Stroke Incidence and Outcome Disparity in Rural Regions of Southern West Virginia.

INTRODUCTION: West Virginia has the highest incidence of obesity, smoking, and diabetes within the United States, placing its population at higher risk of stroke. In addition to these endemic risk factors, Appalachia faces various socioeconomic and health care access challenges that could negatively impact stroke incidence and outcomes. At present, there are limited data regarding geographic variables on stroke outcomes in rural Appalachia. We set out to quantify Appalachian geographic patterns of stroke incidence and outcomes.
METHODS: This is a retrospective analysis of all patients hospitalized with a diagnosis of stroke in West Virginia's largest tertiary hospital. During the study (2000-2018), 14,488 patients were analyzed, with an emphasis on those who died from stroke (n = 1022). We first used institutional ICD-9/10 data alongside demographics information and chart reviews to evaluate disease patterns while also exploring emerging hot spot pattern changes over time; we then exploited an emerging time series analysis using temporal trends to assess differing instances of stroke occurrence regionally with hot spots defined as higher than expected incidences of stroke and stroke death.
RESULTS: Data analysis revealed several hot spots of increasing stroke and mortality rates, many of which achieved statistically significant variance compared to expected norms (P = 0.001). Moreover, this study revealed high-risk zones in rural West Virginia wherein the incidence and mortality rates of stroke are suggestively higher and less resistance to economic change than urban centers.
CONCLUSIONS: Stroke incidence and mortality were found to be higher than expected in many areas of rural West Virginia. The higher stroke risk populations correlate with area that may be impacted by socioeconomic factors and limited access to primary care. These high-risk areas may therefore benefit from investments in infrastructure, patient education, and unrestricted primary care. Copyright:

INTRODUCTION

In the United States, the annual incidence of stroke is 795,000,[1] and, on average, a stroke occurs every 40 s. With an estimated annual cost of $73.7 billion, stroke contributes to a significant burden on the national health care cost.[2] Higher regional stroke incidence and mortality rates have been observed in the southeastern United States, which is also known as the Stroke Belt.[34] Landmark studies have confirmed that regional, gender, and racial disparity lead to an increase in these rates: For instance, the Northern Manhattan Study reported that the incidence of stroke in the African American population was 191 per 100,000 compared to 88 and 149 per 100,000 in the Caucasian and Hispanic populations, respectively.[5] Similarly, the Greater Cincinnati/Northern Kentucky Stroke Study revealed that stroke is twice as common in African Americans as in Caucasians.[6]

Stroke incidence in West Virginia has escalated compared to the national average; for instance, 4.8% of the state's population having suffered a stroke compared to 3% of the total US population. With regard to gender, females in West Virginia overreported strokes by 1.5% and males by 3.2% compared to the national average. Furthermore, a significant difference was seen in the age groups of 45–64 years and older than 65, where an excess stroke rate 1.5% and of 1.4% respectively was reported.[7] However, for health care planning and resource utilization, we need population-level, not state-level, data. Disease incidence and mortality can be interpreted differently depending on the methods used, and cluster spatial analysis may be more beneficial when attempting to understand the population-level stroke incidence, mortality, and disparity.[8]

Disease maps can also be used to identify areas of high and low incidence and disease prevalence. In addition, hot spots or spatial clusters can be used to identify areas where neighborhoods share environmental, demographic, or cultural similarities. Utilizing a geographical information system with an associated increase in computing power has transformed the way in which data can be analyzed and interpreted. Spatial data can be classified into three broad categories: Point reference data, point pattern data, and areal data. Areal data are most commonly used in the field of public health; wherein, one can use zip codes to distinguish areas of interest; however, the use of such data to identify locations is not as accurate.[9]

This study uses spatial data analysis to determine stroke incidence and mortality trends in West Virginia, which can be used to guide future health care planning and resource needs.

METHODS

We analyzed stroke incidences from 2000 to 2018 (n = 14,488), including stroke deaths (n = 1022), utilizing the Charleston Area Medical Center (CAMC) data warehouse to evaluate stroke trends within the Center's service area. Moreover, we utilized an emerging hot spot analysis to determine the disease pattern over time. To understand stroke trends, we conducted an emerging time series analysis, which uses a temporal trend to assess differing instances of hot spots, or differing incidences of stroke and stroke mortality. Stroke cases were identified using ICD 9 code 434.91 and ICD 10 code I63.9. Other demographic variables were collected as well, including patients' ages, genders, zip codes, and mortality. Stroke mortality refers to those who died being treated for stroke or a stroke-related condition within the CAMC health care system. Death certificates were used to determine mortality for this analysis.

An emerging hot spot analysis was conducted to identify potential temporal trends of stroke incidence and associated deaths associated with those infarctions within the CAMC service area. Raster data were obtained from the West Virginia GIS house to establish a nested map, while geocoded data were obtained from the CAMC data warehouse and then geocoded based on the address of a given patient using ArcMap 10.7, with a 99% accuracy rate being achieved for the sample rate for the sample (n = 14,488). The raster points were aggregated into space-time bins at half-mile intervals and tallied with an additional temporal trend calculated based on the date of the producer; data were then converted into netCDF data cube format for all emerging hot spots based on 4-month intervals. For each cube with a temporal trend, the GI* equation (see Equation 1) was calculated using the date function to calculate the time series component of the emerging hot spot. Moreover, an emerging hot spot analysis was used to evaluate differing patterns of stroke within the CAMC service area. The GI* equation measures the intensity of the clustering of high and low values. Time values reflected the temporal trend of 4 months for stroke incidence. Stroke deaths (n = 1022) had a temporal distance of 1 mile and a time-step interval of 1 year. Different levels were examined due to the differences among cases, and it appeared to be the most efficient model through multiple interactions. The emerging hot spot analysis used the Mann-Kendall statistic to test for statistical significance through the creation of CDF bins throughout the 18-year analysis period. Each area tested was a 0.5 × 0.5 mi square, with each square being based on the addresses of those examined. Strokes were measured within these areas to define areas of concern where death might be affecting human health. It appears that more hot spots occurred in later years of the analysis. The analysis was initially based on the address level and then subdivided into 0.5 × 0.5 mi squares. It would also be possible to conduct a county-versus local-level analysis if necessary. Finally, to further understand the effect of mortality throughout the entire study period, we calculated an optimized outlier analysis by first calculating the incidence. We then constructed a 2 s optimized outlier analysis using the morality areas defined by the Bounding Polygons of incidences Defining Where Incidents Are Possible within the region. Using this constructed optimized outlier analysis, we could then calculated the at-risk zones in relation to the average median household income calculated by ESRI 2019 median household income layer file for comparison allowed the ability to calculate zones of concern in relation to median household incomes.

RESULTS

Of the 3,415 bins assessed in this analysis, 1642 were statistically significant in three areas [Figure 1], in which new or emerging hot spots were defined as the most recent time-step interval with new areas of stroke incidences. The analysis also indicated 617 areas of consecutive hot time-step intervals. Of the remaining statistically significant zones, 275 sporadically indicated that some of the area intervals were hot and areas of interest. Oscillating zones, of which there were 747, were defined zones of statistical significance; some of these zones were hot and others cold within the 18-year period considered. In this analysis, the incidence of death was also associated with statistically significant zones of concern within the area. Twenty-two new hot spot analysis were observed. Furthermore, zones of consecutive hot time-step intervals arose for 90% of occurrences in 36 consecutive zones.

Figure 1

Emerging hot spot analysis stroke incidence

Sporadic zones were observed at 102 as well, and these were defined as zones that are hot at differing times within the intervals throughout the entire period analyzed. The last interval existed only in one zone that was defined as oscillating, and, in one zone, the area category was defined as one in which some of the time-steps were hot, and others were cold within the zone [Figure 2]. Overall, it appears that stroke incidence occurs at higher levels with multiple neighborhoods in southern West Virginia, but mortality seems to be more suggestively statistically significant in areas with less economic median household income. We identified 110 zones of statistically significant morality zones associated with Stroke in Southern West Virginia. Of these zones 55.5 percent of these zones had a median household income <$47,000 dollars which ranged from $12,100-$47,400 in [Figure 3]. Ninety-eight zones had a statistically significant P = 0.01 in which 49 of these zones where in the $47,400-$82,600 range. Twelve zones which had a statistical significant P value of 0.005 had a 12 areas of income at $47,400.

Figure 2

Emerging hot spot analysis stroke incidence and mortality

Figure 3

Stroke incidence rates (income)

Overall, it seems that zones of incidence of stroke survivability and death occur in both urban and rural environments. This study suggests ongoing educational initiatives and improvement of health care resources are essential for the improvement in stroke incidence and outcomes in some southern West Virginia areas.

DISCUSSION

Previous studies have considered the incidence and prevalence of stroke at state and county levels;[210] however, very few studies have explored stroke distribution at the neighborhood level.[8] Previous studies also confirm that geographic disparities still exist even after adjusting for race, age, socioeconomic status, behavioral risk factors, and comorbidities,[111213] which highlights the need to obtain neighborhood-level data to better understand the occult factors that might contribute to increased stroke incidence in some focal areas of Southern WV. Pedigo et al. performed one of the first studies to identify the importance of neighborhood-level data; they found that considerable variability existed within the county investigated when they examined risk of stroke and myocardial infarction (MI), which was not evident when considering county-level data.[8]

Conflicting results can be obtained when comparing studies considering neighborhood-level characteristics. Previous studies concerning MI risk characteristics used individual-and neighborhood-level data, an approach referred to as multilevel analysis, which reinforces this notion.[141516] The confounding factors that could explain this discrepancy include geography, social conditions, availability of health care facilities, and neighborhood attitudes. It should also be noted that health care planning is often performed at the population level, and thus, studies focused on at the neighborhood level may provide valuable insights into the population-level geographic differences, social factors, and occult factors that may be contributing to the increased rate of stroke in some regions. It can be postulated that the findings of our study result from a lack of health care resources, poor socioeconomic status, geographical differences, and a lack of schools and recreational facilities. While median household income can be calculated, education level is more difficult to determine.[171819]

A French study determined that geocoded areas with high rates of stroke hospitalization were predominantly rural areas with poor socioeconomic status,[20] whereas a large Japanese study found that deprived neighborhoods (identified based on numbers of blue-collar workers and unemployed had had higher stroke incidences.[21] Future interventions could focus on establishing stroke units and rehabilitative facilities in hot spot area to help address the rise in stroke incidence and mortality. Another intervention that could be undertaken is the implementation of a telestroke networks in the endemic hot spot communities. One such project was piloted in France and increased the number of patients with access to thrombolytics.[22]

Health insurance coverage can affect disease incidence and mortality.[23] One study suggests that states that expanded Medicaid coverage exhibited a statistically significant decline in cardiovascular mortality.[24] Medicaid expansion also revealed a decrease in all-cause mortality in patients with end-stage renal disease, increases in the utilization of cardioprotective medications, improvement in the management of diabetes, and a decrease in cardiovascular hospitalizations.[252627]

Previous studies found an inverse relationship between distance from a hospital and the probability of receiving thrombolytic therapy for stroke.[28] Those who live far from a hospital might not be able to reach it within the critical 3–4.5 h, which could explain the decrease in thrombolytic treatment in certain areas.

A limitation of this study that arose due to the use of neighborhood-level data is the small number problem. Analyzing risks in areas with low populations can induce bias in a study due to increased variance. To address the problem of population heterogeneity, the raw incident rates can be smoothed using spatial empirical Bayes (SEB) smoothing.[29] Moreover, areas with an obvious spatial pattern areas with obvious spatial patterns based on the less reliable estimates for areas with lower populations are adjusted according to the local mean.[8]

This study focused on the local level and individual addresses, and the methodology was characterized by a focus on half-mile segments. The study cited above examined an above level that required a smoothing effect. The methodology of this study used the Gi* equation and was based on half-mile squares, with each segment being considered in its own analysis.

CONCLUSION

Increased stroke incidence in Southern West Virginia was evident in the emerging hot spot analysis reported herein. Further data are needed but these finding suggest health care planners should focus on increasing the availability of health care resources in hotspot areas, including but not limited to; implementing telemedicine, improving education, and increasing primary provider physician availability. In addition, new stroke units could be established in these areas to increase the proportion of patients receiving thrombolytic therapy, which could ultimately translate into a lower overall stroke morbidity and mortality burden.

Research quality and ethics statement

This study was approved by the Institutional Review Board / Ethics Committee vide no 17-348. The authors followed applicable EQUATOR Network (“http://www.equator-network.org/) guidelines during the conduct of this research project.

Financial support and sponsorship

Nil.

Conflicts of interest

There are no conflicts of interest.