Skip to main content

Archived Comments for: Improving patient adherence to lifestyle advice (IMPALA): a cluster-randomised controlled trial on the implementation of a nurse-led intervention for cardiovascular risk management in primary care (protocol)

Back to article

  1. Danger of recruitment bias

    David J Torgerson, York Trials Unit

    24 January 2008

    This cluster randomised trial is at risk of incurring recruitment bias. Because primary care staff are trained in the intervention before randomisation then those in the intervention group may, due to the training, recruit different kinds of patients compared with the control group, thus introducing selection bias. This is a widespread problem in cluster randomised trials (See Puffer S, Torgerson DJ, Watson J. (2003) Evidence for risk of bias in cluster randomised trials: a review of recent trials published in three general medical journals. British Medical Journal 327, 785 and Farrin A, Russell I, Torgerson DJ, Underwood M. (2005) Differential recruitment in a cluster randomized trial in primary care: the experience of the UK Back pain, Exercise, Active management and Manipulation (UK BEAM) feasibility study. Clinical Trials 2, 119-124). The authors should, ideally, identify the patients before randomising the practices to maintain allocation concealment or use independent people in both practices (blind to the allocation) to do the recruiting. Unless this is done, then we cannot be sure that the participants are similar in unknown or unmeasured covariates.

    Competing interests


  2. The tension between internal and external validity

    Marije Koelewijn - van Loon, Maastricht University, School for Public Health and Primary Care

    18 March 2008

    We highly appreciate the discussion about clustered randomised controlled trials and the risk of recruitment bias, as is mentioned by David J Torgerson.

    For clearness of the discussion: the practice nurses in our study are not trained before the randomisation. Practice nurses were allocated to intervention group (with extensive training) or control group (with short training) before the training actually took place. But, we acknowledge that the risk of recruitment bias still exists. This is a recurring problem in pragmatic field studies. We would like to say two things about this. First why we chose this design, and second, what we have done to minimise the risk of recruitment bias.

    We chose for a clustered randomised controlled trial design since the aim of the study is to examine the effect of a strategy for the implementation of a national guideline for cardiovascular risk management. It is therefore important that the external validity is high. The research setting has to be (nearly) similar to real clinical practice to maximise the change of success by implementation in a real practice setting (Eccles, M, Grimshaw, J, Campbell, M, Ramsay, C. (2002) Research designs for studies evaluating the effectiveness of change and improvement strategies. Qual Saf Health Care, 12, p47-52).

    To minimise the danger of post randomisation bias, we asked an independent statistician to perform the allocation of practices to intervention or control group. Secondly, the allocation was a central block randomisation, after stratification of the practices into four geographical regions. This was done to avoid that groups were not comparable on characteristics of practices on the level of the patient population, like social economic status or health status. Third, patients were not informed about an intervention or control group.

    David J Torgerson advised in his comment to identify the patients before randomising the practices into the two arms. We agree that this would have given more security about the comparability of the groups on patient level, but this was not possible. The intervention of this study is to our opinion innovative because it combines aspects of risk estimation, risk communication and adapted motivational interviewing. When a patient is eligible for a cardiovascular risk inventory, he or she will be invited for the study and will visit the practice nurse. The nurse will assess the cardiovascular risk and will inform the patient about his risk, and options for risk reduction will be discussed. These elements could not be performed separately. General practitioners could not identify potential high risk patients, but give them care (intervention or usual) weeks or months later, after the randomisation is executed and the training is given.

    Identification of patients by independent people was not preferred and not possible for practical reasons. Patients were identified during consultation hours with GP or practice nurse. The appointment of an independent GP or nurse in 20 practices would have been too expensive. But more important is that it would have threatened our external validity. The presence of an external person in this study would have decreases the generalisability to real practice. Especially in case of lifestyle change, where many aspects could have an influence of the outcome measure, adaptation of the normal situation has to be avoided as much as possible.

    Summarising, although we take the warning of recruitment bias seriously, we stick to this protocol due to the arguments given. We will prospectively register the characteristics of the patients per group and take potential differences into consideration in the analyses, by integrating such variables in multiple models for analyses.

    Competing interests