Patient recruitment is often reported as the most challenging step in conducting randomised trials (RTs) and data on effective recruitment strategies is lacking [1–3]. This is particularly the case for cluster randomised trials (CRTs), where intact social groups (for example, primary care clinics or schools), rather than individuals, are the unit of randomisation [4, 5]. Lower level evidence in the form of case studies describing recruitment successes and failures experienced in individual patient RTs [6–10] and CRTs which recruit patients to receive an intervention [11–13] exists. Results of these studies suggest that barriers such as time constraints, forgetting to recruit, and too few eligible patients influence recruitment rates [9, 14, 15]. However, to our knowledge, no studies have been undertaken to investigate factors associated with recruitment of patient participants for a particular subtype of CRT known as a professional-cluster trial [16]. In these trials, interventions are targeted at health professionals, and while patients are likely to be effected by any change in health professional behaviour, patients do not directly receive an intervention. The paucity of research for this subtype of CRT is perhaps unsurprising, since it is preferable to have separation in cluster trials between those receiving the intervention (and thus aware of their allocation), and those recruiting patient participants [17]. However, there are instances where due to privacy issues, and financial and feasibility constraints, it becomes difficult to have this separation. Different barriers to recruitment of patient participants in this type of trial may exist compared with other designs. Therefore, the aim of this study was to investigate factors that contribute to GPs' recruitment of patients in a professional-cluster trial.

Professional-cluster trials are frequently used in the field of implementation science to evaluate interventions designed to increase uptake of research into clinical practice [18, 19]. In these trials, patients may be recruited to provide measures of patient outcomes (for example, health outcomes), or measures of practitioner behaviour (for example, whether their GP provided a particular treatment), or both. The latter may occur in instances where it is not possible to measure practitioner behaviour through other sources (for example, clinical records or administrative data).

Recruitment of patient participants into professional-cluster trials can result in selection bias from the selective recruitment of patients by individuals who are often aware of intervention allocation (for example, Farrin 2005 [11]). Strategies to minimise the risk of this selection bias in professional-cluster trials have been recommended, such as identification of patients prior to treatment allocation, blinding of recruiters, and collection of routine data available in patient files [20, 21]. In Australia, collection of data from patient medical files can only be undertaken with the patient's consent for a particular study [22]. Further, health professionals cannot provide researchers with contact details of patients to invite them to give this consent without the patient's permission. This poses a number of challenges to trialists undertaking professional-cluster trials in Australian primary care settings, as the privacy and ethical responsibilities, resources available, and the nature of the clinical condition means it may only be possible to recruit patients through practitioners who are aware of their allocation status.

While it may be supposed that the factors that contribute to health professionals' patient recruitment rates in professional-cluster trials are similar to those in individual patient RTs, differences may also be expected. For example, in professional-cluster trials, patients are less likely to perceive they will directly gain, and so may be less interested in being involved, or their health professional may perceive this to be the case and be less motivated to approach them for participation [14, 23]. Determining whether this poses a barrier to patient recruitment is useful as it suggests that the way to promote participation may have to be more targeted for professional-cluster trials. The IMPLEMENT CRT [24] provided an opportunity to explore factors contributing to poor patient recruitment in a professional-cluster trial.

Recruitment strategies used in the IMPLEMENT CRT

Three recruitment strategies were used during the CRT. No financial incentives were offered to the GPs to recruit patients and GPs were not affiliated with the academic centre that conducted the study. The first strategy was designed to minimise potential selection bias from selective recruitment of patients and did not involve GPs identifying potential participants. Recruitment strategies two and three were more susceptible to selective recruitment since in these strategies the GPs were involved in identifying potential participants. These recruitment strategies, and the numbers of participants recruited, are outlined in Figure 1.

Nine months after patient recruitment began, a questionnaire was mailed to all GPs still participating in the IMPLEMENT CRT (94 GPs from 81 practices). Eleven practices (six from the intervention group and five from the control group) comprising 18 GPs dropped out of the study prior to the start of patient recruitment, so were not posted a questionnaire. GP reasons for withdrawing from the trial before patient recruitment included being too busy to continue, GP left the included practice and GP poor health. Non-responders to this questionnaire were contacted by mail reminders and by telephone.

This questionnaire included seven items to assess possible reasons for the poor recruitment of patients as perceived by the GPs (see Figure 2). The authors developed the questionnaire by creating an item for each of the hypothesised reasons for poor recruitment in the trial, and then piloted the items with the clinical investigators. One item asked GPs to recall the number of patients with acute LBP they invited to participate over the recruitment period. Five seven-point Likert-scale items were used to measure factors which were potentially predictive of recruitment (for example, intention to recruit and forgetting to recruit). An open-ended item was included for GPs to provide comments about the recruitment process and whether there was anything further they thought we could have done to assist them. Double data entry was used to enter the questionnaire data, and discrepancies were resolved via discussion with a third researcher. Descriptive statistics (means, medians, inter-quartile ranges (IQRs) and 95% confidence intervals (CIs)) were calculated for the quantitative items, based on the available data for each item. Data were analysed at the practice level to adjust for correlation that may occur between GPs within the same practice by calculating an average response for practices with more than one GP (9/68 practices).

Comments provided in response to the open-ended question that were interpreted as being sufficiently similar in theme were coded into categories developed by two researchers independently. Coding of comments provided in response to the open-ended question was done freely rather than based on a pre-developed framework. Two researchers independently extracted themes from each comment, for example, if a GP wrote "I am sorry but I just forgot to recruit patients for this study", a possible theme nominated could be 'Forgot to recruit patients'. The researchers listed more than one theme/category resulting from each comment if applicable. Any discrepancies in the themes assigned to each comment by the two researchers were resolved via discussion.

Ethical approval for this study was obtained from the Monash University Standing Committee on Ethics in Research involving Humans (2006/047).

The sample of respondents consisted of GPs with a mean age of 52.2 years (SD = 11.2), who practised on average 38.6 hours per week (SD = 13.4), 52 were males (66%) and 39 GPs were allocated to the intervention. When surveyed at baseline, the GPs reported consulting a median of eight patients with acute LBP per week (IQR 4-10).

General practitioners recalled inviting a median of two patients with acute LBP (interquartile range 0 - 5.5) to participate in the IMPLEMENT CRT, with a maximum of 20 patients. Twenty-six GPs (28%) reported inviting no patients. Across the 71 GPs who responded to this item, the total number of patients they recalled inviting during the recruitment period was 287.

The results of this survey provide some insight into factors that contributed to recruitment issues in the IMPLEMENT professional-cluster trial. The results suggest that few patients with acute LBP were invited to participate in the trial, despite GPs indicating at the beginning of the study that they saw a median of eight eligible patients per week on average. Despite intending to recruit patients, GPs reported forgetting to approach patients to participate, and did not perceive that patients had neither a strong interest nor disinterest in participating. Additional open-ended comments were coded into six categories that were generally consistent with the quantitative data.

Failure to recruit the target number of patients in trials in the primary care setting is common [25]. Frequently reported barriers in RTs and CRTs where patients receive an intervention include time constraints, forgetting to recruit, and few eligible patients treated [9, 14, 15]. Results from the current study suggest that similar recruitment barriers may be faced in professional-cluster trials. Our hypothesis that lack of an intervention being directly targeted at patient participants may pose a major barrier to GPs recruitment of patients was not supported by the data. With the large number of clinical tasks GPs conduct during a consultation, it is likely that even if patients were very interested in participating, and GPs perceived this to be the case, the chances of them having limited time to invite patients into the study, or forgetting to do this, were probably high. Given the identified barriers in this study were similar to those reported in individual patient RTs, interventions which are effective in enhancing patient recruitment in individual patient RTs may also be applicable in professional-cluster trials.

This study highlights an important challenge for trialists planning to recruit patients in professional-cluster trials undertaken in Australian primary care; namely, minimising the impact of the trial on practice staff time, maximizing the recruitment rate, and minimising the chance of selection bias. The use of advertisements in the waiting rooms during the first phase of recruitment in the IMPLEMENT CRT was designed to minimise this risk of selection bias arising from GPs recruiting patients and to reduce the burden on GPs and practice staff. On average, the GPs recalled seeing these advertisements in their practice, but such a passive strategy was not successful. The second two recruitment strategies had a high possibility of selective recruitment of patients, and in addition, required a greater time commitment from GPs and practice staff, and were also unsuccessful. Use of alternate strategies to minimise selection bias were not possible given available resources. To prevent bias from occurring in future professional-cluster trials, the ethics and feasibility of alternative methods of obtaining patient data other than direct approach of patients needs to be considered [19].

A brief questionnaire provided some insight into GPs perceptions about the reasons for the failure to recruit patients with acute LBP into a professional-cluster trial. The barriers to patient recruitment identified in this study were similar to those reported in trials which recruit patients to receive an intervention. Results indicate that while GPs intended to recruit patients, barriers such as time constraints and forgetting to recruit patients may have contributed to poor recruitment. To advance the evidence base for patient recruitment strategies in professional-cluster trials in the primary care setting, trialists need to develop and evaluate patient recruitment strategies that minimise the efforts required by practice staff to recruit patients, meet privacy and ethical responsibilities, while also minimising the risk of recruiting a selective sample of patients.