The predictive power of hemodynamic data on postoperative neurocognitive impairment: a logistic regression and random forest approach

This study assesses hemodynamic data and parameter combinations in predicting neurocognitive impairment post-cardiopulmonary bypass graft (CABG) using logistic regression and random forest algorithms. 28 patients underwent the Montreal Cognitive Assessment (MoCA) test preoperatively and one month postoperatively. Patients were grouped by MoCA score changes: Group 1 (< 2 points decrease) and Group 2 (>= 2 points decrease). Real-time hemodynamic data were recorded during surgery, and after artifact removal, a large dataset was analyzed. Derived parameters included Absolute Maximum Decrease (AMD), areas under thresholds, and duration spent below thresholds. Logistic regression and Random Forest algorithms assessed individual and combined parameter effects. Partial Dependence Plots (PDPs) aided interpretability. Results: Logistic regression and Random Forest analyses indicated hemodynamic data have limited predictive power for neurocognitive impairment. No logistic regression analysis yielded statistically significant results, and no Random Forest model achieved high accuracy. Conclusion: Hemodynamic data alone are insufficient for prediction. Including cerebral oxygen saturation, micro emboli, and hematocrit may improve model performance. Larger sample sizes and long-term follow-up are recommended for better accuracy. This study provides a basis for future research to mitigate postoperative cognitive dysfunction.