Explainable ML Model for COVID-19 Mortality Risk in Older Adults Using Routine Blood Tests

Lin Luo's research dataset underpins a machine learning model for predicting mortality risk in older adults with COVID-19. The model was developed using routine hematological indicators from 2393 patients and optimized via Bayesian optimization. The Light Gradient Boosting Machine (LGBM) model achieved an AUC of 0.973, and the study identifies top 10 key biomarkers like CRP and D-dimer.

Use Cases

• Training mortality risk prediction models based on routine hematological indicators like CRP and D-dimer.
• Analyzing non-linear associations and threshold effects of biomarkers on clinical outcomes.
• Developing simplified, interpretable clinical risk scoring systems using SHAP analysis.
• Benchmarking machine learning algorithms (LR, SVM, RF, XGBoost, LGBM) for clinical tabular data.

Strengths

• Model performance is rigorously evaluated, with the LGBM model achieving an AUC of 0.973 and an accuracy of 0.918 on a held-out test set.
• The study uses a substantial cohort of 2393 COVID-19 patients and employs a detailed methodology including Bayesian optimization and SHAP analysis.
• A simplified model using the top 10 features reduced training time by 58.31% without compromising performance, enhancing practical utility.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the underlying data is unknown, which may limit suitability assessment.
• The dataset is very small at 41.7 KB, indicating it likely contains model documentation or summary results rather than the full patient-level data.

Provenance

Source

Lin Luo, via figshare.

Collection Method

Retrospective study of patient records; missing values were imputed via Random Forest.

Freshness

Last updated 2026-05-25 06:03:38; freshness should be verified.