1. Analyzing the Geospatial Imbalance of the Primary Care Physician Labor Supply in the Contiguous United States By Russ Frith University of W. Florida Capstone Project in Statistics April, 2016

2. Abstract The national availability of primary care physicians remains steady or is declining in many areas of the nation. Even with greater reliance on nurse practitioners and physician assistants to provide primary care services to patients, geographic disparities with the primary care physician (PCP) labor supply persist. National projections of increased PCP supply mask problems that exist in specific geographic areas. Some areas benefit from an excess supply of PCPs well above the national average whereas other areas have a supply that is inadequate to meet the local needs for primary care services. Using publicly available data sets detailing county-based PCP supply, a detailed geostatistical analysis was done to characterize the adequacy of supply within the contiguous United States at the county level. Statistical measures were computed that showed PCP labor distribution aligned along income distribution and ethnic composition of geographical regions. Those measures are reported using both choropleth maps depicting clusters of underserved populations and conventional statistical procedures.

3. Problem Statement

4. Front Matter The Federal government agency responsible for monitoring and remediating the national PCP supply imbalance is the Health Resources and Services Administration, National Center (HRSA) for Health Workforce Analysis division of the Health and Human Services Administration. HRSA is the authoritative source for geographic data sets that identify PCP imbalance. This agency compiles data on PCP supply in states, counties, and census tracts. It makes determinations if a geographic unit has a PCP labor shortage. HRSA has developed a protocol for PCP shortage determination based on localized count data. Remediation grants and policies are formulated based on threshold counts typically tabulated at the county level. The HRSA protocol for shortage calculations acknowledges that ethnic and low-income populations are at higher risk of being underserved and that such demographic cohorts are ubiquitous in nearly all counties. Not taken into consideration through protocol however is the possibility that underserved populations are clustered. Furthermore, PCP supply shortage calculations are not adjusted for other demographic factors such as disease prevalence, educational attainment, and age. These protocols and assumptions have the potential for skewing the distribution of grants as well as the potential that qualified and eligible funding recipients are inadvertently excluded from such awards. In addition, incentives formulated to rectify PCP imbalance may not be effective in that they do not address significant indicators of PCP imbalance.

5. Is this random?

6. Data Exploration

7. Research Work Flow

8. Data Sources  County Health Rankings, http://www.countyhealthrankings.org/rankings/data http://www.countyhealthrankings.org/rankings/data  Center for Disease Control Wonderhttp://wonder.cdc.gov/)http://wonder.cdc.gov/)  United States Census Bureau, American Community Survey, 2012, http://api.census.gov/data/2012/acs5http://api.census.gov/data/2012/acs5  Centers for Medicare and Medicaid Services, http://www.cms.gov http://www.cms.gov Database: http://rfrith.uaa.alaska.edu/ProseminarData.php http://rfrith.uaa.alaska.edu/ProseminarData.php

9. Methodology Factor-SaTScan™ Analysis Ordinary Least Squares and Geographic Weighted Regression Getis-Ord or “Hot-Spot” Analysis Poisson-SaTScan™ Analysis http://www.satscan.org

10. Factor-SaTScan™ Analysis

11. Data Acquisition CDC CHR Medicaid Census Bureau Normal SaTScan™ Factor Scores Factor Analysis Demographic Cluster Ordinary Least Squares Regression Nonstationary? Generate Geographic Weighted Regression Rate Maps Y Getis-Ord Hot Spot Clustering Physician Cluster Intersection not empty? Y Spatially Intersect N Poisson SaTScan™, no covariates, purely spatial, PCP only Spatially Intersect Calibrate Intersection not empty? N 1.Add covariate i 2.Add covariate j 3.Add both covariates 4.Poisson SaTScan™ for PCP clusters Spatially Intersect Output statistically significant clusters for each covariate, accept H A

12. The “diabetes belt” revealed. Higher values imply less chance of random occurrence

17. Data Acquisition CDC CHR Medicaid Census Bureau Normal SaTScan™ Factor Scores Factor Analysis Demographic Cluster Ordinary Least Squares Regression Nonstationary? Generate Geographic Weighted Regression Rate Maps Y Getis-Ord Hot Spot Clustering Physician Cluster Intersection not empty? Y Spatially Intersect N Poisson SaTScan™, no covariates, purely spatial, PCP only Spatially Intersect Calibrate Intersection not empty? N 1.Add covariate i 2.Add covariate j 3.Add both covariates 4.Poisson SaTScan™ for PCP clusters Spatially Intersect Output statistically significant clusters for each covariate, accept H A

20. Nonstationary!

21. GWR Surfaces

22. Data Acquisition CDC CHR Medicaid Census Bureau Normal SaTScan™ Factor Scores Factor Analysis Demographic Cluster Ordinary Least Squares Regression Nonstationary? Generate Geographic Weighted Regression Rate Maps Y Getis-Ord Hot Spot Clustering Physician Cluster Intersection not empty? Y Spatially Intersect N Poisson SaTScan™, no covariates, purely spatial, PCP only Spatially Intersect Calibrate Intersection not empty? N 1.Add covariate i 2.Add covariate j 3.Add both covariates 4.Poisson SaTScan™ for PCP clusters Spatially Intersect Output statistically significant clusters for each covariate, accept H A

23. Hot-Spot Analysis

27. Data Acquisition CDC CHR Medicaid Census Bureau Normal SaTScan™ Factor Scores Factor Analysis Demographic Cluster Ordinary Least Squares Regression Nonstationary? Generate Geographic Weighted Regression Rate Maps Y Getis-Ord Hot Spot Clustering Physician Cluster Intersection not empty? Y Spatially Intersect N Poisson SaTScan™, no covariates, purely spatial, PCP only Spatially Intersect Calibrate Intersection not empty? N 1.Add covariate i 2.Add covariate j 3.Add both covariates 4.Poisson SaTScan™ for PCP clusters Spatially Intersect Output statistically significant clusters for each covariate, accept H A

28. The clusters are regions of statistically low rates of PCP supply.

29. The clusters are regions of statistically high rates of PCP supply.

30. Statistically significant clusters of under-served populations of primary care physicians after adjusting for both age and poverty.

31. Statistically significant clusters of under-served populations of primary care physicians after adjusting for both black and uninsured.

32. Conclusion: Reject the null hypothesis in favor of the alternative hypothesis with a significance level of 95% confidence. In 2012, the primary care physician imbalance was aligned along demographic strata. Implications? Targeted grants Infrastructure remediation ACA Enforcement Detection protocol change?

33. Questions?