Modeling risk using generalized linear models

Area of Science:

• Biostatistics
• Health Economics
• Medical Informatics

Background:

• Linear regression is the standard for medical risk prediction.
• Two-part models are commonly used for healthcare utilization and cost data.
• Limitations exist in traditional methods for modeling complex healthcare data.

Purpose of the Study:

• To present novel extensions of generalized linear models for the second part of two-part models.
• To improve the accuracy and flexibility of medical risk prediction.
• To provide an alternative to traditional linear regression in healthcare modeling.

Main Methods:

• Utilizing extensions of the generalized linear model (GLM).
• Employing maximum likelihood estimation (MLE) as the primary estimation method.
• Discussing quasi-likelihood and extended quasi-likelihood generalizations.

Main Results:

• The proposed GLM extensions effectively model the second part of two-part models.
• Demonstrated application using medical expense data from Washington State employees.
• Incorporated demographic variables and Ambulatory Care Group for enhanced prediction.

Conclusions:

• The extended GLM approach offers a powerful alternative for medical risk prediction.
• This methodology enhances the modeling of healthcare utilization and costs.
• The findings have implications for health economics and biostatistical modeling.