Multilevel modeling of geographic variation in general practice consultations

Area of Science:

• Primary Care Research
• Biostatistics
• Health Services Research

Background:

• General practitioner (GP) consultations are a key indicator of primary care utilization.
• Understanding factors influencing GP consultation frequency is crucial for health service planning.
• Complex data structures, including geographical clustering and individual GP practice, often exist in primary care datasets.

Purpose of the Study:

• To evaluate and compare the performance of different statistical models for analyzing GP consultation data.
• To assess the impact of socioeconomic factors and geographic remoteness on GP consultation frequencies.
• To determine the best-fitting model for data with non-hierarchical structures, such as geographical areas and GP practices.

Main Methods:

• Utilized data from the Sax Institute's 45 and Up Study, linking 261,930 participants to GP consultation records.
• Employed single-level, hierarchical, and multiple membership cross-classified (MMCC) negative binomial and Poisson regression models.
• Adjusted models for age, gender, socioeconomic status, and demographic variables.

Main Results:

• The MMCC negative binomial regression model demonstrated superior fit compared to simpler models and an MMCC Poisson model.
• Between-area variances were consistent across models, even when accounting for between-GP variation.
• Lower GP consultation rates in non-metropolitan areas were only apparent when simultaneously considering between-GP and between-area variations within the MMCC framework.

Conclusions:

• The MMCC model is necessary for accurately estimating variances and effect sizes in large primary care datasets with complex, non-hierarchical clustering.
• Accurate modeling requires simultaneous consideration of geographical areas and individual general practitioners.
• This approach enhances the understanding of primary care utilization patterns and associated factors.