Scope of the Problem 108
Consequences of Poor Recruitment 108
Ethical Concerns 109
Great Expectations? 109
Recruitment Terminology 109
Keeping Tabs 111
Alternatives to the Traditional Randomised Trial 111
Single Randomised Consent 111
Double Randomised Consent 111
Partially Randomised Patient-Preference Trial 112
Cohort Multiple Randomised Controlled Trial 113
Special Populations: Reaching the Hard-to-Reach 114
Language and Cultural Barriers 114
What Works? 114
The Way Ahead 116
Recruitment to randomised controlled trials is a widespread problem. Poor recruitment can lead to inadequate power, nonrepresentative participants, abandonment of trials, and squandered resources. Most importantly, patients may suffer needlessly in the absence of trial results. Eligibility fractions and enrolment fractions describe the recruitment process. Several alternatives to traditional trials have been suggested to skirt recruitment challenges. These include single randomised consent (Zelen method), double randomised consent, partially randomised patient-preference trial, and cohort multiple randomised controlled trial (cmRCT). Special populations, such as the elderly or minorities, can be hard to reach; special efforts are often required. Cochrane systematic reviews indicate that four strategies improve recruitment: open-label designs (participants are not blinded to treatment), opt-out strategies (every potential participant is approached unless expressly declined), telephone contacts, and financial compensation. Use of social media, smartphone applications, and business models may facilitate recruitment in the future.
The greater the obstacle, the more glory in overcoming it. Molière
For researchers around the world, failing to accrue participants in a timely manner and in sufficient numbers remains a daunting challenge. As a result, a voluminous literature exists on strategies for overcoming this hurdle. This chapter describes the scope of the problem, identifies some of its adverse consequences, reviews several alternatives to traditional trials, summarises the evidence regarding what works to improve recruitment, and considers new approaches to the problem.
Scope of the Problem
Failure to reach the intended sample size frustrates many randomised controlled trials. Several surveys of registered trials have documented widespread difficulties. Among 114 trials funded by UK agencies, only 31% met their sample-size target, and 53% requested an extension to enable additional recruitment. About one-third of these trials modified the sample size, usually downward (86%). Regrettably, these extensions rarely translated into substantial improvements in enrolment. Most trials (63%) experienced problems with recruitment early on. In 11% of trials, recruitment was halted before the intended close of recruitment, usually due to slow enrolment.
An update to this UK study found modest improvement in recruitment. Fifty-five percent of publicly funded trials met their planned sample size, but nearly half (45%) had to request an extension. A compilation of all randomised trials published over 6 months in six journals documented that 21% of 133 trials failed to attain intended sample sizes.
A review of more than 6000 cardiovascular trials of all types confirmed that slow recruitment was the most common reason for premature termination of the study. This occurred in 11% of the trials. Trials with US federal funding, studies of dietary or behavioural interventions, and those without a comparison group had a lower risk of premature stopping.
Consequences of Poor Recruitment
Some researchers simply ‘throw in the towel’ and quit. A follow-up survey of randomised controlled trials in Canada, Switzerland, and Germany from 2000 to 2003 found that 25% were discontinued. Once again, poor recruitment was the most common reason. Among these, the median percentage of sample size achieved was only 41%. These discontinued trials were less likely to be published, which is itself unethical. Industry-sponsored research was less likely to be discontinued than were investigator-sponsored trials. Only a minority of investigators (38%) reported the trial discontinuation to the relevant institutional review board (IRB). Surveys from other countries have reported trial discontinuation rates ranging from 11% to 45%.
Poor recruitment has potential harms that are both methodological and ethical. Failure to reach the intended sample size often means less power than intended; the risk of a type II error (saying no difference exists when a difference exists) is increased. Stated alternatively, clinically important findings may be overlooked.
Although some allege that underpowered trials are unethical, we disagree ( Chapter 11 ). If trials are free of bias, then they can contribute important information, especially when the alternative is continuing ignorance about the safety and usefulness of therapy. ‘New’ is not a synonym for ‘improved’. For example, in a sample of publicly funded trials, only about half of new therapies proved better than existing treatments.
Randomised controlled trials inevitably involve trade-offs. Because of methodological rigour, trials have the best internal validity (avoiding bias) in the hierarchy of research types. However, because they include only screened and consented volunteers, the generalisability of results (external validity) may be a concern. Recruitment of representative participants, especially in special populations such as the elderly or visually impaired, is often challenging.
Eligibility criteria vary by the purpose of a trial. ‘Efficacy’ (also termed ‘explanatory’) trials aim to find whether a treatment can work in the best of circumstances. In contrast, ‘effectiveness trials’ (also called ‘pragmatic trials’) hope to determine whether a treatment will work in usual conditions. Less restrictive eligibility criteria are used for the latter, which hope to have good external validity.
Some trials impose exacting eligibility criteria that deter recruitment and diminish external validity. Taken to its extreme, this approach to trials can yield such eclectic participants that although the trial results are internally valid, they may not be capable of extrapolation to the population of interest. This weakness is common in pharmaceutical industry trials, which may exclude participants who are elderly, obese, or complicated by comorbidities. Alternatively, in surgical trials, the skill levels of participating surgeons may be so high that the results cannot be generalised to community surgeons.
In order to recruit participants, some trials have run-in periods before randomisation to identify those with poor compliance or poor retention, who are then excluded from the main study. Although this approach may improve the overall compliance or retention rate of the main study, the results may not relate to the general population, which includes many who struggle with treatment adherence and follow-up. A variation on this theme is ‘enrichment’, in which those who have previously responded well to therapy are preferentially recruited. This, too, undermines external validity because the participants no longer represent the general population.
Slow recruitment frustrates everyone involved. Those recruiting potential participants may get discouraged; funding agencies may be disinclined to provide future funding to research teams that are struggling.
Requests for supplemental funding to achieve target sample size may divert money that could be invested in other trials. If these other trials are more important, then this diversion has a net negative effect on health.
Slow accrual of participants can delay identification of potentially life-saving treatments. For example, slow recruitment of participants to a trial of streptokinase may have led to the preventable deaths of 10,000 patients in the United States.
Researchers routinely underestimate the difficulties recruiting to randomised controlled trials. Louis Lasagna, a pioneering clinical pharmacologist, commented on the challenges he faced in recruiting surgical patients to a trial of a single dose of an analgesic. More than 8000 surgical patients had operations on the surgery service during the recruitment phase; 100 were successfully enrolled. The unjustified optimism about recruitment rates over time subsequently became known informally as ‘Lasagna’s Law’. Other early trialists noted that recruitment takes much longer than anticipated. ‘Muench’s Third Law’ states that the number of potential participants who can be recruited should be divided by 10 to provide a more accurate estimate. In our experience (admittedly more favourable), we have used a rule of π: take the estimate of how long recruitment will require, then multiply by 3.14. For trials in developing countries, where logistical challenges are greater, the multiplier is 2π. These provide a ballpark estimate and help novice researchers with reality testing.
A randomised controlled trial involves only a tiny fraction of the target population. The sampled fraction of the target population gets progressively smaller during recruitment ( Fig. 10.1 ). In this Venn diagram, a subset (A) of the target population is identified as potential participants and approached about the study. A subset of these (B) is then screened and found to be eligible to take part. Yet another subset (C) agrees to participate.
Several terms describe this selection process. The proportion of potential participants (A) who is screened as eligible for the study (B) is called the ‘eligibility fraction’ ( Fig. 10.2 ). In Fig. 10.2 , 700 of 1000 potential participants approached were found eligible, for an eligibility fraction of 70%. The proportion of potential eligible participants (B) who subsequently agrees to enroll (C) is termed the ‘enrolment fraction’. In Fig. 10.2 , 500 of 700 medically eligible potential participants enrolled, for an enrolment fraction of 71%. The ‘recruitment fraction’ is the proportion of potential participants (A) who subsequently enroll (C). In Fig. 10.2 , 500 of 1000 potential participants enrolled, for an overall recruitment fraction of 50%.
A study of 172 randomised trial reports found that a minority provided sufficient detail to calculate these various fractions. The median number of persons needed to screen to enroll one participant was 1.8, with a range of 1 to 68. Stated alternatively, one trial enrolled every person screened, while another had to approach 68 to admit one to the trial. The eligibility fraction in those reporting data had a median of 65% (interquartile range 41–82). The enrolment fraction had a median of 93% (interquartile range 79–100). The overall recruitment fraction had a median of 54% (interquartile range 32–77). These data should be viewed with caution, because 20 reports claimed that they enrolled all eligible patients; this suggests that some reports deemed those who declined participation as ‘ineligible’. Better reporting of the recruitment process through use of the CONSORT flow diagram will help readers judge the external validity of trial results.
Recruitment progress should be monitored as the trial progresses. A graph of anticipated versus actual recruitment numbers over time alerts investigators to problems. Should recruitment lag, researchers need to take corrective actions (e.g., adding new study sites or allocating more participants to sites with fast accrual). Other approaches include easing eligibility criteria, simplifying the trial for participants and clinicians, bolstering funds for recruitment, or extending the recruitment phase. Onerous data-collection forms and ponderous informed-consent documents (some running 20 pages) are common problems needing remedy. Many IRBs do not realise that trial participants often do not read or understand these documents.
Alternatives to the Traditional Randomised Trial
Several alternative research designs have been advanced to skirt some of the challenges to recruitment; these vary widely in their ethical and scientific rigour.
Single randomised consent
Zelen ignited an ongoing controversy with his suggestion to obtain consent after, rather than before, randomisation to treatment. For example, patients would be randomised to a new versus a standard treatment; only those randomised to the new treatment would be told of the experiment and consent requested. Should they decline participation, they would receive the standard therapy instead ( Fig. 10.3 , top). Zelen claimed that any loss in statistical inefficiency (unimportant except with gross disparities in treatment group sizes) would be offset by easier recruitment. He correctly observed that trial participants prefer to know their treatment.
This single randomised consent approach met with stiff resistance for several reasons. First, informed consent was denied to half of the participants. Second, considerable contamination (participants randomised to the new treatment crossing over and receiving the standard treatment) would dilute the experimental treatment effect in an intention-to-treat analysis. This would increase the likelihood of a type II error, unless a compensatory increase in sample size was made. Crossover (also called ‘transfer’) rates of 10% to 36% have been observed in cancer trials. More recently, crossover rates of zero to 74% have been seen. Moreover, neither allocation concealment nor treatment blinding would be possible for clinicians.
Double randomised consent
The ethical defects of the Zelen design are insurmountable. Withholding relevant information from participants in research violates the ethical principles of autonomy and beneficence. To avoid this problem, double randomised consent was suggested ( Fig. 10.3 , bottom). With this design, patients are randomised to treatment, after which both treatment groups learn of the trial and are asked to provide informed consent for their allocated treatment. Those who decline the assigned treatment can cross over to the alternative, or perhaps, still other treatments. In the decades since its introduction, the Zelen approach has been used infrequently and generally inappropriately. We find little to recommend it.
Partially randomised patient-preference trial
In the mid-1980s, a hybrid approach was suggested to address both recruitment and external validity concerns. Initially termed ‘comprehensive cohort study’, this design combines a cohort study and a randomised controlled trial in one study. After recruitment and informed consent, participants are asked if they are willing to be randomised to treatment ( Fig. 10.4 , top). If so, they are randomised in usual fashion. If not, they choose their treatment (cohort study). Participants in the cohort study and randomised trial receive the same follow-up observations. This allows a direct comparison of cohort and randomised trial results. For example, in German studies of breast cancer therapy, only 35% of women were willing to be randomised to loss of a breast; the remainder chose therapy in conjunction with their physicians.