Mortality prediction after major surgery in a mixed population through machine learning: a multi-objective symbolic...

Area of Science:

• Medical research
• Machine learning applications in healthcare
• Surgical outcomes analysis

Background:

• 1-year mortality after major surgery is a critical indicator of patient outcomes and peri-operative care quality.
• Existing models for predicting 1-year mortality are limited in accuracy.
• Complex non-cardiac surgery patients require robust risk stratification tools.

Purpose of the Study:

• To develop a novel predictive model for 1-year mortality in patients undergoing complex non-cardiac surgery.
• To utilize multi-objective symbolic regression, a machine learning technique, for enhanced predictive accuracy.
• To compare the performance of the new model against existing mortality prediction models.

Main Methods:

• A single-institution database of patients with prior cardiopulmonary exercise testing was utilized.
• Data were segmented into pre-operative clinical, cardiorespiratory/physiological, and combined datasets.
• A multi-objective symbolic regression model was developed and validated using the F1 score; Shapley additive explanations identified key predictors.

Main Results:

• The study included 1190 patients (71 years median age, 69% male) from a database of 2145.
• The multi-objective symbolic regression model achieved a robust F1 score of 0.712.
• Key predictors identified were ventilatory equivalents for carbon dioxide, oxygen at peak exercise, and BMI, outperforming surgery type and comorbidities.

Conclusions:

• A multi-objective symbolic regression model can effectively predict 1-year postoperative mortality in a mixed non-cardiac surgical population.
• Cardiorespiratory fitness and physiological data are crucial for accurate surgical risk assessment.
• The model's high sensitivity and F1 score indicate its potential as a valuable tool for peri-operative risk prediction and patient optimization.