Key result
The cost-sensitive support vector classification model demonstrated superior performance in predicting in-hospital mortality for ICU patients with rheumatic heart disease, achieving an AUC of 0.888 and sensitivity of 0.800.
Why the study?
Rheumatic heart disease disproportionately burdens the developing world and receives less attention than other infectious diseases, posing challenges for predicting mortality risk in imbalanced datasets.
Do cost-sensitive machine learning models improve the prediction of in-hospital mortality in ICU patients with rheumatic heart disease?
Observational (n=1,266)
No
Do cost-sensitive machine learning models improve the prediction of in-hospital mortality in ICU patients with rheumatic heart disease?
Cost-sensitive machine learning models, particularly support vector classification, can accurately predict in-hospital mortality in ICU patients with rheumatic heart disease while identifying key physiological and laboratory predictors.
Resampled ML models may aid RHD mortality prediction; hypothesis-generating and requires prospective validation before clinical use.
Linked to poverty, rheumatic heart disease (RHD) disproportionately burdens the developing world, receiving less attention than other infectious diseases. Resampling and cost-sensitive learning techniques are applied to predict the mortality risk of imbalanced RHD datasets. A total of 57 models were constructed, and was comprised of 50 resampled machine learning (ML) models and 7 cost-sensitive learning models. The results from the Friedman and Nemenyi tests highlight the superior performance of the cost-sensitive support vector classification model, with an AUC of 0.888, sensitivity of 0.800, G-means of 0.806, and a Brier score of 0.061. The global and local interpretability are advanced through two post-hoc interpretable ML methods, facilitating the prioritization of key features associated with mortality risk, the determination of thresholds for features, and a comprehension of how variations in these features influence patient mortality rates. These findings may prove to be clinically valuable, assisting clinicians in tailoring precise management that is essential to maximize the survival of RHD patients.
No takes yet. Share an insight, caveat, or question.
Tao et al. (2024) conducted an observational in Rheumatic heart disease (n=1,266). Cost-sensitive support vector classification model vs. Other machine learning models was evaluated on In-hospital mortality prediction (Area Under the Curve) (AUC 0.888, p=<0.05). The cost-sensitive support vector classification model demonstrated superior performance in predicting in-hospital mortality for ICU patients with rheumatic heart disease, achieving an AUC of 0.888 and sensitivity of 0.800.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: