Development and validation of explanatory models of postural stability using a force platform

Abstract

Introduction: Although many determinants of balance have been examined, few studies have integrated task-related and personal characteristics into internally validated multivariable prediction models. Objective: To develop and internally validate mixed-effects models predicting center-of-pressure (COP) 95% elliptical area and mean COP velocity in healthy adults based on personal and experimental factors. Methods: Secondary analysis of a public dataset with 163 adults performing twelve 60-s posturography trials under randomized conditions (firm/foam surface; eyes open/closed). Predictors included age, sex, body height, body weight, vision, surface, and base of support. Linear mixed-effects models were developed in participants without fall history (n=121) and internally validated in those with ≥1 fall (n=42). Performance was evaluated using marginal and conditional R², root mean square error (RMSE), calibration slope/intercept, and limits-of-agreement analyses. Results: Foam surface and visual deprivation were the strongest predictors of postural sway, followed by age. Height showed small but statistically significant effects, whereas sex and body mass had little influence. COP velocity model demonstrated strong explanatory and predictive performance (marginal R²=0.821; conditional R²=0.925; validation R²=0.688; RMSE=0.832). Calibration slope (1.100) and intercept (−0.017) indicated minimal systematic bias. COP area model explained a similar proportion of variance during development (marginal R²=0.820; conditional R²=0.894) but showed weaker validation performance (R²=0.501; RMSE=1.320) and wide limits-of-agreement. Conclusion: COP velocity can be robustly predicted from personal and task-related factors and demonstrated strong internal validity, supporting its use as a reference metric in research and clinical assessment. COP area displayed limited generalizability and requires methodological refinement and external validation before clinical application.