Regression to the mean and alcohol consumption: a cohort study exploring implications for the interpretation of change in control groups in brief intervention trials

Background: Reductions in drinking among individuals randomised to control groups in brief alcohol intervention trials are common and suggest that asking study participants about their drinking may itself cause them to reduce their consumption. We sought to test the hypothesis that the statistical artefact regression to the mean (RTM) explains part of the reduction in such studies. Methods: 967 participants in a cohort study of alcohol consumption in New Zealand provided data at baseline and again six months later. We use graphical methods and apply thresholds of 8, 12, 16 and 20 in AUDIT scores to explore RTM. Results: There was a negative association between baseline AUDIT scores and change in AUDIT scores from baseline to six months, which in the absence of bias and confounding, is RTM. Students with lower baseline scores tended to have higher follow-up scores and conversely, those with higher baseline scores tended to have lower follow-up scores. When a threshold score of 8 was used to select a subgroup, the observed mean change was approximately half of that observed without a threshold. The application of higher thresholds produced greater apparent reductions in alcohol consumption. Conclusions: Part of the reduction seen in the control groups of brief alcohol intervention trials is likely to be due to RTM and the amount of change is likely to be greater as the threshold for entry to the trial increases. Quantification of RTM warrants further study and should assist understanding assessment and other research participation effects.