The Not-So-Distant Future or Just Hype? Utilizing Machine Learning to Predict 30-Day Post-Operative Complications in...

Area of Science:

• Surgical outcomes research
• Clinical informatics
• Predictive modeling in healthcare

Background:

• Machine learning (ML) models offer advanced predictive capabilities in clinical settings.
• The predictive performance of ML for laparoscopic colectomy (LC) morbidity remains under-explored and uncompared to traditional methods.
• Logistic regression (LR) is a standard statistical approach for clinical outcome prediction.

Purpose of the Study:

• To compare the predictive accuracy of ML models against LR for post-operative morbidity in LC patients.
• To evaluate the performance of specific ML algorithms including Random Forests, XGBoost, and L1-L2-RFE.

Main Methods:

• Analysis of laparoscopic colectomy (LC) patient data from the National Surgical Quality Improvement Program (NSQIP) between 2017-2019.
• Definition of post-operative morbidity using a composite outcome of 17 variables.
• Comparison of three ML models (Random Forests, XGBoost, L1-L2-RFE) with logistic regression (LR).

Main Results:

• ML models (Random Forests, XGBoost, L1-L2-RFE) and LR demonstrated comparable predictive performance for 30-day post-operative morbidity, with Area Under the Curve (AUC) values around 0.71.
• All models showed similar, high predictive accuracy (AUC ≤ 0.9) for specific complications like septic shock.
• No significant difference was observed in the predictive capabilities between ML and LR.

Conclusions:

• Machine learning (ML) and logistic regression (LR) exhibit negligible differences in predicting post-operative morbidity following laparoscopic colectomy (LC).
• The potential advantages of ML computational power may not be fully realized with limited dataset sizes.
• Further research with larger datasets may be needed to ascertain the superior utility of ML in this context.