Estimating the incidence of autoimmune inflammatory arthritis after Lyme disease

Abstract

Objective:

A previous case series described the development of new autoimmune, inflammatory arthritis (IA) developing within 2 years of Lyme disease (LD). This study aimed to estimate the incidence of IA following LD using administrative claims data. Influenza, an infection not typically associated with post-infectious IA, was used for comparison.

Methods:

We conducted a retrospective cohort study using the Johns Hopkins Health System administrative claims data from 01/2013-05/2024. Patients with LD and influenza were identified using International Classification of Diseases (ICD) codes, with LD cases further defined by requiring an antibiotic prescription within 30 days of the ICD code. IA cases were identified using 2+ ICD codes for rheumatoid arthritis, psoriatic arthritis or spondyloarthritis. Comparison between infection groups were conducted using chi-square tests and Z-tests as appropriate. Logistic regression analyses were conducted to estimate the odds of IA development within two years of Lyme or influenza diagnosis, controlling for age and sex.

Results:

The incidence of IA was significantly higher following LD compared to influenza (1.67% vs 0.45%, p<0.0001), with the highest incidence of IA occurring within the first year after LD. Regression analysis showed LD was associated with increased odds of IA compared to influenza (OR 3.76, 95% CI: 2.41-5.86, p<0.001) after adjusting for age and sex.

Conclusion:

The incidence of IA was higher within a year of LD infection, higher than that expected in the general population. The temporal association and elevated incidence support the need for prospective studies to elucidate mechanisms linking infection and autoimmunity in LD.

Lyme arthritis (LA) is the prototypic infectious, inflammatory arthritis caused by Borrelia burgdorferi infection¹. While most patients with LA have resolution of inflammation and pain with a return to health after antimicrobial therapy, at least 10% of patients have persistent synovitis, termed post-infectious Lyme arthritis (PILA) when present in the previously affected joint at least one month after completing antimicrobial therapy². PILA is thought to be the immune-mediated sequelae of LA given the lack of evidence for ongoing infection, its strong association with specific HLA-DR alleles, and improvement with immunosuppression^2,3. Recently, other forms of immune-mediated inflammatory arthritis (IA) have been described after Lyme disease (LD)⁴.

Steere et al. reported thirty cases of new-onset rheumatoid arthritis (RA), psoriatic arthritis (PsA) and spondyloarthritis (SpA) that met classification criteria and developed following standard antibiotic treatment for Lyme disease (LD)⁴. These conditions typically emerged within a few months after infection (median 4 months, range 0-2 years). Unlike the mono- or oligoarticular involvement seen in LA and PILA, patients with post-Lyme IA often present with polyarticular inflammation and additional features characteristic of rheumatic disease, such as rheumatoid factor [RF] positivity, psoriasis, enthesitis, dactylitis, or sacroiliitis. Whereas LA only occurs when early manifestations of LD are left untreated and can improve with antibiotics, post-Lyme IA can occur after any stage of LD (e.g., erythema migrans, neuroborreliosis, Lyme arthritis). While this pivotal paper introduced the possibility of systemic autoimmunity following LD, it has remained challenging to discern whether the observed phenomenon of post-Lyme IA represents a causal relationship or if these observations occurred by chance, particularly as roughly 475,000 individuals are diagnosed with LD annually in the United States⁵.

Insurance claims-based data have been used to identify associations between rare events, and our study aimed to use these data to estimate the incidence of post-Lyme IA⁶. To better understand whether the relationship between LD and IA was unique to Borrelia infection, we included a control cohort of patients with influenza, a condition not typically associated with post-infectious IA.

Methods:

Study population sources:

We used the Johns Hopkins Health System (JHHS) administrative claims data to identify adult patients (≥18 years) who were diagnosed with LD and influenza among patients seen in the hospital, emergency department or outpatient clinics between January 2013 and May 2024. JHHS is a not-for-profit healthcare system serving over four million members in the Baltimore and Washington DC metropolitan and suburban areas.

Lyme cohort:

Patients with LD were identified using the International Classification of Diseases, 9^th Revision (ICD-9) and 10^th Revisions (ICD-10), using the following codes: ICD-9 (088.81) and ICD-10 (A69.20, A69.21, A69.22, A69.23, and A69.29), identifying 7,231 patients (Figure 1). We then identified patients who were prescribed antibiotic therapy targeting LD (e.g., doxycycline, amoxicillin, cefuroxime, ceftriaxone), within thirty days after the first ICD-9 or ICD-10 code to improve specificity as performed in previous studies^7,8. If a patient had multiple LD diagnoses at different dates, only the first was used for the purposes of this study. This cohort, with 3,231 patients, was used to estimate the incidence of post-Lyme IA.

Figure 1:

This study flow diagram illustrated the selection for patients included in the Lyme disease (LD) and influenza cohorts using the Johns Hopkins Health System administrative claims data. Inflammatory arthritis is representative of rheumatoid arthritis, psoriatic arthritis, and spondyloarthritis. JHHS: Johns Hopkins Health System.

Influenza cohort:

Patients with influenza were identified using ICD-9 (487-488) and ICD-10 (J09-J11) codes as previously described, identifying 27,171 patients^9–12. We then included only those with positive influenza viral testing, based on detection of viral antigen or RNA from nasal swab or bronchoscopy or based on positive influenza antibody titers (greater than 1:8), identifying 6,887 patients, and this cohort was used for primary incidence estimates. To attempt to isolate the potential relationship between a particular infection with IA, patients with both influenza and Lyme diagnostic codes were excluded from this study (n = 197).

Arthritis cohort:

We identified patients with at least two diagnostic codes within a 12-month period for RA (M05, M06)^13–15, PsA (L40, M07)^16–18 or SpA (M45, M46)^19,20. Previous studies using insurance claims data required IA diagnoses recorded by a rheumatologist and a prescription for immunosuppressive therapies. We considered the possibility that patients may have been evaluated by rheumatologists or filled prescriptions outside JHHS, which would lead to missing cases. We elected to increase sensitivity for detection of patients with IA and did not apply these more restrictive criteria. We evaluate the rate of false positives in a subsequent step using direct chart review to verify rheumatologist confirmation of the IA diagnosis and subsequent treatment with immunosuppression. Limited clinical information, including sex assigned at birth and age at time of IA diagnosis, was also collected from these groups.

Cohort validation:

Claims-based data may lack detailed patient-level information and are primarily intended for billing purposes, potentially leading to misclassification. We performed a chart review for patients diagnosed with IA after LD and influenza infection to confirm diagnoses. We focused on the group of patients who developed IA within two years of Lyme or influenza infection. Using previously published criteria⁶, LD was categorized as “confirmed” with (1) physician-documented erythema migrans (EM), independent of Lyme serologies, or (2) positive serologies paired with either a flu-like illness or late Lyme manifestations (e.g., arthritis, neuroborreliosis, myocarditis). Using the ISDA guidelines, serologies were considered positive if there were at least 2 of 3 reactive IgM bands within 30 days of the onset of symptoms or with at least 5 of 10 reactive IgG bands with symptoms lasting greater than 30 days. A prescription for antimicrobial therapy targeting LD with duration > 7 days was also required. LD was considered “probable” based on laboratory-confirmed disease and treatment but without specific documentation for erythema migrans or other eligible late manifestations⁸. Patients were considered to have “suspected” LD based on antibiotic prescription but no laboratory confirmation and no specific documentation of erythema migrans or other late manifestations. Patients were excluded if unable to meet these criteria. Influenza diagnosis was confirmed by the presence of a flu-like illness in conjunction with positive viral testing. Antiviral therapy was not required for influenza confirmation.

Post-Lyme IA was validated using rheumatology notes, characterizing IA and recommendations for immunosuppressive therapy. When available, information about serologies (e.g., RF, anti-citrullinated protein autoantibodies [ACPA], antinuclear antibodies [ANA], HLA-B27) and exam (e.g., swollen joint count, tender joint count, enthesitis) was abstracted. We then sought to assess whether patients with PILA were misdiagnosed as having post-Lyme IA. PILA is a mono- or oligoarticular IA which only occurs after LA. PILA diagnosis required the following: (1) preceding LA which was adequately treated with antibiotics, (2) persistent inflammation of only the joint(s) affected by LA, and (3) absence of polyarticular disease. Post-Lyme IA was defined as an (1) oligo- or polyarticular IA, (2) including joint(s) not previously affected by LA/PILA, and (3) associated with additional classification criteria for RA (e.g., symmetric, small joint involvement with or without positive serologies), PsA (e.g., psoriasis, nail dystrophy, dactylitis) or SpA (e.g., multifocal enthesitis, sacroiliitis, HLA-B27).

Statistical analysis:

Comparison between infection groups were conducted using chi-square tests to compare frequencies of categorical variables, and Z-tests for continuous variables as appropriate. Unadjusted and adjusted logistic regression analyses were conducted to estimate the odds of development of IA within two years of Lyme or influenza diagnosis, controlling for age and sex. A sensitivity anlaysis was conducted removing data after 2020 to ensure that findings do not reflect surveillance bias related to reduced health care utilization during the COVID-19 pandemic. Data analysis was conducted using Stata 18.0²¹.

Results:

Estimating the incidence of IA after Lyme and influenza:

As outlined in Figure 1, there were 7,231 patients identified with LD based on a single ICD code, of which 3,231 patients (44.7%) received a LD-specific antibiotic within 30 days of the ICD code. Within this cohort, 3.3% [107 of 3,231] of patients had two or more IA ICD codes, and of these patients, 85% (91 of 107) had the first IA ICD code after LD, with 49.5% (54 of 91) occurring within two years after LD.

There were 27,171 patients with influenza identified by a single ICD code, of which 6,887 patients (25.3%) had a paired positive influenza test. Two or more IA ICD codes were present for 3.02% [208 of 6,887] of these patients, of whom 39.9% [83 of 208] had the first IA ICD code after influenza, with 37.3% [31 of 83] occurring within two years after influenza.

There was a higher incidence of a new IA ICD code within two years after LD when compared to influenza (1.67% vs 0.45%, p<0.0001). Compared to influenza, the incidence of IA was highest within the first year of LD diagnosis (1.2% vs 0.33%, p<0.0001) and remained significantly higher in the following years as described in Figure 2. For comparison, we estimated the incidence of new IA before LD. There were 5 patients with the first IA ICD code 0-1 year before the first LD diagnosis, with an estimated incidence of 0.15% [5 of 3,231], and there were 2 patients with the first IA ICD code 1-2 years before LD diagnosis, with an estimated incidence of 0.06% [2 of 3,231]. The incidence of IA was significantly higher 0-1 year after LD when compared to the year preceding LD diagnosis (1.2% vs 0.15%, p<0.0001). Whereas in the influenza cohort, there was no difference in the incidence of IA in the year before influenza infection when compared to the year after infection (0.32% vs 0.33%, p=0.88). We converted the incidence estimates to person-years for comparison with other forms of IA. To account for potential loss to follow-up using EHR data, we used the first year after infection to calculate this incidence. Focusing on the year after LD infection (39 cases of new IA from 3,231 patients), the estimated incidence of any IA after LD was 1,207 per 100,000 person-years. With influenza (23 cases of new IA from 6,887 patients), the estimated incidence was 334 per 100,000 person-years.

Figure 2: Incidence of inflammatory arthritis after Lyme disease and influenza.

The incidence of new inflammatory arthritis diagnostic codes before and after Lyme and influenza infections. There was no difference in the incidence of these diagnostic codes before infection, though there was a stastistically significant difference in the incidence after Lyme disease when compared to influenza. This incidence was highest in the year after infection, although the incidence of new IA remained significantly elevated for years after LD.

* p<0.001

There is suggestion that SARS-CoV2 infection may also be associated with post-infectious IA²². Given ubiquity of this infection, we removed all patients identified from these cohorts after 2020 to determine if SARS-CoV2 infection may account for these findings. After removing these data, we still found a higher incidence of IA within two years of LD compared to influenza (1.2% vs 0.24%, p<0.001).

Demographic characteristics of patients with IA:

The LD cohort (n = 3,231) was older than the influenza cohort (n = 6,887) (50.7 ± 16.8 vs 48.1 ± 18.6 years, p<0.001) as shown in Table 1. There was no difference in age between those who developed IA after LD when compared to those who developed IA after influenza (p = 0.137), specifically with no difference in those developing IA within 2 years of infection (p = 0.077). While there were no differences between the Lyme and influenza cohorts, patients who developed IA within 2 years of LD were younger than those with IA before LD (49.0 ± 16.0 vs. 58.0 ± 13.0 years, p = 0.029).

Table 1:

Demographics of patients with rheumatoid arthritis, psoriatic arthritis, or ankylosing spondylitis after Lyme and influenza infection

	Influenza ICD only	Influenza with positive viral testing	Lyme ICD only	Lyme with antibiotic prescription
Total cohort	27,171	6,887	7,231	3,231
Mean age (SD), years	49.5 (19.3)	48.1 (18.6)	52.1 (17.1)	50.7 (16.8)
Female (%)	61.4%	60.7%	56.8%	54.2%
2+ Arthritis ICD codes	809	208	278	107
Arthritis before infection	480	125	115	16
Mean age (SD), years	60.5 (15.8)	61.8 (15.2)	60.6 (14.9)	58.0 (13.0)
Female (%)	76.3%	72.0%	64.4%	81.3%
Arthritis after infection	329	83	163	91
Mean age (SD), years	58.0 (15.2)	53.9 (14.3)	54.0 (15.4)	50.5 (15.7)
Female (%)	70.5%	66.3%	66.9%	63.7%
Arthritis ≤1 year after	90	23	75	39
Arthritis 12-24mo after	53	8	28	15
Arthritis 0-2 years after	143	31	103	54
Mean age (SD), years	59.2 (16.2)	55.5 (16.1)	53.6 (16.1)	49.0 (16.0)
Female (%)	67.1%	61.3%	66.0%	63.0%
Arthritis 2-5 years after	115	33	30	16
Arthritis 5+ years after	71	19	30	21
Arthritis >2 years after	186	52	60	37
Mean age (SD), years	57.0 (14.4)	52.9 (13.2)	54.7 (14.2)	52.6 (15.1)
Female (%)	73.1%	69.2%	68.3%	64.9%

There was a higher proportion of female patients in the influenza cohort (60.7% vs 54.2%, p<0.001); however, there was no difference in the proportion of female patients who developed IA after influenza when compared to those who developed IA after LD (66.3% vs 63.7%, p=0.720). There was no difference in the proportion of female patients who developed IA before LD, 0-2 years after LD, and >2 years after LD (all p >0.17).

Because of these differences between cohorts, we performed a regression analysis to determine if age and sex at birth affected development of IA within 2 years of infection. Lyme infection was associated with significantly higher odds of developing IA compared to influenza (OR 3.76, 95% CI: 2.41-5.86, p<0.001) after adjusting for age and sex. In the adjusted model, age (OR 1.01, 95% CI: 0.99-1.02, p=0.369) and sex (OR 1.29, 95% CI: 0.83-2.01, p=0.259) were not significantly associated with development of IA.

The type of arthritis (e.g., RA, PsA, SpA) diagnoses were similar when comparing those with IA before LD to those with IA before influenza. However, of those who developed IA, there was a higher proportion of RA diagnoses after LD compared to RA diagnoses after influenza (61.5% vs 43.4%, p=0.017), and this difference was most prominent 0-1 year after LD (74.4% vs 34.8%, p=0.002) when compared to 0-1 year after influenza. After the first year, there was no difference in the types of IA diagnoses 1-2 years after LD compared to 1-2 years after influenza (60.0% vs 50%, p=0.65) and no difference >2 years after LD compared to >2 years after influenza (48.7% vs 46.2%, p=0.82). Within the Lyme cohort, the proportion of RA diagnoses was highest <1 year after infection compared to the proportion of RA diagnosed ≥1 year after infection (74.4% vs 51.9%, p=0.029).

Cohort validation through chart review:

We completed chart review to verify Lyme and IA diagnoses, focusing on the 54 patients who developed post-Lyme IA within two years after LD. Lyme was “confirmed” in 30 patients (55.6%), “probable” in 6 patients (11.1%), and “suspected” in 5 (9.2%). In 13 patients (24.1%), LD could not be confirmed: 4 due to a lack of history and laboratory testing in the EMR and 9 due to a misdiagnosis, most often (n = 8) due to fatigue and/or musculoskeletal pain paired with a positive Lyme ELISA with negative Western blot or due to one or two reactive IgM on Western blot with fewer than 5 reactive IgG bands when symptoms were present >30 days before testing. Of the 30 patients with “confirmed” LD, IA could not be confirmed in 5, most often due to other medical conditions which were mislabeled as IA: “sacroiliitis” diagnosed by non-rheumatologist without supportive axial imaging, disseminated LD with resolution of pain after antibiotics, post-treatment Lyme disease without IA, PILA without systemic autoimmune arthritis, tophaceous gout. Two patients were excluded as IA was diagnosed outside the JHHS EMR before LD, but the timing was not captured using ICD codes. After removing these 7 patients, there were 23 (42.6%, n = 54) patients with confirmed post-Lyme IA.

We completed the chart review for the patients with concomitant influenza ICD code with positive testing who developed IA after infection (n = 31). IA and influenza could only be confirmed from 29% [9 of 31] of patients. Of the 22 patients excluded, the most common reason was due to the inability to confirm IA (n = 12). This was most often due to “sacroiliitis” being diagnosed by a non-rheumatologist, with no imaging or other documentation to support axial spondyloarthritis. Another 10 patients were excluded as arthritis developed before influenza, with rheumatology consultation occurring outside the JHHS EMR, with timing not adequately captured using ICD codes.

We abstracted the clinical manifestations of post-Lyme IA using this cohort (n = 23). LA and neuroborreliosis were the most common initial LD presentations (each with 35%, n=8) followed by erythema migrans (22%, n=5) and flu-like symptoms with positive acute serologies (9%, n=2). All patients received doxycycline therapy, with 56% (n=13) receiving a second antibiotic, 13% (n=3) receiving amoxicillin and 43% (n=10) receiving ceftriaxone. Over half of these patients (56%, 13 of 23) were diagnosed with RA, 39% (n= 9) with PsA, 4% (n=1) with AS. Joint counts were available for 15 patients, of whom 9 had polyarthritis and 6 with oligoarthritis. Enthesitis was documented in roughly half these patients (n=7). RF and ACPA data were available for all but one patient; 54% [7 of 13] of patients with RA were positive for at least one of these tests, with only two patients having both RF and ACPA. Only 20% [2 of 10] with spondyloarthritis had a positive HLA-B27. An antinuclear antibody was present in 8.6% [2 of 23] of all patients with post-Lyme IA, both of whom had PsA.

Adjusting post-infectious IA estimates based on chart review:

Using data from the chart review, we then sought to re-evaluate the incidence of IA after LD. LD was “confirmed” in 55.6% [30 of 54] of the patients reviewed. Applying this proportion to the administrative claims data (e.g., Lyme ICD code paired with an appropriate antibiotic prescription, n = 3,231), we estimate that approximately 1,795 patients would be expected to have “confirmed” LD. Using our confirmed cases of IA (n = 23), this would give a revised estimated incidence of 1.28% [23 of 1795] within 2 years of LD, which is comparable to the 1.67% estimate based on administrative claims. In the influenza cohort, all patients [31 of 31] had a flu-like illness associated with positive viral testing, giving an estimated incidence of 0.13% [9 of 6,887] during the two years after influenza.

Discussion:

Our study is the first to estimate the incidence of new IA after LD using claims-based data. We found that 1.67% of patients developed post-Lyme IA within 2 years of infection, which was significantly higher than the two years preceding LD and also higher compared to the incidence of IA after influenza. The incidence was highest the year after Lyme infection, with an incidence of 1,207 per 100,000 person-years. For comparison with the general population, the incidence of RA is estimated at 40-50 per 100,000 person-years^23,24, with a lower estimated incidence of PsA (7-23 per 100,000 person-years)^25,26 and AS (7-27 per 100,000 person-years)^27,28. While difficult to compare directly with these estimates, these values are numerically much lower than our estimates in the year after Lyme infection.

Claims-based data has several limitations, including lack of clinical specificity (e.g., misclassification of diagnoses), variability in coding practices, missing data (e.g., care outside the healthcare system, incomplete follow-up) and potential for surveillance bias. We attempted to address these issues by utilizing protocols to confirm diagnoses from previously published claims-based research and utilizing chart review to improve clinical specificity.

To assess the representativeness of our cohorts, we compared our data to published LD surveillance estimates. According to the CDC surveillance data, the annual incidence of LD in Maryland is approximately 25 cases per 100,000 person-years²⁹. Other studies have reported higher estimates in Maryland, up to 98 cases per 100,000 person-years. However, Baltimore County – the primary location of JHHS – has a notably lower incidence of 13 cases per 100,000 person-years³⁰. Using the JHHS patient population of approximately 4.25 million over a 12.5-year study period, there would be an estimated 6,900 – 52,000 LD cases expected, depending on the incidence rates applied. Our LD cohort (n = 7,231) using administrative claims falls within the lower range of these projections, which is consistent with several factors. First, continuous enrollment data for all 4.25 million patients were not available throughout the study period, potentially underestimating the cases of LD. Second, JHHS primarily serves an urban population in Baltimore County, where exposure to Ixodes ticks and LD is significantly reduced. These considerations support the plausibility of our estimates in relation to regional epidemiological trends.

Estimating the incidence of influenza is particularly difficult as patients with mild disease may not seek medical care, and several other infections can present with influenza-like symptoms. In one study evaluating the specificity for influenza diagnoses made with claims-based data, ICD-10 codes J09-J11 had high specificity (96%) and high positive predictive value (85%) in patients who also underwent influenza testing¹². However, claims-based data have the potential to underestimate the burden of influenza; for example, in a large outpatient network, influenza was confirmed in 23% of patients with acute respiratory infections, though only 27% (902 of 3381) received an ICD-10 code for influenza³¹. In this study, less than a third of patients with laboratory-confirmed influenza received an ICD code reflecting a diagnosis of influenza. While previous studies described specificity of claims-based data for influenza infection for patients seen in the emergency department and in the hospital, there is little data about its use in the outpatient setting. Using sentinel providers, the Maryland Department of Health estimates roughly 6-8% of outpatient visits are due to influenza-like illness during influenza season, with 1-4% of these patients having positive viral testing³². These estimates for influenza are much larger than our influenza cohort. In our influenza cohort, we found no difference in the incidence of IA after influenza infection, with [83 of 6,887] or without [329 of 27,171] positive influenza viral testing (1.2% vs 1.2%, p = 0.969), respectively. There was also no difference in the incidence of IA 0-2 years after influenza infection, with [31 of 6,887] or without [143 of 27,171] positive influenza viral testing (0.45% vs 0.53%, p=0.428). While this helps support the generalizability of our results, it is possible that patients with influenza, who were not included in this study, developed new IA which would not be captured by our claims-based data.

Previous arthritis studies utilizing claims-based data differed in that they (1) require that IA ICD codes come from a rheumatologist and (2) require a concurrent prescription for immunosuppression¹⁵. For RA, these criteria increase the positive predictive value (PPV) using the ICD codes, increasing from 56% with two claims coded (not specifying that codes come from a rheumatologist), to 66% with two claims coded by a rheumatologist, to 90% with addition of a disease-modifying antirheumatic drug prescription. Similarly, the PPV ranged from 69-86% when there were two or more claims coded by a rheumatologist for PsA¹⁶. As post-Lyme IA was thought to be a relatively rare complication of LD, we removed these criteria (i.e., allowing ICD codes to come from any physician and not requiring a concurrent immunosuppressive treatment) to improve our ability to identify patients with post-Lyme IA. However, these features were specifically addressed during the chart review, allowing confirmation that diagnoses originated from a rheumatologist and that patients were treated with immunosuppression. Using these criteria, IA was confirmed in 76.6% [23 of 30] in the post-Lyme cohort, compared to 29.0% [9 of 31] in the influenza cohort. These findings suggest that administrative claims may overestimate the true incidence of IA, particularly following influenza, and potentially to a lesser extent following LD.

One limitation is the modest confirmation rate of post-Lyme IA (42.6%, 23 of 54) among patients meeting administrative claims criteria. This discrepancy was primarily attributable to the LD diagnosis itself, which was confirmed in only 55.6% [30 of 54] of reviewed cases. When comparing to previously published studies utilizing chart review for LD, we found a similar proportion of patients with “confirmed,” “probable,” and “suspected” LD in addition to a similar proportion of patients misdiagnosed with LD⁸. In the previously published LD chart review study, only 54% (70 of 128) of charts reviewed had confirmed LD, with 12% (15 of 128) having probable LD, 27% (35 of 128) having suspected LD, and 6% (8 of 128) not having LD⁸.

When compared to the case series described by Steere et al., our study found a higher prevalence of female patients developed IA after LD (63.7% vs 40%, p = 0.022), though age at infection was numerically similar (50.5 vs 55 years). Our cohort had an increased prevalence of LA and neuroborreliosis as the initial Lyme manifestation, with fewer patients presenting with erythema migrans or a flu-like illness with positive acute serologies⁴. Our study showed similar trends in antimicrobial therapy, with nearly all patients receiving doxycycline, and roughly half receiving a second antibiotic, though intravenous ceftriaxone was used more frequently in our study, likely reflecting differences in the initial Lyme presentation. Our study also showed similar types of IA diagnoses, with roughly half of the patients diagnosed with RA and half with PsA or SpA. The clinical similarities between these studies support the validity of our outcomes.

We also found that the incidence of post-Lyme IA is highest within the first year after LD and decreases afterward. This time-dependent relationship supports, though does not confirm, post-Lyme IA as a specific infection-mediated sequelae. To assess for the possibility of misclassification of LA and PILA as post-Lyme IA, we conducted the chart review and found only one occurrence (1 of 30). Differentiating LA and PILA from post-Lyme IA can be difficult as all three conditions are associated with synovitis. As described in the Methods, we used operational definitions of PILA (a mono- or oligoarticular IA persisting after antimicrobial therapy in the joint(s) affected by LA) and post-Lyme IA (characterized as oligo- or polyarticular IA including joint(s) not previously affected by LA/PILA and associated with features specific to autoimmune arthritis [e.g., enthesitis, psoriasis, abnormal serologies]) during the chart review to attempt to differentiate these conditions. Using these operational definitions, we found very few cases of misdiagnosis in the chart review. In the work by Steere et al., 3 of 24 patients with PILA developed synovitis in new joints during the post-antibiotic period, potentially making it difficult to differentiate these conditions⁴. Despite our operational definitions, we acknowledge the potential clinical similarities between PILA and post-Lyme IA, which cannot always be fully differentiated by chart review.

Surveillance bias should be considered as a potential explanation for the higher incidence of IA diagnoses after LD, as previous studies have documented increased healthcare utilization after LD, including 87% more outpatient visits over a 12-month period⁶. However, our study design incorporated a detailed chart review, which served to mitigate the effects of this bias. Specifically, the chart review allowed us to validate the IA diagnoses rather than relying solely on administrative claims data. By confirming these diagnoses, we reduced the risk of misclassification which allows us to exclude cases where increased healthcare utilization may have led to incidental or unrelated arthritis coding.

Influenza was selected as a comparator to Lyme disease as it is not thought to be associated with an increased risk of IA. While there were demographic differences between these cohorts, we were able to adjust for these variables and show that the risk of post-infectious IA was higher after LD. While the inclusion of an influenza comparator group helps contextualize these findings, it does not fully account for other unmeasured confounders that may drive the observations described in this study.

A significant limitation of our study is the inability to account for continuous enrollment in the Lyme and influenza cohorts. Patients who transitioned to other healthcare systems or ceased seeking medical care during the study period could be missed. As these cases would not be captured using our administrative claims data, there is a risk of ascertainment bias, potentially leading to inaccurate estimates of IA incidence. We anticipate that the impact of this limitation is likely uniformly distributed across both groups. Further, our analyses focus primarily on the two years following infection, reducing the likelihood of substantial attrition. However, the absence of continuous enrollment data underscores the need for caution when interpreting incidence estimates and highlights the importance of prospective studies with robust patient tracking.

In a separate study, our team has enrolled patients at the time of an erythema migrans rash and then evaluates patients longitudinally for several months after antimicrobial therapy, in efforts to better understand persistent symptoms after Lyme disease. Using unpublished data from this cohort, we found that 1.57% [2 of 127] patients developed an IA after LD, with both patients developing synovitis and multifocal enthesitis within four months of enrollment. While interpretation of these findings is limited based on the smaller sample size, the incidence of IA after LD in this cohort is similar to our estimates based on administrative claims data.

Steere et al. first described post-Lyme IA, though it was difficult to determine if these conditions occurred coincidentally⁴. However, our administrative claims data suggest against the random occurrence of LD and IA. We instead propose that B.burgdorferi infection or the effects of antimicrobial therapy (e.g., changes in the microbiome) may play a role in initiation and propagation of an autoimmune response in a subset of patients. However, this autoimmune response may not be maintained in all patients. In the work by Steere et al, 13% [4 of 30] (2 with RA, 2 with PsA/SpA) attained drug-free remission, which is atypical for most autoimmune arthritides⁴.

Our study highlights that autoimmune IA occurs more frequently than expected after LD. While there is much interest in the relationship between infection and autoimmunity, claims-based data and retrospective chart reviews do not allow us to discern whether LD caused the IA or whether infection exacerbated pre-existing, subclinical disease. Prospective studies, enrolling patients at the time of LD, are ideal to re-assess the incidence of post-Lyme IA but also to explore mechanisms which may contribute to autoimmune initiation and propagation.

Conclusion:

Post-Lyme IA has been considered a rare entity after LD, though our claims-based data suggest that roughly 1% of patients infected with LD will develop a new, systemic IA. We propose that this infection may be a mechanism of autoimmune initiation and propagation which warrants further investigation through prospective translational studies.

Highlights:

1. The incidence of rheumatoid arthritis, psoriatic arthritis, and spondyloarthritis is highest within the year after Lyme infection, and this incidence is higher than expected when compared to another infectious exposure, influenza.

2. For those with new inflammatory arthritis after Lyme disease, rheumatoid arthritis was the most common diagnosis. However, the incidence of new rheumatoid arthritis and psoriatic arthritis was similar >1 year after Lyme infection.

3. Post-Lyme inflammatory arthritis occurred most frequently after late manifestations of Lyme disease (e.g., Lyme arthritis and neuroborreliosis).