Machine-learning model to predict the cause of death using a stacking ensemble method for observational data.

Area of Science:

• Computational biology and bioinformatics
• Medical informatics
• Public health research

Background:

• Access to cause-of-death data is crucial for clinical research but is often limited.
• Existing methods for determining cause of death can be restrictive.

Purpose of the Study:

• To develop and validate a machine-learning model for predicting cause of death from a patient's last medical checkup.
• To address the limitations in accessing cause-of-death data for research.

Main Methods:

• A stacking ensemble method was employed to classify mortality status and specific causes of death.
• Clinical data from national claims and electronic health records were used for model development and validation.
• Cause of death was imputed using data from multiple US claims databases, all formatted to the Observational Medical Outcomes Partnership Common Data Model.

Main Results:

• The model achieved a generalized area under the receiver operating characteristic curve (AUROC) of 0.9511 for predicting cause of death within 60 days.
• External validation demonstrated a strong AUROC of 0.8887.
• Analysis of imputed data revealed that 11.32% of deaths in the Medicare Supplemental database were attributed to malignant neoplastic disease.

Conclusions:

• Machine-learning models show significant potential as an alternative to overcome the challenges of accessing cause-of-death data.
• The developed model demonstrated competent performance and is easily applicable to other institutions.
• Transparency in all processes ensures the model's reliability and adaptability.