Beyond Sarcopenia: An Exploratory Machine Learning Analysis of Fatigue, Sleep Quality and Acute-Phase Severity as...

Background: Handgrip strength has been proposed as an objective marker of post-COVID functional impairment, but its contribution relative to fatigue and sleep quality has not been formally evaluated in Latin American populations. Methods: We conducted an exploratory cross-sectional analysis of 130 adults with post-COVID condition in Palmira, Colombia. The outcome was any functional limitation on the Post-COVID Functional Status (PCFS) scale (grades 1-4 versus grade 0). Eight predictors were pre-specified on clinical grounds before examining outcome associations, and ridge-penalized logistic regression was pre-specified as the primary model. Sample size adequacy followed the criteria of Riley et al. Internal validation used bootstrap optimism correction (B = 1000), re-running the complete pipeline within each replicate. We assessed calibration, decision curves, prediction stability, SHapley Additive exPlanations (SHAP) with bootstrap rank intervals, and subgroup performance. Reporting followed TRIPOD + AI; risk of bias was self-assessed with PROBAST + AI. Results: Functional limitation affected 91/130 participants (70.0%). The minimum sample required for eight parameters was 566; the cohort met 23.0% of this requirement, so the analysis is exploratory. The optimism-corrected area under the curve was 0.748 (95% confidence interval (CI) 0.669-0.830), with calibration slope 1.15 (0.71-1.69). Fatigue ranked first in SHAP importance (54.0% of resamples), whereas grip strength ranked fifth (median rank 6). Net benefit exceeded both default strategies between thresholds 0.38 and 0.88. Thirty percent of participants changed classification in over 20% of resamples. Conclusions: Fatigue, sleep quality and acute-phase severity, rather than grip strength, emerged as the most influential correlates. These hypothesis-generating findings require confirmation in adequately powered cohorts.