Sage Journals: Discover world-class research

Abstract

Infant mortality rates (IMR) and other health disparities (e.g., low provider access; higher obesity rates) exist across rural America, especially in the southeastern United States. These disparities coincide with negative socioeconomic factors such as poverty and low household income. Georgia’s 2020 preterm birth rate was seventh highest among the 50 U.S. states, its rate of low birth weight (LBW) babies was fourth among states.

Methods

This study used a novel, combined spatiotemporal analysis to assess IMR and health trends in Georgia counties (n = 159). We spatially regressed IMR on 14 biopsychosocial health variables and assessed IMR 2012–2022 longitudinal trends, rural versus urban or by Health Professional Shortage Area (HPSA) status, using a two-level hierarchical linear model.

Results

Analyses demonstrated significant associations between IMR and rural counties as well as counties that have high African American populations, high unemployment, and high uninsurance rates. Principal Care Provider Rates showed a negative spatial regression relationship. There was a −0.056 IMR slope for urban counties from 2015 to 2022, whereas the rural county IMR slope was 0.135.

Conclusions

Findings concur with previous research suggesting the need for coordinated county-specific socioeconomic development and corresponding increased health programs. Combined geospatial and multilevel models represent a novel approach to inform public health epidemiology.

Keywords

infant mortality geospatial regression hierarchical linear models rural health Georgia,USA

Background

In 2020, the leading causes of death among U.S. Georgians were (a) heart disease (183.7 per 100,000 people); (b) cancer (147.6); (c) COVID-19 (138.8); (d) accidents (50.8); (e) Alzheimer’s Disease (45.9); (f) stroke (43); (g) chronic lower respiratory diseases (41.6); and (h) diabetes (23.9).^1,2 These rankings closely mirror corresponding 2020 mortality rankings for the United States.³

Of particular concern are U.S. infant mortality statistics that lag behind many other developed nations.⁴ Georgia’s 2020 preterm birth rate was 11.4 per 1000 births, seventh highest among the 50 states, and low birth weight (LBW) was 9.9, fourth among states.^2,5 The infant mortality rate (IMR) was 6.28, sixteenth among states.^2,4–6

Factors impacting IMR

When looking at disparities in IMR, African American IMR have improved but remain approximately twice that of non-Hispanic Caucasians nationally.⁷ Regionally, the Great Lakes states (Illinois, Indiana, Michigan, Ohio, Wisconsin) have the highest regional African-American IMR = 13.77, whereas the New England states were lowest at IMR = 8.78. In this particular study, Kandasamy et al.⁷ identified higher African American marriage rates and higher state Maternal and Child Health (MCH) budgets as protective factors against IMR. Furthermore, IMR was significantly higher in counties that are rural and/or experience high poverty.⁸

Bhatt and Beck-Sagué⁹ demonstrated that 2014–2016 infant mortality rates declined in Affordable Care Act (ACA) Medicaid expansion states but increased in the nineteen non-expansion states. Ten of the nineteen Medicaid non-expansion states, including Georgia, were in the U.S. South, which generally has higher IMR rates. As of November, 2022, 11 states had not participated in Medicaid expansion, including Georgia and seven additional Southern states.¹⁰ In an earlier 2007–2009 study of the Period Linked Birth-Infant Death File, Hirai et al.¹¹ recommended state-centered health promotion strategies given variations in causes of IMR across Southern states. Current IMRs are being targeted by Healthy People 2030 national objectives, including reductions in fetal, infant, and children deaths.¹²

Yang and McManus¹³ applied geospatial analyses to examine IMR and other health/social factors in the 159 Georgia counties. They identified hotspot counties of high IMR in varied clusters that generally concentrated across mostly rural counties in the eastern middle portion of the state, extending westward to the Southwestern counties while including a few Atlantic coast counties. These clusters followed the overall Piedmont/coastal plain geographical characteristics of low socioeconomic conditions, IMR, and other poor health outcomes across the Southeastern United States.¹⁴ Clustering of specific variable frequencies has been a characteristic of complex systems, including physical phenomena and biological systems.^15,16 Likewise, Arcaya et al.¹⁷ used a two-level geospatial model to a combination of within-state and overall spatial processes that impact U.S. county-level longevity rates.

Ehrenthal, Kuo, and Kirby¹⁸ examined the 2014–2016 NCHS Period Linked Birth-Infant Death Files, demonstrating significantly higher odds ratios for neonatal, post-neonatal, and infant deaths in noncore rural and micropolitan counties compared to fringe and central metropolitan U.S. counties.

The spatiotemporal analysis of health trends is of extreme importance for better understanding the biopsychosocial forces that impact individual and population health. The study uses 11 consecutive years of county-level population health data, with longitudinal time and spatial autocorrelation/regression along two analytical pathways: (a) geospatial regression^19,20; and (b) hierarchical linear models.^21,22

Spatial analysis

When examining event data, spatial lattice data, and other geographic data, it is important to examine the effects of neighboring geographic entities in various combinations for a particular observed variable. Correlated geographic effects between neighboring counties can inflate significance and the regression coefficient of determination when Ordinary Least Squares (OLS) regression is used.

Anselin and Bera¹⁹ noted the importance of spatial dependence in economic statistics and modeling because of high/low value regional clustering and neighbor similarities/dissimilarities. Fotheringham and Brunsdon²⁰ also noted scaling differences in statistical reporting, with the realization of local variations in geographic clusters with finer spatial divisions even when there were similarities in spatial trends.^20,23–25

Besides the issues of spatial autocorrelation and heterogeneity, Anselin²⁶ focused on the importance of proper model specification, including latent variables. Earlier, Anselin²⁷ had also emphasized the usefulness of Local Indicators of Spatial Association (LISA) for the identification of geographic hotspots that strengthen overall global spatial statistics, including Moran’s I statistic. Furthermore, Wheeler and Tiefelsdorf²⁸ stressed the importance of proper model weighting and assessments of regression coefficient multicollinearity to avoid biased models.

Temporal analysis

Longitudinally, Bryk and Raudenbush²¹ identified several major problems in social science regression models: (a) variability not just within individuals but between individuals across different, nested groups/treatments at multiple levels; (b) variability within individuals over time; and (c) correlation of regression slopes and intercepts as well as errors. The result was the development of multilevel models, including hierarchical linear models that provided improved estimation and prediction.²¹

Similarly, Singer and Willett²² stressed the importance of multilevel models that: (a) adjust for and center time; (b) address longitudinal change within persons and between persons across layers of nested groupings; and (c) address the limited reliability of traditional OLS regression estimates in complex models, like spatial regression.

Therefore the objectives of this study were to:

1. Spatially map and regress the 2022 Georgia IMR on specific health and biopsychosocial variables as described above in the research literature; and

2. Temporally/longitudinally assess 11 year trends (2012–2022) in IMR and these associated health and biopsychosocial variables, comparing within and between 159 counties nested within rural/urban or low/medium/high HPSA (i.e. Health Professions Shortage Areas).

Methods

This study used GeoDa version 1.18.0.16 (spatial.uchicago.edu/software) for geospatial mapping/statistics and SAS Proc Mixed²⁹ for hierarchical linear models. The data source was the 2012–2022 annual County Health Rankings and Roadmaps (https://countyhealthrankings.org) compiled by the University of Wisconsin Population Health Institute from nationally-representative data sources and from county and state health department reported data. Annual data were merged using county FIPS codes.

Due to some variations in data capture, plus minimizing missing data, across the 11 data collection periods (2012–2022), we used the following county-level variables: (a) Infant Mortality Rate; (b) Percent Low Birth Rate (LBR); (c) Teen Birth Rate (TBR); (d) Percent Adults Obese; (e) Percent Adults Physically Inactive; (f) Percent Adults with Diabetes; (g) Percent Women screened Mammography; (h) Primary Care Physician (PCP) Rate; (i) Percent attended College; (j) Percent Single Parent Households (SPH); (k) Percent Unemployed; (l) Percent Uninsured; (m) Percent Black/African American; (n) Percent Rural; and (o) Violent Crime Rate (VCR). For some variables, missing data in certain years required longitudinal analysis adjustment to shorter time ranges (e.g., 2015–2022). All variables were freely available from the County Health Data and Roadmaps website.

For the outcome variable Infant Mortality Rate, n = 79 rural counties did not report data across the 2012–2022 longitudinal period, thereby limiting geospatial and HLM models in this study to n = 80 Georgia counties, of which n = 37 were urban and n = 43 were rural. Across the reporting counties, IMR ranged between 3.41 and 14.46 infant deaths per 1000 births. All Georgia counties (n = 159) reported the other variables, with the exception of PCP rate (n = 144) and Violent Crime Rate (n = 152).

Statistical models

Geospatial regression

For the single year (2022) geospatial analysis, each variable was mapped by scatterplot onto Infant Mortality Rate, including the Local Moran’s I and EB rates. A multiple regression analysis was conducted:

IMR = B (PctLBR) + B (TBR) + B (PctAdultsObese) + B (PctPhysicallyInactive) + B (PctAdultsDiabetes) + B (PctMammography) + B (PCPrate) + B (PctCollege) + B (PctSPH) + B (PctUnemployed) + B (PctUninsured) + B (PctBlack) + B (PctRural) + B (VCR) + Residual Error .

For the geospatial analyses, spatial regression statistics included spatial R2, Moran’s I for spatial autocorrelation, and Jarque-Bera multicollinearity estimates. Spatial weighting utilized Queen contiguity. Given 14 independent variables, we used a Bonferroni Type 1 adjusted error rate with alpha set at 0.0035.

Hierarchical linear model

For the hierarchical linear model of these variables across the 11 years, 2012–2022, the following analyses were conducted using SAS Proc Mixed.²⁹ The individual level model for each county is as follows^21,22:

Y_{ij} = B_{0 i} + B_{1 i} ({Time}_{ij}) + r_{ij}

where Y_ij is the outcome (i.e., IMR) for county i at time j. B_0i is the intercept or true change in IMR for county i, whereas B_1i is the incremental change or slope of the true change in the outcome IMR for the county. The variable r_ij is the level 1 measurement error.

The level 2 model across all counties and groups built upon and is incorporated within the B_0i and B_1i parameters in the level 1 individual model:

B_{0 i} = γ_{00} + γ_{01} (Group) + u_{0 i}

B_{1 i} = γ_{10} + γ_{11} (Group) + u_{1 i}

where γ₀₀ is the average true initial status for all counties given a group value of 0 (e.g., Urban), whereas γ₀₁ is the difference in average initial status between counties in group 1 (e.g., Rural) and counties in group 0 (e.g., Urban). Likewise, γ₁₀ is the average annual slope or rate of change for all counties given a group value of 0, whereas γ₁₁ is the difference in the annual rate of change/slope between counties in group 1 and counties in group 0. The level 2 residual errors u_0i and u_1i are shown.

IMR and change in IMR were assessed in the level 1 and 2 hierarchical models for n = 159 Georgia counties over 11 years from 2012 to 2022. Given univariate comparisons, we used a Type 1 error rate of 0.05 for the model assessments. One analysis focused on the Group as Rural (1) versus Urban (0), rural counties defined as 40% rural or greater (Variable “Percent Rural”) in the County Health Rankings and Roadmaps. By this criterion, n = 120 of n = 159 Georgia counties meet the rural criterion, consistent with the Georgia Department of Community Health.³⁰

A separate analysis focused on Group as three levels: Health Professions Shortage Areas HPSA low (0) (i.e., high PCP rate), HPSA medium (1), and high HPSA (2), as defined by the United States Health Resources Services Administration.³¹ High HPSA, or high need shortage areas are operationally defined as a Primary Care Physician (PCP) Rate of 3000:1 patients: provider.³¹ Given that the mean U.S. county PCP Rate is 1320:1 and based upon variations in county population densities with respect to access,³² we used a median HPSA range for PCP Rates less than 3000:1 but greater than 1000:1. Low need HPSA was defined as a county PCP Rate <1000:1.

Results

The distribution of Infant Mortality Rates (IMR) across Georgia is shown in Figure 1(a). Whereas 36 counties did not provide information, the higher IMR counties clustered in rural sections of the state, including the east and across the southern portion of the state. One urban area with a high disadvantaged population is included in central Georgia. A few northern counties above the greater Atlanta metropolitan region had higher IMR.

Figure 1.

(a) Infant Mortality Rates (IMR) across Georgia counties, 2016. Moran’s I = 0.251 for IMR by Rurality. The four circled counties had significantly high rurality and IMR, whereas the two counties shown in black had significantly low rurality and IMR. Lightly shaded counties without highlighted boundaries did not provide IMR data. (b) Spatial Plot IMR by Principal Care Provider (PCP) Rate (R2 = 0.02). (c) Spatial Plot IMR by Pct. African American (R2 = 0.276).

Spatial analysis

Table 1 presents univariate regression (i.e., R² and slope) and spatial regression (i.e., Moran’s I) statistics for IMR on each of the 14 independent variables. The statistics reveal the advantage of spatial regression over OLS regression in terms of regional county autocorrelations. For example, Low Birth Rate, Teen Birth Rate, Obesity, Inactivity and Diabetes had contradictory positive and negative slopes with respect to IMR, and none were significant.

Table 1.

Univariate and multivariate spatial regressions of 2022 infant mortality rate on county-level independent variables.

Independent variable	Univariate regression			Multivariate regression (R² = .469, AIC = 817.7)
Independent variable	R²	Slope	Moran’s I	B	SE	p
Low birth weight	.009	.158	.141	−.123	.171	.474
Percent single parent households	.006	.026	−.044	.012	.037	.742
Teen birth rate	.003	−.018	.321	.012	.030	.693
Percent adults obese	.004	−.074	.275	.004	.188	.982
Percent physically inactive	.029	−.146	.351	.366	.217	.092
Percent adults diabetes	.001	−.057	.144	−1.02	.464	.028*
Primary care physician rate	.187	.057**	−.131	.016	.011	.123
Percent mammography screened	.030	.121	.280	.044	.047	.350
Percent rural	.374	−.083**	.251	−.071	.011	<.001**
Percent African American	.045	.048*	.478	.057	.038	.131
Percent attended college	.094	.104**	.282	.064	.034	.060
Percent uninsured	.014	−.179	.467	.278	.128	.030*
Percent unemployed	.085	.826**	.367	.483	.212	.023*
Violent crime rate	.061	.004**	−.053	.001	.001	.626

*Significant at p < .05; **Significant at p < .0035.

However, Moran’s I showed large positive clustering of counties with respect to IMR. This means that these counties tended to be nonrandomly located together. Moran’s I represents the slope of the spatial regression and corrects for correlation overestimates that occur when geographic variations are not addressed in ordinary least squares regression.^33,34 For example, PCP Rate showed an unexpected positive but significant association with IMR on the OLS scatterplot slope, but the Moran’s I showed the expected negative association between clustering of counties with IMR (Table 1).

Besides PCP Rate, Percent Rural (negative), Percent Attended College, Percent Unemployed, and Violent Crime Rate had significant slopes/associations with IMR at p < .0035. Percent African American had a significant positive slope at p = .007. The negative associations for IMR with Primary Care Provider Rate and positive associations with Percent African American population were consistent with previous findings (Figure 1(b) and (c)).

Percent Rural, Percent African American, Percent Uninsured, Percent Unemployed, Percent Physically Inactive, Percent Adults Obese, Teen Birth Rate, Low Birth Rate, Percent Attended College, and Percent Adults with Diabetes each had strong, positive Moran’s I county clustering with IMR (Table 1, Figure 2). The multiple regression results for IMR on all 14 independent variables are shown in Table 1. The model R² = .469 was strong, and additional statistics supported a robust model. Only one independent variable was significant at the Bonferroni-adjusted p < .0035: Percent Rural, but slope was negative. Three variables were significant at p < .03: Adults with Diabetes, Percent Uninsured, Percent Unemployed (Table 1).

Figure 2.

GIS Moran’s I and EB rate of IMR on each of six predictor variables. (a) PCP rate (b) Percent rural (c) Percent African American (d) Percent attended college (e) Percent unemployed (f) Violent crime rate.

Hierarchical linear models

Hierarchical linear model trends for IMR are shown in Figure 3. Rural versus Urban IMR, 2015-2022, is shown in Figure 3(a). Individual county trajectories are shown, some increasing, some decreasing for both county categories. The smoothed mean county trajectories are shown, with the Urban mean slope starting at a lower level than Rural in 2015 and decreasing through 2022. In contrast, the Rural mean slope increased through 2022.

Figure 3.

(a) Infant mortality rates by urban (left) versus rural (right) counties, 2015–2022. Smoothed average county trajectories are shown by the darkened slope lines. (b) Infant mortality rates by low HPSA/High PCP counties (left), between adequate and shortage (middle), and high HPSA/low PCP shortage counties (right), 2015–2022. Smoothed average county trajectories are shown by the darkened slope lines.

Likewise, three categories of HPSA counties are shown in Figure 3(b), with individual county and mean category trajectories shown. Interestingly, the low HPSA (i.e., Adequate PCP Rate <1000:1) counties showed slightly higher IMR with a gradual increase from 2015 to 2022. The Medium and High HPSA counties had lower IMR from 2015 to 2022, relatively flat for high HPSA although there was a steep negative slope for the Medium HPSA counties where the PCP Rate is between 1000 and 3000:1.

Rural/Urban Multilevel Model statistics for the Figure 3 trends are shown in Table 2. For urban/rural status, the rate of change γ₁₀ = −0.056 showed a negative IMR slope for urban counties, as illustrated in Figure 3(a). The difference in rural county IMR slope was γ₁₁ = 0.191 points higher, giving a rural slope of positive 0.135 (i.e., 0.191–0.056; Table 2a). The initial status of rural and urban counties was significantly different, as shown by the γ₀₀ and γ₀₁ statistics (p = .025 and <.0001, respectively; Table 2a). For the variance components of the level 1 and level 2 models, three variance components (σ_ε², σ₀², and σ₁²) representing within county, initial status, and rate of change residual variation were significant, indicating that there is substantial residual variation that remains in the true values of these three components after controlling for rurality/urbanicity. In other words, an additional variable(s) might contribute with rurality/urbanicity to the multilevel time model of IMR.

Table 2.

Infant mortality rate by rural county status (A) and HPSA county status (B) multilevel models for change statistics.

Fixed effects	Parameter	Symbol	Estimate	SE	p
A. Rural status
Initial status π_0i	Intercept	γ₀₀	120.68*	52.78	.025
	Rural/Urban	γ₀₁	−383.8**	87.95	<.0001
Rate of change π_1i	Intercept	γ₁₀	−0.056	2.59	.983
	Rural/Urban	γ₁₁	0.191	3.59	.958
Variance
Level 1	Within county ε_ij	σ_ε²	1.031**	0.077	<.0001
Level 2	In initial status ζ_0i	σ₀²	23.02**	27.96	<.0001
	In rate of change ζ_1i	σ₁²	248.0**	18.63	<.0001
	Covariance ζ_0i & ζ_1i	σ₀₁	0	427.0	1.0
B. HPSA status
Initial status π_0i	Intercept	γ₀₀	−6.85	72.50	.925
	Low/Med/High HPSA	γ₀₁	−12.38	63.92	.846
Rate of change π_1i	Intercept	γ₁₀	0.007	3.42	.998
	Low/Med/High HPSA	γ₁₁	0.006	2.74	.998
Variance components
Level 1	Within county ε_ij	σ_ε²	1.277**	0.095	<.0001
Level 2	In initial status ζ_0i	σ₀²	55.22	11,884	.498
	In rate of change ζ_1i	σ₁²	307.1**	22.74	<.0001
	Covariance ζ_0i & ζ_1i	σ₀₁	0	574.0	1.0

*Significant at p < .05; **Significant at p < .0001.

For the three group HPSA Multilevel Model statistics, the rate of change γ₁₀ = 0.007 showed the relatively flat, slight positive slope for the high HPSA counties, as illustrated in Figure 2(b). The difference in county IMR slopes across the three groups was γ₁₁ = 0.006, roughly flat when all three groups were considered together. Likewise, the initial statuses of the three groups are not significantly different. For the variance components of the level 1 and level 2 models, two variance components (σ_ε² and σ₁²) representing within county and rate of change residual variation were significant, indicating that there was substantial residual variation that remained in the true values of these two components after controlling for HPSA level. As before, an additional variable(s) might contribute with HPSA level to the multilevel time model of IMR, especially given the large variance component for rate of change.

Discussion

The 2022 spatial regression results showed significant IMR disparities between rural and urban Georgia counties, consistent with previous research findings and trends along the southeastern coastal plain for IMR and other health disparities. There were significant associations between IMR and rural counties as well as counties that have high African American populations, high unemployment, and high uninsurance rates. Principal Care Provider Rates showed a negative spatial regression relationship, as expected. The findings support previous studies that indicate a strong need for county-specific social capital interventions, health programs, improved housing, etc.^11,13,14

From a temporal perspective, the Rural/Urban hierarchical linear model reiterates this rural/urban dichotomy from 2012 to 2022. The IMR slope for 120 rural Georgia counties is increasing, whereas the urban IMR slope is decreasing. The HPSA model is less clear, although the medium need counties (PCP Rates between 1000 and 3000:1) demonstrate a period IMR decline. It is possible that the higher IMR rates for low need counties (PCP Rates <1000:1) that are increasing might be due to the presence of high disadvantaged populations within urban counties: a population density effect.

The complementary geospatial and longitudinal, temporal analyses used in this study support previous findings^5–14,17,18 that show regional socioeconomic risk factors that associate with higher infant mortality rates. These trends persist over time, and they show worsening outcomes in many rural counties. Maternal and child health outcomes in rural Georgia and across the southeastern United States are further complicated by the prevalence of health professional shortage areas (HPSAs), with the concentration of pediatric, mental health, and other medical specialists in urban areas. Many rural counties have no such specialists and rely heavily on a few general practitioners.

The study demonstrates the advantages of using both spatial and temporal analyses to assess population health trends. These approaches are straightforward to implement, and they can be valuable to clinicians, policymakers, and researchers to improve the health of rural communities. The spatial approach improves upon OLS regression methodologies to show regional clusters in measured variables. The temporal approach using hierarchical linear models not only displays time trends but also statistically evaluates both within county variation as well as between county variation, coupled with nesting effects within larger groupings (e.g., Rural vs Urban). Such hierarchical linear models have rarely been applied in public health research, and never with county-level data. Such approaches have been extensively used in educational and economic research and planning.^21,35

The high association rates between IMR, rurality, high minority, and low socioeconomic conditions mirror other chronic health problems confronting communities. Health disparities and overall poor health ratings in communities strongly correlate with poor socioeconomic conditions. Public health approaches could benefit from stronger partnerships with business/industry, economic planners, and state and local governments to target rural county infrastructure and opportunities for citizens.

Limitations of this study include: (a) Missing data such that some IMR comparisons were for 2015–2022 instead of all 11 years data 2012–2022; (b) data source limitations in the County Health Rankings and Roadmaps, which vary from nationally-representative data to public health departments, the latter of which might be biased towards certain population groups in varied counties; (c) variations in choice of Queen contiguity levels and alternative methods of weighting that are available for geospatial regression analyses; and (d) nonreporting from 36 or more Georgia counties on IMR or other variables, particularly in rural areas that were strategic to the assessed models.

Future research directions should examine the possibility of truly merging spatial and longitudinal methodologies to better assess spatiotemporal clusters. Applications from this research include state and local government, business, education, and health department partnerships to improve health programs for expectant families to provide resources and supports to reduce IMR and other health risks. Most importantly, there is a need for county-level investments to lift residents out of cycles of poverty and disadvantaged conditions that lead to poor health.

Conclusion

In summary, research indicates that infant mortality and other negative health outcomes cluster in geographic areas both at the state and county level, and they are strongly impacted by geographic-specific socioeconomic factors. Besides geospatial regression of infant mortality factors in static time, the study originally adds a longitudinal component across 11 years of data and tests these temporal trends within potential group nesting effects. Major contributors to Georgia infant mortality rates included high percentage unemployed, high rurality, percentage African American population, and violent crime rates. The study findings strongly indicate the need for public health infrastructure and socioeconomic supports in rural counties to reduce infant mortality.

Footnotes

ORCID iD

David W. Hollar

Ethics approval

This study used secondary analysis of data that is publicly available online.

Author contributions

DH contributed all study and manuscript components, including conceptualization, methodology, analysis, investigation, and writing.

Funding

The author received no financial support for the research, authorship, and/or publication of this article.

Declaration of conflicting interests

The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability Statement

All data is available upon request.

References

Arias

Tejada-Vera

, et al. U.S. state life tables, 2020. Natl Vital Stat Rep 2022; 71(2): 1–18.

CDC Wonder . 2020 death data. CDC National Center for Health Statistics, 2020. https://cdc.gov/nchs/pressroom/states/Georgia/ga.htm (Accessed 7 January 2023).

Murphy

Kochanek

, et al. Mortality in the United States, 2020. NCHS Data Brief 2021; 427: 1–8.

CDC Wonder . Infant mortality rates by state, 2020. CDC National Center for Health Statistics, 2020. https://cdc.gov/nchs/pressroom/sosmap/infant_mortality_rates/infant_mortality.htm (Accessed 8 January 2023).

Georgia Department of Public Health . Infant mortality report. Author, 2022 (Accessed 14 January 2023). https://dph.georgia.gov/infant-mortality

Ely

Driscoll

. Infant mortality in the United States, 2020: data from the Period Linked Birth/Infant Death File. CDC National Center for Health Statistics, 2022.

Kandasamy

Hirai

Kaufman

, et al. Regional variation in black infant mortality: the contribution of contextual factors. PLoS One 2020; 15(8): e0237314.

Mohamoud

Kirby

Ehrenthal

. Poverty, urban-rural classification and term infant mortality: a population-based multilevel analysis. BMC Pregnancy Childbirth 2019; 19: 40.

Bhatt

Beck-Sagué

. Medicaid expansion and infant mortality in the United States. Am J Public Health 2018; 108: 565–567.

10.

Kaiser Family Foundation (2022). Status of state Medicaid expansion decisions: interactive map. Accessed 9 January 2023. https://www.kff.org/Medicaid/issue-brief/status-of-state-medicaid-expansion-decisions-interactive-map/

11.

Hirai

Sappenfield

Kogan

, et al. Contributors to excess infant mortality in the U.S. South. Am J Prev Med 2014; 46(3): 219–227.

12.

Office of Disease Prevention and Health Promotion, US Department of Health and Human Services . Healthy people 2030 framework. U.S. Department of Health and Human Services, 2020. wayback.archive-it.org/5774/20220414112518/. https://odphp.health.gov/healthypeople/about/healthy-people-2030-framework (Accessed 12 January 2023).

13.

Yang

McManus

. Infant mortality and social environment in Georgia: an application of hotspot detection and prioritization. Environ Ecol Stat 2010; 17(4): 455–471.

14.

Hollar

. Evaluating the interface of health data and policy: applications of geospatial analysis to county-level national data. Child Health Care 2015; 45(3): 266–285.

15.

Mandelbrot

. The fractal biology of nature. Freeman, 1977.

16.

West

Brown

Enquist

. A general model for the origin of allometric scaling laws in biology. Science 1997; 276: 122–126.

17.

Arcaya

Brewster

Zigler

, et al. Area variations in health: a spatial multilevel modeling approach. Health Place 2012; 18(4): 824–831.

18.

Ehrenthal

Kuo

HHD

Kirby

. Infant mortality in rural and nonrural counties in the United States. Pediatrics 2020; 146(5): e20200464.

19.

Anselin

Bera

. Spatial dependence in linear regression models with an introduction to spatial econometrics. In: Ullah

David

(eds). Handbook of Applied Economic Statistics, chapter 7. Marcel Dekker, 1998, pp. 237–289.

20.

Fotheringham

Brunsdon

. Local forms of spatial analysis. Geogr Anal 1999; 31(4): 340–358.

21.

Bryk

Raudenbush

. Hierarchical linear models: applications and data analysis methods. Sage, 1992.

22.

Singer

Willett

. Applied longitudinal data analysis. Oxford University Press, 2003.

23.

Anselin

. Thirty years of spatial econometrics. Pap Reg Sci 2010; 89(1): 3–26.

24.

Voss

Long

Hammer

, et al. County child poverty rates in the U.S.: a spatial regression approach. Popul Res Policy Rev 2006; 25: 369–391.

25.

Fotheringham

Zahn

. A comparison of three exploratory models for cluster detection in spatial point patterns. Geogr Anal 1996; 28: 200–218.

26.

Anselin

. Under the hood. Issues in the specification and interpretation of spatial regression models. Agric Econ 2002; 27(3): 247–267.

27.

Anselin

. Local indicators of spatial association – LISA. Geogr Anal 1995; 27(2): 93–115.

28.

Wheeler

Tiefelsdorf

. Multicollinearity and correlation among local regression coefficients in geographically weighted regression. J Geogr Syst 2005; 7: 161–187.

29.

Wang

. Using SAS PROC MIXED to demystify the hierarchical linear model. J Exp Educ 1997; 66(1): 84–93.

30.

Georgia Department of Community Health (2022). State Office of Rural Health (SORH) fact sheet. https://dch.georgia.gov/divisionsoffices/state-office-rural-health/sorh-fact-sheet. Accessed 28 January 2023.

31.

Scannell

Quinton

Jackson

, et al. Primary care health professional shortage area designations before and after the affordable care act’s shortage designation modernization project. JAMA Netw Open 2021; 4(7): e2118836.

32.

Schlak

Poghosyan

Liu

, et al. The association between Health Professional Shortage Area (HPSA) status, work environment, and nurse practitioner burnout and job dissatisfaction. J Health Care Poor Underserved 2022; 33(2): 998–1016.

33.

Anselin

. The Moran scatterplot as an ESDA tool to assess local instability in spatial association. In: Fischer

(ed). Spatial Analytical Perspectives on GIS. Routledge, 1996, pp. 111–125.

34.

Chen

. Spatial autocorrelation equation based on Moran’s index. Sci Rep 2023; 13: 19296.

35.

Angrist

Imbens

. Two-stage least squares estimation of average causal effects in models with variable treatment intensity. J Am Stat Assoc 1995; 90(430): 431–442.

Spatiotemporal clusters of infant mortality risk factors in Georgia

Abstract

Methods

Results

Conclusions

Keywords

Background

Factors impacting IMR

Spatial analysis

Temporal analysis

Methods

Statistical models

Geospatial regression

Hierarchical linear model

Results

Spatial analysis

Hierarchical linear models

Discussion

Conclusion

Footnotes

ORCID iD

Ethics approval

Author contributions

Funding

Declaration of conflicting interests

Data Availability Statement

References