The Predictive Validity of Individualised Load–Velocity Relationships for Predicting 1RM: A Systematic Review and Individual Participant Data Meta-analysis

Abstract

Background: Load–velocity relationships are commonly used to estimate one-repetition maximums (1RMs). Proponents suggest these estimates can be obtained at high frequencies and assist with manipulating loads according to session-by-session fluctuations. Given their increasing popularity and development of associated technologies, a range of load–velocity approaches have been investigated. Objective: This systematic review and individual participant data (IPD) meta-analysis sought to quantify the predictive validity of individualised load–velocity relationships for the purposes of 1RM prediction. Methods: In September 2022, a search of MEDLINE, SPORTDiscus, Web of Science and Scopus was conducted for published research, with Google Scholar, CORE and British Ethos also searched for unpublished research. Studies were eligible if they were written in English, and directly compared a measured and predicted 1RM using load–velocity relationships in the squat, bench press, deadlift, clean or snatch. IPD were obtained through requests to primary authors and through digitisation of in-text plots (e.g. Bland–Altman plots). Risk of bias was assessed using the Prediction model Risk Of Bias ASsessment Tool (PROBAST) and the review conducted in accordance with PRISMA-IPD guidelines and an a priori protocol. Absolute and scaled standard error of the estimates (SEE/SEE%) were calculated for two-stage aggregate analyses, with bootstrapping performed for sampling variances. Estimates were pooled using three-level hierarchical models with robust 95% confidence intervals (CIs). One-stage analyses were conducted with random intercepts to account for systematic differences across studies and prediction residuals calculated in the absolute scale (kg) and as a percentage of the measured 1RM. Moderator analyses were conducted by including a priori defined categorical variables as fixed effects. Results: One hundred and thirty-seven models from 26 studies were included with each identified as having low, unclear or high risk of bias. Twenty studies comprising 434 participants provided sufficient data for meta-analyses, with raw data obtained for 8 (32%) studies. Two-stage analyses identified moderate predictive validity [SEE% 9.8, 95% CI 7.4% to 12.2%, with moderator analyses demonstrating limited differences based on the number of loads (β 2Loads:>2Loads = 0.006, 95% CI − 1.6 to 1.6%) or the use of individual or group data to determine 1RM velocity thresholds (β Group:Individualised = − 0.4, 95% CI − 1.9 to 1.0%)]. One-stage analyses identified that predictions tended to be overestimations (4.5, 95% CI 1.5 to 7.4 kg), which expressed as a percentage of measured 1RM was equal to 3.7 (95% CI 0.5 to 6.9% 1RM). Moderator analyses were consistent with those conducted for two-stage analyses. Conclusions: Load–velocity relationships tend to overestimate 1RMs irrespective of the modelling approach selected. On the basis of the findings from this review, practitioners should incorporate direct assessment of 1RM wherever possible. However, load–velocity relationships may still prove useful for general monitoring purposes (e.g. assessing trends across a training cycle), by providing high-frequency estimates of 1RM when direct assessment may not be logistically feasible. Given limited differences in predictions across popular load–velocity approaches, it is recommended that practitioners opting to incorporate this practice select the modelling approach that best suits their practical requirements. Registration: https://osf.io/agpfm/ .