A Systematic Review Examining the Approaches Used to Estimate Interindividual Differences in Trainability and Classify Individual Responses to Exercise Training

Background: Many reports describe statistical approaches for estimating interindividual differences in trainability and classifying individuals as “responders” or “non-responders.” The extent to which studies in the exercise training literature have adopted these statistical approaches remains unclear. Objectives: This systematic review primarily sought to determine the extent to which studies in the exercise training literature have adopted sound statistical approaches for examining individual responses to exercise training. We also (1) investigated the existence of interindividual differences in trainability, and (2) tested the hypothesis that less conservative thresholds inflate response rates compared with thresholds that consider error and a smallest worthwhile change (SWC)/minimum clinically important difference (MCID). Methods: We searched six databases: AMED, CINAHL, EMBASE, Medline, PubMed, and SportDiscus. Our search spanned the aerobic, resistance, and clinical or rehabilitation training literature. Studies were included if they used human participants, employed standardized and supervised exercise training, and either: (1) stated that their exercise training intervention resulted in heterogenous responses, (2) statistically estimated interindividual differences in trainability, and/or (3) classified individual responses. We calculated effect sizes (ES IR ) to examine the presence of interindividual differences in trainability. We also compared response rates ( n = 614) across classification approaches that considered neither, one of, or both errors and an SWC or MCID. We then sorted response rates from studies that also reported mean changes and response thresholds ( n = 435 response rates) into four quartiles to confirm our ancillary hypothesis that larger mean changes produce larger response rates. Results: Our search revealed 3,404 studies, and 149 were included in our systematic review. Few studies ( n = 9) statistically estimated interindividual differences in trainability. The results from these few studies present a mixture of evidence for the presence of interindividual differences in trainability because several ES IR values lay above, below, or crossed zero. Zero-based thresholds and larger mean changes significantly (both p < 0.01) inflated response rates. Conclusion: Our findings provide evidence demonstrating why future studies should statistically estimate interindividual differences in trainability and consider error and an SWC o