Lyme disease cases that meet the case definition by clinical criteria and/or laboratory criteria are reported to local and state health departments by health care providers. Once personal identifiers are removed, these data are reported to the Centers for Disease Control and Prevention (CDC). The most recent case definition, published by the CDC in 2022, characterizes Lyme disease by 1 or more of the following clinical criteria in the absence of another known etiology: erythema migrans, lymphocytic meningitis, cranial neuritis including facial palsies, radiculoneuropathy, encephalomyelitis, recurrent joint swelling in 1 or a few joints, and acute onset of high-grade atrioventricular conduction defects that resolve within days to weeks [11]. Laboratory criteria include the isolation of the causative pathogens B burgdorferi or B mayonii in culture, detection of the causative pathogens by nucleic acid amplification test assay, detection of B burgdorferi antigens by immunohistochemical assay on biopsy, or positive serologic testing for B burgdorferi immunoglobulin M or G [11].

Data on reported numbers of Lyme disease cases were extracted from the cases reported each month in the United States by CDC Lyme Disease Surveillance data sources from 2008 to 2020 [12]. Our study also focused on North Carolina, a state in the southeastern United States, where Lyme disease cases are publicly reported regularly, including at the county level. Monthly North Carolina data were extracted from the North Carolina Department of Health and Human Services (NCDHHS) monthly communicable disease reports for the years from 2017 to 2020. As monthly data were no longer reported using similar formats, a plot digitizer was used to extract the monthly Lyme disease cases from the annual reports from 2021 to 2022. Data on the yearly number of cases by North Carolina county were extracted from NCDHHS’s “Disease Data Dashboard” [13]. Population estimates for the entire United States for each year analyzed were retrieved from the United Nations Division of Population country estimates [14]. North Carolina estimates used the estimated population by county for each year of analysis, using data from the North Carolina Office of Management and Budgets [15].

This study was approved by the Wake Forest University School of Medicine Institutional Review Board (IRB00117564). All authors had permission to use these data.

We used a Bayesian regression framework fitting data prior to 2020 to estimate the under-ascertainment of cases by month and year. Cyclic cubic regression splines with 12 knots representing each month of the year were used to account for the seasonal dynamics of infection during the year. We fitted these models using both Poisson and negative binomial generalized linear models with log link functions, including population estimates as offsets. We included a random effect to account for the changing case definition when modeling the national case data [16]. Models were fitted using the brms R package (R Foundation for Statistical Computing) [17] and assessed using Pareto smoothed importance sampling [18]. In each case, weakly informative priors of Student t distribution with 3 degrees of freedom, a mean of 0, and a scale of 1046 were used, with the exclusion of a unit normal distribution prior placed on the case definition random effect. Model convergence was assessed using the Gelman-Rubin diagnostic with values of Rhat less than 1.01 and sufficient effective samples in the posterior estimates. Case ascertainment was defined as the difference between the model estimated number of cases and the reported number of cases: (Estimated number of cases – Actual number of reported cases) / Estimated number of cases).

In a geospatial analysis, we fitted generalized regression models for negative binomial distributed data with a log link function, using an offset for population on the yearly number of Lyme disease cases by North Carolina county. To account for spatial autocorrelation among the counties, we included a smoothed term using the centroid of each county, accounting for both the longitude and latitude of the county centroid. A yearly fixed effect was included to account for the secular increase in the reported number of Lyme disease cases. Model fitting was conducted using the mgcv R package [19]. Case ascertainment rates were calculated as defined above.

We conducted all analysis in R (version 4.1.3).

We found evidence of an increase in the under-ascertainment of Lyme disease during the pandemic period, using the historical case rates, in both the United States and North Carolina in 2020. Following February 2021, as COVID-19 cases began to decline in North Carolina (Figure S3 in Multimedia Appendix 1), reported cases of Lyme disease returned to expected levels. This under-ascertainment of Lyme disease demonstrates the importance of a robust surveillance system with well-defined diagnostic criteria of this well-documented emerging infectious disease in the southeastern United States [5,20]. The unique reporting structure of North Carolina allows us to characterize the dynamics of the ascertainment rate at the county level during this period. With the risk of acquiring Lyme disease being the same as years prior, if not higher due to the increased popularity of outdoor recreational activities, it is reasonable to suggest that there was an increased number of cases of Lyme disease that went undiagnosed and untreated during this period. Our estimates should likely be taken as the upper bound of the likely ascertainment rate of Lyme disease cases.

Our findings support early research by McCormick and colleagues [6], who estimated that fewer Lyme disease laboratory tests were reported during the year 2020 than 2019. However, our findings quantitatively establish likely bounds for the numbers of under-ascertained cases of Lyme disease at a national level and at the finer county level in North Carolina, as well as estimate when the normal under-ascertainment of cases resumed. Understanding the dynamics of case ascertainment is vital, especially for passive surveillance systems that are already biased toward persons who seek health care [9,21]. Passive disease surveillance data are often used to feed models to identify potential outbreaks of infectious diseases [10]. Large variations in the ascertainment rates could potentially mask outbreaks of disease or similarly trigger action for case investigations when it is not warranted.

In 2020, the most populous counties in North Carolina, including Wake, Mecklenburg, and Guilford, showed the highest under-ascertainment of cases (Figure 2 and Table S2 in Multimedia Appendix 1), suggesting that individuals in higher-populated counties demonstrated behavioral changes, resulting in lower cases being reported to local health departments. During the COVID-19 pandemic, many outpatient medical offices switched to telehealth platforms to protect patients and staff, and hospitals became overwhelmed with increased numbers of patients. People in densely populated areas may have avoided medical centers to decrease their risk of exposure to COVID-19, and by not receiving medical care, a diagnosis for Lyme disease may have been missed. Globally, there was wider adoption of telehealth as a consequence of the pandemic [22]. However, a systematic review of the challenges of telehealth during the pandemic found that the inability to conduct a comprehensive physical examination was the top barrier to diagnosis during a telehealth encounter [23]. This could be further compounded by the delay or inability to receive laboratory diagnostics for Lyme disease during the pandemic period. Furthermore, an estimated 41% of adults in the United States avoided or delayed medical care during the COVID-19 pandemic by June 30, 2020 [24], further underlining the high likelihood of under-ascertainment we detected.

In 2020, we also found that the counties that had the highest number of reported cases compared to their predicted cases included Ashe, Yancey, and Buncombe counties (Figure 2). These counties, found in Western North Carolina, include many popular outdoor travel and recreational spots such as the Blue Ridge Parkway, Asheville, and Mount Mitchell. This suggests that behavioral changes, such as increased travel and utilization of outdoor recreational activities during the COVID-19 pandemic, could have resulted in more exposure to vectors, including ticks, and a higher-than-normal number of reported cases of Lyme disease.

Untreated Lyme disease can pose complications for patients, with symptoms of late-stage Lyme disease including Lyme arthritis, most commonly in the knees; neurological complications; and carditis, which may be life-threatening [4,25]. Late-stage Lyme disease may occur months to years following the initial infection, which could pose significant health concerns in the coming years for patients who were never diagnosed and treated appropriately. One could hypothesize that the lack of treatment for Lyme disease during this period could result in a subsequent increase in chronic Lyme disease, which shares many symptoms with post–COVID-19 condition [26]; however, this is development unknown at this time.

There were several limitations, particularly that exposures to Lyme disease differ by region, both throughout North Carolina and the United States in general. Lyme disease surveillance and the available data are captured by county of residence, which may not represent where the exposure occurred. Lyme disease includes a broad spectrum of symptoms and presentations across the different stages, suggesting that some reported cases may be due to another cause. As described above, Lyme disease case reports only represent those individuals who seek health care and are subsequently diagnosed; thus, these numbers more generally represent an underestimate of the true burden of Lyme disease. Additionally, the case definition for the reporting of Lyme disease has been modified multiple times over the years, and the lack of good diagnostic tests impacts surveillance.

A key limitation to our study is the assumption of using historical reported Lyme disease cases prior to the pandemic period to generate the counterfactual estimate of Lyme disease cases in the absence of the pandemic. While this is a strong assumption, it serves to establish a likely lower bound for the number of under-ascertained Lyme disease cases due to variations in exposure and health care–seeking behaviors during the pandemic period. Furthermore, we were unable to conduct our analysis of month-to-month, estimated under-ascertainment rates at the national level due to a lack of available data. Other unrecorded and unobserved confounders are likely present in the recorded data. Our analysis could be improved by incorporating measures of greenspace utilization at appropriate spatial and temporal scales, both before and after the pandemic, to better refine the estimate for the likely number of cases given the changing exposure. Previous teaching emphasizing the lack of Lyme disease cases in North Carolina to minimize overdiagnosis and overtreatment is now superseded by the growth of this emerging infection.

The COVID-19 pandemic had substantial effects on the passive surveillance of Lyme disease and likely ascertainment of cases. Public health officials should be wary of using data that have substantial variation in the ascertainment rates of cases to estimate secular trends and detect outbreaks. Future surveillance approaches that are less sensitive to wide variation in ascertainment rates should be developed.