Muscle-related parameters-based machine learning model for predicting postinduction hypotension in patients...

Area of Science:

• Anesthesiology
• Medical Informatics
• Surgical Oncology

Background:

• Postinduction hypotension (PIH) is a common complication in patients undergoing colorectal tumor resection under general anesthesia.
• Identifying predictive factors for PIH is crucial for improving patient outcomes and anesthetic management.

Purpose of the Study:

• To develop and validate a machine learning model for predicting PIH in patients undergoing colorectal tumor resection.
• To identify key clinical and demographic variables associated with PIH.

Main Methods:

• Utilized machine learning algorithms, including Boruta for feature selection and Random Forest (RF) for prediction, on data from 318 patients.
• Employed repeated cross-validation and hyperparameter optimization to build and select the best predictive model.
• Evaluated model performance using R², Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE).

Main Results:

• Key predictive variables identified by Boruta screening included age, sex, body mass index, L3 skeletal muscle index, and HUAC.
• The optimal RF model achieved high performance, with R² values of 0.7708 (training) and 0.7591 (test).
• The model demonstrated strong predictive accuracy with low error rates (e.g., MAE of 0.0408 on the test set).

Conclusions:

• A high-performance machine learning algorithm was successfully developed and validated for PIH prediction.
• The model aids in controlling important characteristic variables, thereby reducing the occurrence of PIH in surgical patients.
• This approach offers a valuable tool for proactive management of blood pressure during anesthesia induction.