Continuous imputation of blood gas and metabolic panel laboratory values in the intensive care unit using machine learning
Authors' conclusion
Our real-time imputation models led to significant error reduction of most ABG and BMP targets, improving deep learning model performance for the need of RRT.
Study context
- Population
- Critically ill ICU patients
- Comparison
- Machine learning (ML)-based hourly laboratory imputation vs previous-value imputation
- Outcome
- Prediction of need for renal replacement therapy (RRT) (Bi-LSTM models)
Methodological detailsXGBoost imputation, Bi-LSTM prediction models
- XGBoost imputation
- Bi-LSTM prediction models
Main finding
As reported in the abstract
| Group | AUROC | Area under the precision-recall curve (AUPRC) |
|---|---|---|
| ML-based hourly laboratory estimates | 0.951 (95% CI 0.942-0.960) | 0.419 (95% CI 0.375-0.465) |
| Previous-value laboratory imputation | 0.923 (95% CI 0.910-0.935) | 0.343 (95% CI 0.301-0.386) |
AUROC
ML-based hourly laboratory estimates0.951 (95% CI 0.942-0.960)
Previous-value laboratory imputation0.923 (95% CI 0.910-0.935)
Area under the precision-recall curve (AUPRC)
ML-based hourly laboratory estimates0.419 (95% CI 0.375-0.465)
Previous-value laboratory imputation0.343 (95% CI 0.301-0.386)
Further findings
Accuracy of laboratory imputation (XGBoost vs previous value)
- Average error reduction in mean absolute error (MAE), arterial blood gas (ABG) variables
- 33%
- Average error reduction in root mean squared error (RMSE), ABG variables
- 32%
- Average error reduction in MAE, basic metabolic panel (BMP) variables
- 19%
- Average error reduction in RMSE, BMP variables
- 21%
Original publication
Continuous imputation of blood gas and metabolic panel laboratory values in the intensive care unit using machine learning. J Clin Monit Comput. Published 23 July 2026. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →