Development of a machine learning-based risk prediction model for cerebral infarction and comparison with nomogram...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Big Data Analytics

Background:

• Cerebral infarction (CI) risk prediction is crucial for early intervention.
• Routine blood test data offers a rich resource for developing predictive models.
• Machine learning algorithms can effectively mine large datasets for complex patterns.

Purpose of the Study:

• To develop and validate a machine learning-based risk prediction model for cerebral infarction (CI).
• To compare the performance of the developed model against traditional nomogram models.
• To assess the model's efficacy across different age groups.

Main Methods:

• Utilized Extreme gradient Boosting (XgBoost), Logistic Regression, Support Vector Machine, and Random Forest algorithms.
• Trained and validated models using routine test data from three distinct cohorts (2017-2021).
• Compared the performance of the optimal machine learning model (CI-Lab8) with a nomogram model using AUC, calibration, and decision curve analyses.

Main Results:

• The XgBoost-derived CI-Lab8 model, incorporating eight key features, achieved an AUC of 0.823 in cohort 2.
• CI-Lab8 demonstrated superior performance compared to the Fibrinogen (FIB) marker (AUC=0.737) and the nomogram model.
• The model showed high diagnostic accuracy in both younger (<50 years, AUC=0.800) and older (≥50 years, AUC=0.856) CI patient subgroups.

Conclusions:

• The CI-Lab8 model, developed using XgBoost, offers superior diagnostic accuracy for cerebral infarction compared to nomogram models.
• This machine learning model effectively predicts CI risk across different age groups.
• The CI-Lab8 model shows potential to enhance clinical decision-making for CI assessment.