Association between wearable inertial-sensor-derived training-load metrics and musculoskeletal injury in athletes: a systematic review and meta-analysis

Wearable inertial sensors are widely used to monitor training load in sport, yet whether these proxy measures of external mechanical loading are associated with musculoskeletal (MSK) injury has not been systematically evaluated as a distinct exposure class. This review synthesised the association between inertial-sensor-derived load metrics and MSK injury and appraised the certainty of the evidence. Six databases were searched from inception through June 2026. Longitudinal studies reporting odds ratios, relative risks, or hazard ratios linking accelerometer-derived load metrics to physician-diagnosed or time-loss MSK injury were eligible. Risk of bias was assessed using ROBINS-E and evidence certainty using GRADE. Random-effects meta-analysis used restricted maximum likelihood estimation with the Hartung-Knapp-Sidik-Jonkman adjustment. Eleven studies (508 athletes, six sports plus one military cohort) were included. Risk of bias was Some Concerns in six and High in five studies. The exploratory pooled estimate combining different effect-measure types was 2.74 (95% confidence interval 1.89 to 3.96; I-squared 38%); however, injury proportions exceeded 10% in several studies, making the rare-outcome approximation questionable. Effect-type-stratified analyses, which should receive greater interpretive weight, yielded 2.54 (0.84 to 7.68) for odds-ratio-only and 2.98 (1.59 to 5.60) for relative-risk-only studies. The cumulative load-magnitude subgroup (k = 5) showed a pooled effect size of 2.86 (2.08 to 3.93; I-squared 0%), whereas the acute-to-chronic workload ratio subgroup (k = 5) was non-significant under conservative inference (2.64, 0.86 to 8.13; I-squared 69%). All GRADE outcomes were rated Very Low certainty. The evidence is of Very Low certainty and should be considered hypothesis-generating. A possible positive association was observed between elevated inertial-sensor-derived training loads and MSK injury, but pooled values should not be interpreted as causal effect magnitudes or as evidence of predictive ability. Because different constructs were examined in different populations, no inference about construct superiority can be drawn. The current evidence does not support the establishment of reliable load thresholds for injury prevention. PROSPERO CRD420251135945