Trial flow

The flow of participants through the trial is detailed in Figure 3.

FIGURE 3

Flow of participants through the trial. GP, general practitioner.

Identification of patients (screening)

Patients were identified by clinicians working in the urology or accident and emergency departments of participating sites, who were supported by local clinical research teams. Approved trial publicity material in the form of posters was used to help alert staff that the trial was taking place at specific sites.

Recruitment process

Clinicians assessed patients presenting with suspected ureteric calculi in accordance with standard practice. A screening log was completed and included all patients assessed at participating sites to document the reasons for inclusion or non-inclusion in the trial. Following adequate pain relief and confirmation of a single ureteric stone by CT KUB, identified eligible patients were given a patient information leaflet (PIL; see Appendix 1) to inform them of the purpose and need for the trial as well as the uncertainties around the clinical usefulness of MET. The PIL was developed in conjunction with the BAUS Section of Endourology Patient Group. Following receipt of the PIL, a member of the local research team asked the patient if they were interested in the trial and ensured any questions that the patients had were answered appropriately. Further checking against eligibility criteria, particularly around the use of tamsulosin and nifedipine as MET, was performed by local research staff. When a patient was eligible and happy to take part in the trial they were asked to sign a trial consent form (see Appendix 2).

Randomisation and intervention allocation

Eligible and consenting participants were allocated using minimisation to one of the two intervention groups or the placebo group on a 1 : 1 : 1 basis using the telephone interactive voice response randomisation application hosted by the CHaRT, HSRU, at the University of Aberdeen. The minimisation algorithm used the trial centre (site), stone size (≤ 5 mm, > 5 mm) and stone location (upper, middle, lower ureter) as covariates.

Blinding

At randomisation, the participant was allocated a unique participant study number and assigned a numbered participant pack. The packs were provided by an independent supplier containing the overencapsulated trial medication to ensure that the participant, local investigator and trial personnel remained blinded to treatment allocation.

Unblinding

The treatment code was broken only in the case of a serious adverse event (SAE), when it was necessary for the principal investigator at site or treating health-care professional to know which intervention the participant was receiving to determine a management plan.

Each participant was given a card to carry with details of a contact telephone number at the site to be used in the event that unblinding was necessary. Contact information was also available in the participant’s hospital records. If unblinding was necessary, a member of the research team or a member of clinical staff at the local recruiting site telephoned the dedicated randomisation service at the CHaRT in Aberdeen on the number provided using the trial centre identification and the participant study number. In the unlikely event of the randomisation service not being able to field the query, the on-call pharmacist at Aberdeen Royal Infirmary was contacted and the same procedure followed.

Following any unblinding via the telephone randomisation service, automatic e-mails were sent to the chief investigator, trial manager and members of the CHaRT management team. If an on-call pharmacist performed the unblinding they would e-mail the same list of people. These e-mails did not contain the treatment code, and the trial team remained blinded as far as was practicable. The chief investigator then ascertained why unblinding had taken place. If the patient was unblinded because of a SAE this was then reported.

Interventions

The trial interventions were:

1. tamsulosin hydrochloride in the form of 400-µg MR capsules
2. nifedipine in the form of 30-mg MR capsules
3. placebo (lactose-filled capsules).

A summary of product characteristics for each of the investigational medicinal products (tamsulosin hydrochloride and nifedipine) used in the trial is included in Appendix 3.

All the medicinal products were overencapsulated to maintain the blinding of the trial. Trial medication was presented as capsules in amber plastic containers with a childproof closure and labelled in accordance with Annex 13 of Volume 4 of The Rules Governing Medicinal Products in the European Union: Good Manufacturing Practices.⁵⁷ All disguised drug packs were stored at site pharmacies under temperature-controlled conditions until dispensed to participants. The medicinal products and the placebo were overencapsulated, packaged and labelled by Tayside Pharmaceuticals, Ninewells Hospital, Dundee, UK, in accordance with Good Manufacturing Practice.

Participants were instructed to store the medication in accordance with the manufacturer’s instructions. Unused medication and/or empty packaging were returned to the site by the participant at the 4-week follow-up visit or returned directly to the pharmacy; alternatively, if participants did not attend the 4-week visit, they were instructed to dispose of surplus trial medication appropriately.

Data collection

Questionnaires were designed to obtain information on stone passage or further intervention, pain, HRQoL and resource use, including NHS and personal costs. Participants were asked to complete trial questionnaires at baseline, 4 weeks post randomisation and 12 weeks post randomisation. The baseline questionnaire was completed in hospital before randomisation. Further questionnaires were sent to each participant by post from the trial office (CHaRT, Aberdeen) with pre-paid envelopes at 4 weeks and 12 weeks post randomisation (see Appendix 4). If a participant did not return the questionnaire a reminder letter was sent out approximately 2 weeks later with a short form of the questionnaire containing the EQ-5D only (see Appendix 4).

In addition, CRFs were completed by the research team at the recruiting site at baseline and at the follow-up visit 4 weeks after randomisation (see Appendix 5). If the participant did not attend the follow-up visit, the CRF was completed from the participant’s health-care records. If participants indicated on their 12-week questionnaire that they had received further intervention for their stone since their 4-week questionnaire, or if they completed only a short form of the questionnaire, or if no 12-week questionnaire was returned, a further CRF was completed 12 weeks post randomisation from their health-care records (see Appendix 5). The outcome measures collected and their timings of measurement are described in Table 8.

TABLE 8

Source of timing of outcome measures

Safety reporting

Non-serious adverse events were not collected or reported. Planned hospital visits for conditions other than those associated with the ureteric stone were not collected or reported. Hospital admissions (planned or unplanned) associated with the treatment of the ureteric stone diagnosed at the time of entry to the trial were expected to occur for a proportion of participants. These were recorded as an outcome measure, but were not recorded or reported as SAEs.

All suspected SAEs were assessed in respect of severity, potential relationship to trial medication and whether they were expected or unexpected. Confirmed SAEs were reported to the Trial Office and then to the chief investigator and sponsor, who subsequently provided their assessment and action plan.

Participants who had left hospital were advised to contact their general practitioner (GP) should they experience an adverse event. This is current standard clinical practice for participants receiving tamsulosin or nifedipine within the NHS. As part of their notification that one of their patients was participating in the trial, GPs were asked to inform the trial office of any SAEs or reactions. This provided a robust system for the notification of any serious adverse reactions or SAEs occurring outside hospital research sites.

Change of status/withdrawal

The trial status of some participants changed during the trial for a number of reasons. These included post-randomisation exclusion, participant withdrawal and medical withdrawal. Participants were free to withdraw from the trial at any time without giving a reason. If a participant withdrew from receiving the trial questionnaires, permission was sought for the research team to continue to collect outcome data from their hospital records. In the event that a participant was told to stop taking trial medication by clinical or trial staff for any reason, the participant continued in the trial and was asked to complete the trial documents unless he or she did not wish to do so.

Data management

Clinical data were entered into the electronic SUSPEND database through the trial web portal (https://viis.abdn.ac.uk/HSRU/suspend/), together with data from participant questionnaires, by the research team working at each hospital site. Questionnaires returned by post to the trial office were entered into the database by the central research team. Staff in the trial office worked closely with the local research teams to ensure that the data were complete and accurate. All trial staff and the statistician responsible for analysing the data remained blinded to allocation until completion of the trial and locking of the database.

Data collected during the course of the research were kept strictly confidential and accessed only by members of the trial team. Participants’ details were stored on a secure database under the guidelines of the Data Protection Act 1998, including encryption of any identifiable data.⁵⁸ Participants were allocated an individual specific trial number and all data, other than personal data, were identified only by this unique study number.

A random 10% sample of all trial data was generated by the database for re-entry by the trial office to validate correct data entry input. Any discrepancies between original data entry and the re-entered data were reviewed against the original paper copy and incorrect entries corrected accordingly. An initial data entry error rate of > 5% would have triggered a requirement to re-enter the entire data set from that questionnaire. This was not found to be necessary.

Trial oversight committees

The trial Data Monitoring Committee (DMC) comprised three independent individuals who met initially at the beginning of the trial when terms of reference and other committee procedures were agreed. The DMC then met a subsequent four times during the course of the trial to monitor unblinded trial baseline and outcome data provided by the trial statistician and details of SAEs. The DMC reported any recommendations to the chairperson of the Trial Steering Committee (TSC).

The TSC was chaired by a clinician independent from the trial and consisted of two other independent members as well as the grant holders. The TSC met five times over the duration of the trial.

Patient and public involvement

A patient representative was involved in the study design and conduct, with input into production of the PIL and other trial documentation, and membership of both the trial management group and the TSC. The patient representative contributed to, and reviewed, the trial protocol and final report. Additionally, the PIL for the trial (see Appendix 1) was developed in conjunction with the BAUS Section of Endourology Patient Group.

Important changes to methods after trial commencement

Sample size and power calculation

We combined the data from two meta-analyses,^24,41 which suggested a RR of approximately 1.50 comparing MET (either alpha-blocker or calcium channel blocker) against ‘standard care’ as the primary outcome. These reviews indicated that the percentage of spontaneous stone passage was approximately 50% in control groups of included RCTs. Only three of the included RCTs directly compared a calcium channel blocker and an alpha-blocker, and these suggested that alpha-blockers were potentially superior to calcium channel blockers. From an analysis of data from Singh et al.⁴¹ and Hollingsworth et al.,²⁴ we estimated that proportions of stone passage in the alpha-blocker and calcium channel blocker groups were approximately 85% and 75%, respectively. The most conservative sample size was required to detect superiority between the two active treatments, and the trial was powered on this basis. To detect an increase of 10% in the primary outcome (spontaneous stone passage) from 75% in the calcium channel blocker group to 85% in the alpha-blocker group, with type I error rates of 5% and 90% power, required 354 participants per group (1062 in total); adjusting for 10% loss to follow-up inflated this to 400 per group. We aimed to recruit 1200 participants (randomising 400 to each of the three treatment groups: alpha-blocker, calcium channel blocker and placebo) to provide sufficient power (> 90%) for all other comparisons of interest, and allowing for an anticipated 10% loss to follow-up.

General methods

Treatment groups were described at baseline and follow-up using means [with standard deviations (SDs)], medians (with interquartile ranges) and numbers (with percentages) where relevant. Primary and secondary outcomes were compared using generalised linear models. Treatment effects were estimated from unadjusted and adjusted models. Adjusted models included the trial centre (random effect), stone size (≤ 5 mm, > 5 mm) and stone location (upper, middle or lower ureter). All estimates of treatment effect are presented with 95% CIs. The analysis strategy was by allocated group (intention to treat). Two a priori comparisons were considered for the primary trial analysis:

1. MET [an alpha-blocker (tamsulosin) or a calcium channel blocker (nifedipine)] versus placebo
2. an alpha-blocker (tamsulosin) versus a calcium channel blocker (nifedipine).

We also made two post-hoc comparisons between tamsulosin and placebo, and nifedipine versus placebo. All analyses were carried out using Stata^® 13 (StataCorp LP, College Station, TX, USA).

Primary outcome

The primary outcome was analysed using logistic regression. We summarised treatment effects as odds ratios (ORs) and absolute percentage differences, from both adjusted and unadjusted models and presented with 95% CIs. Subgroup analyses (appropriately analysed by testing treatment by subgroup interaction) explored the possible effect modification of stone size (≤ 5 mm or > 5 mm to 10 mm), location in ureter, (upper, mid or lower) and sex, all using stricter levels of statistical significance (99% CIs; p-value < 0.01).⁶¹ Subgroup analyses were also summarised visually using forest plots.⁶² There was no correction for multiple testing.⁶³ During the planning of the SUSPEND trial it was anticipated that there would be few or no missing primary outcome data (owing to the algorithm specifying the primary outcome) and the primary outcome was analysed using complete-case analysis. The pragmatic nature of the trial made assessing the adherence unreliable, and no attempt was made to incorporate any analysis of treatment received.

Secondary outcomes

The secondary outcomes were analysed in a similar manner to the primary outcome using the appropriate link functions. Quality-of-life data were analysed using a mixed model that allowed treatment effects to vary at each time point.

Timings and frequency of analysis

The DMC considered interim inspection of the data on four occasions during the trial. The committee met to review and consider data on outcome measures and SAEs after randomisation of 300, 600 and 900 participants had occurred. Having seen and considered these data, the DMC did not make any recommendations to alter the progression of the trial on any of the occasions on which they met.

Algorithm for primary outcome

The primary clinical outcome is spontaneous passage of ureteric stones at 4 weeks (defined as no further intervention required to facilitate stone passage at up to 4 weeks). The algorithm to create this outcome can be found in Appendix 6.

Missing data

Baseline data were collected prior to randomisation. Where baseline data were missing, no imputation was undertaken for the reporting of the baseline covariates of the trial cohort. If there were missing data for covariates that were used in the analyses of the trial outcomes, single imputation was performed using the guidelines set out in White and Thompson⁶⁴ (i.e. centre-specific means for continuous variables and an indicator for categorical variables). It was anticipated that the nature of the clinical condition and the algorithm to generate the final outcome would result in few cases of missing primary outcome data and, as such, no plan was made to impute missing primary outcome data. Participants with missing primary outcome data were excluded from analysis of the primary outcome. For other outcomes, participants were included where they provided data under a missing-at-random assumption. Sensitivity analyses were planned to follow guidelines laid out by White et al.⁶⁵ to assess the impact of any missing outcome data on quality-of-life data from patient questionnaires. The analysis of quality-of-life outcomes was repeated using multiple imputation models with predictions based on all baseline covariates collected. Results were combined across 10 imputed data sets. The robustness of the results was then tested using pattern mixture models, which imputed missing data across a range of potential values from minus half of the observed SD to plus half of the SD of the outcome being analysed.

Introduction

Given that the condition under study was anticipated to be a short-term resolving problem for patients and the NHS, the main planned economic analysis was a ‘within-trial’ economic evaluation using data collected during the 12 weeks of individual participation. The question addressed was: ‘What is the cost-effectiveness of medical expulsive therapy using either tamsulosin or nifedipine compared to no treatment (placebo)?’ The trial was set within the perspective of the NHS, although it included both the NHS costs as well as those health-care costs falling on the participants.

Measurement of resource utilisation

Resource use and costs were estimated for each participant. Resource data collected included the costs of the intervention drugs and simultaneous and consequent use of primary and secondary NHS services by participants. Personal health-care costs, such as purchase of medication, were also estimated.

At recruitment, data were collected on the intervention that the participants received, the diagnostic tests conducted and the medications prescribed at the admission. At 12 weeks post randomisation, participants were asked to provide information by questionnaire of their primary and secondary health-care service use. They were asked for details of medications purchased, the cost of these and whether or not they had any visits to non-NHS health-care providers.

The consequential use of health services was recorded prospectively for each participant in the trial. Resource utilisation data were based on responses to the participant questionnaires and the CRFs completed by the local research teams. The CRFs recorded information on non-protocol visits (protocol visits are those scheduled for the purposes of data collection), outpatient visits and readmissions relating to the use and consequences of drug treatment. Use of primary care services, such as prescription medications, and contacts with primary care practitioners (e.g. GPs and practice nurses) were collected via the health-care utilisation questions administered in the participant 12-week questionnaire. Details of the sources used to estimate resource utilisation are included in Table 9.

TABLE 9

National Health Service resource use during the 12 weeks of participation

Identification of unit costs

Unit costs were obtained from published sources such as the British National Formulary (BNF)⁴⁰ and NHS reference costs.⁶⁶ The unit cost data source year was 2012–13 and the currency was British pounds.

The cost of the trial intervention included the cost of the drug to which the participants were allocated, the costs of diagnostic tests performed to confirm ureteric stone and the cost of the medications or antibiotics prescribed at diagnosis. The unit costs of medications were obtained from the BNF⁴⁰ and the diagnostic tests costs were obtained from NHS reference costs.⁶⁶ The unit costs of medicines given on admission were assumed to be those of the most commonly used drugs. The unit cost of NSAIDs was that of diclofenac (50 mg) given as a tablet, and the cost of opiates was based on the cost of morphine (10-mg injection). Antibiotic costs were based on the unit cost of a 3-day course of ciprofloxacin (500 mg). The initial secondary care attendances prior to and at recruitment were not included as costs because they were considered to be the same across all trial groups. The unit cost for the diagnostic test received was based on the average NHS reference cost for a computerised tomography (CT) scan ordered by the urology department.

The costs of subsequent resource use comprised costs to both the primary (GP appointments) and secondary (outpatient appointments and admissions) NHS care services. Unit costs for GP visits were obtained from the Personal Social Services Research Unit costs of primary care.⁶⁷ Outpatient visit unit cost was based on the average NHS tariff for a urology department consultant-led outpatient appointment obtained from the reference costs.⁶⁶ A summary of the unit costs of the resources used is provided in Table 10.

TABLE 10

Unit costs and their sources

Unit costs of further active intervention for the ureteric stone were derived from costs associated with different urology Healthcare Resource Group codes as detailed in Table 10. For occasions when a participant received two interventions on the same day, unit cost use was defined as the average cost of treatment of urinary tract stone disease with intervention. For those that had an admission with no intervention, the cost of urinary stone disease without intervention was used. As the median stay in the urology department was 1 day, any extra admissions days were costed using the long-stay excess days tariff.⁶⁶

The participant resource use data and unit cost were combined for each of the primary and secondary NHS care services to give an estimate of the total health-care cost per participant, as well as the average cost for each identified resource and the average total cost for each group of the trial.

Participant costs

Participant costs comprised self-purchased health care, such as prescription costs (for participants who pay prescription charges), over-the-counter medications and visits to non-NHS health-care providers. Information about participant resource use was collected using the 12-week health-care utilisation questionnaire (see Appendix 4).

Health status

Health-related quality-of-life measures were collected at baseline, 4 weeks and 12 weeks by participant completion of the EQ-5D and the SF-36 questionnaires. The EQ-5D divides health status into five dimensions (mobility, self-care, usual activities, pain/discomfort and anxiety/depression). Each of these dimensions has three levels, so 243 possible health states exist. Responses on the participants’ EQ-5D questionnaires were transformed using a standard algorithm to produce a health-state utility at each time point for each participant. The utility scores obtained at baseline, 4 weeks and 12 weeks were used to estimate the mean QALY score for each group⁵² over the 12-week (approximately 0.25 years) period of observation.

Responses from the SF-36 questionnaire were also used as the basis of QALYs as a sensitivity analysis to validate the EQ-5D scores. They were mapped onto the existing Short Form questionnare-6 Dimensions (SF-6D) measure using a standard algorithm⁶⁸ to allow utility values to be estimated for each time point. These utility scores were transformed to QALYs using the methods described above to provide an alternative measure of QALYs for each participant.

Data analysis

Resource use, cost and QALY data were summarised and analysed using Stata 13. As data were collected over a short (12-week) period, discounting was not carried out. The main cost-effectiveness analysis reports the results of participants with complete data. All the difference estimates are presented with 95% CIs. Data reported as mean costs for both active treatment groups and the placebo group were derived for each item of resource use and then compared using unpaired Student’s t-test and linear regression adjusted for baseline values. The mean incremental costs were estimated using general linear models, with adjustment for minimisation variables [centre at which participant was recruited, stone size (≤ 5 mm, > 5 mm), stone location at diagnosis (lower, mid or upper ureter) and sex]. The general linear model allowed for heteroscedasticity by specifying a distributional family which reflects the relationship between mean and variance.⁶⁹ A modified Park’s test was conducted to identify the appropriate family, which identified a gamma family. This allows for the skewness of cost data and assumes that variance is proportional to the square of the means as appropriate. A link function needs to be identified for the general linear model to specify the relationship between the set of regressors and the conditional mean. The link test recognised the identity link as the appropriate link function. The identity link leaves the interpretation of the coefficients unchanged from that of the ordinary least squares, as the covariates act additively to the mean. The mean incremental QALYs were estimated using ordinary least squares and were adjusted for minimisation factors, as well as for the baseline EQ-5D score.

Incremental cost-effectiveness

Cost-effectiveness of the trial interventions from the perspective of the NHS during the period of observation was measured in terms of the number of participants needing further treatment within 12 weeks, and in terms of QALYs accrued by participants in each group at 12 weeks. The results are presented as point estimates of mean incremental costs, number of further treatments needed, QALYs, incremental cost per further treatment needed and incremental cost per QALY. Measures of variance for these outcomes required bootstrapping of the point estimates. Incremental cost-effectiveness data are presented by cost-effectiveness acceptability curves (CEACs). Forms of uncertainty (e.g. concerning the unit cost of a resource from the different centres) are addressed using deterministic sensitivity analysis. As the data were not normally distributed, non-parametric bootstrapping was used to generate 1000 estimates of mean costs and QALYs for each treatment group. CEACs were generated using these 1000 estimates, using the net monetary benefit (NMB) approach. The NMB associated with a given treatment option is given by the formula:

where effects are measured in QALYs and Rc is the ceiling ratio of willingness to pay (WTP) per QALY. Using this formula, the strategy with greatest NMB is identified for each of the 1000 bootstrapped replicates of the analysis, for different ceiling ratios of WTP per QALY. Plotting the proportion of bootstrap iterations favouring each treatment option (in terms of the NMB) against increasing WTP per QALY produces the CEAC for each treatment option. These curves graphically present the probability of each treatment strategy being considered optimal at different levels of WTP per QALY gained.

The degree of missing data for the variables used in the derivation of costs was very low, and the data that were missing were considered to be missing completely at random. However, the number of participants with completely missing data for EQ-5D scores at 4 weeks and 12 weeks used for the derivation of QALYs was high (available data: tamsulosin group = 164/383; nifedipine group = 165/383; and placebo group = 157/384). Several methods of imputation were used as described in the sensitivity analysis.

Sensitivity analysis

There are elements of uncertainty owing to the lack of available information; therefore, various sensitivity analyses were conducted to explore the importance of such uncertainties. One-way sensitivity analyses using extreme values were performed around the QALY estimates. As the base-case analyses were performed using participants with complete cases, multiple imputation was carried out using chained equations in Stata 13 to replace missing cost and EQ-5D variables with a plausible value in 20 imputed data sets.

There was uncertainty around the QALY estimates as they were derived using the EQ-5D. There was some uncertainty over whether or not the dimensions in the EQ-5D are sensitive enough to capture the loss in quality of life, particularly in reference to acute pain. Therefore, SF-36 responses were mapped on the SF-6D measure using the algorithm by Brazier et al.⁶⁸ to validate the estimate of utility value for each time point derived from the EQ-5D. These scores were used in the same way as the EQ-5D to provide an alternative measure of QALYs for each participant.

A modelling exercise had been planned to extrapolate the estimates of the cost–utility analysis to a longer time horizon than that considered by the trial. However, the decision was taken not to perform any further analysis as the trial data suggested that there were very few cases that had not resolved by the end of the 12-week trial period and there was no chance of recurrence of the same stone.