Is Personality a Key Predictor of Missing Study Data? An Analysis From a Randomized Controlled Trial

First, an admission of bias: It is hard to believe that these personality factors are not predictive of compliance with protocols in any setting. This bias is not formally evidence-based and little informed by behavior of participants in RCTs, but on observed non-compliance of clinical researchers with their own research protocol. It is but a short step from recognizing that the highly open, agreeable and, most of all, conscientious clinical researchers are least likely to bend their own protocol rules, to thinking that the same qualities in RCT participants might also predict their willingness to abide by protocol requirements.

Nevertheless, there are numerous statistical reservations about this “proof”. Above all, 415 were randomized, but only 381 analyzed. Thus 8.1% of the randomized subjects were missing from the analysis for missing data, while only 4.5% of those analyzed had missing data. Why all that “adjustment”, but no consideration of obvious collinearities or important interactions (moderating effects)(1)? With the sparsity of “missingness” at each time point, why a random effects model, and, if so, with what covariance structure? Why was the Odds Ratio used as an effect size (2,3)? In light of these reservations, I would consider this, not a hypothesis-testing study resulting in a conclusion, but an interesting exploratory study leading to a hypothesis that might be tested in a future RCT, giving information and ideas that might be used to design a powerful such RCT.

However, let me focus not on reservations but on this question: Why might the conclusion reported here be of importance to either designing future RCTs, or to a medical consumer?

To exclude otherwise eligible participants from participation? Not likely or wise. To implement procedures to forestall such non-compliance? That might be a good idea to be tested in future RCTs, but since routine personality testing is not practical in clinical practice, such components would be more effective if implemented without requiring identification of personality types.

Should assessment of personality types be explicitly considered in the design of RCTs? If “missingness” is not associated with treatment outcome, it doesn’t matter whether or not personality predicts missing data. On the other hand, if “missingness” is associated with treatment outcome, and if personality predicts both “missingness” and treatment outcome, and particularly if personality moderates the effect of treatment on both “missingness” and treatment outcome, then ignoring personality in RCT designs and analysis or in clinical decision making is risky. However, these issues were not considered by Jerant et al. In short, the analysis here is overly complicated, but the thinking behind the analysis overly simplistic.

1. Kraemer HC, Wilson GT, Fairburn CG, Agras WS. Mediators and Moderators of Treatment Effects in Randomized Clinical Trials. Archives of General Psychiatry. 2002;59:877-883. 2. Newcombe RG. A deficiency of the odds ratio as a measure of effect size. Statistics in Medicine. 2006;25:4235-4240. 3. Sackett DL. Down with odds ratios! Evidence-Based Medicine. 1996;1:164- 166.

Competing interests:   None declared