Selection of regression models for health care data

Area of Science:

• Biostatistics
• Health Services Research
• Epidemiology

Background:

• Traditional research often favored linear regression models due to computational limitations.
• Advancements in computing power and algorithms now facilitate the use of nonlinear models.

Purpose of the Study:

• To evaluate the appropriateness of linear and nonlinear regression models for analyzing cohort study data on medical service utilization.
• To guide researchers in selecting regression models that best represent underlying phenomena.

Main Methods:

• Discussion of linear and nonlinear regression models.
• Application to a cohort study dataset on medical service use within a prepaid health plan.
• Emphasis on utilizing modern nonlinear minimization algorithms for model fitting.

Main Results:

• Nonlinear models are now computationally feasible and offer advantages over linear models.
• Model selection should prioritize the intrinsic properties of the phenomena being studied.

Conclusions:

• Researchers should strongly consider nonlinear regression models when they better represent the data and phenomena.
• Model evaluation should focus on descriptive accuracy and predictive validity for future trends.