Subgroup State Prediction under Different Noise Levels Using MODWT and XGBoost.

Area of Science:

• Medical informatics
• Computational biology
• Biostatistics

Background:

• Traditional medical state prediction assumes identical data distributions across individuals.
• Precision medicine demands accurate subgroup analysis for personalized healthcare.
• Existing methods may lack accuracy in diverse patient subgroups.

Purpose of the Study:

• To propose and evaluate an aggregated method for medical state prediction.
• To enhance subgroup analysis accuracy in precision medicine.
• To compare the aggregated method against the original approach across various denoising levels.

Main Methods:

• Developed an aggregated method by combining results from different subgroup models.
• Compared the aggregated method with the original method.
• Assessed performance using metrics like Area Under the Curve (AUC), F1-score, and sensitivity.
• Analyzed variable importance at different denoising levels.

Main Results:

• The aggregated method demonstrated superior performance, achieving an AUC of 0.95, F1-score of 0.87, and sensitivity of 0.82.
• Performance improvements were particularly notable at a denoising level of 2.
• Variable importance analysis revealed increased significance of factors like heart rate and arterial lactate with higher denoising levels.

Conclusions:

• The proposed aggregated method is effective for medical state prediction, offering enhanced accuracy in subgroup analysis.
• The method shows significant promise for precision medicine applications.
• Denoising levels critically influence prediction performance and variable importance.