Key result
Adding clinical data improves in-hospital mortality prediction vs administrative models to a C-index of 0.863.
Why the study?
To assess the validity of the hospital standardized mortality ratio risk-adjusted model by comparing models incorporating clinical information with the current administrative data-based model in South Korea.
Does incorporating clinical information into the hospital standardized mortality ratio (HSMR) risk-adjustment model improve its predictive validity compared to using administrative data alone in inpatients?
Observational (n=53,976)
Yes
Does incorporating clinical information into the hospital standardized mortality ratio (HSMR) risk-adjustment model improve its predictive validity compared to using administrative data alone in inpatients?
Absolute Event Rate: 0.863% vs 0.785%
Incorporating clinical data, such as present-on-admission comorbidities and cancer stage, significantly improves the predictive validity of hospital standardized mortality ratio models compared to using administrative data alone.
May refine HSMR mortality prediction with clinical data; leaves open effects on hospital benchmarking pending validation.
PURPOSE: This study assessed the validity of the hospital standardized mortality ratio (HSMR) risk-adjusted model by comparing models that include clinical information and the current model based on administrative information in South Korea. MATERIALS AND METHODS: The data of 53976 inpatients were analyzed. The current HSMR risk-adjusted model (Model 1) adjusts for sex, age, health coverage, emergency hospitalization status, main diagnosis, surgery status, and Charlson Comorbidity Index (CCI) using administrative data. As candidate variables, among clinical information, the American Society of Anesthesiologists score, Acute Physiology and Chronic Health Evaluation (APACHE) II, Simplified Acute Physiology Score (SAPS) 3, present on admission CCI, and cancer stage were collected. Surgery status, intensive care in the intensive care unit, and CCI were selected as proxy variables among administrative data. In-hospital death was defined as the dependent variable, and a logistic regression analysis was performed. The statistical performance of each model was compared using C-index values. RESULTS: There was a strong correlation between variables in the administrative data and those in the medical records. The C-index of the existing model (Model 1) was 0.785; Model 2, which included all clinical data, had a higher C-index of 0.857. In Model 4, in which APACHE II and SAPS 3 were replaced with variables recorded in the administrative data from Model 2, the C-index further increased to 0.863. CONCLUSION: The HSMR assessment model improved when clinical data were adjusted. Simultaneously, the validity of the evaluation method could be secured even if some of the clinical information was replaced with the information in the administrative data.
No takes yet. Share an insight, caveat, or question.
Park et al. (2024) conducted an observational in In-hospital mortality (n=53,976). Risk-adjustment model incorporating clinical data (Model 4) vs. Current administrative data-based model (Model 1) was evaluated on C-index for predicting in-hospital mortality. A risk-adjustment model incorporating clinical data such as ASA score, present on admission comorbidities, and cancer stage improved the C-index for predicting in-hospital mortality from 0.785 to 0.863 compared to the current administrative model.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: