To develop an accurate, efficient, clinically interpretable machine learning model for predicting mortality risk in elderly COVID-19 patients using routine hematological indicators at admission, thereby avoiding extra medical costs and radiation exposure.
Key Findings:
The LGBM model achieved the best performance with AUC of 0.973, recall of 0.924, accuracy of 0.918, and F1-score of 0.918. Non-linear associations and threshold effects were particularly noted with CRP and D-dimer levels.
Interpretation:
The TPE-LGBM model demonstrated favorable accuracy, efficiency, and interpretability, suggesting its potential application in clinical settings for mortality risk assessment in elderly COVID-19 patients, which could enhance patient management.
Limitations:
The study may be limited by its retrospective design, potential biases in data collection, and the generalizability of findings to broader populations.
Conclusion:
The study highlights the value of using explainable machine learning to address unmet medical needs in predicting mortality risk among older adults with COVID-19.
