) and usual care, although there were 29% who discontinued treatment and 32% loss to follow up in the paroxetine arm. This conclusion does not agree with the p values reported in Creed (2003) (p=0.007).

Figure 13

c. Change in SF36 physical health score at 3 months

The change in the physical health component also showed no significant difference at 3 months. This did not agree with the p value reported in Creed (2003) (p=0.24).

Figure 14

B. Comparison of different doses of Tricyclic antidepressant

One study (Myren 1984) compared different doses of Trimipramine in 428 participants.

1. Global symptoms

The study reported the ‘total effect of treatments’ in the opinion of the physicians. Medians with their 95% confidence intervals were reported for each group. We used this to calculate the standard deviation, which we used with the median value to compare groups. Figure 16 shows there was no significant difference between any of the doses.

Figure 16. Patient pathway for Tricyclics (TC) and SSRIs (PDF, 51K)

2. Individual symptoms

a. Number of patients with abdominal pain

The study reported a statistically significant reduction in pain in patients that were given 50 mg per day either as a single dose or in two divided doses (p < 0.01). Patients taking 35 mg in a single dose had significantly less pain than placebo (p< 0.05), and for patients taking 30 mg per day in three divided doses there was stated to be no difference between the drug and placebo (p= 0.10).

Figure 15

Modelled response rates

Figures 18 to 23 below give the modelled response rates for six different strategies for prescribing tricyclics and SSRIs for the management of chronic pain in IBS.

Figure 18

Response rates for a tricyclic up to 30 mg followed by second tricyclic up to 30 mg in non-responders compared to no treatment.

Figure 19

Response rates for a tricyclic up to 30 mg followed by an SSRI up to 20 mg in non-responders compared to no treatment.

Figure 20

Response rates for an SSRI up to 20 mg compared to no treatment.

Figure 21

Response rates for an SSRI up to 20 mg followed by a tricyclic up to 30 mg in non-responders compared to no treatment.

Figure 22

Response rates for an SSRI (up to 20 mg) followed by a second SSRI in non-responders compared to no treatment.

Figure 23

CEAC for up to 30 mg of tricyclic compared to no treatment.

1. Tricyclic up to 30 mg
2. Tricyclic up to 30 mg followed by second tricyclic up to 30 mg
3. Tricyclic up to 30 mg followed by SSRI up to 20 mg
4. SSRI up to 20 mg
5. SSRI up to 20 mg followed by tricyclic
6. SSRI up to 20 mg followed by second SSRI up to 20 mg

The RR of response for SSRIs versus placebo (or usual care) was much higher than the RR of response for tricyclics versus placebo (see Table 1) giving much larger increases in the number responding at each dose. However, as there was no head-to-head comparison of SSRIs and tricyclics, it was not possible to tell whether the apparent superiority of SSRIs is simply because they have been tested in a population that is more likely to respond to a pharmacological intervention. Indirect comparisons between SSRIs and tricyclics should be interpreted with caution due to the potential for bias.

Table 1

Intervention specific parameters – Tricyclics.

Figure 17Response rates for a tricyclic (up to 30 mg) compared to no treatment

Amitriptyline tablets were the lowest cost tricyclic at doses of 10 mg to 20 mg. Doxepin had a lower cost per 10 mg but the smallest tablet size is 50 mg, so it was not considered practical to assume that doxepin is commonly prescribed at a doses of 10 to 30 mg per day. It was assumed in the model that 10 mg amitriptyline tablets (non-proprietary) are prescribed as the first line tricyclic, with trimipramine (Surmontil®) prescribed second line. The highest cost tricyclic was amitriptyline solution (non-proprietary) at £5.03 per month for a dose of 10 mg per day, so this preparation was used in the high drug cost sensitivity analysis.

Of the two SSRIs considered, the lowest cost preparation is fluoxetine (non-proprietary) and it was assumed that this is prescribed first line, with the lowest cost paroxetine preparation (non-proprietary) prescribed second line. A sensitivity analysis was carried out assuming that the highest cost preparation is prescribed (Branded liquid preparations of fluoxetine and paroxetine at £14.40 and £9.62 per month respectively).

Tricyclic up to 30 mg but no further treatment if no response

A strategy of using up to 30 mg of tricyclic is estimated to provide 0.46 additional QALYs (difference between QALYs gained for intervention and no treatment), at a cost of £4,459, compared to no treatment for a cohort of 100 patients over a 6 month period, provided treatment is not continued in those who do not respond to a dose of 30 mg. This strategy has a cost per QALY of £9,762, compared to no treatment, in the first 6 months after initiating treatment. Treatment can be continued in the next 6 months for a cost per QALY of £3,395 provided that treatment is reviewed every 6 months and discontinued in patients who no longer experience a therapeutic benefit, either due to lack of effectiveness or a change in their symptom profile.

The probabilistic sensitivity analysis estimates the uncertainty in the cost-effectiveness due to uncertainty surrounding the efficacy estimate, the utility gain, and the response rate in the no treatment arm. The results of this analysis are presented in Figure 24 as a CEAC, which shows the probability of the cost per QALY falling under various thresholds. The CEAC shows that a strategy of prescribing up to 30 mg of tricyclic has a 76% probability of being under £20K compared to no treatment over a 6 month timeframe.

Figure 24

CEAC for up to 30mg a tricyclic with switch to a second tricyclic if no response compared to up to 30mg tricyclic without switch for non-responders.

Univariate sensitivity results for up to 30 mg of tricyclics

The basecase cost-effectiveness estimates assume that an equal number of patients respond to the 10 mg, 20 mg and 30 mg doses. However, if we assume that patients only respond after reaching the 30 mg dose, then the cost per QALY is higher at £11,296. If patients are allowed to increase their dose up to 50 mg in order to achieve a response and we assume that no patient responds to a lower dose then the cost per QALY is significantly higher at £17,937.

If the highest cost preparation is prescribed instead of the lowest cost preparation, then the cost per QALY is increased to £14,022. If the GP follow-up is twice as frequent as modelled in the pathway (i.e. every two weeks until an effective dose is established instead of every four weeks and two follow-up appointments in the next 12 weeks instead of one), then the cost per QALY is increased to £17,826. These sensitivity analyses suggest that using tricyclics at doses of up to 30 mg is likely to be cost-effective compared to no treatment, even if the costs of care are higher than estimated in the modelled care pathway.

The threshold analysis on the utility gain associated with an improvement in global symptoms shows that the cost per QALY would be over £20,000 if the utility gain is less than 0.035. For comparison, we used a utility gain of 0.071 in the basecase and a utility gain of 0.135 would be equivalent to a complete resolution of IBS symptoms.

Tricyclic up to 30 mg with switch to a second tricyclic if no response to first

If patients who do not respond to 30 mg of tricyclic are allowed to switch to a second tricyclic drug, and we assume that the second tricyclic is just as likely to be effective, then the cost rises to £5,858 per 100 patients over the first 6 months, but the total QALY gain is increased to 0.60. The cost per QALY of treating patients with up to two tricyclics to gain a response is £9,789 per QALY compared to no treatment, and £9,873 compared to stopping treatment after the first tricyclic. Even if the chance of response to the second tricyclic is half the chance of response to the first, the cost per QALY is £10,912 compared to no treatment and £18,324 compared to stopping after a failure on the first tricyclic.

A probabilistic sensitivity analysis was carried out on the cost-effectiveness of allowing non-responders to switch to a second tricyclic compared to no further tricyclic therapy for those who do not respond to the first tricyclic. The CEAC in Figure 25 shows that a strategy of prescribing up to 30mg of tricyclic with a switch to a second tricyclic for non-responders has an 87% likelihood of being under £20K compared to no further treatment for non-responders to 30mg of tricyclic.

Figure 25

CEAC for up to 30mg of tricyclic with switch to an SSRI if no response compared to up to 30mg of tricyclic without switch for non-responders.

Tricyclic up to 30mg followed by an SSRI if no response to tricyclic

If patients who do not respond to 30mg of tricyclic are allowed to switch to an SSRI (up to 20mg), and we assume that the SSRI effectiveness is independent of the response to the tricyclic, then the cost rises to £5,648 per 100 patients over the first 6 months, but the total QALY gain is increased to 0.84. The cost per QALY of treating patients with a tricyclic at doses of up to 30mg followed by an SSRI of up to 20mg is £6,703 per QALY compared to no treatment, and £3,080 compared to stopping after the tricyclic. Even if the chance of response to the SSRI is half the chance of response seen in the SSRI trials, then the cost per QALY is £8,450 compared to no treatment and £5,342 compared to stopping after a failure on the tricyclic.

A probabilistic sensitivity analysis was carried out on the cost-effectiveness of allowing non-responders to switch to a SSRI compared to no further SSRI therapy for those who do not respond to the first tricyclic. The CEAC in Figure 26 shows that a strategy of prescribing up to 30mg of tricyclic with a switch to an SSRI for non-responders has a 95% likelihood of being under £10K compared to no further treatment for non-responders to 30mg of tricyclic.

Figure 26

CEAC for SSRI up to 20mg (with no switching for non-responders) compared to no treatment.

SSRI up to 20mg but no further treatment if no response

A strategy of using up to 20mg of SSRI is estimated to provide 1.23 additional QALYs, at a cost of £3,708, compared to no treatment for a cohort of 100 patients over a 6 month period, provided treatment is not continued in those who do not respond to a dose of 20mg. This strategy has a cost per QALY of £3,020 compared to no treatment, in the first 6 months after initiating treatment. Treatment can be continued in the next 6 months for a cost per QALY of £1,483 provided that treatment is reviewed every 6 months and discontinued in patients who no longer experience a therapeutic benefit, either due to lack of effectiveness or a change in their symptom profile.

These estimates assume that an equal number of patients respond to the 10mg and 20mg doses. However, if we assume that patients only respond after reaching the 20mg dose, then the cost per QALY is higher at £3,209. If patients are allowed to increase their dose up to 40mg in order to achieve a response, then the cost per QALY is significantly higher at £3,790 when using the dose distribution from Tabas (2004). However, if no patient responds until a dose of 40mg is reached then the cost per QALY is higher still at £4,698.

If the highest cost preparation is prescribed instead of the lowest cost preparation, then the cost per QALY is increased to £9,799. If the GP follow up is twice as frequent as modelled in the pathway (i.e every two weeks until an effective dose is established instead of every four weeks and two follow-up appointments in the next 12 weeks instead of one), then the cost per QALY is increased to £5,667. These sensitivity analyses suggest that using SSRIs at doses of up to 20mg is likely to be cost-effective compared to no treatment, even if the costs of care are higher than estimated in the modelled care pathway.

The threshold analysis on the utility gain associated with an improvement in global symptoms shows that the cost per QALY would be over £20,000 if the utility gain is less than 0.011. For comparison, we have used a utility gain of 0.071 in the basecase and a utility gain of 0.135 would be equivalent to a complete resolution of IBS symptoms.

A probabilistic sensitivity analysis was carried out on the cost-effectiveness of treatment with an SSRI (up to 20mg) compared to no treatment. The CEAC in Figure 27 shows the probability of the cost per QALY falling under various thresholds. The CEAC shows that a strategy of prescribing up to 20mg of SSRI has a 95% likelihood of being under £10K compared to no treatment over a 6 month timeframe.

Figure 27

CEAC for up to 20mg of SSRI with switch to tricyclic if no response compared to up to 20mg SSRI without switch for non-responders.

SSRI up to 20mg with switching to a tricyclic (up to 30mg) if no response to SSRI

If patients who do not respond to 20mg of SSRI are allowed to switch to a tricyclic drug (up to 30mg), and we assume that the tricyclic effectiveness is independent of the response to the SSRI, then the cost rises to £4,526 per 100 patients over the first 6 months, but the total QALY gain is increased to 1.30. The cost per QALY of allowing patients to try an SSRI followed by a tricyclic is £3,477 per QALY compared to no treatment, and £11,073 compared to stopping after the SSRI. Even if the chance of response to the tricyclic in patients who haven’t responded to an SSRI is half the chance of response seen in the tricyclic trials, the cost per QALY is £3562 compared to no treatment and £21,574 compared to stopping after a failure on the first tricyclic. If the frequency of follow-up is twice that estimated in the basecase pathway, then the cost per QALY for treatment with an SSRI, followed by a TCA in non responders, is £6,523 compared to no treatment and £20,750 compared to stopping treatment after the SSRI.

A probabilistic sensitivity analysis was carried out on the cost-effectiveness of allowing non-responders to switch to a tricyclic compared to no further tricyclic therapy for those who do not respond to the SSRI. The CEAC in Figure 28 shows that a strategy of prescribing up to 20mg of SSRI with a switch to a tricyclic for non-responders has a 59.6% probability of being under £20K compared to no further treatment for non-responders to 20mg of SSRI. In the probabilistic analysis, there is a 12.8% probability that all of the patients respond to the first SSRI resulting in no further benefit to be gained by switching patients to a second SSRI. In drawing the CEAC we have assumed that the cost per QALY of allowing patients to switch would be above any reasonable threshold when no benefit can be achieved for that switch.

Figure 28

CEAC for up to 20mg SSRI with switch to a second SSRI if no response compared to up to 20mg SSRI without switch for non-responders.

SSRI up to 20mg with a second SSRI (up to 20mg) if no response to first SSRI

If patients who do not respond to 20mg of SSRI are allowed to switch to a second SSRI (up to 20mg), and we assume that the effectiveness of the second SSRI is independent of the response to the first SSRI, then the cost rises to £4,677 per 100 patients over the first 6 months, but the total QALY gain is increased to 1.43. The cost per QALY of allowing patients to try an SSRI followed by a tricyclic is £3,275 per QALY compared to no treatment, and £4,843 compared to stopping after the first SSRI. Even if the chance of response to the second SSRI in patients who haven’t responded to the first SSRI is half the chance of response seen in the SSRI trials, the cost per QALY is £3,361 compared to no treatment and £7,544 compared to stopping after a failure on the first SSRI.

A probabilistic sensitivity analysis was carried out on the cost-effectiveness of allowing non-responders to switch to a second SSRI compared to no further SSRI therapy for those who do not respond to the first SSRI. The CEAC in Figure 29 shows that a strategy of prescribing up to 20mg of SSRI with a switch to a second SSRI for non-responders has a 75% likelihood of being under £10K compared to no further treatment for non-responders to 20mg of SSRI. Again, there is a 12.8% probability that all of the patients respond to the first SSRI resulting in no further benefit to be gained by switching patients to a second SSRI. In drawing the CEAC we have assumed that the cost per QALY of allowing patients to switch would be above any reasonable threshold when there is no benefit can be achieved for that switch.

Figure 29

Indirect comparison of tricyclics and SSRIs

The results presented above (see Tables 3 and 4) for tricyclics compared to no treatment and SSRIs compared to no treatment, suggest that SSRIs are more cost-effective than tricyclics as they have a larger QALY gain and lower cost compared to usual care. However, as discussed earlier, this conclusion should be treated with caution as it is based on an indirect comparison and these have a high potential for bias.

Table 3

Results for up to 30 mg tricyclic (amitriptyline) with no switches compared to no treatment in a cohort of 100 patients with IBS (all subtypes).

Table 4

Results for up to 20mg SSRI (fluoxetine) with no switches compared to no treatment in a cohort of 100 patients with IBS (all subtypes).

SELECTION CRITERIA

The selection criteria described in the general methodology section were used, but some were specific to the evaluation of adverse effects and are reported in the following sections.

Assessment and Reporting of Adverse Effects

The results for the included trials are shown in Table 4 to Table 9.

Table 4

Details of RCTs of Antispasmodic agents.

Table 5

Details of RCTs of Antispasmodic agents.

Table 6

Details of RCTs of Antimotility agents.

Table 7

Details of RCTs of Antimotility agents.

Table 8

Details of RCTs on Laxatives.

Table 9

Details of RCTs of Antimotility agents.

Although some trials stated that they specifically enquired about adverse effects, none of them actually stated whether this was an open question e.g. ‘Are you having any problems with the treatment?’ of if the enquiry was targeted at particular symptoms e.g. ‘Have you had any diarrhoea?’

Moreover, some trials stated that people were asked to record problems in a diary, but the trial report gives no details about whether people had to record specific events (including when and how often) or if it was left to their discretion.

RESULTS

RESULTS OF SPECIFIC CLASSES OF IBS THERAPIES

CRITERIA FOR CONSIDERING STUDIES FOR THIS REVIEW

PARAMETERS OF THE REVIEW

It is recognised that these drugs are typically used to treat large populations with psychiatric morbidities. The context for this review is to recognise the potential harmful effects of these drug types when prescribed at low dose for the treatment of IBS symptoms, namely pain and discomfort. In the last twenty years, tricyclics and SSRIs have been prescribed in the treatment of functional GI disorders such as IBS. The prevalence of anxiety and depressive disorders is high in people with severe and/or intractable IBS and may be present to some degree in all people with IBS. A recognised pharmokinetic effect from these drug types is an analgesic effect that is separate from inhibitor effects typically desired in the treatment of depression. There is growing evidence to suggest that visceral pain syndromes such as IBS may be effectively treated using these drugs, that appear to modulate the interactions between the central and enteric nervous systems.

The two drug classes (tricyclics, e.g. trimipramine, amitriptyline, doxepin, and SSRIs, e.g. paroxetine, fluoxetine) are among the medications that have been used as long-term maintenance therapy (i.e. for 3 months or more) for IBS.

Treatment (for example with amitriptyline, the lowest cost tricyclic) is generally initiated at 10mg with dose increases of 10mg no more frequently than every 2 weeks. Patients are encouraged to increase the dose until an effective dose is established or side-effects become problematic, up to a maximum dose of 30mg for tricyclics and 20mg for SSRIs. An exception to this is doxepin, where the smallest tablet size is 50mg. Patients who do not respond to maximum dose are switched to an alternative drug, with switches from first to second tricyclic considered and switches between tricyclics and SSRIs considered.

A. Tricyclics

1. Adverse effects

All TCAs cause, to varying degrees, anticholinergic side effects (dry mouth, blurred vision, constipation, urinary retention, sweating), sedation and postural hypotension. These side effects necessitate starting with a low dose and increasing slowly. Mianserin is a sedating tricyclic.

Evidence is available for the adverse effects occurring for doxepin (Boerner 1988) and mianserin (Tanum and Malt 1996). Tanum and Malt (1996) included the adverse effect of mild sedation, and it is known that mianserin is a sedating tricyclic. All the trials compared tricyclics with placebo control.

In Boerner (1988), a dose of 50mg doxepin was compared with placebo. In Tanum and Malt (1996), the dose of mianserin was up to 120mg; we note that in this study only 60% of the participants had IBS. These doses are higher than those generally recommended for the use of these drugs in IBS.

Figure 1

The Tanum and Malt (1996) study reported that significantly more patients were found to have adverse effects of mild sedation. In both studies the confidence intervals were wide.

One further study (Myren 1984) compared different doses of trimipramine (30 to 50mg). There were no significant differences between any of the trimipramine groups and placebo for palpitations, dizziness or dryness in the mouth. There was a significant increase in tiredness and morning drowsiness in the first 1 to 2 weeks for the antidepressant groups compared with placebo, but there were no significant differences at the end of treatment.

2. Number of patients withdrawing

Two studies recorded the numbers of withdrawals from the trials, for people receiving 120mg minaserin (Tanum and Malt 1996) and 10mg trimipramine (Tripathi 1983), compared with placebo. Meta-analyis showed a statistically significant effect in favour of placebo, but the confidence interval was wide.

Figure 2

B. SSRIs

1. Adverse effects

Evidence is available for the adverse effects occurring for fluoxetine, in comparison with placebo (Kuiken 2003). It was used at a dose of 20mg in this study. The confidence interval was fairly wide, but there was no significant difference between interventions.

Figure 3

2. Number of people withdrawing because of side effects

One study also investigated the number of participants withdrawing because of side effects for paroxetine compared with placebo (Tabas 2004); and Creed 2003 recorded the number who did not complete the treatment, for paroxetine compared with usual care. The confidence intervals were either wide or very wide, so conclusions could not be drawn about the non-significant results, but we noted there were significantly more people discontinuing treatment in the paroxetine group of the Creed (2003) study than in the usual care group. In Creed (2003), the dose of paroxetine was 20mg; and in Tabas (2004), the dose was up to 40mg.

Figure 4

EVIDENCE STATEMENTS

Evidence statements for this review are mostly based on those in NICE Clinical Guideline 23 ‘Depression’ (statements 2 to 4).

1. There is a moderate amount of good quality evidence to show there are significantly more patients discontinuing treatment with SSRIs compared with usual care.
2. Patients started on antidepressants who are not considered to be at increased risk of suicide should normally be seen after 2 weeks. Thereafter they should be seen on an appropriate and regular basis.
3. In people in primary care, there is insufficient evidence to determine whether there is a clinically significant difference between other antidepressants and amitriptyline (TCA) on reducing the likelihood of leaving treatment early either for any reason or due to side effects.
4. There is good evidence in trials of eight weeks and longer that there is no clinically significant difference between SSRIs and placebo on reducing the likelihood of leaving treatment early. This is not consistent when analysing the reasons for leaving treatment, which demonstrate a clinically significant difference favouring placebo over SSRIs in relation to leaving the treatment early due to side effects.