Resistance exercise has the clearest pooled signal on frailty scores, but the range of true effects includes no benefit

A systematic review and meta-analysis of randomised evidence examined three standalone exercise modalities for frailty in older adults: resistance exercise, walking and Tai Chi Meta-analysis (Li et al., PMID 42891303). It followed PRISMA 2020, was prospectively registered, searched five databases from inception to October 2025, and assessed risk of bias with the Cochrane tool. Ten articles contributed 13 exercise-control comparisons and 1,361 participant observations (682 intervention, 679 control).

The overall random-effects estimate favoured exercise on frailty scores (standardised mean difference −0.57, 95% confidence interval −0.91 to −0.24). Heterogeneity was substantial (I² 88%) and the approximate 95% prediction interval, which describes where the effect in a new setting might fall, crossed the null (−1.73 to 0.57). Resistance-exercise comparisons showed a significant pooled reduction in frailty scores (−0.78, −1.22 to −0.35); walking and Tai Chi estimates were imprecise and not statistically significant. Leave-one-comparison-out estimates stayed in the same direction (−0.65 to −0.46).

The authors state that the review does not establish the superiority of resistance exercise, because no head-to-head or network comparison was performed, and that the dose-related findings rest on small numbers of comparisons and should generate hypotheses rather than prescribe a regimen. The outcome is a frailty score; the abstract describes no patient-reported experience measure, and it does not describe the settings or delivery mode of the included trials. Whether resistance training can be delivered in groups or by remote supervision is therefore a service-design question this review does not test.

Against the record, the direction agrees with the published pattern on resistance exercise and remote delivery, but a wide prediction interval and 88% heterogeneity mean the pooled effect is less certain than its point estimate suggests. It does not meet that pattern’s condition for lifting its qualification, which is a patient-centred trial of remote delivery.