Comparing characteristics and outcomes of individuals who participated in a trial versus those that did not

Background
Participants in trials of pain treatments may differ from those who are not recruited. It is important to understand how non-recruited patients, and their outcomes, differ from participants.
Aims
To compare participants in a randomised controlled trial (RCT) of telephone-delivered cognitive behavioural therapy (tCBT) to prevent chronic widespread pain (CWP) with those who did not participate, at screening and outcome.
Method
The MAmMOTH Study was an RCT of tCBT for prevention of CWP in those at risk. To recruit, screening questionnaires were mailed to patients registered at GP practices in Scotland. Respondents were considered at risk if they had pain that was not CWP and met at least two of three risk factors: illness behaviour score of at least 4 (range 0-24); somatic symptoms score of at least 2 (0-5); sleep problems score of at least 5 (0-20). Eligible respondents consenting to contact were sent invitation letters. Those recruited were randomly allocated to tCBT or usual care (UC). TCBT consisted of assessment with a therapist by telephone, followed by six weekly-sessions, and boosters at three and six months. Those allocated UC received no further intervention. Follow-ups were at 3, 12, and 24 months. CWP at 12 months was the primary outcome. High-risk patients returning screening questionnaires who were not recruited but consented to contact about future studies, were sent a questionnaire at a time matched to the 12-month follow-up measuring pain, risk factors, and global change in health.
Differences at screening were expressed as difference in medians, percentages, or mean differences as appropriate, with 95% confidence intervals (CIs). Outcomes at 12 months were compared between those not recruited and those in the two trial arms using linear, binary, and ordinal logistic regression models adjusted for age, gender, GP practice, number of risk factors, and baseline outcome scores, to match analyses in the published trial paper. Mean differences and adjusted odd ratios (aOR) were reported comparing those in tCBT and those in UC to those not recruited as the reference group.
Result
Of 18035 completed screening questionnaires, 4435 met criteria for being high-risk, and 996 were recruited. Compared to those not recruited, participants were older (median 59.1 vs 54.8 years, difference 4.2, 95% CI 2.7-5.6), and more likely to be female (41.5% vs 35.8%, 5.7%, 2.2-9.2%). Participants had higher risk factor scores than those not recruited – illness behaviour score 9.83 vs 8.34 (mean difference 1.49, 1.26-1.72), 19.4% with 2 to 5 somatic symptoms vs 16.1% (3.2%, 0.5-6.0%), sleep problems score 10.18 vs 9.70 (0.48, 0.15-0.81).
Follow-ups were available for 827 of those not recruited and 825 participants - 441 in UC, 384 in tCBT. CWP was not significantly different between those not recruited and either those in tCBT or those in UC, and neither were illness behaviour scores or somatic symptoms. Participants reported greater global improvements in health compared to those not recruited (in UC, aOR of 1 point increase in global impression of change score indicating worsening health 0.75, 0.60-0.93, in tCBT, 0.38, 0.30-0.48). Those in tCBT had lower sleep problem scores, but not those in UC, compared to those not recruited (8.20 in tCBT vs 9.31 in those not recruited, -1.13, -1.83 to -0.42).
Conclusion
Trial participants differed from those not recruited in age and pain risk factors. At follow-up, those in tCBT reported better sleep than those not recruited confirming the main trial results. Those in both trial arms reported better health at follow-up compared to those not recruited, perhaps due to participation effects. Using unrecruited participants as a comparison group confirms tCBT as effective for some pain risk factors, i.e., sleep problems, and this is important for people with pain.