Swallowing Recovery Trajectories

The three-class solution was chosen because it provided the best balance of model fit, entropy above 0.9, interpretability, and expert judgement. Nearly half of patients followed a consistently low-risk pattern, about one-quarter recovered rapidly within 24 hours, and nearly one-third had persistently high-risk scores. LCGM allowed recovery patterns to be separated into clinically interpretable subgroups rather than treating all patients as having the same average trend. Latent class growth modelling identified three distinct swallowing function trajectories in adult ICU patients after extubation. Models with one to four latent classes were tested, with selection criteria including AIC, BIC, adjusted BIC, log-likelihood, entropy, interpretability, and expert opinion. The heterogeneity of swallowing recovery trajectories suggests post-extubation management should not rely on a single fixed assumption about recovery timing.