Key result
Valve surgery utilization in infective endocarditis increases modestly despite steep rises in predicted mortality risk.
Observational (n=125,100)
Yes
At the population level, valve surgery utilisation in infective endocarditis varies minimally across a wide spectrum of predicted in-hospital mortality risk and is driven more by age and hospital characteristics than by illness severity or injection drug use status.
Answers are grounded in this paper and cite their sources.
Non-clinical factors appear to drive valve surgery decisions in infective endocarditis; hypothesis-generating and should not yet change practice.
Infective endocarditis (IE) carries high mortality, and valve surgery is indicated in selected patients on the basis of specific clinical circumstances assessed by a multidisciplinary Endocarditis Team. Predicted mortality risk is not equivalent to surgical indication; nevertheless, how surgical utilisation is distributed across the spectrum of illness severity at the population level has not been described. We used machine learning (ML) to describe population-level variation in valve surgery utilisation across strata of predicted in-hospital mortality risk, with a secondary focus on injection drug use (IDU)-associated IE. We analysed adult hospitalisations with an IE diagnosis code in the National Inpatient Sample (2016–2022). Co-primary outcomes were in-hospital mortality and valve surgery. Weighted multivariable logistic regression identified factors independently associated with each outcome. Machine learning models (LASSO, random forest, and XGBoost) were trained using demographic, clinical, and hospital characteristics recorded during the hospitalisation, with temporal validation (2016–2020 training; 2021–2022 testing) and supplementary 5-fold cross-validation. Because diagnoses are recorded for the entire hospitalisation, the models were framed as retrospective, discharge-level risk classification rather than prospective prediction. Model performance was evaluated using AUROC, AUPRC, Brier score, and F1 score, and SHAP analysis was used for model interpretability. Among 125,100 IE hospitalisations (weighted ≈ 625,500), in-hospital mortality was 10.1% and the valve surgery rate was 8.3%. In the 2021–2022 temporal test set, the XGBoost mortality model achieved an AUROC of 0.842 (95% CI 0.836–0.848) and an AUPRC of 0.395 (event rate 11.8%), and the surgery model an AUROC of 0.857 (0.850–0.864) and an AUPRC of 0.403 (event rate 8.5%). Observed mortality increased approximately 80-fold across predicted-risk quintiles (Q1: 0.4% to Q5: 35.5%), whereas surgery rates varied only modestly (Q1: 5.6%; Q4: 10.8%; Q5: 9.5%). In SHAP analysis, mortality predictions were most strongly influenced by clinical severity markers (APR-DRG mortality risk, age, APR-DRG severity), whereas surgery predictions were most strongly influenced by age and hospital characteristics (bed size, teaching status, transfer-in), together with arrhythmia. IDU-associated IE (23.3%) had lower adjusted mortality (aOR 0.64, 95% CI 0.62–0.65) and was only weakly associated with surgery after adjustment (aOR 0.97); surgery rates were comparable between IDU and non-IDU hospitalisations at similar predicted risk. Mortality increased from 8.2% (2016) to 11.9% (2022; P-trend < 0.001), while surgery rates remained stable. Findings were unchanged after excluding APR-DRG variables. At the population level, valve surgery utilisation in IE varied little across a wide gradient of predicted in-hospital mortality and was more closely associated with patient age and hospital characteristics than with markers of illness severity. Because administrative data cannot establish surgical indication, operability, or patient preferences, these findings describe variation in utilisation rather than under-treatment of individual patients. IDU status was not a dominant correlate of surgical utilisation once predicted risk was considered. These hypothesis-generating findings warrant confirmation in clinically detailed cohorts that capture guideline-defined surgical indications. Not applicable.
No takes yet. Share an insight, caveat, or question.
Elbana et al. (2026) conducted an observational in Infective endocarditis (n=125,100). Machine learning-predicted in-hospital mortality risk and injection drug use-associated infective endocarditis vs. Lower predicted mortality risk quintiles and non-injection drug use-associated infective endocarditis was evaluated on In-hospital mortality and valve surgery utilization (aOR 0.64, 95% CI 0.62-0.65). Valve surgery utilization in infective endocarditis varied modestly across predicted mortality quintiles despite an 80-fold mortality increase.
Synapse has enriched 10 related papers. Consider them for comparative context: