Using Machine Learning Algorithms to Predict Hospital Acquired Thrombocytopenia after Operation in the Intensive Care...

Area of Science:

• Medical Informatics
• Clinical Prediction Models
• Hematology

Background:

• Hospital-acquired thrombocytopenia (HAT) is a frequent postsurgical complication.
• Accurate prediction of HAT risk is crucial for timely intervention and improved patient outcomes.

Purpose of the Study:

• To develop and evaluate seven machine learning (ML) algorithms for predicting HAT risk in intensive care unit (ICU) patients post-surgery.
• To compare the predictive performance of different ML models using metrics like AUC and DCA.

Main Methods:

• A retrospective cohort study included 10,369 adult patients transferred to the ICU post-surgery.
• Patients were divided into derivation (70%) and test (30%) sets.
• Ten-fold cross-validation was used for hyperparameter tuning, and model performance was assessed using sensitivity, specificity, AUC, and DCA.

Main Results:

• HAT occurred in 13.1% of patients (1354/10,369).
• All seven ML models achieved an AUC > 0.7.
• Gradient Boosting (AUC=0.834) and Random Forest (AUC=0.828) demonstrated the highest predictive performance, with no significant difference between them.
• All models showed high net benefits on decision analysis curves.

Conclusions:

• ML models utilizing preoperative variables can effectively predict HAT in postsurgical ICU patients.
• These models offer a valuable tool for enhancing risk stratification and guiding clinical management.
• The findings support the integration of ML-based prediction tools into routine clinical practice for better patient care.