Promising solution for standardised length of hospital stay based on time-to-event models and contemporary Australian administrative data

Le résumé fourni par la source

OBJECTIVE: Hospital length of stay (LOS) is a key indicator of hospital efficiency and quality of care, but a reliable metric for benchmarking LOS remains problematic. This report describes a time-to-event methodology to generate a hospital standardised LOS ratio (HSLR). DESIGN: Retrospective observational analysis of LOS from a jurisdictional administrative dataset using a time-to-event (hazard of discharge) analytic approach to generate risk-adjusted LOS (predicted LOS-pLOS), and the HSLR (= (sum observed LOS)/(sum total pLOS)). SETTING: 219 (public and private) acute-care hospitals in the State of Victoria, Australia, adult population 5.28 million. PARTICIPANTS: 2.73 million adult multiday separations and 15.53 million bed-days from July 2019 to June 2024. INTERVENTIONS: Nil. OUTCOME MEASURES: Descriptive statistics for annual mean LOS (aLOS), pLOS and HSLR at the hospital level with model fit assessed for calibration (Cox-Snell residuals), classification (aLOS and HSLR results for hospital-years compared to benchmark), variance (intraclass correlation coefficient (ICC) at provider level) and model dispersion (value (φ) and random effect SD (τ)) characteristics. RESULTS: Observed LOS was markedly right skewed and autocorrelated (p<0.001); population aLOS 5.68±4.98 days, median 3.94 (IQR: 2.48-6.76) days. LOS prediction model included six demographic covariates (age, sex, aged-care residency, emergent, admission source, unplanned transfer) and 12 145 separate principal diagnoses aggregated into nine ranked LOS risk-categories. 572 (61% of 940 hospital-year) aLOS values were outliers (>3 SD of benchmark); whereas 936 (99.5%) HSLR values were inliers (<±3 SD); 98% within ±2 SD. Some overdispersion (φ=14.5 ±1.7, τ=0.09 ±0.001) remained, but ICC at provider level (0.025) was low. CONCLUSIONS: aLOS is a simple descriptor but poor comparator. Time-to-event survival analytic models furnish risk-adjusted pLOS and HSLR metrics which indicate that the majority of LOS variation is due to patient-related, not hospital, factors.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.