Data-driven reweighting of the Charlson and Elixhauser comorbidity indices using German claims data: development for in-hospital mortality only, external validation across clinical and health-economic outcomes.
5 affiliations
- 1Institute for Medical Biometry and Statistics, Faculty of Medicine and Medical Center, University of Freiburg, Stefan-Meier-Str. 26, Freiburg, 79104, Germany. [email protected].
- 2Center for Big Data Analysis in Cardiology (CeBAC), Department of Cardiology and Angiology, Medical Center, Faculty of Medicine, University Heart Center, University of Freiburg, University of Freiburg, Freiburg, Germany. [email protected].
- 3Institute for Medical Biometry and Statistics, Faculty of Medicine and Medical Center, University of Freiburg, Stefan-Meier-Str. 26, Freiburg, 79104, Germany.
- 4Department of Cardiology and Angiology, Faculty of Medicine, University Heart Center, Medical Center University of Freiburg, University of Freiburg, Freiburg, Germany.
- 5Center for Big Data Analysis in Cardiology (CeBAC), Department of Cardiology and Angiology, Medical Center, Faculty of Medicine, University Heart Center, University of Freiburg, University of Freiburg, Freiburg, Germany.
Abstract
Background
Comorbidity scores such as the Charlson and Elixhauser indices are widely used for risk adjustment in health-economic and clinical outcomes research. Yet the existing weighting schemes, originally designed for mortality prediction, may no longer reflect today's patient populations or perform equally well across diverse outcomes. This study aimed to reweight both indices using a data-driven approach and to evaluate their predictive validity for multiple clinical and health-economic outcomes.Methods
Using administrative data from a large tertiary care teaching hospital in Germany (N = 198,819 hospitalizations), we applied adaptive lasso regression to recalibrate the Charlson (Deyo weights) and Elixhauser (van Walraven weights) comorbidity indices. Reweighting was performed for in-hospital mortality as the development endpoint. Penalized coefficients were converted into integer-based weights. Model performance was assessed for three dichotomous endpoints (in-hospital mortality, severe bleeding, prolonged mechanical ventilation) and two continuous outcomes (length of hospital stay, total reimbursement). External validation was performed in three independent cohorts: Elective endometrial cancer (N = 85,204), interventional left atrial appendage closure (N = 28,039), and emergency ST-elevation myocardial infarction (STEMI; N = 248,062). AUC, R², and RMSE were used as complementary measures of discrimination and predictive accuracy. Pooled relative improvements of AUC, R², and RMSE across the three validation cohorts was calculated using random effects meta-analyses.Results
Reweighting consistently improved predictive performance across endpoints and cohorts, with modest absolute gains for in-hospital mortality and more pronounced gains for severe bleeding and continuous outcomes. When analysing in-hospital mortality in the three validation cohorts, the reweighted Charlson improved AUC from 0.679 to 0.714 in left atrial appendage closure, 0.588 to 0.594 in STEMI, and 0.798 to 0.810 in endometrial cancer (pooled relative increase: 2.05% [95% CI 0.56-3.53]); the reweighted Elixhauser showed larger gains, particularly in STEMI (0.708 to 0.766) and left atrial appendage closure (0.780 to 0.818; pooled relative increase: 4.86% [0.07-9.65]). The Elixhauser reweighting demonstrated consistently larger gains for severe bleeding (pooled AUC increase: 6.57% [4.23-8.92]). For continuous outcomes, the reweighted Elixhauser index yielded the largest gains in explained variance, with a pooled relative R² increase of 39.66% [33.72-45.60] for total reimbursement. Corresponding reductions in RMSE were modest and often not statistically significant.Conclusions
Adaptive lasso reweighting consistently enhances the predictive validity of the Charlson and Elixhauser indices across both clinical and health-economic endpoints. Our multi-endpoint evaluation shows consistent performance gains across the three examined German validation cohorts, offering a practical basis for risk adjustment in health-economic studies using German claims data. Because development and validation were based exclusively on German data and a single coding system (ICD-10-GM), international validation across other coding systems and healthcare systems is required before broader applicability can be assumed.
Related
- Release-aware, SAS-equivalent Elixhauser comorbidity scoring in R: cross-implementation validation and application to Texas inpatient discharges
- Development and validation of an interpretable machine learning model for early prediction of mortality in ICU patients with chronic obstructive pulmonary disease and heart failure
- Development and validation of a novel metabolic-inflammatory health-related nomogram for predicting in-hospital mortality in older patients with type 2 diabetes mellitus combined with cardiovascular disease
- Admission-time prediction of 7-day mortality in clinician-attributed hypertensive intracerebral hemorrhage: development and external validation against the ICH score
- Development and validation of an interpretable machine learning model for predicting in-hospital mortality among critically ill patients with liver cirrhosis
BibTeX
@article{kaier2026datadriven,
title = {{Data-driven reweighting of the Charlson and Elixhauser comorbidity indices using German claims data: development for in-hospital mortality only, external validation across clinical and health-economic outcomes.}},
author = {Klaus Kaier and Derek Hazard and Adrian Heidenreich and Alexander Maier and Dirk Westermann and Constantin von Zur Mühlen and Vera Oettinger},
journal = {Health economics review},
year = {2026},
month = oct,
doi = {10.1186/s13561-026-00888-9},
url = {https://europepmc.org/article/MED/42955616},
note = {CC BY},
}
RIS
TY - JOUR
TI - Data-driven reweighting of the Charlson and Elixhauser comorbidity indices using German claims data: development for in-hospital mortality only, external validation across clinical and health-economic outcomes.
AU - Klaus Kaier
AU - Derek Hazard
AU - Adrian Heidenreich
AU - Alexander Maier
AU - Dirk Westermann
AU - Constantin von Zur Mühlen
AU - Vera Oettinger
PY - 2026
DA - 2026/10/27
JO - Health economics review
DO - 10.1186/s13561-026-00888-9
UR - https://europepmc.org/article/MED/42955616
AB - Background Comorbidity scores such as the Charlson and Elixhauser indices are widely used for risk adjustment in health-economic and clinical outcomes research. Yet the existing weighting schemes, originally designed for mortality prediction, may no longer reflect today's patient populations or perform equally well across diverse outcomes. This study aimed to reweight both indices using a data-driven approach and to evaluate their predictive validity for multiple clinical and health-economic outcomes.Methods Using administrative data from a large tertiary care teaching hospital in Germany (N = 198,819 hospitalizations), we applied adaptive lasso regression to recalibrate the Charlson (Deyo weights) and Elixhauser (van Walraven weights) comorbidity indices. Reweighting was performed for in-hospital mortality as the development endpoint. Penalized coefficients were converted into integer-based weights. Model performance was assessed for three dichotomous endpoints (in-hospital mortality, severe bleeding, prolonged mechanical ventilation) and two continuous outcomes (length of hospital stay, total reimbursement). External validation was performed in three independent cohorts: Elective endometrial cancer (N = 85,204), interventional left atrial appendage closure (N = 28,039), and emergency ST-elevation myocardial infarction (STEMI; N = 248,062). AUC, R², and RMSE were used as complementary measures of discrimination and predictive accuracy. Pooled relative improvements of AUC, R², and RMSE across the three validation cohorts was calculated using random effects meta-analyses.Results Reweighting consistently improved predictive performance across endpoints and cohorts, with modest absolute gains for in-hospital mortality and more pronounced gains for severe bleeding and continuous outcomes. When analysing in-hospital mortality in the three validation cohorts, the reweighted Charlson improved AUC from 0.679 to 0.714 in left atrial appendage closure, 0.588 to 0.594 in STEMI, and 0.798 to 0.810 in endometrial cancer (pooled relative increase: 2.05% [95% CI 0.56-3.53]); the reweighted Elixhauser showed larger gains, particularly in STEMI (0.708 to 0.766) and left atrial appendage closure (0.780 to 0.818; pooled relative increase: 4.86% [0.07-9.65]). The Elixhauser reweighting demonstrated consistently larger gains for severe bleeding (pooled AUC increase: 6.57% [4.23-8.92]). For continuous outcomes, the reweighted Elixhauser index yielded the largest gains in explained variance, with a pooled relative R² increase of 39.66% [33.72-45.60] for total reimbursement. Corresponding reductions in RMSE were modest and often not statistically significant.Conclusions Adaptive lasso reweighting consistently enhances the predictive validity of the Charlson and Elixhauser indices across both clinical and health-economic endpoints. Our multi-endpoint evaluation shows consistent performance gains across the three examined German validation cohorts, offering a practical basis for risk adjustment in health-economic studies using German claims data. Because development and validation were based exclusively on German data and a single coding system (ICD-10-GM), international validation across other coding systems and healthcare systems is required before broader applicability can be assumed.
SN - 2191-1991
ER -