Population focus

Four studies were UK based: three within the general population^{83,103–106,121–123} and one with the Bangladeshi community;^101,102 four were US-based, with a population of people over 40,¹¹⁰ a Hispanic population,¹¹¹ Catholic Mexican-American¹¹² and African-American^98–100 communities. In Gary 2003^98–100 the study took place in East Baltimore, a particularly deprived inner city community. A total of 629 people with any chronic condition were recruited in Kennedy 2007;^104–107 476 Bangladeshis with diabetes, CVD, respiratory disease or arthritis were recruited in Griffiths 2005;^101,102 and 602 people with arthritis in Barlow 2000;⁸³ 1140 over-40s with a diagnosis of heart disease, lung disease, stroke or arthritis were recruited in Lorig 1999;¹¹⁰ and 551 Hispanics in the northern California area, with heart disease, lung disease or type 2 diabetes (other diagnoses were allowed) were recruited in Lorig 2003.¹¹¹ The three diabetes studies recruited 150,¹¹² 186^98–100 and 591^121–123 people with diabetes.

Location

The intervention was generally delivered in non-NHS community settings,^104–107 in general practices or community centres,^101,102 in community-based setting,⁸³ such as churches, neighbourhood centres and clinics,¹¹⁰ in a faith-based community clinic with telephone follow-up¹¹¹ or in participants' homes with additional telephone contact^98–100 or by telephone alone.^121–123

Referral/recruitment

Participants were recruited via general practitioner (GP) registers^101,102 or people with self-defined long-term conditions were recruited within Strategic Health Authorities (SHAs), using community-based recruitment strategies, including posters in GP surgeries and media advertisements.^104–107 In Griffiths 2005,^100,101 a further 14 volunteered after hearing about the programme by word of mouth or local media. Barlow 2000⁸³ recruited through Arthritis Care's trainers, via the Arthritis Care Branch Network, information was placed in GP practices and rheumatology departments, and public service announcements were made in the local media. In Lorig 1999,¹¹⁰ subjects had to have a physician-confirmed diagnosis and were refereed using public service announcements in the mass media, flyers left in physicians' offices, community clinics, posters at senior citizen centres, announcements in patient newsletters and from government employees. In Lorig 2003,¹¹¹ community outreach to churches, community centres and clinics were used. Participants for the three diabetes studies were recruited via care providers: these were GP registers^121–123 and a faith-based community clinic¹¹² or medical chart review from two outpatient medical centres.^98–100

Intervention components

Role/training

Intensity of intervention

Health status

General health was measured by a single question in Kennedy 2007^104–107 and did not change significantly. It was not measured in Griffiths 2005^101,102 Neither study showed significant effect on health-related quality of life (HRQoL) as measured by the European Quality of Life-5 Dimensions (EQ-5D) instrument.^132,133 Psychological well-being was measured with five items in Kennedy 2007^104–107 and improved significantly (effect size 0.25 cited by authors). The trial was powered to have a 90% probability of detecting a standardised effect size of 0.25, and, subsequently, the target sample size of n = 600 was exceeded by 4.8%. Depression and anxiety were measured using the Health Assessment Questionnaire [Hospital Anxiety and Depression Scale (HADS)]¹³⁴ in Griffiths 2005,^101,102 but neither changed significantly.

Pain was measured on a five-point Likert scale in Griffiths 2005^101,102 and on a five-item questionnaire in Kennedy 2007.^104–107 Neither measure changed significantly. In Gary 2003,^98–100 Lujan 2007¹¹² and Young 2005^121–123 energy significantly improved in the study group compared with controls (effect size 0.18, p = 0.004), but fatigue did not change significantly in Griffiths 2005.^101,102 Physiological measures and adverse events were not assessed in either study.

Gary 2003,^98–100 Lujan 2007¹¹² and Young 2005^121–123 did not assess general health, QoL, psychological well-being, pain, fatigue or adverse events. However, in Young 2005^121–123 > 90% of intervention participants agreed that the intervention improved their well-being.

Using self-administered mailed questionnaires, Lorig 1999¹¹⁰ reported significant improvement in treatment subjects compared with controls in five variables: self-rated health, disability, social/role activities limitations, energy/fatigue and health distress (p < 0.02). Post-trial assessment of ability reveals 95% power to detect an effect size (Cohen's d¹³¹) of 0.238, equivalent to detecting, for example, a difference in means between groups of 0.3 assuming a common SD of 1.26 and a 0.05 (two-sided) significance level. No significant difference was demonstrated for pain or physical discomfort, shortness of breath or psychological well-being. Validated outcomes were used.^135–140 Using the self-rated health item from the medical outcomes studies and visual numeric scales for pain and fatigue, Lorig 2003¹¹¹ reported improvements in health status with usual care controls (p < 0.05). Post-trial power was 90% to detect a standardised mean difference of 0.279, implying an ability to detect a difference in means of, for example, 1.0, where the common SD is 3.6 and a 0.05 (two-sided) significance level.

Barlow 2000⁸³ used validated measures, including Health Assessment Questionnaire¹⁴¹ HADS,¹³⁴ and Positive and Negative Affect Scale,¹⁴² and reported statistically significant mean decreases in fatigue (effect size 0.17), anxiety (effect size 0.21) and depression (effect size 0.27) and an increase in positive mood (effect size 0.29) when compared with the control group. No significant changes were reported in the control group. No statistically significant mean changes or between group differences were found on EQ-5D¹⁴³ visual analogue scale measures. Power was predetermined at 90% to detect an effect size (difference in means) of 0.35 between groups.

Gary 2003,^98–100 Lujan 2007¹¹² and Young 2005^121–123 measured HbA_1c levels, for which the reference range (that found in healthy persons) is about 4%–5.9%.¹⁴⁴ Young 2005^121–123 found a significant (p = 0.003) difference between groups of −0.31% HbA_1c (95% CI −0.11 to 0.52) and effect size 0.25 (95% CI 0.07 to 0.43 with estimated pooled variance) at 1 year. The trial was powered to have a 90% probability of detecting a difference of 1% in HbA_1c level, assuming a SD of 2% between groups. While there were no significant differences at the 3-month assessment, Lujan 2007¹¹¹ found a significant difference of −0.25% HbA_1c levels at 6 months (effect size 0.41, 95% CI 0.08 to 0.73). However, there was a difference in the mean baseline HbA_1c level between the intervention and control groups, the intervention mean being 0.45% higher. Levels of HbA_1c increased markedly in the control group over 6 months (0.3%). It is generally accepted that HbA_1c levels rise over time, but at a typical rate of 0.2% per year. Pretrial power was set at 90%, based on unspecified differences in HbA_1c levels and Diabetes Knowledge Questionnaire scores.

Gary 2003^98–100 found a similar-sized difference between their groups at 2 years – −0.30% HbA_1c (± 0.48%, insufficient information for an effect size calculation) – but this was not statistically significant. Gary 2003^98–100 measured other surrogate markers of cardiovascular health, such as low-density lipoprotein (LDL) and high-density lipoprotein (HDL) cholesterol levels, triglycerides, systolic and diastolic blood pressure, and body mass index (BMI). Unfortunately, the absolute changes in outcomes are not reported in Gary 2003^98–100 only the difference between arms, making it difficult to assess whether the reported effects are due to decreases in the trial arms or increases in the controls. The reported changes in systolic and diastolic blood pressure, HDL and LDL cholesterol and triglycerides are mostly in a similar direction to the primary outcome measure, but are not significant. Neither a target difference nor a difference in power were specified; however, defining a clinically significant difference would seem to be the main issue in studies of this nature. That said, the authors claim that the observed difference of −0.8% between the collective treatment group and controls was clinically significant. Therefore, given the statistically non-significant p-value, one might conclude that the study may have been underpowered. Nonetheless, the study may prove useful and encouraging to anyone planning further work in this area.

Health behaviours

Self-care behaviour was assessed in Griffiths 2005^101,102 using the Cognitive Symptom Self-Management Scale from the Chronic Disease Self-Efficacy Scale¹⁴⁵ and improved significantly (effect size 1.16, p = 0.047). The authors state that their study was 80% powered to detect ‘… an effect size associated with improvements in behaviour, heath status and healthcare …’. Subsequent recruitment figures confirm that the trial was sufficiently powered to detect relatively small effect sizes. Kennedy 2007^104–107 measured exercise (six items) and diet (one item), neither of which changed significantly. Lorig 1999¹¹⁰ reported significant improvement in four behaviour variables (p < 0.01): number of minutes exercise per week of stretching/strengthening exercise and aerobic exercise, increased practice of cognitive symptom management and improved communication with physician. At 4-month comparison, Lorig 2003¹¹¹ reported improvements in health behaviours compared with usual care controls (p < 0.05).

Barlow 2000⁸³ used scales developed by the Stanford Arthritis Centre¹⁴⁵ and reported statistically significant mean increases in cognitive symptom management relative to controls (effect size 0.46), communication with physician (effect size 0.24) and no mean change on dietary habit or fluid intake. No significant changes were found in the control group.

Gary 2003^98–100 measured dietary practices using a validated food frequency questionnaire designed to guide cholesterol reduction in low-income individuals;¹⁴⁶ this did not change significantly between the groups. Physical activity was measured using a validated questionnaire about habitual physical activity during leisure time¹⁴⁷ and this increased significantly in the CHW group and CHW/nurse case manager group compared with the control group, all p < 0.05 (mean change +0.26 ± 0.18 and +0.34 ± 0.18, respectively).

Participation

Kennedy 2007^104–107 found that social role limitation (assessed with four items) improved significantly in the expert patient group (effect size 0.19). Griffiths 2005^101,102 did not assess participation.

Health-care beliefs and knowledge

The primary outcome for Kennedy 2007^104–107 was self-efficacy, which both studies claimed to improve significantly: and Griffiths 2005,^101,102 effect size of 1.47; Kennedy 2007^104–107 effect size of 0.44. Kennedy 2007^104–107 found no significant differences in self-efficacy among groups with different chronic conditions. Griffiths 2005^101,102 assessed communication with physicians using the communication strategies scale of the Chronic Disease Self-Efficacy Scale,¹⁴⁴ but it did not change significantly. Kennedy 2007^104–107 assessed partnership with clinicians (with four items), which improved significantly in the Expert Patients Programme group (effect size 0.25). Lorig 2003¹¹¹ assessed physician visits, which remained statistically unchanged. In Griffiths 2005,^101,102 51% of intervention participants attended three or more sessions, whereas 21% attended none. The attendance in Kennedy 2007^104–107 was higher, with 60% attending four or more sessions. Neither Griffiths 2005^101,102 nor Kennedy 2007^104–107 measured any other aspects of health-care beliefs and knowledge. At 4-month comparison, Lorig 2003¹¹¹ reports improvements in self-efficacy compared with usual care controls (p < 0.05). Barlow 2000⁸³ used the Arthritis Self-Efficacy (ASE) Scale¹⁴⁸ and reported statistically significant mean increases on ASE: other symptoms (effect size 0.43) and pain (effect size 0.41). Small, but statistically significant, increases in ASE pain score (effect size 0.14) were also found in the control group (unverified statistics).

Lujan 2007¹¹² measured diabetes knowledge and health beliefs using validated questionnaires. The DKQ¹⁴⁹ score mean change of the intervention group was significantly higher than that of the control group at the 6-month assessment, with effect size 0.63, 95% CI 0.29 to 0.97 (baselines of original adjusted for health insurance). With the diabetes health beliefs measure,¹⁵⁰ a higher score indicates a higher belief in the ability to manage diabetes. The mean changes of the two groups decreased, without a significant difference at the 3-month assessment, the decrease was significantly less [F(1, 148) = 5.97, p < 0.01] for the intervention group than for the control group at the 6-month assessment. The consistent decrease in the diabetes health beliefs mean scores of both of the groups at the two points of assessment indicates that the participants did not experience an increase in their belief about their ability to manage diabetes, although the intervention group demonstrated more knowledge.

In Gary 2003^98–100 it was expected that individuals would complete six intervention visits before the 2-year follow-up. Their actual participation fell far short of that goal, primarily because of insufficient staff support and participant non-compliance (although figures were not provided). Overall, more individuals were seen in the health advisor groups, which may be related to the fact that they saw the participants in the convenience of their homes. This may be a surrogate indicator of acceptability, which appears to be better in the health advisor group than in the usual care group.

In Lujan 2007¹¹² 96% of participants completed the classes (i.e. attended at least six of the eight classes) and the overall attrition rate was 6% (n = 9). Two of these nine participants also failed to complete the education phase of the intervention. One of the participants, who did not attend either the 3- or 6-month assessment interview, died from pneumonia, two moved to another city, and six reported that they were unable to attend the assessment interviews because of a lack of time. The very high attendance rate of the classes suggests that it was acceptable to most participants.

Young 2005^121–123 noted that withdrawal from the study occurred in 10.7% of usual care subjects and 15.7% of telephone-support patients. This suggests that there may have been some negative issues regarding acceptability in the telephone-support group. They assessed satisfaction with treatment using the validated Diabetes Satisfaction and Treatment Questionnaire,¹⁵¹ and acceptability of the approach with a purposely designed self-completion questionnaire. Over 90% of participants found the intervention acceptable and agreed that it improved their knowledge and control of diabetes. However, only 50% of intervention participants would rather have this approach than seeing a health professional face to face. Participants generally described the development of strong bonds with the LAs, and liked the personalised format of the intervention. A total of 33% thought it had enhanced their self-knowledge and helped with changes in attitudes and behaviours.

Health-care use

There was no significant difference in health-care visits over 6 months in Kennedy 2007^104–107 or in primary care visits over the previous 3 months in Griffiths 2005.^101,102 Kennedy 2007^104–107 also measured the number of counsellor visits, outpatient appointments, day-case appointments and inpatient days, none of which differed significantly between the two groups. Lorig 1999¹¹⁰ reported that the treatment group had fewer hospitalisations (p < 0.05) and spent, on average, 0.8 fewer nights in hospital (p = 0.01). There were no significant differences in visits to physicians (p = 0.11). At 4-month comparison, Lorig 2003¹¹¹ reports no difference in days of hospitalisation, but the treatment group did show a trend to fewer physician visits. Barlow 2000⁸³ reports on number of physician visits where arthritis was discussed, but did not find any difference at 4 months between intervention and control groups. However, at 12 months they found significantly fewer mean number of visits to the GP, though these data were uncontrolled.

Costs

The delivery of the Expert Patients Programme cost £123 per participant in Griffiths 2005,^101,102 and £250 per participant in Kennedy 2007.^104–107 Kennedy 2007^104–107 found lower overall costs in the intervention arm that more than compensated for the estimated cost of the intervention (£250). This difference was driven by a marked (but not statistically significant) reduction in inpatient length of stay. The difference, 0.8 days, has a large impact on overall costs owing to the high cost of inpatient stays (cost of £203–486 applied). It is possible that this difference has arisen from a few patients with extended hospital stays.

In Lorig 1999¹¹⁰ the treatment group reduced visits to physicians slightly more than control group but the difference was not significant. The decreases in the number of hospitalisations and in the number of nights of hospitalisation were significant (p < 0.05). Assuming a cost of US$1000 per day of hospitalisation, the 6-month health-care costs for each control participating were > US$820 for each treatment subject. The costs of providing the programme for treatment subjects who completed the 6 months were calculated to be US$70 per participant. This includes US$26 for training leaders. No costs are reported in Lorig 2003¹¹¹ and Barlow 2000.⁸³

Population focus

Two studies^26,97,120 were based in the USA, and two in the UK.^30,113 West 1998³⁰ recruited 172 smokers based in the general population; May 2006¹¹² recruited 564 smokers from three sites across London, UK. Woodruff 2002¹²⁰ recruited 313 smokers in the Latino community, and Emmons 2005^26,97 recruited 796 smokers who were childhood cancer survivors.

Location

The intervention was delivered in people's homes via visits and telephone calls¹²⁰ or telephone calls alone.^26,30,97,113

Referral/recruitment

West 1998³⁰ recruited smokers from their GP records in south-east London. Participants in May 2006¹¹² were a subset of those participating in a larger RCT of glucose as an aid for smoking cessation. In this study, smokers were recruited through advertisements in local papers, word of mouth and GP referrals. Emmons 2005^26,97 recruited smokers from the Childhood Cancer Survivors Study^152,153 register, and Woodruff 2002¹²⁰ used 11 trained recruiters, who worked at community events, popular neighbourhood shopping centres, and within their own social networks, to identify Latino smokers.

Intervention components

Role/training

Intensity of intervention

Health status

No measures of health status were assessed in any of the studies.

Health behaviours

Woodruff 2002¹²⁰ found that attrition rates were significantly different by condition, with 4.5% of comparison group participants dropping out versus 15.4% of the intervention participants (χ²[1] = 10.47, p < 0.001) (effect size 0.18). Participation in the intervention varied from zero to seven sessions with an average of 3.44 sessions. In total, 24% of the control group reported using the smoking cessation helpline.

Emmons 2005^26,97 uses self-reported smoking status at 8 and 12 months; both West 1998³⁰ and Woodruff 2002¹²⁰ use expired air CO₂ monitoring to verify self-reported abstinence, at 1 month and 3 months, respectively. Significantly more remained abstinent from smoking in the buddy support group after 4 weeks [15% difference, effect size 2.79 (95% CI 1.26 to 6.22)] in West 1998³⁰ but not in May 2006¹¹³ (adjusted OR 1.16, 95% CI 0.76 to 1.78). Woodruff 2002¹²⁰ reports that 20.3% of the intervention group had quit at 3 months compared with 8.7% of the comparison group, the comparison group being a statistically significant 2.5 times more likely than the intervention group to be smoking at the 3-month assessment after adjusting for gender and amount smoked per day at baseline. However, these results are based on a per-protocol analysis that ignores differential attrition in the intervention arm. Applying a more conservative intention-to-treat analysis, and assuming that all of those lost to follow-up have relapsed, gives a quit rate of 17.3% in the intervention group and 8.3% in the control group. Emmons 2005^26,97 reported that the quit rate was significantly higher in the peer counselling group than the control group (16.8% vs 8.5%, p < 0.0003) at 8 months. This difference was maintained at the 12-month follow-up (15% vs 9%, p < 0.01). Controlling for baseline self-efficacy and depression, the peer counselling group was more likely to quit smoking by the 12-month follow-up compared with the control group (12-month OR 1.99, 95% CI 1.27 to 3.14). Post hoc power for this trial was around 87% for a difference of 8.5% (quit rate) between groups, albeit that power to detect differences in proportions was dependent on the location of the difference.

Participation

None of the studies assessed any measures of participation.

Health-care beliefs and knowledge

Emmons 2005^26,97 reported that 74% of the control participants responded that they had indeed received the self-help smoking cessation materials. Of that group, 67% reported having read either a lot, or all, of the materials sent; 56% of participants reported that they found the materials to be somewhat useful, and 21% reported that they were very useful. As expected, recall of receipt and rates of use of the materials were higher among the peer-counselled intervention group participants (95% reported receiving the materials, 79% reported reading a lot or all of the materials).

None of the studies reported any other measures of health-care beliefs and knowledge.

Health-care use

West 1998³⁰ and Emmons 2005^26,97 reported the rates of use of NRT. There was no significant difference in the use of NRT in the two groups in West 1998,²⁹ with about 50% of both groups using it. Emmons 2005^26,97 reported that at the 8-month follow-up, 33% of participants in the peer-support condition reported that they had used NRT during the previous 6 months, compared with 8% of the control (self-help) participants. At the 12-month follow-up, 16% of the provider counselling participants indicated that they had used NRT in the previous 4 months compared with 6% of self-help participants. No significance values were given. A total of 14% of those in the self-help group who used NRT reported that they had quit compared with 26% of the peer counselling group, although this difference did not reach significance using intention-to-treat analyses.

Emmons 2005^26,97 does not appear to have recorded NRT use in the control group for the first 2 months of the trial, and this oversight may explain why there is no attempt to adjust for it in the results. NRT has well-established effectiveness data with an OR for the patches of 1.67. There is no mention of NRT in Woodruff 2002,¹²⁰ and it does not appear to form a planned constituent of the intervention.

None of the studies showed any other data for health-care use.

Costs

No costs were given in West 1998,³⁰ May 2006¹¹³ or Woodruff 2002.¹²⁰ Emmons 2005^26,97 stated that the total intervention delivery cost per person was US$298.17 for the peer counselling group and US$1.25 for the self-help group. Therefore, the incremental cost-effectiveness of the peer counsellor (PC) condition compared with the self-help (SH) control [(cost_pc − cost_sh)/(quit rate_PC − quit rate_SH)] was US$5371 per additional quit at 12 months.

Population focus

The studies were based in semiurban settings in North America: Mexico City, Mexico^114,115 and Toronto, ON, Canada.⁸⁸ Morrow 1999^114,115 recruited 130 pregnant women, whereas Dennis 2002⁸⁸ recruited 258 primiparous breastfeeding mothers.

Location

The interventions were delivered in mothers' homes by home visits ^114,115 or via telephone.⁸⁸

Referral/recruitment

Study mothers were identified in Morrow 1999^114,115 by a semiannual door-to-door census and continuous reporting of new pregnancies in the community by study staff and mothers. Eligible mothers were identified within two community hospitals for the Dennis 2002⁸⁸ study.

Intervention components

Role/training

Intensity of intervention

Health status

Neither study assessed any measure of health status of the mothers. Morrow 1999^114,115 measured rates of diarrhoea in infants 0–3 months of age, which was reduced significantly in the supported group (RR = 0.47, i.e. the probability of a baby in the intervention group having diarrhoea is 0.47 that of the control group, or less than a half.)

Health behaviours

Dennis 2002⁸⁸ and Morrow 1999^114,115 reported that their intervention groups were significantly more likely to be breastfeeding at 3 months [Morrow 1999^114,115 exclusive breastfeeding only (RR = 1.11); Dennis 2002⁸⁸ all breastfeeding, (p = 0.01) RR = 1.21] but this effect was not observed at 3 months when it was measured in Morrow 1999.^114,115 Dennis 2002⁸⁸ also noted that the rates of exclusive breastfeeding were significantly higher in the intervention group up to 3 months post partum (p = 0.01, RR = 1.21). Morrow 1999^114,115 also provides details of the differential responses in breastfeeding outcomes to peer counselling, finding that multiparous women and those with uncertainty about infant feeding plans were more likely to respond to peer counselling by initiating breastfeeding. It was also demonstrated that peer counselling had the ability to counteract the negative effects of early supplementation on breastfeeding among the subgroup of breastfeeding mothers who have introduced formula within the first day post partum. The study by Morrow 1999^114,115 was powered on a hypothesised difference between a combined intervention group (three and six visits) and a control group; however, results are also reported in relation to differences between the three groups. Post hoc power was 95% in relation to the observed difference of 20% between the combined study group and controls. Dennis 2002⁸⁷ powered their study at 90% to detect a difference of 20% located at 60% and 80%; post hoc recruitment figures confirmed power at 90%.

Neither study assessed any other measure of health behaviour.

Participation

Neither study assessed any measure of participation.

Health-care beliefs and knowledge

Dennis 2002⁸⁸ found no significant difference in mean satisfaction scores between the two groups on maternal satisfaction, but significantly fewer mothers in the intervention group reported dissatisfaction [RR (intervention vs control) = 0.63, 1.5% vs 10.5%, p = 0.02]. Significantly fewer mothers in the peer support group in Dennis 2002⁸⁸ indicated that they would breastfeed their next infant differently (RR = 0.68, 23% vs 34%, p = 0.05). Morrow 1999^114,115 stated that nearly all (98%) intervention group mothers reported that the peer counsellor was helpful and supportive. In Dennis 2002⁸⁸ three mothers indicated dissatisfaction with the peer support, most indicating a preference for a higher frequency of contact. However, a few mothers responded that they did not like a specific aspect of their peer volunteer. For example, only one mother requested to discontinue her participation in the intervention, stating that the peer volunteer frightened her about the potential hazards of not breastfeeding. The peer volunteer's comments made her anxious and diminished her feelings of confidence, despite the fact that breastfeeding was going well. Another mother felt her right to confidentiality was violated when her peer volunteer contacted the public health department without her consent. Although this mother did require professional assistance, the peer volunteer should have discussed the referral with the new mother. Neither study assessed any other measure of health-care beliefs or knowledge.

Health-care use

Morrow 1999^114,115 noted the number of visits to the doctor due to infant diarrhoea but did not compare between the two groups. Dennis 2002⁸⁸ reported on health service utilisation but in a format inaccessible to the reviewers in the timescale available.

Costs

Neither study assessed any measure of costs.

Population focus

The study was based in the USA and recruited 161 mothers whose children aged 7–11 years had been diagnosed as having diabetes, sickle cell anaemia, cystic fibrosis or moderate-to-severe asthma, living within a 80-km range of Baltimore, MD, USA.

Location

The intervention for this study was delivered in participants' homes or in nearby coffee shops if requested, as well as in the community (for events organised, such as bowling parties or small group lunches).

Referral/ recruitment

Participants were identified by 11 specialty clinics and five general paediatric clinics.

Intervention components

Role/training

Intensity of intervention

Health status

Levels of anxiety were measured using an 11-item anxiety subscale of the Psychiatric Symptom Index (PSI).¹⁶⁶ Whereas participants in the control group reported higher levels of anxiety after baseline, participants in the experimental group reported lower levels of anxiety post intervention compared with baseline scores. The interventions' effect (reduction in anxiety scores) was especially pronounced for mothers who were highly anxious at baseline, with mean anxiety scores for the highly anxious experimental group mothers decreasing from 33.3 at baseline to 26.4 at 12 months post baseline, and those for the highly anxious mothers in the control group remaining unchanged. Maternal physical health was also an important factor in determining effects of the intervention. The mean anxiety score for mothers in the experimental group who reported that they were in good, fair or poor health at baseline decreased from 26.4 to 23.9 during the intervention period, whereas for those mothers in the control group who reported being in good, fair or poor health the mean anxiety score increased. Whereas mothers in the experimental group who reported being in very good or excellent health also showed a decrease in anxiety (from 13.4 to 11.5), those in the control group reporting very good or excellent health reported an increase in anxiety in this period (from 15.2 to 17.9). No relationship was found between the effects of the intervention and the number of reported stressful life events or the dose of the intervention. No effect was demonstrated on symptoms of depression as reported on the Beck Depression Inventory.¹⁶⁷ The second step model (using baseline as a covariate) resulted in a standardised ‘B coefficient’ of 0.145 (p ≤ 0.05); however, this effect disappears when other covariates are included in the model. Specifically, the effect of each and all of the stage 3 factors make a substantial contribution to the regression coefficient (R² = 0.51), suggesting that 51% of the variance in post-test PSI anxiety score is explained by the stage 3 model, i.e. it is the most predictive model. In other words, the intervention group was no longer a significant factor.

Health behaviours

Not measured.

Health-care beliefs and knowledge

Not measured.

Health-care use

Not measured.

Costs

No details given.

Population focus

All four studies were based in the USA. Bird 1998^84–87 surveyed 645 Vietnamese-American women in two urban communities (San Francisco, CA – intervention; Sacramento, CA – control). In Andersen 2000,^42,81,82 a cohort of 352 women aged 50–80 years from each of the 40 communities (giving a total of 14,080 participants) was randomly selected and surveyed to assess intervention effectiveness. The communities were located predominantly in rural areas of Washington state. Earp 2002,^16,63,93,94 surveyed 993 rural African-American women in 10 counties in NC, USA; five counties were allocated to each group and they were also geographically separated by the Pamlico Sound. The studies of Bird 1998^84–87 and McPhee et al.,^85,86 were conducted in the context of the Breast and Cervical Cancer Control Program (BCCCP), which covers screening fees for all age-eligible, low-income women. Paskett 2006^116,117 assessed the impact of LHAs who were randomly assigned to 453 women individually, with a control group of 444 women receiving normal care. These 897 women were from a rural, low-income, triracial (white, Native American and African-American) population within a county ranked the eighth poorest of the 100 counties in North Carolina and in which one-half of the adults are high school graduates.

Location

The interventions were delivered in participants' homes (Bird 1998,^84–87 by telephone, Paskett 2006^116,117), in community settings (beauty parlours, churches, bingo halls, clubs, stores, libraries, golf courses)^{16,63,84–87,93,94} or in health settings within the community (health fairs, mobile mammography van days).^16,63,93,94

Referral/recruitment

The LHAs spoke to any woman with whom they came into contact in their social group.^{16,63,84–87,93,94} In Paskett 2006,^116,117 women who had been clients of the clinic for at least 2 years and had not had a mammogram in the prior year were randomly selected from the health records of their health-care provider. Participants were randomly selected from a list of women purchased from a mailing list company in Andersen 2000^42,81,82

Intervention components

Role/training

Intensity of intervention

Health status

Not measured in Andersen 2000,^42,81,82 Bird 1998,^84–87 Earp 2002^16,63,93,94 or Paskett 2006.^116,117

Health behaviours

Bird 1998^84–87 distinguishes regular users from those who have ever had a mammogram, defining regular users as those who have had at least two mammograms in the previous 5 years with the most recent within 18 months. Andersen 2000^42,81,82 defines regular users as those reporting at least two mammograms with one in the last 2 years (50% of sample), and all other women as underusers.

Bird 1998^84–87 reports the largest gain in regular mammography users. The unadjusted data show an increase of 18% in the intervention arm compared with a fall of 4% in the control. The rates of ever having had mammograms (intervention OR 2.2) and Pap smear (OR 4.5) were significantly raised in the intervention community.^84–87 In addition, the rates of having had more than one screen in the last 5 years were, again, significantly raised in the intervention population for mammograms (OR 2.4) and cervical cancer screening (OR 2.4). The trial had in excess of 80% power to detect clinically significant differences for all primary and secondary outcomes.

Earp 2002,^16,63,93,94 showed that self-reported mammography use in past 2 years increased in the intervention group compared with the controls by a statistically significant 7% (adjusted for age, medical visits, physician recommendation for mammography and perceived susceptibility to breast cancer). The difference between the two populations was even greater when just the low-income (< US$12,000 per year) women in each community were compared: 11% (adjusted, p = 0.02 − insufficient data reported to calculate effect size). The high-income women in the two communities did not differ significantly in their use of mammograms. Post hoc power was not assessed owing to the diversity of outcomes.

Paskett 2006^116,117 showed that those in the LHA group were significantly more likely to have reported having a mammogram in the 12 months before the follow-up assessment (RR = 1.56, 95% CI = 1.29 to 1.87, p < 0.001). When assessed by racial group (African-Americans, Native Americans and white people), all three groups improved rates of mammography use and there were no statistically significant differences in screening rates observed between racial groups or clinics. A total of n = 820 women completed the study resulting in > 80% power to detect a prespecified difference of 10% overall and 20% within racial groups.

Andersen 2000^42,81,82 studied the effect of IC and (IC + CA)/or CA on women who were underusers of mammography at baseline, and on the prevention of relapse for those women who had had mammograms at regular intervals at the baseline interview. Each intervention demonstrated increases in mammography use in both regular users (relapse prevention) and underusers relative to the control communities. The only statistically significant difference is observed among regular users in the CA arm, where 2.9% more women report a mammogram. Andersen 2000^42,81,82 combines the impact of the intervention among regular and underusers to obtain a percentage increase in the number of women using mammography of 2.5% in the CA arm. Given the similar costs for each intervention the authors conclude that CAs are the most cost-effective. This was a large-scale study (n = 6592), providing high power to detect small differences between three treatment groups and controls.

Participation

Not measured in Andersen 2000,^42,81,82 Bird 1998,^84–87 Earp 2002^16,63,93,94 and Paskett 2006.^116,117

Health-care beliefs and knowledge

Bird 1998^84–87 measured whether the women had ever heard of mammography (OR 7.0) or Pap smears (OR 52.7), both of which were significantly increased in the intervention population.

Earp 2002^16,63,93,94 measured increase in awareness of mammography-promoting interventions and materials over the period of the intervention (3 years), which did not differ significantly between the two groups. However, there was a difference between high- and low-income groups, in that although women with a high income had more exposure to the intervention the changes for this group were smaller.

Paskett 2006^116,117 measured knowledge (12 items), barriers (12 items) and beliefs (four items) of mammograms and breast cancer with an unvalidated questionnaire developed for their study. The knowledge scores did not differ significantly between the groups. The barrier score was significantly smaller in the LHA group (insufficient data reported to calculate an effect size). The proportion of women reporting inaccurate beliefs was statistically significantly reduced (p = 0.034) in the LHA group (insufficient data reported to calculate an effect size).

Andersen 2000^42,81,82 did not measure health-care beliefs or knowledge.

Health-care use

Not measured in Andersen 2000,^42,81,82 Bird 1998,^84–87 Earp 2002^16,63,93,94 and Paskett 2006.^116,117

Costs

Not measured in Bird 1998^84–87 and Paskett 2006.^116,117 stated that the total cost of the intervention was US$329,054. The difference in mammography rates between the two groups was 15.2%, which translates into 66 additional mammograms in the LHA group; therefore, each additional mammogram in the advisor group cost US$4986.

Although no exact intervention costs were given in Earp 2002,^16,63,93,94 the programme has entailed ‘substantial direct costs’ due to staffing costs as a result of the large size of the LA network and the area that it covers (although the LAs volunteered their services, paid staff were involved in the stages of implementation, most intensively in the training phase); the materials for the training workshops and LA activities; consultant expenses, incentive payments, refreshments, tape recorders, tapes and transcription costs associated with the focus groups; and, finally, consultant expenses, development of mock-ups, photography costs and printing costs associated with brochure development.

Although no exact intervention costs were detailed in Bird 1998,^84–87 it is mentioned that the free services that were available at the time of the study allowed the trial to be conducted in a cost-free environment, and therefore, if participants had been subject to fees for screening, increases in receipt and maintenance of tests might have been smaller.

Population focus

All five studies were conducted in North America: in a Canadian Aboriginal community;⁸⁰ in uninsured women, over the age of 50 years, from a mainly Hispanic community;¹¹⁹ in a Latinas community;^95,96 in African-American church communities;¹¹⁸ and in African-American women with type 2 diabetes^108,109). In Anand 2007,⁸⁰ 57 households (174 individuals) were recruited; 357 participants in Elder 2006;^95,96 200 in Keyserling 2002;^108,109 1022 participants in Resnicow 2004;¹¹⁸ and 326 in Staten 2004.¹¹⁹ The rural or urban nature of the studies was not well defined in any of the studies but was probably rural in Anand 2007⁸⁰ (on the reservation) and probably urban in Staten 2004¹¹⁹ (clinics in Tucson). In Anand 2007⁸⁰ the household structure was chosen to build upon the strength of family ties and promote healthy lifestyle role modelling. Two of the studies restricted the age of their participants: over 40 years of age^108,109 or over 50 years of age.¹¹⁹

Location

The interventions were delivered at home,^80,95,96 home and clinic,^108,109 home, clinic and community,¹¹⁸ and in church and the home.¹¹⁸

Referral/recruitment

Participants were recruited via clinics in two studies,^108,109,119 by telephoning people in the region with Hispanic surnames,^95,96 by recruiting within church communities on a first-come first-served basis,¹¹⁸ and by recruiting eligible households within the reservation.⁸⁰ In Staten 2004¹¹⁹ the clinics from which participants were recruited were participating in the National Breast and Cervical Cancer Early Detection Program.

Intervention components

Role/training

Intensity of intervention

Health status

None of the studies identified measured general health status, QoL, pain, fatigue or adverse events. One study measured psychological outcomes: Keyserling 2002^108,109 measured mental well-being on a validated scale,¹⁹⁷ but this did not differ significantly between the groups. Three of the studies measured specific physiological measures: Staten 2004¹¹⁹ found no significant reductions between waist measurements in the CHW group versus the group that had provider counselling and health education (linear regression adjusted for BMI, ethnicity and age, – insufficient information to calculate effect size). There were significant reductions in systolic blood pressure (approximately 5.4 mmHg – insufficient information to calculate effect size) the CHW group versus both other groups (linear regression unadjusted). They also measured BMI, diastolic blood pressure, total cholesterol, glucose and triglycerides, but these physiological measures did not differ significantly between the groups. Keyserling 2002^108,109 measured HbA_1c levels, total cholesterol levels, HDL cholesterol levels and weight, but these physiological measures did not differ significantly between the groups. Anand 2007⁸⁰ found that, overall, there were no statistically significant changes in body weight, waist circumference, skinfold thickness or body fat percentage in the intervention versus the usual care group. Resnicow 2004¹¹⁸ did not assess any measures of health status. Elder 2006^95,96 measured BMI, but failed to report the results.

Health behaviours

Three studies assessed physical activity levels: Keyserling 2002^108,109 measured exercise over a period of 1 week using Caltrac accelerometers – devices worn on the hip and designed to detect and record movement; Staten 2004¹¹⁹ and Anand 2007⁸⁰ rely on self-reported data. Keyserling 2002^108,109 determined that the CHW group was significantly more active at 12 months than the minimal intervention group [effect size approximate (unadjusted 3.0) ∼52 kcal/day], but not significantly different from the group that had access to the clinic intervention. Staten 2004¹¹⁹ and Anand 2007⁸⁰ found no significant difference between the groups' physical activity levels.

Dietary changes were measured in all five studies:

• Total food energy Elder 2006^95,96 showed a significant difference in energy immediately after the intervention. This effect did not last to the 6- and 12-month follow-up time points. Keyserling 2002^108,109 and Anand 2007⁸⁰ showed no differences in total daily energy intake.
• Fats Resnicow 2004¹¹⁸ found significant reductions in the amount of fat eaten (effect size = 0.26). Anand 2007⁸⁰ also found a significant reduction in trans fatty acid consumption [p = 0.02, mean difference (D) = 0.8, effect size (ES) = 0.34, 95% CI −0.65 to −0.02] and a reduced consumption of ‘fats, oils and sweets’ by approximately two servings per day (p = 0.006, D = 1.9, ES = 0.12, 95% CI −0.44 to 0.19, compared with the control). However, there were no differences in the percentage of daily calories from fats or in the consumption of ‘milk, yoghurt and cheese’. Elder 2006^95,96 found significant reductions in dietary total fat and total saturated fat consumption. These effects did not last to the 6- and 12-month follow-up time points. Keyserling 2002^108,109 found no significant differences between the groups' percentage of calories from saturated fat or dietary cholesterol.
• Carbohydrates Elder 2006^95,96 showed a significant reduction in total carbohydrates, glucose and fructose immediately after the intervention, but this difference was not seen at 6 and 12 months' follow-up. Anand 2007⁸⁰ found no difference in percentage of calories from carbohydrates or the types of food served (such as ‘bread, cereal, rice, pasta’).
• Proteins Anand 2007⁸⁰ found no difference in percentage of calories from protein or the types of food served (such as ‘meat, poultry, fish, dried beans, eggs, nuts’).
• Fruit and vegetables Resnicow 2004¹¹⁸ found significant improvements in the levels of fruit and vegetables eaten (effect size of 0.39 for the two-item measure and 0.18 for the 17-item measure), but Staten 2004¹¹⁹ and Anand 2007⁸⁰ found no significant differences in the consumption of fruit and vegetables between their groups.
• Drinks Only Anand 2007⁸⁰ measured consumption of drinks and they found that water consumption increased by ∼0.4 of a serving per day (p = 0.04, D = 0.04, ES = 0.35, CI 0.03 to 0.67), and carbonated drink consumption decreased by ∼0.2 servings per day (p = −0.02, D = 0.02, ES = 0.16, 95% CI −0.15 to 0.48).

Participation

Resnicow 2004¹¹⁸ measured the levels of social support to eat more fruit and vegetables on a scale developed for this study and found it was significantly improved (effect size 0.39). Keyserling 2002^108,109 measured social well-being on a validated scale but this did not differ significantly between the groups.

Three studies^80,95,96,119 did not measure participation.

Health-care beliefs and knowledge

One study measured health-care beliefs: Resnicow 2004¹¹⁸ measured autonomous/intrinsic motivation and controlled/extrinsic motivation with a validated outcome measure and self-efficacy with a measure developed for this study and found that these all significantly improved in the intervention group (effect sizes of 0.21, 0.33 and 0.22, respectively).

One study measured health-care knowledge: Keyserling 2002^108,109 measured diabetes knowledge with a validated scale (see Dunn et al.¹⁹⁸). Although they stated that there was a significant overall group effect (p = 0.037) they did not conduct the analysis to determine which group(s) produced this effect and whether it was significant.

The patient acceptability of the intervention was measured specifically in two studies.^108,109,118 Resnicow 2004¹¹⁸ measured satisfaction with the programme; 77% of participants reported being very satisfied with the cookbook and educational materials, and 72% of those receiving at least one call reported being very satisfied with their volunteer advisors. Keyserling 2002^108,109 measured programme acceptability. For clinic-based IC, 94% of 117 respondents reported being very satisfied with the amount of information and help the nutritionist gave about diet, and 88% were very satisfied with the counselling provided to enhance physical activity, whereas 15% reported having some difficulty getting to the clinic for these visits. For the community diabetes advisor component, 85% of 59 respondents felt the number of telephone calls was appropriate, 86% felt the role of community diabetes advisors in the programme was important, and 83% strongly agreed that talking to someone else with diabetes was very helpful. One study measured attrition rates between the groups, which can be suggested to be a surrogate marker of acceptability: Elder 2006^95,96 found that the total attrition rate over 12 months was 21%: 23% in the promotora group, 24% in the tailored print group and 18% in the control group.

Two studies^80,119 did not assess any measure of health-care knowledge or beliefs.

Health-care use

None of the studies identified assessed any measure of health-care use.

The studies by Keyserling 2002,^108,109 Staten 2004¹¹⁹ and Elder 2006^95,96 were not powered to detect specific differences. Anand 2007⁸⁰ powered their trial at 80% to detect modest changes in total calories and increase in physical activity.

Costs

Elder 2006^95,96 detailed the costs to be US$9 per participant for the control condition, US$45 per participant for the tailored condition and US$135 per participant for the intervention group. In looking at simple costs per unit of pre–post change for the control, tailored and promotores groups, respectively, these costs were US$1.30, US$5.11 and US$8.28 per reduced gram of fat; US$3.21, US$17.31 and US$21.09 per reduced gram of saturated fat; and US$0.07, US$3.21 and US$0.36 per reduced calorie.^95,96 Resnicow 2004,¹¹⁸ although not detailing the costs of the intervention, do state that larger-scale dissemination of the intervention would require ‘a considerable cadre of trainers to implement the intervention, which would involve substantial costs’. The other studies did not identify any assessed costs.

Population focus

The studies were conducted in the USA but within different project settings. The two studies were conducted in an urban community of active drug users,^89–92 in which the focus was work with these drug users as LAs. In Dickson-Gomez 2003^89,90 and 2006^91,92 some participants were also HIV sero-positive (20% in Dickson-Gomez 2006^91,92), or homeless or had a history of sexually transmitted disease (STD) or hepatitis.

Location

The interventions were delivered in the community,^89,90 through outreach^91,92 or in an unspecified training location (Dickson-Gomez 2003^89,90 for the training per se). Settings had a particular impact on intervention effectiveness and acceptability in Dickson-Gomez 2006^91,92 as interactions could take place in the streets or in other public or private places, which could be very transitory in nature.

Referral/recruitment

Participants were identified from a previous study and through street outreach and by direct invitation by the project staff,^91,92 and through outreach, ethnographic observations, focus group and geographical coding of drug-related arrests.^89,90

Intervention components

Role/training

Intensity of intervention

Health status

In Dickson-Gomez 2006^91,92 many LAs reported positive experiences related to their own health and well-being, including their knowledge about risk and prevention.

Health behaviours

Dickson-Gomez 2003^89,90 report significant differences between intervention and control participants in overall drug use and unsafe practices: reduction in injection drug use (48% intervention vs 25% control, p < 0.05); increase in cessation (44% intervention vs 22% control, p < 0.05); and reductions in unhygienic needle use (69% intervention vs 30% control, p < 0.10). Some success in reducing risky sexual behaviour is also reported: reduction in unprotected vaginal sex with casual partners (16% intervention vs 4% control, p < 0.05); reduction in number of casual partners (18% intervention vs 7% control, p = 0.05). There were no changes observed in condom use with regular partners. Regression modelling suggested that the intervention condition was almost three times more likely to result in a reduction in injection drug use than the control condition (OR 2.8) and a significant reduction in the use of unhygienic needles (χ² = 3.57, p < 0.01) at follow-up. The experimental condition was found to be more than seven times as likely to result in the increased use of condoms with casual partners. However, these results were based on a regression model that ignored any reported increases in risky behaviour (those reporting the same level or increased levels of risky behaviour were coded 0).

Dickson-Gomez 2006^91,92 presents a comparison of pre- and postintervention self-reported data on risk behaviours. The experimental condition was found to report a greater decrease of the number of casual sexual partners (χ² = 3.33, p = 0.05), and in multiple logistic regression analysis the experimental condition was found to be more than seven times as likely to report increase use of condoms with casual partners.

In Dickson-Gomez 2006^91,92 the programme had a positive impact on many LAs who sought to reduce their drug consumption or stop all together (p ≤ 0.001). Outreach work provided them with an alternative means of engaging with other drug-using community members. LAs also gained a greater sense of self in doing something useful for their community; many saw outreach work as a first step towards employment and a stable housing arrangement. LAs reported increased usage of condoms (p = 0.000), a reduction in the number of sex partners (p ≤ 0.001), increases in cooking of drug solutions (p = 0.007), use of rubber tips among crack users (p ≤ 0.001), and stopping sharing cookers/drug solutions (p = 0.35). A total of 21.3% of LAs reported having entered a drug treatment programme in the 2 months prior to the closing interview.

Participation

In Dickson-Gomez 2003^89,90 participants' attitudes towards outreach were examined; given that the intervention put emphasis on social belonging, responsibility and participation, engagement in outreach activities can be taken as an approximate of social participation. At 6 months' follow-up, participants in the experimental condition were significantly more likely to report talking about HIV with family members (χ² = 6.42, p < 0.05), sex partners (χ² = 6.7, p < 0.05), non-drug users (χ² = 3.92, p < 0.05), and drug users (χ² = 5.32, p < 0.05). In Dickson-Gomez 2003^89,90 there were no statistically significant differences between the experimental and control groups in the outreach self-efficacy score [t(219) = 1.10, p = 0.27). In Dickson-Gomez 2006^91,92 many LAs reported in closing interviews that they had engaged in other activities – for example, independent community action, such as volunteering in homeless shelters or soup kitchens – and working with youth and pastors in their neighbourhood.

Health-care beliefs and knowledge

In Dickson-Gomez 2003,^89,90 99% of experimental condition participants declared themselves proud to be LAs; 94% thought that they gained respect by doing outreach and were glad to show that they were doing something positive; and 95% and 94% of participants reported that their family and friends were supportive of their outreach respectively.

Health-care use

No study in this category assessed health-care use.

Costs

Although Dickson-Gomez 2003^89,90 and 2006^91,92 reported some of the costs incurred by the study (payment for training attendance, for example), neither of the two studies reported any costs for running an HIV infection prevention programme.^89-92

Power calculations are not appropriate given the nature of the above two studies.

Practitioner type

See Box 16.

BOX 16

Role/training.

None of the studies included described a professionally driven intervention. Equally, none distinguished between peer and lay roles. Table 15 describes three kinds of peer roles: peer with common personal experience; peer with a shared community; peer with both a common experience and community; and not a peer.

TABLE 15

Practitioner type.

It is of note that studies that tended to use peers with a common cultural/socioeconomic background were more often conducting general health promotion activities (Table 15). The detail of peership is unclear in Gary 2003^98–100 as although the LA is described as ‘local’; the details of this locality (geographical or cultural) or their gender (75% of the participants were female) or life experience with regards to diabetes are not given.

In Dickson-Gomez 2003^89,90 LAs were older, previous or current drug-using African-Americans, targeting younger people often involved in selling drugs. The study highlights how intervention by men could lead to a struggle for respect, whereas there was more chance of a successful outreach encounter when the LA was more mature woman who could be perceived as a mother figure.

In Earp 2002^16,63,93,94 women who were interviewed about their experiences of interactions with LAs indicated that ‘the LAs own mammography behaviour did not influence whether they listened to the LAs' advice or decided to get mammograms’, perhaps questioning the need for health-related LAs to have common personal experience. In Earp 2002,^16,63,93,94 however, those who had received counselling from LAs did assign credibility to the LAs for having had personal or professional experience of breast cancer.

Although in Resnicow 2004¹¹⁸ efforts were made to recruit LAs with a college degree or graduate-level education, and a background in a helping profession, they were classified as peers with a shared community as, in common with the study participants, they were African-American churchgoers. Where possible, advisors in Ireys 2001¹⁰³ were also in close geographical proximity to the participants.

Interventions using peers with a shared community tended to be the most successful.

Level of training

See Box 17.

BOX 17

Level of training.

LAs' qualification status prior to the study was not always described, and most LAs were paid a fee for their participation in the study, but this could not be described as a salary. However, some level of training was most often described and is categorised in Table 16:

TABLE 16

Level of practitioner training.

• No training
• Moderate technical (health-related) training If training was < 10 hours overall, or if the training was purely related to intervention delivery or communication skills (as opposed to more in-depth knowledge about health or disease)
• Intensive training If it was 10 hours or more, and focused on technical health/disease related issues
• Not described
• Professionally trained If the health advisor had had previous professional training (as a nurse for example) or if they had a minimum of 1 year's practice experience in a field directly relevant to the intervention.

The level of LA training was only partly related to their experience, as both people who were not peers^{116,117,121–123} and people who both had a personal experience and a cultural/socioeconomic background in common with study participants^{91,92,108,109,114,115} received intensive training before the start of the intervention. This is true too of their professional background, as, although the LAs were provided with a moderate technical training, they had previously been professionally trained in Resnicow 2004.¹¹⁸ Of note is Paskett 2006^116,117 in which some LAs were professionally qualified (as a nurse and social worker), but nevertheless received intensive training in order to enable them to increase awareness of the importance of mammogram screening and increase the uptake of it.

May 2006,¹¹³ Staten 2004¹¹⁹ and West 1998³⁰ are the only studies in which LAs received no training, as they were ‘buddies’ in a smoking cessation intervention, and attempting to stop smoking themselves in May 2006¹¹³ and West 1998,³⁰ and women previously trained as CHWs in Staten 2004.¹¹⁹ The training in Elder 2006^95,96 is different from most of the other studies, as the sessions were based on informal discussions between those training to become promotores.

Interventions that used moderate or no technical training tended to be the most successful.

Introduction

Three of the five studies reviewed are UK based, and Kennedy 2007^104–107 evaluates the Expert Patients Programme, which has been implemented across the UK. Griffiths 2005^101,102 and Kennedy 2007^104–107 provide costs and all three UK studies provide outcomes measured with EQ-5D. In addition, a sister publication to Kennedy 2007^104–107 provides a robust cost-effectiveness analysis.

Assessing evidence of effectiveness

All of the studies provide evidence of significant improvements in patient self-efficacy and self-care behaviour. In addition, there is evidence of an impact of the intervention on participants' perceptions of their conditions. Griffiths 2005^101,102 and Barlow 2000⁸³ demonstrate improvements in anxiety and depression using HADS, although these changes were not statistically significant in Griffiths 2005,^101,102 Kennedy 2007^104–107 and Lorig 2003¹¹¹ find significant improvements in psychological well-being and health distress attributable to the intervention. The evidence of an impact on physical health is mixed. All three UK studies applied the EQ-5D, although in Barlow 2000⁸³ this was limited to a subset of the participants. Only Kennedy 2007^104–107 observed a difference that was statistically significant, after allowing for baseline characteristics, in favour of the intervention. Only Lorig 2003¹¹¹ finds a significant reduction in pain.

Evidence is limited on whether health improvements are maintained following the intervention. Barlow 2000⁸³ and Lorig 2003¹¹¹ applied outcome measurements after the control group received the intervention, and Lorig 2003¹¹¹ demonstrates that improvements in the intervention group are maintained. Barlow 2000⁸³ presents plots of several outcome measures for the intervention group at 4 and 12 months, and for the control group at 4 months. Results at 12 months suggest a slight deterioration in the improvements observed 8 months after the intervention. The controls in this study showed improvements in the outcomes measured at 4 months, albeit not as great as in the intervention arm, and may have continued to improve at 12 months without the intervention. Hence it is possible that the additional benefits from the intervention are short term. The intervention may have accelerated the acquisition of long-term disease management skills that would have been acquired through experience over time.

Reviews of professionally-led Expert Patients Programmes concur with these findings. In their analysis of self-management patient education programmes, Warsi et al.²²⁰ found modest improvements in clinical outcomes, although there was evidence of publication bias. Chodosh et al.²²¹ report similar findings and suggest that the modest benefits observed derive from increased medication compliance.

Hence findings that the Expert Patients Programme improves self-efficacy and symptom management appear uncontroversial. The evidence for an improvement in HRQoL is weak. The findings by Kennedy 2007^104–107 were not replicated in Griffiths 2005^101,102 or Barlow 2000,⁸³ and are not supported by the literature. It is possible that Expert Patients Programmes provide a very small improvement in HRQoL.

Evidence from the studies of a reduction in health-care utilisation is inconsistent. Kennedy 2007^104–107 examined a comprehensive range of health-care utilisation and reports reductions in both primary and secondary care. Analysed by category, none of the differences are statistically significant, but the reduction in inpatient days in the intervention arm is sufficient to offset the cost of the programme. Griffiths 2005^101,102 and Barlow 2000⁸³ examined only primary care contact, and found no evidence of a reduction in health-care utilisation attributable to the intervention. Lorig 1999¹¹⁰ found a reduction in hospital stay but no reduction in primary care. Lorig 2003¹¹¹ found a reduction in physician and emergency room visits attributable to the intervention, but no change in hospital stay. These results may reflect the diverse morbidities of participants in these studies, with considerable heterogeneity in resource use.

Evidence from the literature on costs of profession-led Expert Patients Programmes is mixed. A number of studies, including evidence from the CDSM programme²²² and evaluations of self-care programmes in CVD,²²³ and asthma,²²⁴ have suggested that patient self-management programmes are cost saving. However, in their review of the cost-effectiveness of interventions to support self-care, Richardson et al.²²⁵ conclude that most are methodologically flawed or limited in scope. They cite evidence from the UK in which only one out of six studies found evidence of cost-effectiveness.

Estimating the cost-effectiveness

Richardson et al.¹⁰⁷ undertake a cost-effectiveness analysis of the trial results reported by Kennedy 2007.^104–10 The authors used bootstrapped samples of the trial data²²⁶ to produce a cost-effectiveness acceptability curve²²⁷ and conclude that there is a 94% probability the intervention is cost-effective at a threshold of £20,000 per QALY. While this analysis appears to be robust, some caution needs to be exercised in interpreting the findings. The majority of cost savings observed in the intervention arm derive from a reduction in length of hospital stay. We might expect a reduction in primary care contacts rather than hospitalisations following improvements in self-efficacy and symptom awareness. The possibility remains that the cost differences observed by Kennedy 2007^104–107 were driven by a few resource-intensive patients who may not be truly representative of their populations.²²⁸ This possibility is supported by examination of the baseline characteristics of participants in Lorig 2003¹¹¹ Despite random assignment of 443 participants, those in the intervention arm report more than twice the number of hospital days in the previous 4 months compared with the controls. Nevertheless, it is quite possible that improved disease management results in reduced health-care utilisation that entirely offsets the small costs of these programmes.

Discussion

Expert Patients Programmes offer the possibility of combining patient empowerment with long-term savings for the NHS. Per-patient costs are fairly small; Kennedy 2007^104–10 uses estimates from the Department of Health of £250 per patient. While direct evidence of cost savings is mixed there is evidence that patient self-management courses can lead to measurable improvements in clinical indicators of disease control for diabetes²²⁹ and CVD.²²³ The potential for cost savings from improved disease control in these two areas is likely to be considerable.^230,231 In areas such as arthritis management the scope for savings may be small. While it is tempting to conclude that, overall, these programmes lead to small reductions in resource use that offset their cost, there is insufficient evidence to conclude that the costs of the Expert Patients Programme are offset by savings across all major chronic disease areas.

It is unclear whether improvements in self-efficacy and symptom management translate into gains in HRQoL. However, the impact of chronic diseases on HRQoL is likely to be considerable,²³² and Expert Patients Programmes may provide support, reassurance and coping strategies that are valued by participants, particularly those without extended networks of support in the community. For these patients the value of the programme may well outweigh the cost before any considerations of long-term cost savings. Further research on Expert Patients Programmes might consider ways to capture the value participants place on their experience of these programmes.

Introduction

All three interventions use lay-led counselling to improve lifestyle and disease management in poor, urban populations with type 2 diabetes. The impact of each intervention is measured by assessing the level of the glycated haemoglobin marker HbA_1c, a well-recognised marker of diabetes control.^233,234 Although the population in Young 2005^121-123 is a little older than that of the American studies, baseline HbA_1c levels are similar in each study (7.9%–8.6%). For the purposes of the economic analysis we assume that the target populations are the same. The three interventions report different methods of delivering lay-led lifestyle and disease management advice to disadvantaged patients with diabetes. The telephone-led intervention described by Young 2005^121–123 appears to be less resource intensive than the face-to-face interventions. An economic analysis of the intervention in Young 2005^121–123 has been published¹²³ and provides data on costs. The economic analysis presented here will consider all three interventions as alternative programmes to promote diabetes disease management in marginalised urban populations.

Assessing evidence of effectiveness

Effect sizes based on differences between intervention and control arms will be applied for the modelling of interventions based on Gary 2003^98–100 and Young 2005^121–123 As noted earlier, regression to the mean may have exaggerated the upwards trend observed in the control arm in Lujan 2007¹¹² Regression to the mean would act to exaggerate the treatment effect in the intervention arm, but this might be offset by an expected rise of 0.1% in HbA_1c level over 6 months.²³⁵ Hence, the effect size for the intervention was taken as the absolute fall observed in the treatment arm over the 6-month period.

Estimating the health gain from changes in effectiveness

Two large trials in the UK²³⁵ and USA²³⁶ have examined the long-term impact of control of HbA_1c level. The United Kingdom Prospective Diabetes Study (UKPDS) ran from 1977 to 1991 and examined the benefits of intensive blood glucose control in patients with type 2 diabetes. It also contained a nested trial examining the impact of lower blood pressure. Intensive drug treatment initiated a rapid fall in HbA_1c level of 0.9%. This difference between intervention and control was maintained over the 10-year observation period, although HbA_1c level steadily increased in both arms with time. The trial reported a 25% risk reduction in microvascular complications and a 16% (non-significant) reduction in coronary heart disease (CHD) events in the intervention arm.

A number of models of diabetes have been published,^237–244 many utilising the UKPDS data. The Centers for Disease Control and Prevention (CDC) model²⁴¹ examined the cost-effectiveness of intensive drug treatment in a typical cohort of newly diagnosed type 2 diabetes patients. In the base-case analysis, a reduction in HbA_1c level from 7.9% to 7.0% yielded a lifetime QALY gain of 0.192 (discounted at 3%). The base case ignored any impact of lowered HbA_1c level on CHD. Including CHD events increased the QALY gain to 0.333. Based on typical US practice, intensive drug treatment to reduce HbA_1c level increased overall costs by US$7927 (‘1997’ US$). However, under UK management style, costs were US$1309 lower in the intervention arm. The UKPDS Outcomes Model²⁴² examined the same data and predicted a lifetime gain of 0.27 QALYs for a 0.9% reduction in HbA_1c level. Bagust et al.^237,238 sought to examine the impact of improved HbA_1c level control for health providers and their cost estimate (£2026 increase) explicitly includes the indirect medical costs arising from prolonged longevity.

Two groups have modelled the impact of lowering HbA_1c levels. The CORE²⁴⁰ model (Table 21) uses data from the National Health and Nutrition Examination Survey (NHANES)²⁴⁵ and the INITIATE²⁴⁶ study. The ‘typical’ patient in the model is 59 years old and has had diabetes for 12 years; hence the model simulates the effects of an intervention to reduce HbA_1c levels in the existing diabetes population rather than from diagnosis. Costs and outcomes for three scenarios are reported: lowering HbA_1c level from 9.5% to 8.0%, from 8.0% to 7.0% and from 7.0% to 6.5%. A similar modelling exercise using the DiabForecaster model²³⁹ (Table 22) provides health outcomes and costs from a UK perspective for 1% reductions in HbA_1c level over the range 6%–11%. The results are tabulated below.

TABLE 21

The CORE model predictions for HbA_1c level change.

TABLE 22

The DiabForecaster predictions for HbA_1c level change.

Hence, estimates of the health gain of a reduction in HbA_1c level from 8% to 7% from the CDC, UKPDS, CORE and DiabForecaster models fall in the range of 0.3–0.4 QALYs. The health gain from the CORE model (0.38 QALYs), which is slightly lower than that from DiabForecaster, will be assumed for a 1% reduction in HbA_1c level. The DiabForecaster model estimates of cost, which exclude specific drug treatment costs and indirect medical care costs, will be applied. The study suggests a £600 saving for a fall from 8% to 7% in HbA_1c level. The cost year is unclear in the report but appears to be 2004, hence this value was inflated to pounds sterling in 2008 (£686).

Costing the interventions

Mason et al.¹²³ provides detailed costs of the telephone counselling intervention. First-year costs are £93,700, including one-off commissioning costs of £9000 (‘2003’ GBP). Long-term running costs inflated to pounds sterling in 2008 are £101,800, giving a cost per participant of £258. The intervention supported the 394 patients randomised to the centre but could have supported 600. If we assume the centre operates at 90% capacity (540 patients) then the cost per patient would be £189.

Cost data from Gary 2003^98–100 and Lujan 2007¹¹² are limited. The intervention in Gary 2003^98–100 appears to have employed a full-time nurse and a CHW for the 2-year duration of the intervention, with the CHW spending half of his/her time with the participants in the CHW arm. A yearly cost of £31,043 was assumed for the CHW based on the cost of a social work assistant,²¹⁹ giving a per-participant cost of £757. Lujan 2007¹¹² reports that two promotores provided the 3-month intervention to four groups. It seems unlikely that these ran concurrently, hence it assumed that the promotores were employed for 1 year. In addition, transport was provided for the participants. Applying the same cost estimated for the CHW to the promotores gives estimated staff costs of £62,086 for the year. It was assumed that 50% of the participants utilised the transport with costs of £20 per trip, giving total transport costs of £12,000. Hence the overall cost was estimated at £74,086 (£988 per participant).

Consideration of relapse rates

The three studies report HbA_1c levels over different time periods. The long-term effectiveness of the interventions in sustaining reduced HbA_1c levels is unclear. The economic analysis in Mason et al.¹²³ applied best- and worst-case scenarios and a ‘best guess’. The best-case scenario assumes that the intervention effects last for the life of the participants. The worst-case scenario assumes that they last for only the duration of the intervention. The ‘best-guess’ scenario assumed that maintaining reduced levels of HbA_1c required 50% of the intervention costs in each subsequent year. The model based estimates assume that the changes in physiology are maintained for the patient's life.

In principle, each of the interventions could be repeated yearly (or biannually for Gary 2003^98–100). In practice that might be overkill. We assumed that repeating the intervention at 3, 6 and 10 years ensures that 50% of participants maintain the behaviour change. This is in line with the estimates that 50% of those who quit smoking subsequently avoid relapse. The benefit of the interventions for the other 50% of participants was assumed to be zero. The base-case analysis applied the trial costs reported in Mason et al.¹²³ to the telemedicine intervention (first-year costs and 394 participants). Sensitivity analysis explored the impact of changing these assumptions by varying the proportion of those who relapse between 25% and 75%. Further sensitivity analysis explored the impact of applying long-term costs by ignoring set-up costs and assuming 90% capacity for the telemedicine intervention, and halving the per-participant costs calculated for Gary 2003^98–100 and Lujan 2007.¹¹²

Estimating the cost-effectiveness

Cost-effectiveness estimates for each intervention are presented in Table 23. Costs are calculated assuming that the intervention is repeated at year 3, 6 and 10 (discounted at 3.5%). The assumption that 50% of participants relapse was implemented by halving the estimated benefit (0.38 QALYs) and costs avoided (£686) for the reported changes in HbA_1c level. The benefits and costs of lowering HbA_1c level were assumed to vary linearly with the magnitude of the change.

TABLE 23

Cost-effectiveness calculations for the Chronic Care diabetes interventions.

In the base-case analysis the telephone-based counselling intervention, Young 2005^121–123 is cost-effective if decision-makers apply a threshold of £30,000 per QALY. The CHW intervention in Gary 2003^98–100 is more expensive per participant than the telephone counselling intervention, and less effective in reducing HbA_1c level. It is dominated, and consequently not effective. The ICER for the promotora intervention¹¹² suggests that it is highly unlikely to be cost-effective at a threshold of £30,000. These conclusions are robust to the sensitivity analysis applied here. Assuming that only 25% of participants subsequently relapse and return to previous HbA_1c levels, or halving the estimated costs for the promotora intervention, fails to bring the ICER for Gary 2003^98–100 below £30,000. If 75% of participants relapse (i.e. only 25% of participants benefit from the programme) the ICER for the telephone intervention is just below £30,000.

Costs saved by the intervention have little impact on the overall costs; hence, conclusions on cost-effectiveness rest on the estimates of the health gain from the interventions. The benefit estimate we used is likely to be conservative as it is estimated purely on the change in HbA_1c levels. In reality, lifestyle improvements by the participants may have resulted in falls in blood pressure and cholesterol levels, too. Nevertheless, it is prudent to consider how the results change when the estimate of the health gain from reducing HbA_1c level is varied. The base-case analysis assumes that only 50% of participants benefit. This is equivalent to assuming that 100% of the participants gain half of the estimated health gain (0.19 QALYs). The sensitivity analysis varies the number of patients who relapse (and gain no health improvement) between 25% and 75%. Assuming that 75% of participants relapse reduces the benefit gained from the intervention by 50% compared with the base case. Assuming that 25% of participants relapse increases the benefit gained from the intervention by 50%. Hence the sensitivity analysis where the relapse rate is varied between 25% and 75% is equivalent to varying the benefit of HbA_1c level control by ± 50% (0.19–0.57 QALYs).

Economic analysis in Mason et al.¹²³

Mason et al.¹²³ provided an evaluation of the intervention in Young 2005^121–123 using the framework elaborated in the introduction (Equation 1). Estimates of the underlying cost-effectiveness of reducing HbA_1c (ΔCE_t, US$7927/0.1915 QALYs = US$41,400) and the health gain (Δb_t, 0.1915 QALYs) were taken from the CDC model.²⁴¹ Mason et al.¹²³ converted the ICER to UK pounds sterling (£26,900 per QALY) to estimate ΔCE_t. The programme costs Δc_i were calculated under the assumption that 50% of the costs were required on an ongoing basis per participant to maintain adherence and discounted at 5%. The resulting ‘loading factor’ for the programme of £16,500 per QALY was added to ΔCE_t (£26,900) to generate an ICER for the telemedicine programme of £43,400 per QALY. The authors conclude that the programme is unlikely to be cost-effective.

As noted earlier, the estimate of ΔCE_t from the CDC model is based on US costs and the authors note that under a UK cost scenario the cost of intensive drug management is less than the costs of complications averted. This would give a negative value for ΔCE_t in a UK setting, but a value of zero might be a reasonable, conservative assumption. If we apply a value of zero for ΔCE_t then the cost-effectiveness of the telemedicine intervention is simply the loading factor £16,500 (as estimated in Mason et al.¹²²). This would indicate that the intervention in Young 2005^121–123 is cost-effective in a UK setting, although this rests on a series of assumptions about the long-term cost and effectiveness of the programme required to calculate the loading factor. An ICER of £16,500 is close to our calculation.

Discussion

This analysis suggests that the telephone intervention described by Young 2005^121–123 is cost-effective at a threshold of £30,000 per QALY. That conclusion rests on a number of assumptions, chiefly that the intervention needs to be repeated four times over a typical participant's lifetime, and that doing this ensures that 50% of participants maintain improved control of their HbA_1c levels. Conclusions are sensitive to the assumptions on relapse rates. It would appear feasible to reapply the telephone counselling intervention in subsequent years to maintain and reinforce behavioural change. In practice, a low-intensity telephone contact might be maintained with each participant after the initial intervention. The same assumptions have been applied to Gary 2003^98–100 and Lujan 2007¹¹² and these interventions are not cost-effective at a threshold of £30,000 per QALY. The trial data suggest that the promotora intervention in Lujan et al.¹¹² is more effective, but far more expensive. The reduction in HbA_1c levels does not justify the additional resources.

Estimating the number of HIV cases avoided

Translating risk reduction behaviours into health gains requires an estimate of the number of infections avoided. Estimates of infections avoided are usually based on a Bernoulli process model of transmission, where probability of infection is a function of the number of unsafe acts, the risk of transmission from an unsafe act and the general prevalence of disease.³⁰⁶ Application of trial data on risk behaviour is combined with literature estimates of the risk of transmission and survey data on the prevalence of disease, to estimate the number of infections averted through the reported reduction in risk behaviour. A conservative assumption that the observed reduction in risk behaviour occurs only for the duration of the intervention is generally applied, but the analysis assumes that those protected from infection do not subsequently become infected.

A simplified Bernoulli model is presented in Equation 2. The equation estimates the probability of infection from sharing injection equipment, assuming that shared injection equipment has previously been used by one other user. Using this equation we can calculate the risk of infection and the number of infections in the absence and the presence of the programme. The difference represents the estimate of the number of infections averted by the programme.

US setting

where:

• π = HIV prevalence = 0.2^89,90
• α_d = risk of infection of needle used by seropositive user = 0.9³⁰⁷
• α_i = risk of infection from infected needle = 0.0067³⁰⁸
• n = number of incidences of needle sharing.

Without the programme, n = 12 and p = 0.0144, but with the programme n = 9 and p = 0.0108. Hence the programme reduces the probability of infection by 0.0144 − 0.0108 = 0.0036. For the 10 users who report a reduction in risk behaviour, the number of HIV cases averted is 0.036. Application of the control programme would have averted 0.0036 × 4 = 0.0144 cases. The additional gain from the programme is an additional 0.0216 cases averted.

UK setting

The prevalence of HIV in the UK is much lower than in the USA (around 0.13%³⁰⁹). A prevalence of around 4.0% is observed among injecting drug users in London, but outside London HIV prevalence among injection drug users in England is low (0.6% in 2007).³¹⁰ These values have a dramatic effect on the number of HIV cases averted by the programme (Table 28) and, consequently, its cost-effectiveness. Applying Equation 2 with the same parameter and programme estimates, but applying a background HIV rate (π) of 0.04 (London) and 0.006 (outside London), allows calculation of the potential HIV cases averted in a UK setting (Table 28).

TABLE 28

HIV cases averted by risk reduction behaviours in drug-using LHAs.