Increasing participation of cancer patients in randomised controlled trials: a systematic review

We searched fifteen databases for published and unpublished studies, with no language restrictions: MEDLINE, EMBASE, CINAHL, PsycINFO, Cochrane Database of Systematic Reviews, Cochrane Database of Methodology Reviews, Database of Abstracts of Reviews of Effects, Health Technology Assessment database, American Society of Clinical Oncology Website, Health Management Information Consortium, System for Information and Grey literature in Europe, ISI Science Citation Index, SI Social Science Citation Index, Sociological Abstracts, Applied Social Sciences Index and Abstracts. The search strategy combined groups of search terms representing cancer trials and patient participation. (See Additional file 1 for full details of the MEDLINE search strategy which was amended as necessary for the other databases searched.) We also searched the reference lists of all retrieved articles.

Two reviewers independently assessed titles and abstracts and full papers, where these were obtained. We included studies of any interventions to improve cancer patient participation in RCTs, which reported change in participation rates. The primary outcome of interest was patient participation. Therefore, for example, interventions aimed at increasing physician participation in trials, or interventions targeted at organisational change were eligible for inclusion provided the impact on patient participation rates was also assessed. The definition of participation varied between studies and the definition used by individual studies was accepted. We included controlled trials and also before and after studies, provided baseline participation rates were reported specific to the population being investigated. We contacted authors for clarification when it was unclear whether the intervention was directed at randomised or nonrandomised clinical trials.

One reviewer extracted data from included studies into structured summary tables and assessed the methodological quality of the studies. A second reviewer checked the data and quality assessment with disagreements resolved by discussion and consensus. We developed separate quality checklists to assess RCTs, for each relevant study design, based on CRD Report No.4.[10] We assessed whether measures had been taken by the study authors to avoid or minimise selection bias, attrition bias, performance bias and whether the study design protected against contamination between the intervention and the comparison. Studies were also assessed as to whether the nature of the intervention was clear and whether the target of the intervention was clearly defined. It was not appropriate to undertake a statistical synthesis as the included studies were very diverse; therefore we conducted a narrative synthesis.

We screened 3385 references and assessed 136 full papers. Eight studies met our inclusion criteria: three RCTs; two nonrandomised controlled prospective studies; two observational prospective studies with a comparison group and one before and after study (see Figure 1). The majority of studies were excluded because there was no intervention targeted at improving patient participation in cancer trials. Three studies were excluded because data on participation in randomised and non randomised trials were reported together and, following contact with the authors, no separate data were available for patient participation in RCTs. [11‐13] It was not possible to assess six papers: one had not been received and for five papers it was unclear whether the intervention was directed at an RCT. [14‐18]

Table 1 summarises the main characteristics of the eight included studies. We found three studies that had been conducted in the UK or Europe, [19‐21] one of which was an RCT.[19] Two of the UK studies were concerned with participation in the same cancer treatment trial.[19, 20] The remaining five studies were conducted in the US [22‐25] and Australia.[26] The majority of studies were concerned with some aspect of the consent process. In four of the seven studies of adults, the majority of participants were women. Four studies included predominantly white participants[22, 23, 25, 26] and this information was not reported in the remaining studies. The majority of studies included individuals with different cancers. Only two studies provided details of disease stage or severity and only for subgroups of patients.[22, 23] The included studies varied in quality. A summary of individual study quality is provided in Table 1 and the full quality appraisal for individual studies is available in the full report.[27] Care needs to be taken when comparing rates of trial participation between studies due to between study variation in how participation was defined. In four studies, trial participation was defined as the number of patients accrued or enrolled but it was not clear whether this referred to the proportion who agreed to randomisation or the proportion who actually accepted their allocation. [22‐25] One study that used both measures illustrated that there can be a considerable difference between these figures, with actual accrual being lower than consent to randomisation.[20] A further study based participation rates on questionnaires completed by patients, following their meeting with the doctor, stating whether they wanted to participate. This may have overestimated the number who actually started the trial.[21]

We synthesised two studies separately from the others as the interventions could not be implemented in the UK; these are only briefly summarised here. One was an RCT conducted almost 20 years ago in Australia which investigated a uniform policy of full disclosure of all relevant information when seeking patient consent to trial participation compared with disclosure of information at the discretion of the consultant[26] A written consent document was completed for the former condition whereas only verbal consent was obtained for the latter. In the UK setting, clinical trials regulations require full disclosure of information with written consent.[28] The second study investigated the effect of legislation requiring health insurers to cover clinical trial patient care costs on trial participation rates in the US.[25] Although the funding of trials is an important issue this study is only relevant in settings where health insurance is widespread.

Across the remaining six studies there was no evidence that any of the experimental interventions evaluated led to an increase in cancer patient participation in RCTs compared with the comparison intervention (see Table 2). A good quality RCT, conducted in a UK setting, did find that nurses and urologists were equally effective in recruiting men with prostate cancer to a treatment trial with a two and three-arm comparison though recruitment levels varied between the three centres (94%, 61% and 45%).[19] Based on a cost minimisation analysis, recruitment by nurses was more cost-effective. This finding was unchanged in six out of seven sensitivity analyses exploring different resource scenarios, though the size of the cost difference did change. An uncontrolled qualitative study involving patients and recruitment staff was also undertaken in relation to the same trial. This study found increased participation rates following amendments to the nature and emphasis of information provided to potential trial participants.[20] However, given that this was not a controlled study, the influence of other factors on the recruitment rates cannot be excluded.

The following interventions were not associated with an increase in trial participation: a two-stage process for seeking parental consent for their child's participation in a leukaemia trial compared to the standard approach;[24] a written consent document designed to be easy to read compared with the standard consent form; [22] providing doctors with information on patients' individual information needs and attitudes to trials prior to seeking consent compared with the doctor not having this information;[21] and a multi-component, system level intervention compared with no intervention.[23] However, the evidence is not sufficient to conclude that the interventions investigated are ineffective.

In most of the studies participation levels were high in both the intervention and the control group. Apart from one study with low participation levels, [23] participation rates in the control groups ranged from 68% to 88%. This raises the question of whether there was a Hawthorne effect i.e. that the experience of participation in a study per se led to an increase in participation in the cancer trial. This could have been sufficient to mask an effect of the experimental intervention, especially given the fairly small sample sizes in these studies. Alternatively, the particular cancer trials to which patients were being recruited in these studies may have been trials which were easy to recruit to and that would have had high recruitment levels anyway. This may have lead to a ceiling effect in individual trials. None of the interventions appeared to be in response to recruitment problems being encountered either with the specific trials where individual trials were being targeted or with the health professionals conducting the recruitment where a large number of different trials were involved. For two studies related to a trial for treatment of prostate cancer the trial was described as controversial and difficulties in recruitment were anticipated. [19, 20]

There is the possibility that the specific interventions investigated do not work in the particular contexts in which they were used. They may prove effective with a different patient group or in relation to a different trial. For example, if the effect on participation levels of an 'easy to read' informed consent form, as used in the study conducted by Coyne et al., [22] had been investigated with patients with a lower level of literacy than the women in the study, it may have been found to be effective.

Most of the studies had methodological weaknesses. Apart from one good quality RCT with appropriate randomisation and concealment of allocation, [19] the remaining studies were vulnerable to selection bias. Across the studies there seemed to be a risk of underestimating the effect of the interventions due to the possibility of contamination between the experimental and comparison intervention. Contamination refers to the situation where the control or comparison group receives part of the experimental intervention leading to unplanned similarities between the two conditions; for example, where the same clinicians are responsible for delivering the experimental intervention and comparison, knowledge of the experimental intervention may influence how the comparator is delivered. This can lead to a dilution of any effect of the experimental intervention. Apart from one included study that minimised the risk through study design, [22] there was a risk of contamination across all studies.

We searched for evidence from a wide range of sources on any interventions targeted at improving patient participation in randomised cancer treatment trials or interventions aimed at making the process easier or more efficient. A range of study designs were included. However, given the nature of the topic, no relevant indexing terms were available for any of the databases searched, and the search strategy was heavily reliant on textword searching. This meant that the searches were limited to the terms used by authors in the title and abstract fields of each reference. Because of this, there is always the possibility that studies may have been missed.

We focused on interventions to improve participation in trials involving cancer patients. Studies of interventions with other patient groups may provide useful information that might be transferable to cancer treatment trials. Therefore the review may have excluded studies of patients with other conditions that might have highlighted interventions worthy of further investigation with cancer patients.