Effect sizes and statistical testing in the determination of clinical significance in behavioral medicine research.

BACKGROUND: The interpretation of clinical significance continues to be an obstacle for researchers in behavioral medicine.
PURPOSE: To review selected behavioral medicine research to critically examine the perception among investigators that behavioral effects on health are small based on common metrics of clinical significance.
METHODS: Using quantitative findings from recent behavioral medicine research in medical and psychiatric journals, we explored results in terms of several statistical metrics to assess potential clinical significance: r coefficients, risk ratios, risk difference measures, and attributable risk.
RESULTS: Translated into r coefficients, even established health predictors such as smoking, obesity, and fitness had only modest effects (rs =.03-.22), and the range of effect sizes were comparable with those based on psychological predictors including depression and stress-reactivity (rs =.06-.22). In contrast, effects for both classes of predictors were suggestive of clinical significance based on public health statistics.
CONCLUSIONS: Our choice of statistics for defining "small" and "large" effect sizes affects the perceived importance of behavioral health findings. In the assessment of health outcomes with low incidence rates, effects expressed as correlations using even the most robust predictors will often appear small. In these instances, we challenge researchers to move beyond conventional data analysis approaches and to expand their clinical interpretation efforts by employing additional statistical methods favored in medicine and public health.