Introduction
The incidence of vector-borne diseases (VBDs) is a rising concern in the United States, with West Nile virus (WNV) and Lyme disease being the two most commonly reported VBDs.1 WNV is a mosquito-borne arbovirus that is primarily maintained in a transmission cycle between mosquitoes and birds, with humans and other mammals serving as incidental hosts. Lyme disease results from Borrelia burgdorferi bacteria transmitted by infected black-legged ticks.2,3 Over the past 2 decades, both diseases have experienced an alarming increase in the number of reported cases.2–5 The Centers for Disease Control and Prevention (CDC) revealed that more than 59,000 cases of WNV were reported in the United States from 1999 to 2023, resulting in nearly 3000 deaths.6–9 Lyme disease cases increased by 68.5% in 2022 compared with the average number of cases during 2017–2019, with an estimated 476,000 cases reported each year.10,11
There are spatial differences in the distribution of WNV and Lyme disease across the United States, and their geographic range and distribution have continued to expand in recent decades. Since 1995 for Lyme disease and 1999 for WNV, the geographic range of both diseases has expanded, mainly due to favorable climate conditions for ticks and mosquitoes responsible for the spread of these diseases.11,12 In regard to WNV’s spatial spread, Western and Midwestern states report higher incidence during outbreaks, with some states reporting upwards of 2% of their population infected.13 Regarding Lyme disease distribution, areas that have recently reported higher incidence are Maine with 132.7, New Hampshire with 109.4, Rhode Island with 97.6, and Vermont with 141.6 per 100,000 of the population.14,15 In 2023, Lyme disease cases surged in high-incidence states, including Pennsylvania, which accounted for 29% of US cases, followed by New Jersey and New York, with approximately 12% of cases each.15 In addition, Wisconsin reported a record of 6379 Lyme cases in 2023, up from 4,000 per year over 2018–2022.16 WNV and Lyme case reports show seasonal patterns. Both diseases can develop seasonal spikes while expanding their geographical range, especially in warmer areas or warmer months from late spring to summer, with higher mosquito and tick populations.5,15,17,18
WNV and Lyme disease present understudied mortality patterns in the United States. While incidence trends are well documented, mortality data remain limited, and the true burden may be underestimated. It is reported to the CDC that WNV deaths from 1999 to 2020 were 2,456, with 227 in 2021 in the United States.19 Higher mortality rates have been observed among older male adults.19 A significant challenge in estimating the burden of WNV is that approximately 75% of infections are asymptomatic, resulting in substantial underascertainment of infections and complicating disease surveillance.9,13 Consequently, WNV-associated mortality may be underestimated because infections may go unrecognized or deaths may not be attributed to WNV.7,8 Similarly, the number of reported Lyme disease–related deaths is low, in part because fatal disease is uncommon and deaths are often attributed to complications of infection, such as Lyme carditis, rather than Lyme disease itself.
For example, a CDC review of death certificates from 1999 to 2003 identified 23 cases in which Lyme disease was listed as the underlying cause of death, despite over 96,000 cases reported during that time.20,21 These suggest that WNV and Lyme mortality are likely underrecognized in the true population burden of these VBDs. Although numerous studies have explored the prevalence and incidence of WNV and Lyme disease in relation to environmental conditions, less attention has been given to how these factors influence mortality patterns.
While Lyme disease is diagnosed using clinical and serological tests, WNV relies on detecting the virus in blood or cerebrospinal fluid.22–24 Depending on the hospital, health insurance covers acute care, but chronic conditions are often excluded, which creates a financial burden on patients.22,24 These inequalities are more pronounced in rural regions with poor healthcare systems. A smaller number of hospitals and specialized care practitioners delay diagnosis and management, which worsens the disease and increases mortality.25 Simple measures, such as the administration of a single dose of doxycycline to prevent Lyme disease or rapid admission or hospitalization of patients diagnosed with WNV, can minimize complications and reduce mortality.25–28 However, rural residents often face more significant barriers to accessing such care.25,26
In addition, ecological shifts, such as increased irrigation that fosters mosquito breeding and habitat disruption that drives vectors closer to human settlements, indicate the importance of examining broader sociodemographic factors, such as urban–rural status and health insurance coverage, to better understand the disparities in disease mortality outcomes.29 No study to date has assessed the relationship between mortality from WNV and Lyme disease and factors such as urban–rural disparities and health insurance coverage.
With this study, we address essential gaps in knowledge of WNV and Lyme disease mortality patterns in the United States through two main objectives. First, we conducted an analysis of spatial and temporal mortality trends to examine disease death distribution patterns across the United States. Second, we conducted regression analysis to identify the factors that may contribute to higher mortality rates, including sociodemographic characteristics, urban–rural status, and health insurance coverage.
Unlike prior research on incidence and prevalence, this is the first study, to our knowledge, to prioritize mortality outcomes with environmental and demographic factors. Findings from this study will improve public health by providing new evidence to support better resource allocation in high-risk areas, to improve surveillance and monitor disease mortalities, and to develop better intervention strategies.
Materials and Methods
Data Sources and Variables of Interest
Using Multiple Cause of Death microdata from the NCHS from January 2003 to December 2021, we generated county-level counts of WNV and Lyme disease deaths based on the International Classification of Diseases, 10th revision (ICD-10 codes A92.3, A92.30-A93.33, and A93.39 for WNV and ICD-10 codes A69.2, A69.20-A69.23, and A69.29, for Lyme disease) in the conterminous United States. We filtered the data for both the primary and contributing causes of WNV and Lyme disease based on the NCHS mortality dataset. Furthermore, sociodemographic variables, including race (White vs. non-White) and age (≤64 vs. ≥65 years), were included to adjust for differences in mortality risk.
The county of death occurrence, as well as the month and year from each death certificate in the NCHS mortality dataset, was also retained for additional adjustment. Age and race variables were converted into binary variables for the representativeness of each of the categories. County-level annual population estimates for 2012, representing the midpoint of the 2003–2021 study period, were obtained from the U.S.
Census Population and Housing Unit Estimates.30 These population estimates were utilized to calculate mortality rates per county, per 100,000 population, and reflect the spatial distribution of WNV and Lyme disease mortality patterns over time. Further, using Geographic Information Systems, the county-level mortality rates of WNV and Lyme disease were mapped, indicating the spatial distributions of these mortality rates.
The NCHS 2013 binary urban–rural classification codes were used to classify counties with WNV- and Lyme disease–related deaths into metropolitan and nonmetropolitan counties. Per the NCHS urban–rural definition, which is mainly based on county population size, metropolitan counties were those under the categories of large central metro, large fringe metro, medium metro, and small metro. Nonmetropolitan counties were those categorized as micropolitan and noncore.31 Using this standardized classification, metropolitan counties were categorized as urban, and nonmetropolitan counties were identified as rural.
We used county-level economic characteristics data from the U.S. Census Bureau’s American Community Survey, focusing specifically on health insurance coverage32 The dataset spans from 2010 to 2021 and provides estimates on the percentage of insured and uninsured individuals at the county level. Health insurance coverage in the American Community Survey is measured by determining the proportion of the civilian noninstitutionalized population with or without health insurance, making it a critical indicator of access to healthcare services.
For this research, we obtained the mean percentage of uninsured individuals per county across the 12-year period. This variable was used to investigate its association with WNV and Lyme disease mortality rates across the identified counties.
Statistical Approach
We examined the relationship between health insurance coverage, urban–rural status, and mortality from WNV and Lyme disease. Negative binomial regression models were conducted for overdispersion of the death counts, incorporating sociodemographic factors, including race, age, and sex, to control potential confounders. To account for variability across regions, county-level random effects were included in the model.
The primary analyses were conducted using a pooled dataset that combined WNV and Lyme disease, consistent with the study objective of evaluating mortality patterns across the two leading VBDs in the United States. To account for the distinct epidemiologic characteristics of these diseases, disease-specific multivariable models were also conducted as sensitivity analyses. In addition, interaction terms between disease type and the model covariates were evaluated to assess whether the associations differed by disease.
We followed a sequential modeling process using PROC GLIMMIX. The first model included only age, race, and sex as predictors of death counts. The second model added the urban–rural variable to assess how living in an urban versus rural area influenced mortality outcomes.
In the third model, we introduced health insurance status, represented by the percentage of uninsured individuals in each county. We further ran model four to account for the random effect of the counties. This strategy is implemented to improve the heterogeneity that may exist across the counties.
For all models, we included the natural logarithm of the 2012 population as an offset term, which adjusts death counts by population size, allowing for more accurate comparisons across counties with different population densities. As a sensitivity analysis, all negative binomial regression models were repeated using the average county population from 2010 to 2019, the longest continuous period for which complete county-level population estimates were available, to evaluate the robustness of the findings to the choice of population denominator. All statistical analyses were conducted using SAS software, version 9.4,33 and the spatial visualization and mapping were conducted using ArcGIS Pro.34
Results
Our study included 3694 total deaths attributed to WNV and Lyme disease during 2003–2021, with 2628 (71.2%) deaths attributed to WNV and 1066 (28.8%) attributed to Lyme disease (Table 1). Males accounted for most deaths from both WNV (66.2%) and Lyme disease (56.7%). (Table 1).
Those aged 65 and above accounted for the highest number of deaths for both WNV and Lyme disease, comprising 77.1% (2027 deaths) and 61.6% (657 deaths), respectively (Table 1). Most deaths (91.0%) occurred among White individuals, with 89.0% attributed to WNV and 96.0% to Lyme disease (Table 1).
Most deaths occurred in urban areas, with 89.5% of WNV deaths and 80.9% of Lyme disease deaths occurring in urban counties. Rural areas accounted for 10.5% of WNV and 19.1% of Lyme disease deaths, representing 13% of combined WNV and Lyme mortality (Table 1).
In counties where WNV deaths occurred, 11.82% of the county population was uninsured. In counties where Lyme disease-related deaths occurred, 9.92% of the county population was uninsured. Among the insured population, when compared with other counties at the national average, those reporting Lyme disease deaths had a slightly higher percentage of individuals with private insurance (70.08%), while counties with WNV-related deaths showed a higher percentage of individuals with public insurance (32.91%) (Table 1).
WNV and Lyme disease temporal mortality trends indicated variability across the years. WNV death counts peaked in 2003 (246 deaths, 9.36%) and in 2012 (281 deaths, 10.69%), whereas Lyme disease death counts were highest in 2021 (103 deaths, 9.66%) (Figure 1).
Overall, the death count was highest in 2012 (343 deaths, 9.32%) and lowest in 2009 (105 deaths, 2.57%). Figure 2 illustrates the monthly distribution of deaths attributed to WNV and Lyme disease. The data reveal a seasonal pattern for WNV, with mortality rates rising sharply in July, peaking in September at approximately 750 deaths, and then declining through the fall months.
In contrast, Lyme disease mortality remains relatively stable throughout the year (Figure 2).
In Figure 3A, the map presents the distribution of WNV mortality rates, which mainly affect counties in the western and central parts of the United States. High mortality clusters are evident in states such as California, Arizona, Texas, and Colorado, with additional hot spots in the Midwest, especially along the Mississippi River (Figure 3A). In Figure 3B, the map indicates the Lyme disease mortality rate distributions, with a concentration of high mortality counties in the northeastern United States.
States like New York, New Jersey, and Pennsylvania stand out, along with some areas in the upper Midwest, particularly in Minnesota and Wisconsin (Figure 3B).
Table 2 presents the results from the negative binomial regression models analyzing the association between WNV and Lyme disease death counts and county-level predictors, including urbanicity, health insurance status, and demographic variables such as age, sex, and race. With a sequential modeling approach, we explored the impact of various sociodemographic and geographic factors on mortality outcomes.
We examined the distributional properties of the count data. Specifically, we computed the mean and variance of the death counts and found that the estimated variance of 1.54 slightly exceeds the mean of 1.20. This indicates the presence of overdispersion, a condition where the variability in the data is greater than what is assumed under a Poisson distribution.
The negative binomial model is particularly well-suited for such data because it includes an additional dispersion parameter that accounts for this extra variability, leading to more accurate and robust inference. Therefore, the observed overdispersion supports our choice of the negative binomial model over standard Poisson regression.
Model 1 examined the effects of age, sex, and race on mortality. The results indicate that individuals aged 65 and older have a 45% higher risk of death compared with those aged 64 and younger (rate ratio [RR] = 1.45, P < 0.0001). Similarly, males had a 16% higher risk of death than females (RR = 1.16, P = 0.007), and individuals identified as White had a 73% higher risk of death compared with non-White individuals (RR = 1.73, P < 0.0001).
These RRs suggest that age, sex, and race are significant predictors of mortality, with older age, male sex, and White race associated with increased risk.
In model 2, the introduction of urban/rural counties significantly altered mortality predictions. Those living in rural counties had a 93% lower risk of death from WNV and Lyme disease (RR = 0.07, P < 0.0001). The other predictors remained significant, though slightly reduced in magnitude.
Model 3 incorporated health insurance status, focusing on counties with higher percentages of uninsured populations. Counties with higher rates of uninsured individuals tended to experience fewer deaths from WNV and Lyme disease (RR = 0.98, P = 0.0007). Urban/rural indicators remained a significant predictor of mortality in this model (RR = 0.18, P < 0.0001).
Finally, model 4 accounted for county-level random effects to control for heterogeneity across counties. The inclusion of this random effect slightly reduced the magnitude of most associations, but key predictors such as age, sex, and race remained statistically significant. The effect of uninsured counties was no longer significant in this model (RR = 1.01, P = 0.488), suggesting that the variation in death counts could be partially explained by county-level factors.
Sensitivity analyses were conducted to evaluate the robustness of the primary findings. Disease-specific models yielded results generally consistent with the pooled analysis (Table S1, https://links.lww.com/EE/A469). Similarly, models using the average county population from 2010 to 2019 as the population denominator produced findings materially unchanged from those of the primary analysis (Table S2, https://links.lww.com/EE/A469).
Discussion
This study explored the sociodemographic and geographic factors influencing mortality from WNV and Lyme disease in the United States between 2003 and 2021. Regarding temporal patterns, WNV-related deaths peaked in 2003 and 2012, while Lyme disease mortality showed a steady, gradual increase since 2003, reaching its highest level in 2021. In addition, WNV deaths were mainly distributed in the West and Midwest of the United States, while Lyme-related deaths were mainly manifested in the Northeastern United States.
Demographic results indicated that older age, male sex, and the White population were consistently associated with a higher mortality risk for both diseases. Moreover, WNV and Lyme disease mortality rates were higher in urban areas compared with rural areas. Although the percentage of uninsured individuals at the county level showed a positive association with mortality rates, after accounting for county-level heterogeneity, this association was not statistically significant.
Our findings indicate that older adults, particularly those aged 65 and above, present significantly higher mortality rates from these diseases compared with younger populations. This result aligns with previous literature that links age with a greater risk of severe outcomes in VBDs, likely due to a decline in immune responsiveness with age.35,36 Additionally, older adults often experience higher rates of comorbidities such as cardiovascular disease, diabetes, or chronic respiratory conditions, which can exacerbate the severity of these diseases and contribute to higher mortality rates.36 As a large proportion of Lyme and WNV infections are asymptomatic (an estimated 75% for Lyme disease and 70% for WNV),13,20 these diseases may go undiagnosed, increasing the risk of delayed detection and associated complications, which are of particular concern in vulnerable populations. These findings emphasize the critical need for age-specific interventions, including educational campaigns and tailored vector control programs.
Future studies should focus on the role of comorbidities in influencing mortality risks among older populations. Understanding these interactions can help refine intervention strategies and improve clinical management for vulnerable groups.
Our results also demonstrate that males experience higher mortality rates than females for both WNV and Lyme disease. These results are consistent with existing research suggesting that biological and behavioral differences between sexes can affect both exposure to infectious agents and immune response to infections.37,38 Behavioral factors, such as a greater likelihood of outdoor work or recreational activities among males, could increase their exposure to disease vectors. Addressing this gap requires public health messaging that emphasizes risk mitigation among different sex groups, particularly those in outdoor occupations or activities in endemic areas.
Race emerged as another important factor associated with mortality, with White populations exhibiting higher death rates from WNV and Lyme disease compared with non-White populations. This finding aligns with other studies indicating that White populations have a higher incidence of WNV and Lyme disease.39,40 This consistency with prior research further suggests the importance of environmental exposure and sociobehavioral factors. Factors including outdoor occupational or recreational activities, commonly undertaken by White population, may contribute to a heightened vulnerability to these VBDs.
Additionally, geographic clustering of these diseases in regions with a higher proportion of White populations may amplify exposure risks.41,42 Therefore, our findings not only support this observed trend but also underscore the need for targeted interventions in areas where both disease prevalence and demographic vulnerability intersect.
For WNV, annual outbreaks in the United States are frequently reported in Western and Midwestern states such as California, Texas, and Arizona. These regions experience higher transmission due to prolonged mosquito breeding seasons, elevated temperatures, and environmental conditions favorable for vector growth.43,44 For Lyme disease, most cases are concentrated in the Northeastern and Upper Midwestern states, aligning with established patterns observed in prior studies.10,41 Even though previous studies have primarily addressed spatial patterns of these diseases at the state level and focused mainly on case numbers, our findings, which analyze county-level data nationwide and focus on mortality rates, show a notable degree of consistency with these broader patterns. Moreover, the seasonal mortality patterns observed in our analysis generally reflected the known geographic distribution of WNV and Lyme disease.
However, because mortality records capture the timing of death rather than infection or exposure, these findings should be interpreted as mortality patterns rather than direct measures of disease transmission or outbreak timing. Furthermore, individuals residing in rural, nonmetropolitan counties had lower recorded mortality rates than those in metropolitan counties. However, this finding should be interpreted cautiously because the county of residence or death registration may not reflect the actual location of vector exposure.
This distinction is particularly important for Lyme disease, where exposure often occurs in rural, peri-urban, or wooded environments, even among individuals residing in metropolitan counties.45 In rural areas, limited healthcare resources may hinder timely detection and reporting of cases, which can make mortality appear lower than it actually is, even though delays in diagnosis and treatment may increase the risk of severe outcomes.23,46 Additionally, higher population density in urban settings may elevate the risk of disease transmission, particularly for mosquito-borne diseases like WNV, which thrive in areas with standing water and larger human populations.47,48 Studying the interplay of urban–rural residency and health insurance access is essential to understanding disparities in mortality rates and developing equitable strategies to reduce the public health burden of WNV and Lyme disease.
Health insurance status, as measured by the percentage of uninsured individuals per county, was negatively associated with mortality, indicating that counties with higher uninsured rates experienced lower WNV and Lyme mortality. However, this unexpected finding is likely influenced by unmeasured confounding factors or structural differences across counties. Therefore, after adjusting for county-level heterogeneity, the direction of the association reversed, representing a positive, though nonsignificant, relationship between uninsured county-level rates and the mortality outcomes.
This shift underscores the importance of considering county-specific factors and unobserved heterogeneity when interpreting ecological level associations. The diminished significance in our final model implies that other county-level factors, such as healthcare infrastructure and public health capacity, may play a more central role in determining mortality. This finding is consistent with research that highlights the complexity of the relationship between health insurance coverage, access to care, and health outcomes, particularly in underserved communities where healthcare accessibility may be limited regardless of insurance status.28,49,50
Although WNV disease and Lyme disease differ in their epidemiology and transmission dynamics, the primary objective of this study was to evaluate mortality patterns across the two leading VBDs in the United States. Disease-specific sensitivity analyses yielded findings that were generally consistent with the pooled model, and no significant interactions were observed between disease type and the model covariates. These findings support the use of the pooled model while acknowledging that disease-specific analyses remain important for understanding the unique epidemiology of each disease.
The robustness of the findings was further supported by sensitivity analyses using the average county population from 2010 to 2019, which produced results nearly identical to those obtained using the 2012 midpoint population.
Our study is limited by using county-level data, which may obscure individual-level variations in risk factors, healthcare access, and socioeconomic characteristics. Additionally, while our mortality data spans from 2003 to 2021, health insurance data are available only from 2010 to 2021. This discrepancy could affect the observed associations, as the introduction of the Affordable Care Act (ACA) in 2010 significantly expanded health insurance coverage across the United States, particularly through Medicaid expansions and subsidized insurance plans.32 The ACA led to a marked reduction in the uninsured population, with notable improvements in healthcare accessibility for vulnerable groups, including low-income individuals and those with preexisting conditions.
This shift could influence the relevance of health insurance status in our earlier data and should be considered when interpreting our findings.
This research provides valuable insights into the sociodemographic and geographic factors influencing mortality from WNV and Lyme disease across the United States between 2003 and 2021. A key strength of the research lies in its comprehensive approach, integrating descriptive analyses and predictive modeling over an extended period to identify critical trends and associations. By examining variables such as age, sex, race, urban versus rural residence, and health insurance coverage, this study contributes to a deeper understanding of the disparities in mortality outcomes and highlights areas for targeted public health interventions.
Despite these strengths, the study has several limitations that should be considered when interpreting the findings. The use of county-level data, while useful for identifying broader geographic patterns, may obscure individual-level variations in risk factors, healthcare access, and socioeconomic characteristics. This limitation could affect the precision of the observed associations and underscores the need for future research to incorporate individual-level data.
Additionally, the temporal availability of health insurance data from 2010 to 2021 presents another limitation. This discrepancy means that the earlier years of mortality data (2003–2009) could not account for health insurance coverage, potentially affecting the observed relationships. The implementation of the ACA in 2010, which significantly expanded health insurance coverage, may also have introduced temporal biases in the analysis.
It is important to note that both WNV and Lyme disease case definitions have undergone revisions over time, which complicates the comparability of reported cases and deaths over the years. For Lyme disease, the 2008 revision introduced a “probable case” category and broadened the clinical and laboratory criteria, which likely increased the number of cases that could be counted.14,51 More recently, in 2022, high-incidence states were permitted to report cases based solely on laboratory evidence, a change that contributed to a nearly 70% increase in national case counts compared with prior years. Similarly, WNV became nationally notifiable in 2002, with surveillance expanded in 2005 to include both neuroinvasive and non-neuroinvasive cases.
Later, in 2013, the definition of neuroinvasive disease was revised to remove “fever” as a required criterion, which may have altered case counts without reflecting actual changes in transmission.52 Importantly, these evolving case definitions also introduce the possibility that deaths occurring between 2003 and 2021 were underreported, since mortality surveillance depends on accurate case recognition and classification. Taken together, these changes in definitions highlight that observed trends in mortality or incidence may reflect changes in surveillance practices and diagnostic criteria; therefore, caution is essential when making longitudinal comparisons.
Another potential limitation is the possibility of underreporting or misclassification of mortality data, particularly in rural areas where healthcare infrastructure and surveillance systems may be less robust. Lyme disease and WNV may be underdiagnosed or underreported due to differences in diagnostic practices and healthcare provider awareness. Death certificates may reflect these diseases inconsistently, as the inclusion of WNV or Lyme disease as a cause of death often depends on local practices and provider judgment.
Additionally, mortality records represent the timing of death rather than the timing of infection or vector exposure. Because there may be a delay between infection, symptom onset, disease progression, and death, the recorded month of death may not correspond to the month in which exposure occurred. Consequently, the seasonal patterns observed in this study should be interpreted as mortality patterns rather than direct measures of transmission dynamics or outbreak timing.
This variability could lead to an underestimation of the true burden of these diseases, particularly in regions where diagnostic resources are limited or clinical awareness is low. Nonetheless, this study’s findings underscore the importance of addressing systemic healthcare disparities, improving disease surveillance systems, and tailoring public health interventions to meet the needs of high-risk populations. By identifying critical gaps and limitations, this research provides a foundation for future studies to build upon and refine our understanding of VBD mortality.
Conclusion
The results of this study suggest that age, sex, race, urbanicity, and health insurance status are important factors influencing mortality from WNV and Lyme disease. Targeted interventions, particularly for older adults and urban populations, are essential for reducing the public health burden of these diseases. Future research should aim to integrate individual-level data, including comorbidities and healthcare utilization, to provide a more detailed understanding of the drivers of mortality in WNV and Lyme disease cases.
Supplementary Material