Key result
A full connected layer model with self-attention yielded the lowest mean absolute error (0.9388 in test cohort) for predicting hourly heparin dosage in cardiac surgery patients.
Why the study?
Anticoagulant therapy after cardiac surgery is challenging with risk of fatal complications, and previous studies failed to provide real-time guidance for heparin pump speed adjustment.
Does an FC + self-attention machine learning model accurately predict hourly heparin dosage to reach target aPTT in post-cardiac surgery ICU patients?
Observational (n=1,080)
No
Does an FC + self-attention machine learning model accurately predict hourly heparin dosage to reach target aPTT in post-cardiac surgery ICU patients?
A machine learning model using a full connected layer with self-attention can accurately predict hourly heparin dosage adjustments to maintain target aPTT in post-cardiac surgery patients.
ML model may optimize post-cardiac surgery heparin dosing; leaves open prospective validation before clinical adoption.
BACKGROUND: Anticoagulant therapy for patients who underwent cardiac surgery is a challenge. Both inadequate and excessive anticoagulation can cause fatal complications. Previous studies failed to provide real-time guidance for heparin pump speed adjustment. This study intended to provide a quantitative prediction model to optimize heparin dosage for cardiac surgery patients based on machine learning method. METHODS: Patients who underwent cardiac surgery and admitted to intensive care unit in Peking Union Medical College Hospital (PUMCH) from January 2013 to December 2023 were retrospectively analyzed. In order to reach target activated partial thromboplastin time (aPTT), linear regression, SVM, XGBoost, LSTM, GRU, FC (Full Connected Layer) and FC + self-attention models were used to make hourly adjustment recommendation for administrations of heparin pump speed. Mean absolute square, and absolute percentage errors were used to evaluate the reliability of the models. SHAP method and feature cumulative effect were used to interpret the features of the FC + self-attention model. Safety and economic evaluation based on clinical compliance of this real-world data-oriented model was further analyzed. RESULTS: A total of 1080 patients including 112,554 heparin pump administrations were included in this study. Among seven candidate models, FC + self-attention model yielded the lowest mean absolute error of 0.9388 and 1.1325 in test and validation cohort. Gap to target aPTT, thrombin time, history of coronary heart disease, previous duration of arterial fibrillation and prothrombin activity were identified as important features affecting heparin adjustment. High compliance to FC + self-attention model may increase percentage of normal therapeutic time and decrease supratherapeutic therapeutic time and reducing blood draw until two consecutive normal therapeutic stabilization of aPTT. CONCLUSIONS: This FC + self-attention model is potentially applicable for giving recommendation for healthcare providers to optimize heparin dosage for cardiac surgery patients.
No takes yet. Share an insight, caveat, or question.
Liu et al. (2024) conducted an observational in Cardiac surgery (n=1,080). FC + self-attention model vs. Other machine learning models was evaluated on Mean absolute error. A full connected layer model with self-attention yielded the lowest mean absolute error (0.9388 in test cohort) for predicting hourly heparin dosage in cardiac surgery patients.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: