The regression effect as a neglected source of bias in nonrandomized intervention trials and systematic reviews of observational studies.

Health care providers depend on the findings of observational intervention studies and systematic reviews of those studies to make evidence-based decisions about their clients' care. The nonrandom methods of group formation in observational studies necessitate carefully assessing threats to the validity of conclusions. Regression to the mean is a source of change in clinical outcome measures that has escaped widespread notice as a potential threat to the accuracy of conclusions from observational studies and systematic reviews thereof. Failure to assess the degree to which regression confounds study results elevates the risk of making clinical decisions using biased estimates of intervention effectiveness. Because the change in average outcome scores due to regression can be quantified, it is a type of bias whose direct influence can be known. Yet determining and reporting change due to regression is uncommon in observational studies or systematic reviews thereof. The means to quantify change due to regression in average outcome scores is illustrated by example in this article.