How machine learning facilitates decision making in emergency departments: Modelling diagnostic test orders

Area of Science:

• Emergency Medicine
• Health Informatics
• Predictive Analytics

Background:

• Emergency departments (EDs) face challenges with patient delays, often exacerbated by diagnostic test ordering.
• Efficient decision-making in EDs is crucial for managing patient flow and care.

Purpose of the Study:

• To identify factors influencing diagnostic test orders in EDs.
• To develop an efficient predictive model for laboratory and imaging test frequencies.

Main Methods:

• Linear regression models were used to analyze daily test frequencies based on time variables and ICD-10 codes.
• Machine learning techniques were integrated with regression models to create hybrid models for improved prediction.
• Input variables included day of the week and patient counts across various ICD-10 diagnostic categories.

Main Results:

• Day of the week and specific ICD-10 codes (e.g., 'A00-B99', 'I00-I99') were significant predictors for both laboratory and imaging tests.
• The hybrid model achieved high prediction accuracies, reaching 99.997% for laboratory and 99.995% for imaging services.
• Specific ICD-10 codes showed varying significance for laboratory versus imaging service utilization.

Conclusions:

• Identified factors can predict increased demand for laboratory and imaging services, enabling proactive resource allocation.
• The proposed hybrid model can effectively reduce ED patient delays and mitigate overcrowding.
• This predictive model offers a valuable tool for enhancing ED operational decision-making.