Presentation (CTI)

OBJECTIVES: Hospital length of stay (LOS) is a key indicator of healthcare utilization and patient outcomes. Understanding the factors that influence LOS can inform resource planning and targeted interventions. We aim to identify and quantify the impact of age, health condition, and their interactions on LOS using regularized regression models, with a focus on comorbidity dynamics.
METHODS: We used four regression models to predict LOS. Each model was run using LASSO regression with H2O’s GLM tool, which helps select the most important predictors by shrinking less useful ones to zero. The predictors included age categories, grouped condition indicators (such as mental, respiratory, and cardiovascular), and combinations of these variables to capture interaction effects. The resulting coefficients show how much each factor is estimated to change LOS, assuming all other factors stay the same.
RESULTS: Age was consistently associated with LOS, with older groups predicting longer stays. Mental health burden was the strongest predictor across models (+14.29 to +18.82). Other conditions such as respiratory (+2.93), infections (+1.84), and cardiovascular (+1.52) also contributed positively. Interaction terms revealed complex comorbidity effects, with notable negative synergies (e.g., Maternal and neonatal disorders + Mental disorders: -14.42, Injury + Mental disorders: -10.49). Year indicators were consistently excluded or negligible, suggesting stable LOS patterns over time.
CONCLUSIONS: Aggregated condition burdens and their interactions significantly influence LOS, with mental health and age being dominant factors. The presence of multiple conditions does not always compound LOS linearly, highlighting the importance of modelling comorbidity dynamics. These findings support the use of regularized models for uncovering nuanced patterns in healthcare utilization.