An Evolutionary Computation Approach for Optimizing Multilevel Data to Predict Patient Outcomes.

Area of Science:

• Health Informatics
• Computational Biology
• Biostatistics

Background:

• Electronic health records (EHR) adoption and meaningful use objectives drive demand for data-driven healthcare predictions.
• Large-scale, heterogeneous, and multilevel patient data present challenges for developing accurate predictive models.
• Optimizing data specificity is crucial for effective clinical decision support systems.

Purpose of the Study:

• To develop and evaluate a method for optimally specifying multilevel patient data for prediction tasks.
• To present a general evolutionary computational framework for optimizing data specificity in predictive modeling.
• To improve the prediction of individual patient outcomes in healthcare.

Main Methods:

• An evolutionary computational framework was developed to optimally specify multilevel patient data.
• The method was evaluated for both flattening data to a single level and retaining hierarchical structures.
• The approach was tested using data from emergency department patients across five populations to predict critical outcomes.

Main Results:

• Both flattened and hierarchical predictor structures significantly improved prediction performance compared to baseline single-level models (p < 0.001).
• The proposed framework for optimizing multilevel data specificity outperformed traditional single-level predictor structures.
• The developed method demonstrated enhanced accuracy in predicting critical outcomes for emergency department patients.

Conclusions:

• The evolutionary computational framework effectively optimizes the specificity of multilevel patient data for predictive modeling.
• This approach offers significant improvements over traditional methods, enhancing the accuracy of clinical decision support.
• The framework is adaptable to various healthcare prediction problems and other data-intensive domains.