Interpretable machine learning model to predict the acute occurrence of delirium in elderly patients in the intensive care units: a retrospective cohort study
Guardado en:
| Publicado en: | Journal of Big Data vol. 12, no. 1 (Feb 2025), p. 47 |
|---|---|
| Publicado: |
Springer Nature B.V.
|
| Materias: | |
| Acceso en línea: | Citation/Abstract Full Text - PDF |
| Etiquetas: |
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| Resumen: | BackgroundDelirium is a severe complication in critical elderly patients. This study aimed to develop interpretable machine-learning (ML) models to predict acute delirium and identify risk factors for medical intervention in elderly patients in the intensive care unit (ICU).Patients and Methodselderly patients (age ≥ 65 and ≤ 89) were selected from electronic intensive care unit collaborative research database (eICU-CRD). Data of demographics and laboratory tests were collected on the first day of admission to ICU. Delirium in 7 days after admission was identified. Difference between delirium and non-delirium groups was demonstrated. Association between delirium and mortality was proved through Kaplan–Meier survival curve. Participants were randomly distributed into a training set and a validation set without replacement at a ratio of 7:3. Recursive feature elimination (RFE) was used to determine the number of variables adopted in the model. The predictive capability of the ML models was demonstrated by receiver operating characteristic (ROC) analysis and calibration curve analysis. The interpretability of the model was demonstrated with SHapley Additive ExPlanations (SHAP).Resultsa total of 66263 elderly patients were selected, and in which 6299 patients (9.5%) were identified as acute delirium (within 7d after admission). Hospital mortality in delirium group was higher than that in non-delirium group (16.32% vs. 10.63%, p = 0.000). The cumulative survival probability of non-delirium patients were significantly higher than that of delirium patients (p < 0.001). When 20 variables were adopted, RandomForest and Xgboost models showed the highest predictive capability with the area under curve (AUC) = 0.91. Calibration curve analysis also proved this result. Glascow Coma Scale (GCS), acute physical and chronic health evaluation IV (APACHE IV), and sepsis had the highest importance in ML models. Mechanical ventilation and temperature were also important risk factors of acute delirium.ConclusionAcute delirium is an independent risk of mortality in elderly patients in the intensive care unit. APACHEIV, sepsis, mechanical ventilation and temperature were important risk factors of acute delirium, which were potential targets for medication. |
|---|---|
| ISSN: | 2196-1115 |
| DOI: | 10.1186/s40537-025-01107-8 |
| Fuente: | ABI/INFORM Global |