The external validity of published randomized controlled trials in primary care

We reviewed RCTs published in four primary care journals in the years 2000–2004: BMJ (primary care section), British Journal of General Practice, Family Practice and The Journal of Family Practice.

We collected the following data fields from each RCT:

• Citation details- title, journal citation; author contact email.

• Characteristics of RCT and funding source- individual, cluster based, factorial, cross-over RCT; number practices/patients recruited per arm of RCT; number and type of interventions; country and setting of RCT; funding source- public (NIH, NHS, MRC, Wellcome Trust, Charity), pharmaceutical, self funding.

• Features concerning patient population and recruitment process- inclusion/exclusion criteria in RCT; statement concerning target population; potential population; eligible population and participation; details concerning target population engagement; details concerning eligibility screening; details concerning enrolment of eligible population; statement whether incident cases or prevalent cases were recruited; how patient population was identified- during consultation, disease/practice register, other method; statement of how and who approached eligible cases – by family physician, researcher, other; statement of method of where and in what circumstances eligible population were enrolled- family physician during clinic, family physician by post, practice nurse, waiting room, other; control group event rate.

Two data extractors (RJ and ROJ) independently assessed and extracted relevant data from each published RCT. A Microsoft ACCESS^® database was used to collate relevant details and the completed dataset was imported into Stata^® statistical package for analysis. Disagreement was resolved involvement of a third author (TF) and a consensus decision reached. CM and AM analysed the dataset, independent of the two data extractors. Table 1 provides definitions of the RCT recruitment terminology. Eligibility, enrolment and recruitment fractions were calculated in the same way as Gross et al.[12]

We summarized RCT characteristics as proportions and compared groups using the chi-squared test. We calculated the median Number Needed to Screen (NNS) to randomize one participant and assessed if the NNS, and reporting of data required to calculate the NNS, was related to the following explanatory variables: journal in which the RCT was published; number of enrollees in the RCT; whether the RCT was single or multicentre; funding source and the type of intervention studied.

A total of 148 RCTs were reviewed. As three of the four journals assessed are published in the UK, it was not surprising to find that the majority of RCTs originated from there (56%), with other significant contributions from the US (10%) and the Netherlands (8%). The main funding source was government agency (57%), most were multi-centre (79%), and 17% had at least 650 participants. One hundred and three RCTs (70%) reported the number of individuals assessed by investigators for eligibility, 119 (80%) reported the number eligible for participation, and all reported the actual number recruited (Table 2).

Of the trials that reported quantitative recruitment information, the median eligibility fraction was 83% (IQR 40% to 100%), and the median enrolment fraction was 74% (IQR 49% to 92%). The recruitment fraction could be calculated for 103 trials (70%), with a median participant recruitment of 41% (IQR 16% to 71%) (Table 3). The median NNS was 2.43 (IQR 1.41 to 6.25), with some trials reportedly recruiting every patient or practice screened for eligibility, and one trial screening 484 for each patient recruited. We found no association between magnitude of the NNS and journal publication, RCT size, multi- or single-centre RCT, funding source or type of intervention (Table 4). The data suggested that trials funded by government of interventions other than pharmaceutical therapies may be more likely to provide sufficient recruitment data for calculation of NNS (Table 5).

We found that the reporting of recruitment to RCTs based in primary care is inconsistent and frequently incomplete. Of the 148 RCTs published, 103 (70%) reported the number of individuals who were screened for eligibility and 80% of published RCTs reported the number of individuals who were eligible. In those RCTs that did report on the recruitment process there appears to be marked variation in terms of the proportion of individuals who are recruited. These findings suggest that reporting of RCTs should be improved and that for some RCTs external validity is limited because only a low proportion of eligible participants are successfully recruited.

Our findings differ with those of a recent study that report on the recruitment process in 172 RCTs published in four general medical journals- Annals of Internal Medicine, JAMA, The New England Journal of Medicine and The Lancet.[12] In this study half of the published RCTS were industry sponsored and over two thirds were of a pharmaceutical intervention, 52% of RCTs reported the number of persons who were evaluated by investigators for eligibility compared with 70% of primary-care based RCTs; similarly 43% of the RCTs published in general medical journals reported on the number of persons who were actually eligible for participation, compared with 80% of RCTs published in primary care journals.[12] These higher figures for screening eligibility and proportion being eligible in primary care journals need to be viewed cautiously. Higher screening and eligibility may represent under-reporting concerning the processes used to identify the eligible individuals in primary care-based RCTs. Qualitative research has shown that difficulty in recruiting patients is the most frequently mentioned problem in primary care-based RCTs.[13] The administrative processes that are used to identify and quantify the eligible population of patients may be incomplete with the consequence that those RCTs that report complete recruitment of all eligible patients may be based on an under-estimated denominator population. Consequently, the reported median eligibility (65), enrolment (93) and recruitment (54) fractions in this study should be treated with some caution (Table 3).

It has been recognized that patients seen in primary care settings may be significantly different to patients seen in a hospital (secondary or tertiary) care setting. Primary care patients are more likely to have a lower probability of disease; have a milder severity of disease; and have more undifferentiated symptoms.[14] These factors may influence both the estimated benefits and harms from treatment, making the risk:benefit ratio for some interventions less compelling for patients in primary care compared to those in secondary care.[3] More recent evidence has have shown that RCTs published in major medical journals do not always clearly report exclusion criteria.[15] Furthermore, women, children, the elderly and those with common medical conditions are frequently excluded from RCTs; and pharmaceutical-sponsored and multi-centred RCTs in both primary and secondary care are also likely to report higher numbers of exclusion criteria compared to other forms of RCTs.[15] Rather than rely on intuition and judgement, it would be better to include

The implications of this study relate to improved reporting and recruitment in primary care-based RCTs, and in assisting planning of future studies. Flow diagrams are associated with improved reporting of RCTs, [16] and implementation of the CONSORT statement has been shown to improve the quality of published RCTs, particularly in relation to dimensions of internal validity.[17] Findings from this study suggest that greater attention to the details of the patient recruitment process may similarly influence and improve reporting of RCTs in terms of external validity.[4, 18] There are specific issues that relate to recruitment of patients in primary care that are challenging, such as reliance on family physicians to recruit at the time of a consultation, engaging and running an RCT across many different family practice centres, lack of patient sampling frames and low rates of patient referral.[11, 19] More explicit reporting allied to strategies that are known to be important ingredients in terms of improved recruitment- good organization, simplified documentation and study procedures, and anticipating doctor and patient concerns about prior beliefs relating to efficacy and side effects – are issues that need to be considered at the outset when starting up a primary care-based RCT.[20] Complete and accurate reporting of recruitment in published trials will also greatly assist in planning future studies. Researchers can make more realistic estimates of the resources required if the likely eligibility and recruitment fractions are known to be low. Lastly, a low enrolment fraction may not necessarily represent an RCT with poor external validity. If the sample of patients recruited is representative of the eligible population in terms of demographic and clinical characteristics, then external validity is high. This requires investigators to be explicit in describing the recruitment process as well as reporting the characteristics of those eligible individuals included in the RCT and those eligible individuals who were excluded from the RCT.

Not withstanding concerns about identification of the full eligible population in some RCTs mentioned above, this study may have other shortcomings. The findings of our study relate to three primary care journals that have a UK/European focus. Only one of the journals selected has a North American focus. It may be that RCTs relevant to primary care are published in general medical and North American based journals and may provide better reporting of issues that relate to external validity. For this reason further assessment concerning the external validity of RCTs in other countries and journals is needed. On a more positive note, investment in primary care research and development in the UK has meant that in absolute terms the number of RCTs published in these three UK journals has nearly doubled compared to a five year period in the early 1990's.[21]