Prediction models for dynamic respiratory muscle strength in COPD and asthma patients based on fixed-pressure inspiratory muscle performance

Abstract

Dynamic inspiratory muscle strength, assessed using the S-Index, reflects pressure generation during flow-dependent inspiration and may provide functional information beyond static respiratory muscle strength measures. However, its routine clinical use is limited by the need for specialized equipment. This study aimed to develop and internally validate prediction models for estimating S-Index from accessible pulmonary function, body composition, and fixed-pressure inspiratory performance measures in male patients with chronic obstructive pulmonary disease (COPD) or asthma. In this cross-sectional predictive modeling study, 123 male patients with COPD or asthma underwent S-Index assessment using a single-breath test. Demographic characteristics, body composition, spirometry variables, and inspiratory performance data from a fixed-pressure task-to-failure protocol at 30 cmH₂O were evaluated as candidate predictors. Linear, regularized, robust, support vector, quantile, and multilayer perceptron regression models were developed. Missing spirometry values were handled using 20 multiply imputed datasets. Model performance was evaluated using repeated nested five-fold cross-validation with 10 repetitions and assessed using mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), and coefficient of determination (R²), with bootstrap-derived 95% confidence intervals. Peak inspiratory flow rate and the number of inspiratory repetitions showed the strongest correlations with S-Index ( r = 0.83 and r = 0.70, respectively). In the full predictor scenario, regularized and linear regression models showed broadly comparable internal predictive performance. LassoLars numerically achieved the lowest prediction error, with an RMSE of 7.156 cmH₂O (95% CI: 6.108–8.285), MAE of 5.538 cmH₂O (95% CI: 4.778–6.352), and R² of 0.758 (95% CI: 0.651–0.829). Lasso and Elastic Net produced similar results. Models using only demographic, anthropometric, and body composition variables explained approximately 44–46% of the variance, whereas adding spirometry and inspiratory flow variables increased explained variance to approximately 70%. Complete-case sensitivity analysis yielded broadly consistent findings. In this pooled single-center cohort of male patients with COPD or asthma, S-Index could be estimated with useful internal predictive performance from fixed-pressure inspiratory performance, spirometry, and body composition variables. These models should be interpreted as complementary to, rather than substitutes for, direct S-Index measurement. External validation is required before clinical implementation.