Journal Article

Comparison of established comorbidity scores using administrative data of patients undergoing surgery or interventional procedures in Massachusetts

Schaefer, Maximilian S., Ahrens, Elena, Shay, Denys, Wachtendorf, Luca J., Tenge, Theresa, Munoz-Acuna, Ricardo, Chen, Guanqing, von Wedel, Dario, Redaelli, Simone

2025

Overview

Previous studies proposed comorbidity-based prediction tools to facilitate patient-level assessment of mortality risk, which are essential for confounder adjustment in epidemiologic studies. We compared established comorbidity indices using real-world administrative data of a broad surgical population. Adult patients undergoing surgical or interventional procedures between January 2005 and June 2020 at a tertiary academic medical center in Massachusetts, USA, were included. The Elixhauser Comorbidity Index (van Walraven modification), Combined Comorbidity Score, and Charlson Comorbidity Index were compared regarding the prediction of 30-day mortality. Age and sex were included in all models. Discriminative ability was quantified by the area under the receiver operating characteristic curve (AUROC), and calibration was assessed using the Brier score and reliability plots. A total of 514,282 patients were included, of which 5849 (1.1%) died within 30 days. A model including age and sex alone had an AUROC of 0.73 (95% CI 0.72-0.74). The Elixhauser Comorbidity Index–based model showed the best discriminative ability with an AUROC of 0.86 (95% CI 0.86-0.87) compared to models, including the Combined Comorbidity Score (AUROC, 0.85 [95% CI 0.84-0.85]) and the Charlson Comorbidity Index (AUROC, 0.82 [95% CI 0.81-0.83], P < .001, respectively). The Brier score was 0.011 for all scores. Overall, score performances were similar or improved after the implementation of the 10th Revision International Classification of Diseases (Clinical Modification) coding system. The primary findings were confirmed for in-hospital, 7-day, 90-day, 180-day, and 1-year mortality and when including score comorbidities as separate indicator variables (P < .001, respectively). Patient and procedural characteristics were predictive of mortality (AUROC, 0.91 [95% CI 0.91-0.91]), with confirmatory findings and slightly improved performances when adding comorbidity scores (AUROC, 0.93 [95% CI 0.93-0.93] for the Elixhauser Comorbidity Index; AUROC, 0.93 [95% CI 0.93-0.93] for the Combined Comorbidity Score; AUROC, 0.92 [95% CI 0.92-0.93] for the Charlson Comorbidity Index, P < .001, respectively). All 3 comorbidity indices predicted mortality with excellent discrimination; however, they showed only slightly improved performance when incorporated into a model including patient and procedural characteristics. When surgical data are unavailable and in surgical setting–specific subgroups, the Elixhauser Comorbidity Index consistently performed best. [Display omitted] •Comorbidity-based prediction tools enable patient-level assessment of mortality risk.•We compared established prediction tools using electronic health records.•Among 514,282 surgical patients, the Elixhauser Comorbidity Index performed best.•The Elixhauser Comorbidity Index may be used preferably for mortality prediction in broad surgical populations.

• Publisher: Elsevier Inc,Elsevier Limited
• Subject: Adult / Aged / Aged, 80 and over / Anesthesia / Charlson Comorbidity Index / Combined Comorbidity Score / Comorbidity / Comorbidity risk index / Elixhauser Comorbidity Index / Epidemiology / Female / Health care facilities / Health risks / Hospital Mortality / Hospitals / Humans / Internal Medicine / Male / Massachusetts - epidemiology / Middle Aged / Mortality / Mortality risk / Patients / Perioperative risk / Population studies / Postoperative outcome / Prediction / Regression analysis / Retrospective Studies / Risk Assessment - methods / Sex / Subgroups / Surgery / Surgical Procedures, Operative - mortality / Surgical Procedures, Operative - statistics & numerical data / Variables