County-Level Variables
County-level variables were downloaded from the 2020 CHRD.5,6 An explanation of the CHRD model and the factors used in this study are shown in the eAppendix (available at doi: 10.12788/fp.0279).6 County-level aggregated HISA user data and the CHRD data were matched using county Federal Information Processing Standards codes. Access to care measures collected from CHRD included percentages uninsured and ratios of population to primary care physicians, dentists, mental health professionals, and other primary care professionals. Other CHRD measures included those for quality of care (rate of preventable hospital stay) and housing quality (percent of households with high housing costs, percent of households with overcrowding, percent of households with lack of kitchen or plumbing, percent of households with severe housing cost burden, percent of homeownership). Of secondary interest was county population rurality, as previous research findings showed the annual average of HISA users who are from rural areas ranged from 30 to 35%.
Analysis Methods
SAS (v9.4), R (v4.0.2), and RStudio (v1.3.1093) were used for data preparation and analysis.18 Multiple regression analysis was used to predict county-level utilization rate from county-level variables. Sociodemographic characteristics from CHRD and HISA data were included as important control predictors in the regression model, though our focus for this paper are the access to care and housing quality predictors.
Model diagnostics (examination of model residuals, Breusch-Godfrey test, Breusch-Pagan test) revealed significant heteroskedasticity of the model; thus, robust standard errors and associated P values were computed using the R estimatr package (v0.30.2).19 Some predictor variables of interest (eg, ratio of mental health professionals) were removed during the model building process either due to problems of multicollinearity or excessive missingness that would have resulted in listwise deletion.
Results
County-level HISA utilization rate per 1000 EOY VHA patients ranged from 0.09 to 59.7%, with a 6.6% mean and 5% median (Figure 2). The data were highly positively skewed. The final model included 33 total predictor variables (Table 1). The final regression model was a significantly better predictor of county-level HISA utilization rate than a null model (F[33-2184], 10.18; P < .001). The adjusted model R2 showed that the overall model accounted for approximately 23% of variance in county-level HISA utilization rate (Table 2).
Among the primary variables of interest, percent uninsured adults and rate of preventable hospital stays emerged as significant predictors of county-level HISA utilization rate. Specifically, county percentage of uninsured adults was negatively related to county-level HISA utilization rate (b = -8.99, P = .005), indicating that the higher the proportion of uninsured adults—with all other predictors held constant—the lower the HISA utilization rate. Percent uninsured adults ranged from 2.7 to 42.4% across counties, with a mean (SD) of 12.7% (5.8%) and 11.4% median.
County rate of preventable hospital stays, however, was significantly and positively related to county-level HISA utilization rate (b = .0004, P = .009), indicating that the higher the rate of preventable hospital stays—with all other predictors held constant—the higher the HISA utilization rate. The direction of this effect is the opposite of the direction of the effect of percent uninsured adults (positive rather than negative), even though both could be considered measures of access to care. The standardized β for these 2 predictors indicate that county rate of preventable hospital stays is a somewhat stronger predictor of county-level HISA utilization rate than is county percent of uninsured adults (β = .11 and β = -.09, respectively). Rate of preventable hospital stays ranged from 683 to 16,802 across counties included in this analysis, with a mean (SD) of 4,796.5 (1659.9) and a 4669 median.
Of secondary interest was county rurality. The county-level percentage of rural residents was significantly and positively related to county-level HISA utilization rate, indicating that the higher the proportion of individuals within county considered rural—all other predictors held constant—the higher the HISA utilization rate. The mean (SD) percentage of rural residents per county was 52.3% (30.2) and 52.7 % median.