Examining Health Inequities in A1C Control over Time across Individual,Geospatial,and Geopolitical Factors among Adults with Type 2 Diabetes: Analyses of a Sample from One Commercial Insurer in a Southern State

Abstract

Introduction:

Type 2 diabetes impacts millions and poor maintenance of diabetes can lead to preventable complications, which is why achieving and maintaining target A1C levels is critical. Thus, we aimed to examine inequities in A1C over time, place, and individual characteristics, given known inequities across these indicators and the need to provide continued surveillance.

Methods:

Secondary de-identified data from medical claims from a single payer in Texas was merged with population health data. Generalized Estimating Equations were utilized to assess multiple years of data examining the likelihood of having non-target (>7% and ≥7%, two slightly different cut points based on different sources) and separately uncontrolled (>9%) A1C. Adults in Texas, with a Type 2 Diabetes (T2D) flag and with A1C reported in first quarter of the year using data from 2016 and 2019 were included in analyses.

Results:

Approximately 50% had A1Cs within target ranges (<7% and ≤7%), with 50% considered having non-target (>7% and ≥7%) A1Cs; with 83% within the controlled ranges (≤9%) as compared to approximately 17% having uncontrolled (>9%) A1Cs. The likelihood of non-target A1C was higher among those individuals residing in rural (vs urban) areas (P < .0001); similar for the likelihood of reporting uncontrolled A1C, where those in rural areas were more likely to report uncontrolled A1C (P < .0001). In adjusted analysis, ACA enrollees in 2016 were approx. 5% more likely (OR = 1.049, 95% CI = 1.002-1.099) to have non-target A1C (≥7%) compared to 2019; in contrast non-ACA enrollees were approx. 4% more likely to have non-target A1C (≥7%) in 2019 compared to 2016 (OR = 1.039, 95% CI = 1.001-1.079). In adjusted analysis, ACA enrollees in 2016 were 9% more likely (OR = 1.093, 95% CI = 1.025-1.164) to have uncontrolled A1C compared to 2019; whereas there was no significant change among non-ACA enrollees.

Conclusions:

This study can inform health care interactions in diabetes care settings and help health policy makers explore strategies to reduce health inequities among patients with diabetes. Key partners should consider interventions to aid those enrolled in ACA plans, those in rural and border areas, and who may have coexisting health inequities.

Keywords

Health inequity health disparity social and structural determinants of health inequities rural health border health diabetes

Introduction

Diabetes affects nearly 40 million US residents, of which 90%-95% are diagnosed with Type 2 Diabetes (T2D).¹ A patient’s ability to maintain A1C levels within target ranges is essential to overall diabetes management.² Maintaining target ranges for A1Cs is critical to reduce the risk of preventable complications.³ Target ranges for A1C can be identified by the recommendation of a physician based on the patients’ diabetes history and/or general health status, among other items.⁴ Further, considerations as to a patient’s age, comorbid conditions, and life expectancy have been suggested in setting target A1C goals.³ While these individualized target ranges may vary patient-by-patient or provider-by-provider, more universal A1C thresholds are frequently utilized in public health services research.

Population-level thresholds for A1C that can be used more generally can be pulled from a variety of resources including the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK),⁴ American Diabetes Association (ADA),⁵ and Centers for Disease Control and Prevention (CDC).⁶ For example, non-target ranges for A1C have been identified as values ≥7% within NIDDK and ADA resources^4,5 and >7% within posted guidance by the US CDC.⁶ Separately, “uncontrolled” or “poor control” of diabetes has been identified as >9%⁷ in a CMS report⁸ cited by the US CDC⁹ with an ADA resource¹⁰ providing guidance on the degree of control indicating that A1Cs above 9% were considered poor (10%), very poor (11%), and extremely poor (≥12%).¹⁰ Furthermore, Healthy People 2030 also includes a goal for adults with diabetes of reducing the percentage of those with an A1C value >9% given the increased risk of complications,¹¹ further reinforcing this additional threshold’s practical use. Using these thresholds can assist public health surveillance efforts aimed at monitoring A1C levels within at-risk communities and populations. Because A1C levels outside target ranges can contribute to medical complications (e.g., death, amputation, eye and heart problems)¹² and high A1C is associated with higher rates of all-cause mortality,¹³ surveillance efforts are becoming increasingly important to identify factors associated with successful diabetes management and to address potential inequalities and inequities contributing to uncontrolled A1C values (e.g., socioeconomic status, insurance coverage and type, residing in rural areas). Furthermore, the US-Mexico border area is a unique geopolitical boundary that is among the world’s busiest borders, with some areas having cities on either side of the border close enough to be considered, by some, as a single metropolitan area for local populations.¹⁴ Additionally, a more recently published study reported significant disparities in life expectancy for individuals reported as Hispanic and American Indian¹⁵ who resided within the US border region along the US-Mexico border.¹⁵ In terms of health insurance types, there is also evidence that some of the largest groups enrolling in the ACA’s marketplace insurance between 2013 and 2016 were those reporting fair to poor health, Hispanic individuals, individuals aged 50-64 years, and those reporting their highest education as a high school diploma.¹⁶ Furthermore, efforts to identify factors linked to A1C thresholds (i.e., target ranges and controlled levels) are valuable because they can assess a critical, and potentially modifiable aspect of T2D that may be influenced by individual-level, geospatial, and geopolitical factors.

Theoretical Framework

The WHO Framework for Action on the Social Determinants of Health¹⁷ guided the development of several statistical models to identify factors linked to non-target or uncontrolled A1C among adults with T2D, emphasizing theoretically-informed variables at both the individual-level (age, sex) and contextual levels (rurality). For example, this framework highlights several social and structural determinants of health inequities (e.g., governance, macroeconomic policies, public policies, and culture and societal values),¹⁷ of which we consider the geopolitical indicator, border status, within this along with the broader contextual factors of rurality and Health Professional Shortage Areas (HPSA). Additionally, this framework also includes, among the social and structural determinants of health inequities, gender, and ethnicity,¹⁷ which our measures of sex and, to some extent, border status point to. Given this theoretical framework, we hypothesize that inequities for non-target A1Cs and separately uncontrolled A1Cs will be informed with the above variables.

Aims

To examine the common and unique factors associated with three distinct A1C thresholds, we assessed the contributions of individual-level, geospatial, and geopolitical variables on the likelihood of patients having non-target A1C levels (vs target) and separately uncontrolled A1C levels (vs controlled).

Methods

Data and Study Design

Longitudinal analyses were conducted using medical claims data obtained from a single payer with service areas in each Texas county, then merged with publicly available data using geospatial identifiers (county FIPS codes). This resulted in a de-identified analytical sample used in this study. The medical claims data included data from 2016 and 2019, which allowed us to measure change over time. Time in the analyses was treated binarily, 2016 or 2019. Data linked to the medical claims data included multiple disparate data that were linked using county FIPS codes. These data included: the County Health Rankings¹⁸; the Area Health Resource File¹⁹; and the National Center for Health Statistics (NCHS) Urban-Rural Classification Scheme for Counties.²⁰ The final analytical file is characterized as secondary existing data that is de-identified and without geospatial identifiers below 3-digit ZIP codes, which resulted in an institutional review board (IRB) determination of “Not Human Subjects Research.”

Primary Outcome Variables

Individual-level data included the primary dependent variables derived from A1C levels. A1C levels were used to create three separate variables: Two variables to assess non-target versus target A1C levels; and a third variable to assess uncontrolled versus controlled A1C levels. These different thresholds are not necessarily meant as comparisons to one another, but rather meant to reflect a broader set of results for those in different A1C statuses.

Target A1C

Non-Target Definition A: Using published thresholds from the NIDDK⁴ and ADA,⁵ we used the threshold of < 7% A1C, operationally defined as within target A1C levels, versus ≥7%, operationally defined as non-target or outside of the target A1C levels. Non-Target Definition B: Using published thresholds from the CDC,⁶ we used the threshold of ≤7% A1C, operationally defined as within target A1C levels, versus >7%, operationally defined as outside target A1C levels. Uncontrolled: We created a measure for uncontrolled diabetes, which was meant to compare levels previously indicated as “uncontrolled”⁷ or “poor control”^8,9 A1C status in prior research⁷ using thresholds of >9% operationally defined as uncontrolled, versus ≤9%, operationally defined as controlled.

Additional Variables

Medical claims data

Age (in years), sex (binary: male, female), ACA insurance status (binary: enrolled in an Affordable Care Act-compliant individual and small group insurance coverage purchased on or off of the exchange [ACA, 2010], vs not) were included.

Geospatial population health data

Rurality (binary: rural, urban) was included to provide an assessment of variation across resource-limited versus other areas. The NCHS indicator of rurality²⁰ was operationally defined as metropolitan (codes 1-4; referred to in this study as “urban”) versus non-metropolitan (codes 5-6; referred to in this study as “rural”). Health Professional Shortage Areas (HPSAs) were also included to assess variation in our outcomes across resource-limited versus other areas taken from the Area Health Resource File (AHRF),¹⁹ and treated as binarily, HPSA versus non-HPSA. Border status of Texas counties were also included to provide an assessment of variation in our outcomes across resource-limited areas versus other areas and to provide an indicator of border status, given the geopolitical boundary¹⁴ along the US-Mexico border (binary: border, non-border).

Inclusion Criteria

Individuals aged 18 and older, residing in Texas, with an A1C reported in the first quarter of the year, without a type 1 diabetes flag, and with a T2D flag within the study timeline, with the years 2016 and 2019 being used in this analysis, were included in the sample. This included all those meeting the inclusion criteria enrolled in a single payer. For this study, the sample included individuals from one of the largest insurers with one of the most comprehensive geospatial coverages in the state. Thus, these analyses likely represent among the largest geographically distributed samples of individuals with commercial insurance coverage in the state. Further, based on these available data, this study can provide estimates from a sample likely representative of the experience of many of those insured individuals (from a single payer) aged 18 and older residing in Texas with T2D.

Statistical Analyses

SAS version 9.4 (Cary, NC) and Stata version 18 (College Station, TX) were used for all analyses. We employed the SAS procedure GENMOD with the REPEATED statement²¹ to account for correlation among those with repeated measures. The Generalized Estimating Equations (GEE)^22
-24 provided odds ratios (OR) and associated 95% confidence intervals (CI) and considered an alpha of 0.05 (P < .05) to be considered statistically significant. GEE has been considered appropriate for longitudinal analyses,²¹ with certain assumptions for missing data (i.e., missing at random), and with correlated data²² such as with repeated measures over time. Unadjusted odds ratios were taken from models with the dependent variable and a single independent variable. Adjusted analyses include multivariable models (with identical independent variables across models), adjusted for multiple variables simultaneously with additional interaction terms where specified. Stata’s margins command was used to generate predicted probabilities^25,26 used within interaction plots (See Figure 1 and 2 and Table 3B) gained from adjusted models (with identical variables as those used within SAS) using Stata’s xtlogit procedure.

Figure 1.

(A-C) Interaction plots for definition A.

Figure 2.

(A-C) Interaction plots for uncontrolled A1C.

Results

Descriptive Analyses

As presented in Table 1, the majority of the sample resided in urban areas (89%), non-border areas (90%), and HPSAs (94%). The majority of the sample was also male (54%), not enrolled in ACA coverage (66%), and had an average age of 55 years (range 18-89 years). Fifty percent of patients had A1Cs within target range using the <7% threshold, and 53% had A1Cs within target range using the ≤7% threshold. Additionally, 83% were considered to have A1C levels within the controlled range (≤9%), as operationally defined in this study.

Table 1.

Distribution (2016, 2019, and Pooled Sample) by Key Characteristics.

	2016		2019		Pooled
	N	%	n	%	n	%
Time
2016	26245	100	-	-	26245	33.5
2019	-	-	52088	100	52088	66.5
Rurality
Urban	23378	89.1	46148	88.6	69526	88.8
Rural	2867	10.9	5940	11.4	8807	11.2
Border
Border	2489	9.5	5199	10.0	7688	9.8
Non-border	23756	90.5	46889	90.0	70645	90.2
ACA marketplace insurance
Non-ACA	13833	52.7	38147	73.2	51980	66.4
ACA	12412	47.3	13941	26.8	26353	33.6
HPSA
Non-HPSA	1412	5.4	3168	6.1	4580	5.9
HPSA	24833	94.6	48920	93.9	73753	94.2
Sex
Male	13687	52.2	28610	54.9	42297	54.0
Female	12558	47.9	23478	45.1	36036	46.0
A1C status
Within target (A1C range 4 to <7)	13083	49.9	26203	50.3	39286	50.2
Non-target (A1C range ≥7 to 14)	13162	50.2	25885	49.7	39047	49.9
Within target (A1C range 4 to ≤7)	13851	52.8	27577	52.9	41428	52.9
Non-target (A1C range >7-14)	12394	47.2	24511	47.1	36905	47.1
Controlled (A1C range 4 to ≤9	21685	82.6	43529	83.6	65214	83.3
Uncontrolled (A1C range >9-14)	4560	17.4	8559	16.4	13119	16.8

Age mean (range, std, n): 2016 at 53.68 years (18-89, 9.42, 26 245); 2019 at 55.22 years (18-89, 10.08, 52 088); Pooled at 54.70 years (18-89, 9.89, 78 333).

Pooled sample size includes n = 78 333 observations, inclusive of (while not restricted to) those with data at time 1 and/or time 2.

Note: Target A1C may ultimately be determined on an individual basis with one’s physician. The thresholds used in this study are population-based metrics.

Unadjusted Analyses

Table 2 presents unadjusted analyses using odds ratios (OR) with 95% confidence intervals (CI). Significant variation (P < .05) in our outcome over time was observed for the likelihood of uncontrolled (>9%) A1C, where a higher likelihood of reporting uncontrolled A1C (OR = 1.073, 95% CI = 1.033-1.114) was observed in 2016 versus 2019.

Table 2.

Unadjusted Analyses for Non-Target A1C and separately Uncontrolled A1C.

	Non-target A1C (Definition A)				Non-target A1C (Definition B)				Uncontrolled A1C
		95% confidence interval		Score statistics for type 3 GEE analysis		95% confidence interval		Score statistics for type 3 GEE analysis		95% confidence interval		Score statistics for type 3 GEE analysis
	Odds ratio	Lower	Upper	Pr > ChiSq	Odds ratio	Lower	Upper	Pr > ChiSq	Odds ratio	Lower	Upper	Pr > ChiSq
Time
2016 vs 2019	1.008	0.980	1.036	0.5913	0.998	0.970	1.026	0.8638	1.073	1.033	1.114	0.0003
Rurality
Rural vs Urban	1.132	1.082	1.185	<.0001	1.115	1.066	1.166	<.0001	1.109	1.046	1.176	0.0008
Border
Border vs non-Border	1.117	1.064	1.171	<.0001	1.137	1.084	1.192	<.0001	1.225	1.153	1.302	<.0001
ACA marketplace insurance
Enrollment in ACA Marketplace (ACA) vs non-ACA	1.044	1.013	1.075	0.0056	1.040	1.010	1.072	0.0100	1.022	0.982	1.064	0.2877
HPSA
HPSA vs non-HPSA	0.945	0.889	1.005	0.0704	0.949	0.893	1.009	0.0953	1.008	0.930	1.093	0.8459
Sex
Female vs male	0.730	0.709	0.751	<.0001	0.730	0.709	0.751	<.0001	0.771	0.742	0.802	<.0001

Bolded odds ratios indicate significance using 95% confidence intervals (CI).

In terms of rurality, the risk of non-target A1C was higher among those patients residing in rural (vs urban) areas (see Table 2). This was similar for the likelihood of reporting uncontrolled A1C, where those residing in rural areas were more likely to report uncontrolled A1C (OR = 1.109, 95% CI = 1.046-1.176). In terms of border status, the likelihood of non-target A1C and uncontrolled A1C was higher among those individuals residing in border (vs non-border) areas, respectively (see Table 2). In terms of ACA enrollment, the likelihood of non-target A1C was higher among those individuals enrolled in the ACA coverage (vs non-ACA coverage) (see Table 2). In terms of sex, the likelihood of non-target A1C was lower among female (vs male) patients (see Table 2). This was also true for uncontrolled A1C, where the likelihood of non-target A1C was lower among female (vs male) patients (female vs. male; OR = 0.771, 95% CI = 0.742-0.802).

Adjusted Analyses

Table 3A presents adjusted analyses with multiple interaction terms added to each model. While indicators of non-target A1C using definition A and B are both presented in the Tables, we chose to highlight only Definition A in the majority of the Results section in the adjusted analyses, given the similarities in the major findings between these two definitions (see Table 3A).

Table 3.

A. Adjusted Analyses for Non-Target A1C and Separately Uncontrolled A1C.

	Non-target A1C (Definition A)				Non-target A1C (Definition B)				Uncontrolled A1C
		95% confidence interval		Score statistics for type 3 GEE analysis		95% confidence interval		Score statistics for type 3 GEE analysis		95% confidence interval		Score statistics for type 3 GEE analysis
	Odds ratio	Lower	Upper	Pr > ChiSq	Odds ratio	Lower	Upper	Pr > ChiSq	Odds ratio	Lower	Upper	Pr > ChiSq
Time
2016 vs 2019	1.005	0.975	1.035	0.7495	0.992	0.963	1.022	0.5926	1.040	0.999	1.083	0.0575
Rurality
Rural vs urban	1.225	1.149	1.307	<.0001	1.199	1.124	1.278	<.0001	1.227	1.135	1.326	<.0001
Border
Border vs non-border	1.214	1.138	1.295	<.0001	1.231	1.154	1.312	<.0001	1.342	1.242	1.451	<.0001
ACA marketplace insurance
Enrollment in ACA marketplace (ACA) vs non-ACA	1.144	1.074	1.218	<.0001	1.148	1.079	1.223	<.0001	1.171	1.077	1.272	0.0003
HPSA
HPSA vs non-HPSA	0.924	0.868	0.983	0.0129	0.926	0.870	0.985	0.0156	0.965	0.888	1.048	0.4007
Sex
Female vs male	0.720	0.700	0.741	<.0001	0.720	0.699	0.741	<.0001	0.743	0.715	0.773	<.0001
Interactions
ACA × HPSA
ACA + HPSA vs ACA + Non-HPSA	0.849	0.773	0.933	0.0075	0.849	0.773	0.933	0.0061	0.850	0.752	0.960	0.0028
ACA + HPSA vs Non-ACA + HPSA	1.051	1.018	1.087		1.053	1.019	1.088		1.031	0.987	1.077
ACA + HPSA vs Non-ACA + Non-HPSA	1.057	0.973	1.149		1.063	0.978	1.156		1.129	1.007	1.266
ACA + Non-HPSA vs Non-ACA + HPSA	1.238	1.128	1.359		1.240	1.130	1.361		1.213	1.076	1.368
ACA + Non-HPSA vs Non-ACA + Non-HPSA	1.245	1.104	1.404		1.252	1.110	1.412		1.329	1.133	1.559
Non-ACA + HPSA vs Non-ACA + Non-HPSA	1.005	0.928	1.090		1.010	0.931	1.095		1.095	0.980	1.224
Rural × Border
Border + Rural vs Non-Border + Rural	1.361	1.210	1.531	0.0005	1.371	1.220	1.540	0.0009	1.553	1.352	1.785	0.0003
Border + Rural vs Border + Non-Rural	1.374	1.220	1.547		1.335	1.187	1.503		1.420	1.234	1.633
Border + Rural vs Non-Border + Non-Rural	1.487	1.334	1.658		1.476	1.325	1.644		1.647	1.452	1.867
Non-Border + Rural vs Border + Non-Rural	1.009	0.941	1.082		0.974	0.909	1.044		0.914	0.834	1.001
Non-Border + Rural vs non-Border + non-Rural	1.093	1.039	1.149		1.076	1.024	1.131		1.060	0.992	1.133
Border + non-Rural vs non-Border + non-Rural	1.083	1.026	1.142		1.105	1.048	1.166		1.160	1.082	1.243
Time x ACA
2016 + ACA vs 2016 + Non-ACA	1.195	1.111	1.285	0.0044	1.201	1.117	1.293	0.0032	1.229	1.116	1.355	0.0183
2016 + ACA vs 2019 + ACA	1.049	1.002	1.099		1.038	0.990	1.087		1.093	1.025	1.164
2016 + ACA vs 2019 + Non-ACA	1.150	1.074	1.231		1.139	1.064	1.219		1.218	1.112	1.333
2016 + Non-ACA vs 2019 + ACA	0.878	0.818	0.943		0.864	0.805	0.927		0.889	0.808	0.977
2016 + Non-ACA vs 2019 + Non-ACA	0.962	0.927	0.999		0.948	0.913	0.984		0.991	0.942	1.042
2019 + ACA vs 2019 + Non-ACA	1.096	1.026	1.170		1.098	1.028	1.172		1.114	1.021	1.217

Adjusted models include discrete age as a variable and as such are adjusted for age. Fully adjusted statistical models included age, time, rurality, border status, an indicator of ACA marketplace insurance, HPSA status, sex, multiple interactions (ACA Status * HPSA; Rural* Border; and Time * ACA Status) with the dependent variable being A1C status.

Bolded odds ratios indicate significance using 95% confidence intervals (CI).

B. Predicted Probabilities for Figures 1 and 2.
	Predictive margins	Contrasts	Contrast P>\|z\|
Non-target A1C (definition A)
Non-HPSA + Non-ACA	0.52	Non-HPSA + ACA vs Non-HPSA + Non-ACA	.0003
Non-HPSA + ACA	0.57	HPSA + Non-ACA vs Non-HPSA + Non-ACA	.8947
HPSA + Non-ACA	0.52	HPSA + ACA vs Non-HPSA + Non-ACA	.1903
HPSA + ACA	0.53	HPSA + Non-ACA vs Non-HPSA + ACA	<.0001
		HPSA + ACA vs Non-HPSA + ACA	.0006
		HPSA + ACA vs HPSA + Non-ACA	.0027
Non-Rural + Non-Border	0.50	Non-Rural + Border vs Non-Rural + Non-Border	.0035
Non-Rural + Border	0.52	Rural + Non-Border vs Non-Rural + Non-Border	.0005
Rural + Non-Border	0.52	Rural + Border vs Non-Rural + Non-Border	<.0001
Rural + Border	0.60	Rural + Non-Border vs Non-Rural + Border	.7969
		Rural + Border vs Non-Rural + Border	<.0001
		Rural + Border vs Rural + Non-Border	<.0001
2016 + Non-ACA	0.51	2016 + ACA vs 2016 + Non-ACA	<.0001
2016 + ACA	0.56	2019 + Non-ACA vs 2016 + Non-ACA	.0424
2019 + Non-ACA	0.52	2019 + ACA vs 2016 + Non-ACA	.0003
2019 + ACA	0.54	2019 + Non-ACA vs 2016 + ACA	.0001
		2019 + ACA vs 2016 + ACA	.0422
		2019 + ACA vs 2019 + Non-ACA	.0065
Uncontrolled A1C
Non-HPSA + Non-ACA	0.17	Non-HPSA + ACA vs Non-HPSA + Non-ACA	.0006
Non-HPSA + ACA	0.21	HPSA + Non-ACA vs Non-HPSA + Non-ACA	.0988
HPSA + Non-ACA	0.18	HPSA + ACA vs Non-HPSA + Non-ACA	.0310
HPSA + ACA	0.18	HPSA + Non-ACA vs Non-HPSA + ACA	.0028
		HPSA + ACA vs Non-HPSA + ACA	.0122
		HPSA + ACA vs HPSA + Non-ACA	.1696
Non-Rural + Non-Border	0.16	Non-Rural + Border vs Non-Rural + Non-Border	.0001
Non-Rural + Border	0.18	Rural + Non-Border vs Non-Rural + Non-Border	.0899
Rural + Non-Border	0.17	Rural + Border vs Non-Rural + Non-Border	<.0001
Rural + Border	0.24	Rural + Non-Border vs Non-Rural + Border	.0541
		Rural + Border vs Non-Rural + Border	<.0001
		Rural + Border vs Rural + Non-Border	<.0001
2016 + Non-ACA	0.17	2016 + ACA vs 2016 + Non-ACA	.0000
2016 + ACA	0.20	2019 + Non-ACA vs 2016 + Non-ACA	.7132
2019 + Non-ACA	0.17	2019 + ACA vs 2016 + Non-ACA	.0148
2019 + ACA	0.19	2019 + Non-ACA vs 2016 + ACA	<.0001
		2019 + ACA vs 2016 + ACA	.0065
		2019 + ACA vs 2019 + Non-ACA	.0164

Non-Target (≥7) versus Target A1C (<7%)

Interaction of ACA by HPSA

The interaction of ACA by HPSA (see Figure 1a and Tables 3A and B), provided evidence of a differential effect for the overall interaction term (P < .05). Figure 1a presents the interaction plot. For ACA enrollees, a lower likelihood of having a non-target A1C was identified among those within a HPSA versus a non-HPSA (OR = 0.849, 95% CI = 0.773-0.933) in contrast to non-ACA enrollees with no statistically significant difference between those within a HPSA versus a non-HPSA (OR = 1.005, 95% CI = 0.928-1.090). Among HPSAs, a higher likelihood of having a non-target A1C was identified among those enrolled in ACA versus non-ACA (OR = 1.051, 95% CI = 1.018-1.087), similar in direction to among non-HPSAs with a higher likelihood of having a non-target A1C identified among those enrolled in ACA versus non-ACA (OR = 1.245, 95% CI = 1.104-1.404).

Interaction of Rurality by Border status

The interaction of rurality by border status (see Figure 1b and Tables 3A and B), provided evidence of a differential effect for the overall interaction term (P < .05). Figure 1b presents the interaction plot. Among Rural areas, a higher likelihood of having a non-target A1C was identified among those in Border areas compared to non-Border areas (OR = 1.361, 95% CI = 1.210-1.531), similar in direction to among non-Rural areas with a higher likelihood of having a non-target A1C identified among those in Border areas compared to non-Border areas (OR = 1.083, 95% CI = 1.026-1.142). Among border areas, a higher likelihood of having a non-target A1C was identified among those in Rural areas compared to non-Rural areas (OR = 1.374, 95% CI = 1.220-1.547), similar in direction to among non-Border areas with a higher likelihood of having a non-target A1C identified among those in Rural areas compared to non-Rural areas (OR = 1.093, 95% CI = 1.039-1.149).

Interaction of Time by ACA

The interaction of time by ACA (see Figure 1c and Tables 3A and B), provided evidence of a differential effect for the overall interaction term (P < .05). Figure 1c presents the interaction plot. For ACA enrollees a higher likelihood of having a non-target A1C was identified among those in 2016 compared to 2019 (OR = 1.049, 95% CI = 1.002-1.099), in contrast to non-ACA enrollees with a lower likelihood having a non-target A1C identified among those in 2016 compared to 2019 (OR = 0.962, 95% CI = 0.927-0.999). In 2016, a higher likelihood of having a non-target A1C was identified for those enrolled in ACA versus non-ACA (OR = 1.195, 95% CI = 1.111-1.285), similar in direction to 2019 with a higher likelihood of having a non-target A1C identified among those enrolled in ACA versus non-ACA (OR = 1.096, 95% CI = 1.026-1.170).

Uncontrolled (>9%) versus Controlled A1C (≤9%)

Interaction of ACA by HPSA

The interaction of ACA by HPSA (see Figure 2a and Tables 3A and B), provided evidence of a differential effect for the overall interaction term (P < .05). Figure 2a presents the interaction plot. For ACA enrollees, a lower likelihood of having an uncontrolled A1C was identified among those within a HPSA versus a non-HPSA (OR = 0.850, 95% CI = 0.752-0.960), in contrast to non-ACA enrollees with no statistically significant difference between those within a HPSA versus a non-HPSA (OR = 1.095, 95% CI = 0.980-1.224). Among HPSAs no statistically significant difference between those enrolled in ACA versus non-ACA (OR = 1.031, 95% CI = 0.987-1.077) was identified, in contrast to among non-HPSAs with a higher likelihood of having an uncontrolled A1C identified for those enrolled in ACA versus non-ACA (OR = 1.329, 95% CI = 1.133-1.559).

Interaction of Rurality by Border status

The interaction of rurality by border status (see Figure 2b and Tables 3A and B), provided evidence of a differential effect for the overall interaction term (P < .05). Figure 2b presents the interaction plot. Among Rural areas a higher likelihood of having an uncontrolled A1C was identified among those in Border areas compared to non-Border areas (OR = 1.553, 95% CI = 1.352-1.785), similar in direction to among non-Rural areas with a higher likelihood of having an uncontrolled A1C identified among those in Border areas compared to non-Border areas (OR = 1.160, 95% CI = 1.082-1.243). Among Border areas a higher likelihood of having an uncontrolled A1C identified among those in Rural areas compared to non-Rural areas (OR = 1.420, 95% CI = 1.234-1.633), in contrast to among non-Border areas with no statistically significant difference between those in Rural areas compared to non-Rural areas (OR = 1.060, 95% CI = 0.992-1.133).

Interaction of Time by ACA

The interaction of time by ACA (see Figure 2c and Tables 3A and B), provided evidence of a differential effect for the overall interaction term (P < .05). Figure 2c presents the interaction plot. For ACA enrollees a higher likelihood of having an uncontrolled A1C was identified for those in 2016 compared to 2019 (OR = 1.093, 95% CI = 1.025-1.164), in contrast to non-ACA enrollees with no statistically significant difference between those in 2016 compared to 2019 (OR = 0.991, 95% CI = 0.942-1.042). In 2016, a higher likelihood of having an uncontrolled A1C was identified for those enrolled in ACA versus non-ACA (OR = 1.229, 95% CI = 1.116-1.355), similar in direction to 2019 with a higher likelihood of having an uncontrolled A1C identified for those enrolled in ACA versus non-ACA (OR = 1.114, 95% CI = 1.021-1.217).

Discussion

This study identified individual-level, geospatial, and geopolitical variations associated with having A1C levels outside target ranges. For example, individuals in rural areas experienced disparities relative to those in urban areas, which is inconsistent with some prior research indicating a lack of variation in A1C by rurality.²⁷ The reasons for this may include, among other things, a different sample, somewhat narrower timelines in the current study, and analyses of a single state (i.e., not a national study) that likely will not represent other states. In addition, the higher likelihood of non-target and separately uncontrolled A1C among border areas overall, and within rural areas, provides additional evidence that those residing along the US-Mexico border may be particularly at-risk of having non-target A1C, an area highlighted in prior work.^28,29 While a variety of reasons are likely related to this variation, prior studies identified higher rates of A1C testing in the US relative to Mexico,³⁰ which may be relevant to the findings in the current study and consistent in several ways with the WHO Framework¹⁷ highlighting to social and structural determinants of health inequities. Furthermore, the higher risk of non-target and uncontrolled A1Cs among males provides detail that may help targeted intervention (e.g., clinical and broader public health messaging) efforts. Finally, findings of higher likelihoods of uncontrolled diabetes among those enrolled in the ACA plans may be linked to our earlier note that some of the largest groups enrolling in the ACA’s marketplace insurance between 2013 and 2016 were those reporting fair to poor health, Hispanic individuals, individuals aged 50-64 years, and those reporting their highest education as a high school diploma.¹⁶ Thus, it is likely that those enrolled in ACA plans were already facing health inequities, where, for example, past work has identified that individuals with lower education had worse health outcomes and were more likely to die earlier.³¹ However, we identified that among those enrolled in the ACA marketplace insurance, that the probability of having non-target A1Cs and separately having uncontrolled A1Cs decreased over time, thereby indicating that those on the ACA marketplace insurance saw gains, even though disparities between those not enrolled persisted.

Implications for Policy and Practice

In the current study, evidence of specific geospatial factors linked to health inequities can provide insights for practice. For example, disparities facing rural and border areas, as well as among ACA enrollees, can help inform tailored health communication strategies to those facing significant health inequities. Furthermore, insights from this study can be used in clinical interactions between providers and patients with T2D, to identify those most at-risk of having non-desired A1C levels by location of residence, personal characteristics, and/or with ACA insurance eligibility (i.e., informing those who may otherwise be uninsured about their eligibility). Furthermore, these findings can also be utilized to help inform the types of programs that are developed for and targeted to patients with a particular set of characteristics or those individuals residing in more at-risk locations.

This information can also be used by those in the policy sector to improve A1C levels. A prior study identified that between 2013 and 2016, Texas saw an increase in insurance coverage post implementation of the ACA Marketplace, meaning that many enrolled in this coverage may have previously been uninsured. Based on this study, one likely consideration would be expansion of health care coverage to currently uninsured or underinsured individuals, through, among other things, potential Medicaid Expansion throughout Texas. For example, while some evidence of improved A1C among individuals with certain characteristics was identified, this was not consistent for all individuals. For example, those individuals with ACA coverage saw improvements (lower likelihood of uncontrolled A1C over time) from 2016 to 2019, which is in contrast to those individuals with non-ACA coverage that saw no improvements over the same timeline. This may hold important policy implications in that enrollment in ACA coverage, while not likely done on a uniform basis for exactly the same reasons for all, does at least provide an indication that ACA coverage may be linked to improvements in A1C for some. While not identical, prior research identified a link between what was referred to as health insurance stability and improvements in A1C,³² making this finding in the current study of particular interest, given the modifiable nature of policies that might better ensure insurance stability. This should be a consideration for future public health surveillance efforts into policy-relevant measures to increase the likelihood of achieving target A1C.

Limitations

The data provided in this study are from a single insurance provider, yet having coverage in every county of Texas. This may limit generalizability to, for example, those adults with T2D with commercial insurance residing in Texas within the study timeline. However, our sample of just under 50% being female was close the that of the state at 50% being female per the US Census.³³ Additionally, we did not have a sufficient selection of variables to predict exposure to insurance enrollment overall or ACA versus non-ACA enrollment and as such were unable to provide additional analyses such as propensity score adjustments within the current study. We do, however, suggest this for future work, as it will enable further clarity on whether enrollment propensity within this commercially insured sample was driven by patients’ needs and complexity, etc. The A1C thresholds chosen in this study are based on published information from several resources and do provide valuable assessment of risk of non-target A1C more generally. That said, target levels of A1C may be tailored to the individual based on communication with one’s physician.⁴ These more personalized or tailored targets, rather than the more population-based thresholds used in the current study, may not be uniform across persons and may also change over time. Furthermore, while we were able to assess change over time, the timeline of data made available (2016-2019) was shorter than we would prefer when providing more long-term assessments of, for example, progress in A1C levels by key characteristics. Furthermore, it is beyond the scope of this study to assess whether or not individuals were moving from potentially more generous plans to less generous plans, while still being commercially insured and what other life events (e.g., employment changes, other major life events) might have occurred within the study’s timeline. Also, we did not include detail of possible changes across plan types immediately prior to the study by enrollees, as we did not have this data, but would recommend that future work consider long-term enrollment patterns in larger samples across longer time frames, given those continuously enrolled may have different results than those recently enrolled in different plan types, especially if previously uninsured. These and other potential limitations should be considered while assessing the implications of the current study. Further, where possible, we recommend future work include analyses of both insured and uninsured individuals, detail on plan changes (if any), and detail on race and ethnicity with potentially longer time frames to assess change over time with larger samples.

Conclusions

We provide evidence of health inequities facing individuals defined using both individual- (e.g., sex, ACA coverage) and contextual-level characteristics (e.g., rurality, border status, HPSA). These differential experiences of individuals can provide relevant insights and actionable information to clinicians (e.g., improving targeted communication strategies), policymakers (local, state, national) for expanding targeted health coverage, and also other key parties (e.g., public health departments located along the US-Mexico border) seeking to ameliorate health inequities threatening at-risk populations. In particular, rural border areas are facing inequities that are critical to address, given the findings of this study. Along with the need for ongoing public health surveillance, we recommend increased funding to support sustained evidence-based solutions, especially as decreasing the percentage of those with uncontrolled A1Cs is a national priority.¹¹

Footnotes

Acknowledgements

We thank Blue Cross and Blue Shield of Texas for tremendous data support throughout the project. We also thank Drs. Carrie Byington and Nancy Dickey for their leadership and support at the Texas A&M Health Science Center for this joint effort.

Abbreviations or Acronyms Used

ACA = Affordable Care Act (ACA)

NIDDK = National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK)

ADA = American Diabetes Association (ADA)

CDC = Centers for Disease Control and Prevention (CDC)

WHO = World Health Organization (WHO)

FIPS = Federal Information Processing Standards (FIPS)

HPSA = Health Professional Shortage Area (HPSA)

Declaration of Conflicting Interests

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

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by a grant from Blue Cross and Blue Shield of Texas to establish the Texas A&M University Health Science Center Rural Health Moonshot Initiative. We note that one of the authors was employed by a commercial company, Health Care Service Corporation, at the time the study was initiated. All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations. This research was supported by a grant from Blue Cross and Blue Shield of Texas. The funders had no direct role in study design, analysis, decision to publish, or preparation of the manuscript.

Ethics

This study was considered Not Human Subjects Research by both the Texas A&M University Institutional Review Board (IRB) and separately the University of Central Florida IRB (STUDY00004192).

ORCID iD

Samuel D. Towne Jr.

Data Availability Statement

The data that support the findings of this study are available from Blue Cross and Blue Shield of Texas (BCBSTX) but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission of Blue Cross and Blue Shield of Texas. We refer all data access inquiries to the BCBSTX point of contact for this collaborative effort (Mark Chassey, Chief Medical Officer. Mark_Chassay@bcbstx.com) or the Texas A&M point of contact for all data (Mr. Jim Colson, Texas A&M Vice President, Digital Health, jim.colson@tamu.edu).

References

Centers for Disease Control and Prevention. Type 2 diabetes. Accessed 2024. https://www.cdc.gov/diabetes/basics/type2.html#:~:text=More%20than%2037%20million%20Americans,adults%20are%20also%20developing%20it

Yang

Kong

Hsue

, et al. Knowledge of A1c predicts diabetes self-management and A1c level among Chinese patients with type 2 diabetes. PloS One 2016;11(3):e0150753.

Qaseem

Vijan

Snow

, et al. Glycemic control and type 2 diabetes mellitus: the optimal hemoglobin A1c targets. A guidance statement from the American College of Physicians. Ann Intern Med 2007;147(6):417-422.

National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK). The A1C test & diabetes. https://www.niddk.nih.gov/health-information/diagnostic-tests/a1c-test

American Diabetes Association (ADA). Understanding A1C. Accessed 2024. https://www.diabetes.org/diabetes/a1c

Centers for Disease Control and Prevention. All about your A1C. Accessed 2024. https://www.cdc.gov/diabetes/managing/managing-blood-sugar/a1c.html

Pantalone

Misra-Hebert

Hobbs

, et al. The probability of A1C goal attainment in patients with uncontrolled type 2 diabetes in a large integrated delivery system: a prediction model. Diabetes Care. 2020;43(8):1910-1919.

CMS. Medicare Shared Savings Program. Accountable Care Organization (ACO) 2018 Quality Measures. Narrative Specifications Document. 2018. Accessed 2024. https://www.cms.gov/Medicare/Medicare-Fee-for-Service-Payment/sharedsavingsprogram/Downloads/2018-reporting-year-narrative-specifications.pdf

CDC. Diabetes Self-Management Education and Support (DSMES) Toolkit. Accessed 2024. https://www.cdc.gov/diabetes/dsmes-toolkit/business-case/quality-measures.html

10.

Your A1C results: what do they mean? Clin Diabetes. 2006;24(1):9. doi:10.2337/diaclin.24.1.9.

11.

Office of Disease Prevention and Health Promotion, Office of the Assistant Secretary for Health, Office of the Secretary, U.S. Department of Health and Human Services. Reduce the proportion of adults with diabetes who have an A1c value above 9 percent — D‑03. Healthy People 2030. Accessed 2024. https://health.gov/healthypeople/objectives-and-data/browse-objectives/diabetes/reduce-proportion-adults-diabetes-who-have-a1c-value-above-9-percent-d-03

12.

Stratton

Adler

Neil

, et al. Association of glycaemia with macrovascular and microvascular complications of type 2 diabetes (UKPDS 35): prospective observational study. BMJ. 2000;321(7258):405-412.

13.

W-Y

Pei

, et al. Variability in hemoglobin A1c predicts all-cause mortality in patients with type 2 diabetes. J Diabetes Complications. 2012;26(4):296-300.

14.

Weinberg

Waterman

Lucas

, et al. The U.S.-Mexico Border Infectious Disease Surveillance project: establishing bi-national border surveillance. Emerg Infect Dis. 2003;9(1):97-102.

15.

Gennuso

Pollock

Roubal

AM.

Life expectancy at the US-Mexico border: evidence of disparities by place, race, and ethnicity: study examines life expectancy at the US-Mexico border by place, race, and ethnicity. Health Aff. 2021;40(7):1038-1046.

16.

Pickett

Marks

Gain in insurance coverage and residual uninsurance under the Affordable Care Act: Texas, 2013–2016. Am J Public Health. 2017;107(1):120-126.

17.

Solar

Irwin

A Conceptual Framework for Action on the Social Determinants of Health. WHO; 2007.

18.

University of Wisconsin Population Health Institute. County Health Rankings & Roadmaps. 2022. Accessed 2024. https://www.countyhealthrankings.org

19.

Health Resources and Services Administration. Area Health Resource File (AHRF). Health Resources and Services Administration; 2022.

20.

Statistics NCfH. Urban-rural classification scheme for counties. Centers for Disease Control and Prevention. 2013.Accessed 2024. https://www.cdc . . . .

21.

Orelien

. Model Fitting in PROC GENMOD. Paper 264-26. Accessed May 19, 2022. https://support.sas.com/resources/papers/proceedings/proceedings/sugi26/p264-26.pdf

22.

SAS. Generalized Estimating Equations. Accessed 2024. https://support.sas.com/rnd/app/stat/topics/gee/gee.pdf

23.

Liang

K-Y

Zeger

SL.

Longitudinal data analysis using generalized linear models. Biometrika. 1986;73(1):13-22.

24.

Diggle

Heagerty

Liang

K-Y

Zeger

Analysis of Longitudinal Data. Oxford University Press; 2002.

25.

UCLA: Statistical Consulting Group. Using Margins for Predicted Probabilities. UCLA: Statistical Consulting Group. Accessed 2024. https://stats.oarc.ucla.edu/other/mult-pkg/faq/general/faq-how-do-i-cite-web-pages-and-programs-from-the-ucla-statistical-consulting-group/

26.

StataCorp. Margins. Stata Press. 2023. Accessed 2024. https://www.stata.com/manuals/cmmargins.pdf

27.

Mercado

McKeever Bullard

Gregg

Ali

Saydah

Imperatore

Differences in US rural-urban trends in diabetes ABCS, 1999–2018. Diabetes Care. 2021;44(8):1766-1773.

28.

Smith

Zhong

Lee

Towne

Jr Ory

MG.

Effectiveness and economic impact of a diabetes education program among adults with type 2 diabetes in South Texas. BMC Public Health. 2021;21(1):1-12.

29.

Ory

Lee

Towne

Jr Flores

Gabriel

Smith

ML.

Implementing a diabetes education program to reduce health disparities in South Texas: application of the RE-AIM framework for planning and evaluation. Int J Environ Res Public Health. 2020;17(17):6312.

30.

Aponte

Demographics, A1C testing, and medication use of Mexican Americans and Mexicans with diabetes. Diabetes Educ. 2019;45(2):155-163.

31.

Fiscella

Williams

DR.

Health disparities based on socioeconomic inequities: implications for urban health care. Acad Med. 2004;79(12):1139-1147.

32.

Brown

Kressin

Terrin

, et al. The influence of health insurance stability on racial/ethnic differences in diabetes control and management. Ethn Dis. 2021;31(1):149.

33.

US Census Bureau. QuickFacts. Accessed 2024. https://www.census.gov/quickfacts/fact/table/TX,US/PST045223