Not so proportional after all?
BackgroundThe proportional recovery rule asserts that most stroke survivors recover a fixed proportion of lost function. Reports that the rule can be used to predict recovery, extraordinarily accurately, are rapidly accumulating. Here, we show that the rule may not be as powerful as it seems.\n\nMethodsWe provide a formal analysis of the relationship between baseline scores (X), outcomes (Y) and recovery (Y-X), to highlight the shortcomings of the proportional recovery rule, and illustrate those problems with simulations in which synthetic recovery data are derived from different types of recovery processes.\n\nFindingsWhen the correlation between baseline scores and recovery is stronger than that between baselines scores and outcomes, the former can create an inflated impression of how predictable outcomes really are given baseline scores. This often happens when outcomes are less variable than baseline scores, as is common in empirical studies of recovery after stroke. Moreover, we cannot use the results of these correlations to distinguish proportional recovery from recovery which is either not consistently proportional, or not proportional at all.\n\nInterpretationAnalyses relating baseline scores to subsequent change are a minefield: our formal analysis applies as consistently outside the area of stroke as it does within it. One implication of our analysis is that the proportional recovery rule is not as predictive of real recovery after stroke as recent empirical studies suggest. Another is that different analytical methods will be required to ascertain whether recovery is even proportional at all.