Predicting length of stay in Australian residential aged care using assessment and diagnosis data and a machine learning framework

Abstract

This study examines length of stay (LOS) in Australian residential aged care using a comprehensive dataset derived from the Aged Care Funding Instrument (ACFI). In addition to demographic characteristics and core care-need measures, the analysis incorporates depression-related appraisal variables and grouped mental, behavioral, and medical diagnosis indicators. To account for right-censoring, non-proportional hazards, nonlinear covariate effects, and strong dependence among predictors, we employ the accelerated oblique random survival forest to model LOS. The results show that admission cohort, age at admission, and functional dependency are the main predictors of LOS at the aggregate level, while richer assessment and diagnostic variables provide predictive value. In particular, the Cornell Scale for Depression emerges as an important predictor beyond the standard ACFI domains. Subsample analyses reveal substantial heterogeneity across care-need profiles, admission cohorts, and age groups. Depression-related and mental and behavioral variables become especially important in higher-need groups, and the relative importance of predictors also changes over time. These findings extend existing Australian evidence on residential aged care use and suggest that more detailed assessment and diagnosis information can improve understanding of LOS patterns, support more accurate planning and resource allocation, and strengthen actuarial assessment of long-term care duration risk and related liabilities.