Improving Hospital-Wide Early Resource Allocation through Machine Learning.

Area of Science:

• Health Services Research
• Health Informatics
• Operations Research

Background:

• Accurate determination of diagnosis-related groups (DRGs) is crucial for hospital resource allocation.
• Current DRG assignment occurs at discharge, limiting proactive resource management for elective admissions.

Purpose of the Study:

• To evaluate the effectiveness of early diagnosis-related group (DRG) determination for optimizing scarce hospital resource allocation.
• To compare a Naïve Bayes classifier's accuracy against existing methods for pre-admission DRG classification.

Main Methods:

• A three-stage approach was implemented: (1) Naïve Bayes classification accuracy assessment, (2) development of a statistical program for admission/scheduling decisions, and (3) simulation evaluating a hybrid classification and mathematical programming model.
• The study compared the Naïve Bayes approach with the hospital's baseline method for DRG classification prior to patient admission.

Main Results:

• The standard DRG grouper demonstrated poor pre-admission classification accuracy.
• The Naïve Bayes classifier significantly enhanced the accuracy of DRG classification before patient admission.
• Integrating the Naïve Bayes classifier with the mathematical programming model led to more effective and efficient resource allocation decisions.

Conclusions:

• Early DRG determination, particularly using machine learning methods like Naïve Bayes, is feasible and beneficial.
• A hybrid approach combining advanced classification techniques with mathematical programming can substantially improve hospital resource management and efficiency.