Substantial underpowering of research investigating exercise interventions for tendinopathy

A quantitative review of the evidence base

Abstract

Background: Tendinopathy is a common musculoskeletal condition, with exercise therapy being the cornerstone of conservative management. An extensive evidence base has compared various exercise therapies and alternative treatments, often using exercise therapy as a control. As in many fields, concerns have emerged of inflated effect sizes leading to underpowered studies and unreliable findings.

Objective: To review sample size determination in experimental trials investigating exercise therapy for tendinopathy management and to assess statistical power given the likely true effect sizes.  

Eligibility criteria for selecting studies: A systematic literature search was conducted for trials involving participants diagnosed with rotator cuff, lateral elbow, gluteal, patellar or Achilles tendinopathy comparing exercise and non-exercise interventions.

Review and analysis: Details regarding the development and setting of target effect sizes, statistical models, and calculated sample sizes were extracted. Between intervention standardised mean difference effect sizes were calculated for studies comparing two exercise therapies, or an exercise therapy with a non-exercise intervention. Effect sizes were adjusted for expected overestimations using a shrinkage method based on the signal to noise ratio of observed values. Symmetric distributions centred on zero were then constructed with small, medium, and large thresholds determined. Sensitivity analyses assessed whether adjusted effect size distributions varied by tendinopathy location, outcome domain, and time from baseline measurement.

Results: The review included 126 studies, with 81 reporting a priori power calculations. The median target effect size used for powering studies was 0.85 (IQR: 0.64 to 1.06, range: 0.25 to 1.7), resulting in a median required group sample size of 21 (IQR: 17 to 34, range: 6 to 176). The most common statistical analysis was a t-test applied across group change scores. Shrinkage analysis was applied to 639 effect sizes from exercise-vs-exercise (EvE) comparisons, and 988 effect sizes from exercise-vs-non-exercise (EvNE) comparisons. The thresholds for small, medium and large effects were equal to 0.04, 0.12, and 0.26 for EvE, and 0.08, 0.21, and 0.50 for EvNE. For medium effect sizes, the typical statistical analysis used suggests required group sizes of 1092 and 357, far exceeding sample sizes reecruited by factors of ~20 to 50. Sensitivity analyses showed distributions were consistent across tendinopathy location, outcome domain, and time from baseline measurement.

Conclusions: The findings highlight a prevalent issue of underpowering in exercise therapy trials for tendinopathy, often due to overestimated effect sizes. This review suggests that future research should incorporate adjusted effect size estimates to ensure adequate power and justify resource allocation. Researchers are encouraged to adopt more realistic effect size thresholds and consider alternative study designs, such as high-frequency data collection and real-world evidence studies, for incremental improvements in treatment. Enhanced methodological approaches and collaboration with clinicians and patients are needed to refine power calculations and improve the quality of tendinopathy research.