Foundations of Machine Learning-Based Clinical Prediction Modeling: Part IV-A Practical Approach to Binary...

Area of Science:

• Machine Learning
• Clinical Prediction
• Biostatistics

Background:

• Clinical prediction models are essential for patient outcomes.
• Developing and validating these models requires robust methodologies.
• The R statistical programming language offers tools for such development.

Purpose of the Study:

• To provide a practical guide for training and validating machine learning clinical prediction models in R.
• To illustrate these steps using a simulated glioblastoma patient dataset for 12-month survival prediction.
• To detail essential preprocessing, training, validation, and reporting techniques.

Main Methods:

• Dataset import, checking, and splitting.
• K-nearest neighbor imputation and recursive feature elimination for preprocessing.
• Bootstrapping for model evaluation and selection based on out-of-sample error.
• Upsampling techniques to address class imbalance in classification.
• Reporting key performance metrics: accuracy, AUC, sensitivity, specificity, slope, and intercept.

Main Results:

• A functional machine learning model for predicting 12-month glioblastoma survival was developed.
• The process included practical implementation of imputation and feature selection.
• Model validation was performed using bootstrapping and out-of-sample error assessment.
• Variable importance was determined using an AUC-based method.

Conclusions:

• The study provides a comprehensive workflow for building and validating clinical prediction models in R.
• The presented methods are applicable to various binary clinical outcomes.
• Availability of code and data facilitates reproducibility and further research in neuro-oncology prediction.