Skip to main content
placeholder image

Multilevel modeling of geographic variation in general practice consultations

Journal Article


Abstract


  • Objective: To test relatively simple and complex models for examining model fit, higher-level variation in, and correlates of, GP consultations, where known nonhierarchical data structures are present. Setting: New South Wales (NSW), Australia. Design: Association between socioeconomic circumstances and geographic remoteness with GP consultation frequencies per participant was assessed using single-level, hierarchical, and multiple membership cross-classified (MMCC) models. Models were adjusted for age, gender, and a range of socioeconomic and demographic confounds. Data Collection/Extraction Methods: A total of 261,930 participants in the Sax Institute's 45 and Up Study were linked to all GP consultation records (Medicare Benefits Schedule; Department of Human Services) within 12¬†months of baseline (2006-2009). Principal Findings: Deviance information criterion values indicated the MMCC negative binomial regression was the best fitting model, relative to an MMCC Poisson equivalent and simpler hierarchical and single-level models. Between-area variances were relatively consistent across models, even when between GP variation was estimated. Lower rates of GP consultation outside of major cities were only observed once between-GP variation was assessed simultaneously with between-area variation in the MMCC models. Conclusions: Application of the MMCC model is necessary for estimation of variances and effect sizes in sources of big data on primary care in which complex nonhierarchical clustering by geographical area and GP is present.

Publication Date


  • 2021

Citation


  • Astell-Burt, T., Navakatikyan, M. A., Arnolda, L. F., & Feng, X. (2021). Multilevel modeling of geographic variation in general practice consultations. Health Services Research. doi:10.1111/1475-6773.13644

Scopus Eid


  • 2-s2.0-85102459229

Abstract


  • Objective: To test relatively simple and complex models for examining model fit, higher-level variation in, and correlates of, GP consultations, where known nonhierarchical data structures are present. Setting: New South Wales (NSW), Australia. Design: Association between socioeconomic circumstances and geographic remoteness with GP consultation frequencies per participant was assessed using single-level, hierarchical, and multiple membership cross-classified (MMCC) models. Models were adjusted for age, gender, and a range of socioeconomic and demographic confounds. Data Collection/Extraction Methods: A total of 261,930 participants in the Sax Institute's 45 and Up Study were linked to all GP consultation records (Medicare Benefits Schedule; Department of Human Services) within 12¬†months of baseline (2006-2009). Principal Findings: Deviance information criterion values indicated the MMCC negative binomial regression was the best fitting model, relative to an MMCC Poisson equivalent and simpler hierarchical and single-level models. Between-area variances were relatively consistent across models, even when between GP variation was estimated. Lower rates of GP consultation outside of major cities were only observed once between-GP variation was assessed simultaneously with between-area variation in the MMCC models. Conclusions: Application of the MMCC model is necessary for estimation of variances and effect sizes in sources of big data on primary care in which complex nonhierarchical clustering by geographical area and GP is present.

Publication Date


  • 2021

Citation


  • Astell-Burt, T., Navakatikyan, M. A., Arnolda, L. F., & Feng, X. (2021). Multilevel modeling of geographic variation in general practice consultations. Health Services Research. doi:10.1111/1475-6773.13644

Scopus Eid


  • 2-s2.0-85102459229