Dimensionality reduction for knowledge discovery in medical claims database: application to antidepressant medication...

Area of Science:

• Health Informatics
• Data Mining
• Biostatistics

Background:

• Healthcare costs are rising, necessitating efficient data analysis for cost savings.
• Feature selection is crucial for reducing dimensionality in large medical databases.
• Traditional stepwise regression often yields excessive variables, impacting efficiency.

Purpose of the Study:

• To apply a filter-based feature selection method to a medical claims database.
• To predict the adequacy of antidepressant medication utilization.
• To compare the efficiency of filter-based methods against traditional stepwise regression.

Main Methods:

• Utilized a filter-based feature selection approach employing inconsistency rate measure and discretization.
• Applied the method to a medical claims database.
• Compared results with traditional stepwise logistic regression.

Main Results:

• The filter-based method selected only two key variables (age, number of claims).
• Stepwise logistic regression identified seven variables for predicting inadequate antidepressant use.
• Both methods achieved similar prediction accuracy, demonstrating the filter-based method's efficiency.

Conclusions:

• Filter-based feature selection is a feasible and efficient method for dimensionality reduction in healthcare databases.
• This approach can streamline the analysis of medical claims data.
• Potential for significant cost savings and improved decision-making in the healthcare industry.