Investigation of geographic disparities and temporal changes of non-gestational diabetes-related emergency department visits in Florida: a retrospective ecological study

Introduction

Diabetes is a major public health problem with a significant economic burden as evidenced by the fact that the average healthcare expenditure of a patient with diabetes is 2.3 times higher than that of a patient without diabetes (Yang et al., 2018). Adults with poorly managed diabetes may suffer from short-term complications such as hypoglycemia, hyperglycemia, diabetic ketoacidosis, and long-term complications such as stroke, heart failure, retinopathy, neuropathy, nephropathy, and diabetes-related foot problems such as gangrene, ulcer, Charcot’s foot, and amputation (Mayo Clinic, 2020; Khan et al., 2022). There is evidence that the risk of Emergency Department (ED) visits is higher among persons with diabetes (68 visits/100 persons) than the national average (42.7 visits/100 persons) (Centers for Disease Control and Prevention, 2021a, 2021b). In 2015, about 24% of all ED visits for patients aged 45 or older involved people with diabetes (Washington, Andrews & Mutter, 2013; Hall, Rui & Schwartzman, 2018). Out of 130 million ED visits in 2018 in the US, approximately 17 million visits were due to diabetes and diabetes-related complications.

Emergency departments play a vital role at the interface between the population and the healthcare system. Investigating patterns in the use of ED visits can help identify patterns of health resource utilization, identify disease trends and emerging threats, as well as assess the magnitude and management of disease problems (Kellermann et al., 2013). Previous studies reported that diabetes-related ED visits varied by age, race, ethnicity, income levels, and types of diabetes-related complications (Washington, Andrews & Mutter, 2013; Ginde, Espinola & Camargo, 2008; Menchine et al., 2012; Benoit et al., 2020). Not surprisingly, the southern states of the US, including Florida, have considerably higher diabetes-related ED visits than the rest of the country (Washington, Andrews & Mutter, 2013). A study identified this region as the diabetes belt, an area where diabetes prevalence was significantly higher than the rest of the US (Barker et al., 2011). However, diabetes management programs are not equitably distributed in this area and significant disparities exist in the use of available diabetes management programs (Khan et al., 2021). Lack of access to appropriate and timely diabetes care for some individuals with diabetes results in poor management of the condition leading to otherwise avoidable diabetes-related ED visits resulting in geographic disparities in ED visits (Ricci-Cabello et al., 2010; Barker et al., 2011; Shrestha, 2012; Walker et al., 2014, 2015; Lord, Roberson & Odoi, 2020). Identifying these disparities is important in guiding health planning for these patients. Although a study investigated geographic disparities in diabetes-related ED visits in the US at the regional level (Menchine et al., 2012), no such studies have been done at lower geographic scales and yet this information is important for guiding resource allocation to address the problem at the local level.

Diabetes is considered an ambulatory care sensitive condition (ACSC), a condition in which appropriate ambulatory/outpatient care can prevent complications and the need for ED visits and hospitalizations (Agency for Healthcare Research and Quality, 2001). Therefore, disparities in diabetes-related ED visit may indicate differential outpatient care access, continuity, and quality (Johnson et al., 2012). Previous studies identified disparities in ED utilization among traditionally underserved groups such as Black, Hispanic, uninsured, and low-income patients (Sun, Burstin & Brennan, 2003; Hong, Baumann & Boudreaux, 2007). Since non-Hispanic Black and Hispanic populations represent almost half of the Florida population and there is evidence of a 54% increase in diabetes-related ED visits in Florida between 2011 and 2016, investigating temporal changes and geographic disparities in diabetes-related ED visits is necessary for a better understanding of diabetes burden in Florida (Florida Diabetes Advisory Council, 2019; Florida Department of Health, 2022). Identifying areas with high diabetes-related ED visit rates in Florida could help identify areas with inadequate access to ambulatory care and poor quality of diabetes management (Dowd et al., 2014). This knowledge is important for planning programs targeted at improving access to primary diabetes care, reducing the burden of the condition and its complications, and improving population health. In addition, socioeconomic and demographic predictors of disparities in diabetes-related ED visits, if identified, would help guide resource allocation geared towards reducing disparities in availability of diabetes care as well as diabetes burden in Florida. Therefore, the objectives of this study were to: a) investigate geographic disparities and temporal changes in non-gestational diabetes-related ED visit rates in Florida between 2016 and 2019; b) identify predictors of geographic disparities in non-gestational diabetes-related ED visit rates in Florida.

Materials and Methods

Study design and area

This retrospective ecological study was conducted between 2022 and 2023. Temporal changes in monthly non-gestational diabetes-related ED visit rates (per 100,000 person-months) were investigated. Annual non-gestational diabetes-related ED visit rates (per 100,000 person-years) were computed at the county level and clusters of counties with high non-gestational diabetes-related ED visit rates were identified and displayed on maps. Socioeconomic and demographic characteristics, health-related behavior, and environmental factors were investigated as potential predictors of non-gestational diabetes-related ED visit rates.

The study area encompassed the entire state of Florida and covered the time period from January 1, 2016 to December 31, 2019 (Fig. 1). As of 2020, Florida was the most populous state in the southeastern US with approximately 21.6 million people (Florida Department of Health, 2020a). Twenty-two percent of Florida’s population was 0–19 years old, 31% was 20–44 years old, 26.1% was 45–64 years old, and the rest (20.9%) were 65 years old or older. Approximately half of the population was female. By race, White represented the majority (77.2%) of the population, Black were 17.0%, and all other races comprised the rest (5.8%). By ethnicity, 26.7% of the population was Hispanic-Latino while the rest were non-Hispanic (of any race) (Florida Department of Health, 2020a). The most urban and populous county was Miami-Dade (population: 2.9 million) located in the southern part of the state, while the most rural and least populous county was Lafayette (population: 8,721) located in the northern part of the state (Florida Department of Health, 2020a). In total, Florida has 67 counties many of which are considered part of the diabetes belt which is an area located in the southeastern part of the US where diabetes prevalence is higher (11.7%) compared to the rest of the US (8.5%) (Barker et al., 2011).

Results

There was a total of 34,285,646 ED visits reported in Florida from January 1, 2016 to December 31, 2019. This study included only 2,871,326 ED visits, which had a diagnosis of non-gestational diabetes or non-gestational diabetes-related complications.

Temporal pattern

Overall, non-gestational diabetes-related ED visit rates in Florida increased significantly (p < 0.001) from 266 visits per 100,000 person-months in January 2016 to 332 visits per 100,000 person-months in December 2019 (Fig. 2). The highest non-gestational diabetes-related ED visit rate (332 ED visits per 100,000 person-months) was observed in December 2019, while the lowest (259 ED visits per 100,000 person-months) was in June 2016.

Spatial distribution

Geographic distribution of non-gestational diabetes-related ED visit rates varied across counties in Florida ranging from 1,448 to 10,211 visits per 100,000 person-years (Fig. 3). Overall, more than half of the counties had high non-gestational diabetes-related ED visit rates (>3,385 ED visits per 100,000 person-years) during the study period. Almost all counties in rural northern Florida, including the entire panhandle area up to the westernmost part of the state, had higher non-gestational diabetes-related ED visits than counties in the southern part of the state (Figs. 1 and 3). However, a few counties in the north-central portion had low ED visit rates in 2016 and 2017. Most of the counties in the central part of the state tended to have high non-gestational diabetes-related ED visit rates, while low rates were consistently observed in the southernmost part and urban coastal areas during the study period (Figs. 1 and 3).

Clusters of high non-gestational diabetes-related ED visit rates

Consistent with high non-gestational diabetes-related ED visit rates observed in the northern and central parts of Florida, significant (p < 0.05) high non-gestational diabetes-related ED visit rate clusters were identified in these areas (Figs. 3, 4, and 5). Overall, clusters were similar in size and location across years (Tables 2 and 3, Figs. 4 and 5). Using Tango’s FSSS, the primary high-rate clusters were consistently identified in the south-central portion of the state and mainly included rural counties (Hardee, Highlands, Okeechobee, and Hendry) (Figs. 1 and 4). Similarly, several small high-rate clusters located across the panhandle area included only rural counties. However, a high-rate cluster was identified in the rural-urban interface of north-central Florida and included both rural and urban counties. Although high-rate clusters were not identified in the southern and coastal urban areas, a single county high-rate cluster (Escambia) was consistently identified in the westernmost urban part of the state (Figs. 1 and 4). Similar to the findings of Tango’s FSSS, Kulldorff’s CSSS identified significant high non-gestational diabetes-related ED visit rate clusters in the panhandle area, north-central, and central portions of the state and included mostly rural counties. However, unlike Tango’s FSSS, larger but fewer clusters were identified by Kulldorff’s CSSS (Tables 2 and 3, Figs. 4 and 5).

Table 2:

Statistically significant non-overlapping spatial clusters of high diabetes-related emergency department visit rates identified in Florida using Tango’s flexible spatial scan statistics, 2016–2019.

Year	Cluster*	Population	Observed ED1 visits	Expected ED visits	No. of counties	RR2	p-value3
2016	Cluster 1	1,578,659	70,233	47,893	7	1.47	0.001
	Cluster 2	135,942	8,871	4,409	5	2.01	0.001
	Cluster 3	859,052	36,193	26,947	7	1.34	0.001
	Cluster 4	971,842	38,837	30,031	1	1.29	0.001
	Cluster 5	322,901	15,028	10,054	1	1.49	0.001
	Cluster 6	14,842	736	469	1	1.57	0.001
2017	Cluster 1	1,578,659	77,918	51,596	7	1.51	0.001
	Cluster 2	2,089,357	93,015	69,367	10	1.34	0.001
	Cluster 3	98,251	7,676	3,374	2	2.28	0.001
	Cluster 4	322,901	14,129	10,661	1	1.33	0.001
	Cluster 5	28,524	1,557	987	2	1.58	0.001
	Cluster 6	45,565	2,128	1,536	2	1.39	0.001
	Cluster 7	37,494	1,721	1,253	2	1.37	0.001
2018	Cluster 1	1,209,981	65,020	41,651	6	1.56	0.001
	Cluster 2	1,809,443	85,977	61,927	9	1.39	0.001
	Cluster 3	98,251	8,324	3,436	2	2.42	0.001
	Cluster 4	322,901	14,011	11,020	1	1.27	0.001
	Cluster 5	129,934	6,300	4,519	7	1.39	0.001
	Cluster 6	45,565	2,106	1,587	2	1.33	0.001
2019	Cluster 1	936,626	57,579	34,683	6	1.66	0.001
	Cluster 2	1,592,334	81,540	58,963	8	1.38	0.001
	Cluster 3	47,926	4,877	1,775	1	2.75	0.001
	Cluster 4	322,901	16,641	11,957	1	1.39	0.001
	Cluster 5	128,345	6,717	4,753	6	1.41	0.001
	Cluster 6	37,494	2,137	1,388	2	1.54	0.001

DOI: 10.7717/peerj.18897/table-2

Notes:

1 Emergency department.

2 Rate ratio = Observed ED visits/Expected ED visits in the potential cluster.

3 Statistical significance was assessed using a critical p = 0.05.

* Cluster 1: Primary Cluster; Cluster ≥2: Secondary Cluster.

Spatial scan parameters: Maximum cluster size = 15 counties; Test = one-tailed (high rates); RR for reporting clusters ≥ 1.2.

Geographic distribution of the clusters is shown in Fig. 4.

Table 3:

Statistically significant non-overlapping spatial clusters of high diabetes-related emergency department visit rates identified in Florida using Kulldorff’s circular spatial scan statistics, 2016–2019.

Year	Cluster*	Population	Observed ED1 visits	Expected ED visits	No. of counties	RR2	p-value3
2016	Cluster 1	1,846,525	81,070	59,761	11	1.41	<0.001
	Cluster 2	2,698,053	109,703	87,319	27	1.31	<0.001
2017	Cluster 1	1,886,441	90,568	64,295	11	1.47	<0.001
	Cluster 2	2,731,991	122,774	93,114	27	1.39	<0.001
2018	Cluster 1	2,759,367	129,329	95,906	27	1.42	<0.001
	Cluster 2	709,127	41,802	24,647	2	1.74	<0.001
	Cluster 3	601,631	27,223	20,911	4	1.31	<0.001
	Cluster 4	39,682	2,160	1,379	1	1.57	<0.001
2019	Cluster 1	716,081	45,210	26,516	2	1.75	<0.001
	Cluster 2	2,798,463	134,345	103,625	27	1.36	<0.001
	Cluster 3	609,119	29,056	22,555	4	1.30	<0.001
	Cluster 4	539,563	25,178	19,980	1	1.27	<0.001
	Cluster 5	40,089	2,353	1,484	1	1.59	<0.001

DOI: 10.7717/peerj.18897/table-3

Notes:

1 Emergency department.

2 Rate ratio = ED visit rate inside the potential cluster/ED visit rate outside the potential cluster.

3 Statistical significance was assessed using a critical p = 0.05.

* Cluster 1: Primary Cluster; Cluster ≥2: Secondary Cluster.

Spatial scan parameters: Maximum window size = 13.5% of total population; Test = one-tailed (high rates); RR for reporting clusters ≥ 1.2.

Geographic distribution of the clusters is presented in Fig. 5.

Predictors of non-gestational diabetes-related ED visit rates

Tables 4 and 5 show the results of the univariable and final multivariable negative binomial models used to investigate associations between county-level sociodemographic variables and county-level non-gestational diabetes-related ED visit rates, respectively. Several of the variables assessed had univariable statistically significant associations with the outcome based on a relaxed critical p-value of 0.15. Almost half of these variables were dropped during the variable selection process due to the presence of high collinearity and the fact that none of the dropped variables were important confounders. Additionally, no statistically significant two-way interactions were identified. Thus, only variables that had statistically significant (p < 0.05) associations with non-gestational diabetes related ED visit rates were kept in the final model.

Table 4:

Results of univariable negative binomial regression models examining the associations between county characteristics and the rate of diabetes-related emergency department visits in Florida, 2019.

Predictor variable	IRR1 (95% CI2)	p-value3
Percent with age less than 20 years	1.034 [1.014–1.056]	0.001
Percent with age 20 to 44 years	1.020 [1.004–1.035]	0.012
Percent with age 45 to 64 years	0.984 [0.947–1.022]	0.397
Percent with age ≥65 years	0.986 [0.976–0.995]	0.003
Percent non-Hispanic White	0.994 [0.989–0.999]	0.023
Percent non-Hispanic Black	1.015 [1.007–1.022]	<0.001
Percent Hispanic	0.999 [0.993–1.005]	0.743
Percent non-Hispanic other races	0.967 [0.924–1.013]	0.155
Percent male	1.017 [0.998–1.036]	0.073
Percent female	0.983 [0.966–1.002]	0.073
Percent of having less than high school education	1.035 [1.025–1.045]	<0.001
Percent of having high school education	1.023 [1.012–1.033]	<0.001
Percent of having some college education	0.964 [0.950–0.978]	<0.001
Percent of having college education	0.975 [0.968–0.982]	<0.001
Percent that income less than 25 k per year	1.030 [1.021–1.039]	<0.001
Percent that income 25 to 50 k per year	1.026 [1.008–1.045]	0.005
Percent that income more than 50 k per year	0.978 [0.972–0.984]	<0.001
Percent unemployed	1.227 [1.092–1.379]	<0.001
Percent married	0.975 [0.962–0.988]	<0.001
Percent divorced/widowed/separated	1.026 [1.003–1.049]	0.028
Percent never married or unmarried couple	1.014 [1.000–1.028]	0.050
Percent of having overall poor health	1.041 [1.028–1.054]	<0.001
Percent of having overall good health	0.961 [0.949–0.972]	<0.001
Percent of being highly active	0.970 [0.959–0.982]	<0.001
Percent of being active	0.969 [0.949–0.990]	0.003
Percent of being insufficiently active	0.990 [0.965–1.015]	0.417
Percent of being inactive	1.032 [1.022–1.041]	<0.001
Percent of having normal weight	0.969 [0.956–0.982]	<0.001
Percent of being obese	1.038 [1.028–1.048]	<0.001
Percent of being overweight	0.962 [0.945–0.980]	<0.001
Percent of having less than normal weight	0.998 [0.919–1.084]	0.960
Percent that eat vegetables ≥once a day	0.980 [0.964–0.995]	0.011
Percent that eat fruits ≥once a day	0.980 [0.967–0.992]	0.002
Percent that lack access of healthy food	1.009 [0.996–1.023]	0.167
Percent with food insecurity	1.010 [1.068–1.132]	<0.001
Percent that have access of exercise opportunity	0.995 [0.992–0.999]	0.005
Percent of being current smokers	1.025 [1.010–1.041]	0.001
Percent of being current tobacco or snuff user	1.029 [1.006–1.054]	0.016
Percent of being current ecig users	0.990 [0.947–1.034]	0.637
Percent of being heavy drinkers	0.944 [0.915–0.975]	<0.001
Percent without insurance coverage	1.042 [1.026–1.059]	<0.001
Percent that could not see a doctor in the last 12 months	1.048 [1.024–1.073]	<0.001
Percent that have a personal doctor	0.992 [0.977–1.007]	0.289
Number of primary care physician per 100 k population	0.995 [0.992–0.997]	<0.001
Percent of houses with no vehicle	1.060 [1.022–1.100]	0.002
Percent of having diabetes	1.054 [1.031–1.078]	<0.001
Average age of diabetes diagnosis	0.961 [0.932–0.990]	0.008
Percent of attending DSME	0.989 [0.983–0.996]	0.001
Percent of being depressed	1.009 [0.986–1.032]	0.455
Percent that have any disability	1.028 [1.015–1.042]	<0.001
Percent of having kidney disease	1.078 [1.011–1.149]	0.023
Percent that have regular checkup	0.984 [0.964–1.005]	0.128
Percent that take medications for high cholesterol	1.010 [0.994–1.025]	0.222
Percent that take medications for hypertension	0.985 [0.965–1.005]	0.130
Percent of having myocardial infarction or heart disease	1.042 [0.989–1.098]	0.119
Percent of having stroke	1.076 [1.016–1.140]	0.012
Percent of having arthritis	1.000 [0.985–1.015]	1.000
Percent that have high cholesterol	0.989 [0.969–1.010]	0.299
Percent of having hypertension	1.019 [1.005–1.033]	0.009
Percent of rural population	1.004 [1.002–1.007]	<0.001
Air quality (Average parts per million)	1.089 [1.003–1.183]	0.042

DOI: 10.7717/peerj.18897/table-4

Notes:

1 Incidence rate ratio.

2 Confidence interval.

3 Statistical significance was assessed using a critical p $\le$ 0.15.

Table 5:

Results of a multivariable negative binomial regression model examining the associations between county characteristics and the rate of diabetes-related emergency department visits in Florida, 2019.

Predictor variable	IRR1 (95% CI2)	p-value3
Percent non-Hispanic Black	1.010 [1.004–1.016]	<0.001
Percent of having diabetes	1.036 [1.018–1.054]	<0.001
Percent without insurance coverage	1.019 [1.007–1.032]	0.003
Percent current smokers	1.019 [1.009–1.030]	<0.001
Percent married	0.986 [0.975–0.997]	0.010

DOI: 10.7717/peerj.18897/table-5

Notes:

1 Incidence rate ratio.

2 Confidence interval.

3 Statistical significance was assessed using a critical p $\le$ 0.05.

Based on the final multivariable negative binomial model, the following independent variables had significant high county-level non-gestational diabetes-related ED visit rates: percentages of population who were non-Hispanic Black, current smokers, had diabetes, and had no insurance coverage (Table 5). However, percentage of population that were married had statistically significant low diabetes-related ED visit rate (Table 5). The p-value of the deviance χ² goodness-of-fit test was not significant (p = 0.259) indicating a good model fit. Based on the Cook’s Distance, Glades and Union counties were identified as highly influential observations. Glades county had the highest diabetes prevalence, while Union County had the highest percentage of smokers.

Geographic distributions of the significant predictors of non-gestational diabetes-related ED visit rates are shown in Fig. 6. Counties in the central and eastern panhandle rural area tended to have high percentages of population that were non-Hispanic Black, current smokers, had diabetes, and had no insurance coverage. These counties overlapped with many counties with high non-gestational diabetes-related ED visit rates (Figs. 1, 3, and 6). On the other hand, counties in the southernmost urban area of the state had relatively low non-gestational diabetes-related ED visit rates but tended to have high percentages of non-Hispanic Black and low percentages of married, current smokers, and those with diabetes. Although counties with high percentages of population with diabetes were concentrated in the mid to south-central parts of the state, these counties had high percentages of uninsured people (Figs. 1, 3, and 6).

Discussion

This study investigated geographic disparities and temporal changes of non-gestational diabetes-related ED visit rates in Florida from 2016 to 2019. Although diabetes prevalence and healthcare access for individuals with diabetes vary across counties in Florida (Ricci-Cabello et al., 2010; Barker et al., 2011; Lord, Roberson & Odoi, 2020), little is known about the geographic disparities of non-gestational diabetes-related ED visits and yet this information is critical for improving the health and quality of life of populations with diabetes. The findings of the current study help to fill this gap and are important for guiding healthcare planning targeted at reducing disparities in non-gestational diabetes-related ED visit rates in Florida. In addition, this study investigated sociodemographic, environmental, and lifestyle-related predictors of non-gestational diabetes-related ED visit rates. The results are useful for guiding evidence-based resource allocation aimed at guiding the implementation of control programs and reducing the burden of diabetes in Florida.

The observed high non-gestational diabetes-related ED visit rates in the northern and southcentral parts of Florida may be related to access to diabetes care in these rural communities. Since diabetes is an ambulatory-care sensitive condition, getting regular primary care could substantially reduce ED visit risks and improve quality of life (Agency for Healthcare Research and Quality, 2001). However, access to primary care for diabetes could be limited due to the lack of health insurance coverage, which could result in higher diabetes-related ED visit rates. This is evidenced by the findings of this study because counties with high percentages of populations with no health insurance coverage had high diabetes-related ED visit rates. Additionally, a study by Khan et al. (2021) reported that northern Florida had lower diabetes self-management education (DSME) program participation rates than urban areas due to fewer DSME centers and limited accessibility to DSME program. The DSME program was developed to educate diabetes patients on disease management and reduce diabetes-related complications and ED visit rates (Powers et al., 2015). However, lack of health insurance coverage among populations in northern Florida might have prevented access to DSME programs, lowered DSME participation rates, and, therefore, resulted in higher ED visit rates.

Another reason of the high non-gestational diabetes-related ED visit rates in northern Florida could be the high percentages of non-Hispanic Black populations in those areas. The findings of this study showed that counties with high percentages of non-Hispanic Black populations tended to have significantly high ED visit rates, which is consistent with the findings of previous studies (Taylor et al., 2017; Uppal et al., 2022). A study by Uppal et al. (2022) reported that diabetes-specific ED use among non-Hispanic Black patients was approximately 3 times higher than among non-Hispanic White patients. This difference was the result of non-Hispanic Black patients having higher risks of diabetes-related complications such as albuminuria, retinopathy, lower extremity amputation, end-stage renal disease (ESRD), and worse glycemic control than their non-Hispanic White counterparts (Osborn, De Groot & Wagner, 2013; Canedo et al., 2018). Moreover, minority groups, such as Black and Hispanic populations, are less likely to receive recommended diabetes preventive care (Taylor et al., 2017) and have low participation in DSME programs (Khan et al., 2021).

The significant low county-level non-gestational diabetes-related ED visit rates among the percentage of married people identified in this study suggests that social support from marriage relationships may be beneficial for diabetes patients. Previous studies reported lower diabetes morbidity and mortality among married persons compared to their unmarried counterparts (Kposowa, Ezzat & Breault, 2021). This relationship is explained by the fact that married people are more likely to get better social and mental support (Umberson, 1992), lead healthy lifestyles (Eng et al., 2005), and have better medication adherence and diabetes management (Gelaw et al., 2014; Ahmed, Abugalambo & Almethen, 2017).

The finding that counties with high percentages of current smokers had higher non-gestational diabetes-related ED visit rates is consistent with reports from previous studies (U.S. Department of Health and Human Services, 2010, 2014). According to a report published by the Centers for Disease Control and Prevention (CDC), people who smoke and have diabetes are more likely to develop serious health problems from diabetes such as heart disease, kidney disease, retinopathy, peripheral neuropathy, and lower leg amputations (U.S. Department of Health and Human Services, 2010; Centers for Disease Control and Prevention, 2022). Moreover, these people tend to have trouble with insulin dosing and managing diabetes (U.S. Department of Health and Human Services, 2010, 2014).

Strengths and limitations

This is the first study investigating geographic disparities of non-gestational diabetes-related ED visits in Florida using rigorous statistical approaches. Identifying areas with high diabetes-related ED visit rates is crucial for guiding resource allocation and improving access to primary diabetes care. This study also investigated sociodemographic predictors of non-gestational diabetes-related ED visits in Florida, the findings of which are important for guiding programs aimed at reducing disparities in the availability of diabetes care and improving the health of populations with diabetes in Florida. Although this study was conducted in Florida, the findings of this study showed that spatial statistics are useful for identifying geographic disparities of ED visit rates across states in the US. However, this study is not without limitation. Coding errors might have occurred in the use of ICD-10 codes to report diabetes and diabetes-related conditions in ED data due to the complex nature of the ICD-10 coding system that might result in coding errors potentially leading to misclassification of the outcome and hence misclassification bias (American Academy of Family Physicians, 2013; Burles et al., 2017). Similarly, errors in identifying geographic location might have resulted in location misclassification bias. However, this is expected to be minimal, if at all present, and is expected to have little to no impact on study findings since only aggregated county-level data were used. Additionally, since this was a retrospective study that used previously collected administrative data, the investigation was limited to only variables that were available in the datasets used for the study. It is important to note that since this study was conducted at the county-level, it is potentially subject to ecological bias so our inferences to the individual-level should be considered with caution. Finally, BRFSS data used in this study may be prone to reporting bias. However, previous research has shown that findings generated from BRFSS data are representative of the population (Stein et al., 1995; Bowlin et al., 1996). These limitations notwithstanding, the findings of this study provide useful information for guiding health planners in allocating resources and reducing diabetes burden in Florida.

Conclusions

This study identified geographic disparities of non-gestational diabetes-related ED visit rates in Florida with high-rate areas being observed in the rural northern and southcentral parts of the state. Lack of healthcare access, high diabetes prevalence, low levels of insurance coverage, and certain demographic factors were identified as significant predictors of high non-gestational diabetes-related ED visit rates. These findings are useful for guiding public health efforts geared at reducing disparities and improving diabetes outcomes in Florida.

Supplemental Information