Reevaluating the relationship between COVID-19 and type 1 diabetes mellitus: Methodological considerations

Abstract

A recent nationwide cohort study reported an increased incidence and altered seasonality of type 1 diabetes mellitus (T1DM) during the coronavirus disease 2019 (COVID-19) pandemic. The study found that new-onset T1DM cases were significantly higher during the pandemic than in prior years, and the typical winter peak in T1DM diagnoses was blunted. This occurred alongside markedly reduced circulation of other respiratory viruses under lockdown measures. Carmon et al noted weak positive correlations between T1DM incidence and certain viruses (e.g., influenza and respiratory syncytial virus), suggesting that reduced exposure to common infections - and possibly severe acute respiratory syndrome coronavirus 2 infection itself - might have contributed to the rise in T1DM. To highlight key methodological limitations of that study, which may affect the interpretation of the findings. We reviewed the study design and data of Carmon et al and discussed potential biases, including ecological inference, confounding factors, delayed diagnoses, lack of COVID-19-stratified analysis, and biases in viral surveillance data, supported by recent literature. The association observed by Carmon et al is at risk of ecological fallacy due to the absence of individual infection linkage. Uncontrolled confounders (healthcare access, socioeconomic changes) and not stratifying by COVID-19 infection status limit causal inference. Pandemic-related diagnostic delays likely inflated apparent T1DM incidence, as evidenced by higher rates of diabetic ketoacidosis in new cases. Biases in virological testing data (reduced testing and non-representative sampling) complicate conclusions about “reduced” viral circulation. The pandemic’s impact on T1DM incidence is important but requires cautious interpretation. Future studies should employ individual-level analyses, adjust for confounders, distinguish true incidence increases from diagnostic delays, stratify by infection status, and use comprehensive viral exposure data to draw more robust conclusions.

Key Words: Type 1 diabetes mellitus; COVID-19; Incidence; Seasonality; Epidemiology; Viral infections; Bias

Core Tip: Carmon et al’s study suggests a link between the coronavirus disease 2019 (COVID-19) pandemic and rising type 1 diabetes mellitus cases. This letter provides a constructive critique of their methodology. The authors point out that ecological analysis without individual infection data can be misleading, unaddressed confounders and lack of COVID-19 stratification weaken causal inference, and pandemic-related diagnostic delays (reflected in higher diabetic ketoacidosis rates) may have inflated case counts. The authors also caution that biases in viral surveillance data complicate the interpretation of “reduced” non-COVID infections. Addressing these issues with more granular data and analyses will improve future research on the type 1 diabetes mellitus-COVID relationship.

TO THE EDITOR

The authors read with great interest the article by Carmon et al[1], investigating the impact of the coronavirus disease 2019 pandemic on type 1 diabetes mellitus (T1DM) incidence and seasonality. The study provides timely epidemiological evidence suggesting a possible viral trigger for T1DM. The authors appreciate the innovative use of nationwide registry data by Carmon et al[1] to generate timely evidence, and emphasize that this critique is intended as a constructive extension of their work rather than a refutation. While Carmon et al’s findings[1] indicate a potentially important link between the pandemic and T1DM (higher T1DM cases and altered seasonal patterns coinciding with reduced non-COVID respiratory infections), several critical methodological limitations could affect the strength of the conclusions drawn.

Ecological limitations

Carmon et al[1] correlated population-level T1DM incidence with overall viral circulation trends, but without linking individual T1DM cases to specific infections, any association remains ecological. This approach risks the well-known ecological fallacy, wherein population-level correlations may not hold true at the individual level. In practical terms, the authors cannot discern whether the children who developed T1DM were the same individuals who had coronavirus disease 2019 (COVID-19) or other infections[2]. The observed rise in T1DM during the pandemic could thus be coincidental or mediated by other factors, rather than a direct viral effect. Indeed, a recent nationwide registry study in Germany found no significant spatial correlation between regional COVID-19 case rates and pediatric T1DM incidence (doubling of COVID-19 rates was not associated with any meaningful increase in T1DM risk), suggesting that a causal link between severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection and T1DM does not support a strong population-level correlation[3]. To establish a more direct relationship, individual-level analyses are needed - for example, case-control comparisons of prior infection rates in T1DM patients vs controls, or cohort studies tracking infection status in children who later develop T1DM[4]. Without person-level data, any inference that viral infections triggered the T1DM increase must remain strictly speculative. The authors recommend that future studies perform data linkage between infection records and new-onset T1DM cases, or apply interrupted time-series and mixed models that can account for baseline trends and sudden changes, to more robustly test causality.

Confounding factors and COVID-19 status

The study did not adjust for potential confounders such as socioeconomic status, healthcare access, or lifestyle changes during the pandemic. This is a significant limitation, because such factors could independently influence both infection exposure and diabetes diagnosis rates. For instance, families’ healthcare-seeking behavior changed under lockdown - some delayed medical visits due to fear or limited availability of services - which could alter when and how T1DM was diagnosed[5]. Pandemic lifestyle shifts (diet, physical activity, stress) and health system strain (clinic closures, overloaded hospitals) might have affected T1DM incidence or detection. Without controlling for these variables, it is unclear whether the observed increase in T1DM was truly due to viral exposures or rather due to these parallel pandemic-related influences. For example, if certain communities had reduced access to medical care during lockdowns, children there might present with T1DM later and at more severe stages [e.g., with diabetic ketoacidosis (DKA)], effectively “bundling” cases into the pandemic period and creating an apparent surge[6]. The authors urge that future analyses incorporate multivariate adjustments or stratifications for socioeconomic and healthcare-access factors to avoid biased associations. Key covariates to consider for adjustment include regional unemployment rates, healthcare utilization metrics (e.g., primary care and emergency visit volumes, telemedicine coverage), timing of school/daycare closures, policy stringency indices (e.g., the Oxford COVID-19 Stringency Index), and viral testing intensity (tests per population and test positivity rates). Many of these indicators can be obtained from public data sources (for example, government labor statistics for unemployment and the Oxford COVID-19 Government Response Tracker for policy stringency).

Relatedly, the authors did not stratify T1DM incidence by COVID-19 infection status, a key oversight if one is hypothesizing a direct role of SARS-CoV-2. A subgroup analysis (or at least reporting how many new T1DM cases had confirmed COVID-19) would directly address whether children who actually contracted COVID-19 were more likely to develop T1DM than those who never had COVID-19. Without this stratification, the report cannot tell us whether the rise in T1DM was concentrated among post-COVID patients or occurred broadly. This represents a missed opportunity to strengthen causal inference. We acknowledge that individual COVID-19 status data may not have been universally available early in the pandemic. If person-level linkage was infeasible, area-level exposure proxies (e.g., cumulative regional SARS-CoV-2 incidence and test positivity rates, with appropriate 1-4 week lags) can be used in spatial or spatiotemporal analyses to partially mitigate this limitation, as exemplified by recent register-based studies[3]. Notably, multiple studies have now reported that pediatric patients with documented COVID-19 have a significantly higher subsequent risk of new-onset diabetes compared to those without COVID-19. For example, a Centers for Disease Control cohort study found that children < 18 with COVID-19 were 2.66 times more likely to be diagnosed with diabetes (not limited to T1DM) in the months following infection than contemporaneous controls without COVID, and notably, non-COVID respiratory infections were not associated with increased diabetes risk[7]. Such evidence indicates SARS-CoV-2 infection itself may contribute to new-onset diabetes in susceptible individuals. If Carmon et al[1] had access to individual health records, it would have been informative to report T1DM incidence specifically among children who had COVID-19 vs those who did not. Even a simple 2 × 2 comparison (COVID-positive vs COVID-negative) of new T1DM case rates could reveal whether the surge was driven predominantly by children who had the virus. Conversely, if children with no history of COVID-19 also showed increased T1DM incidence, that would point to other pandemic-related factors (such as lifestyle or healthcare disruptions) rather than direct viral triggering. Without infection-status stratification, the conclusion that “viruses (including SARS-CoV-2) may serve as triggers for T1DM” remains speculative. We encourage researchers to incorporate individual COVID-19 status in future analyses, as this would greatly clarify the relationship between the pandemic virus and T1DM onset.

Diagnostic delays and DKA severity

Another important consideration is whether the higher recorded incidence of T1DM during the COVID-19 outbreak might partly reflect delays in diagnosis rather than a true surge in new cases. In the early pandemic phases (especially during strict lockdowns), many families were hesitant or unable to seek routine medical care, and elective clinic visits dropped. This likely led to children with diabetes symptoms presenting later and with more severe illness (e.g., diabetic ketoacidosis, DKA) once they finally did get medical attention. In other words, some T1DM cases that ordinarily would have been diagnosed earlier (or with milder symptoms) may have been “bunched up” into the pandemic period due to postponed healthcare visits[8]. The absence of any analysis on the timing of diagnoses in Carmon et al’s study means we do not know if the surge in reported cases was partly an artifact of such catch-up diagnoses[1]. Importantly, multiple reports across different countries documented a rise in the proportion of new-onset T1DM cases presenting in DKA during COVID-19 lockdowns, implicating diagnostic delay as a factor. For example, one multicenter study observed an approximately 30% increase in the proportion of pediatric DKA cases requiring intensive care during the first pandemic wave[8]. A global meta-analysis found that during the first year of the pandemic, the number of pediatric T1DM cases presenting with DKA increased by about 25% compared to pre-pandemic, even larger than the overall case increase[9]. Several regional studies similarly noted that while T1DM incidence rose modestly, the severity of presentation (DKA rates) rose dramatically, consistent with delayed diagnosis rather than solely increased incidence[10]. Carmon et al[1] gave a limited discussion of this possibility. Distinguishing a true increase in disease incidence from a backlog of late diagnoses is crucial for accurate interpretation. We suggest that future studies examine diagnosis trends in finer temporal detail - for instance, comparing monthly or quarterly new-case counts and the rates of DKA at presentation - to detect if a dip in diagnoses during lockdown was followed by a compensatory spike when medical access improved. If many children present with more advanced T1DM (higher DKA percentages) during pandemic peaks, it would support the notion that the pandemic caused diagnostic shifts as much as pathogenic shifts. Incorporating metrics of disease severity at diagnosis (such as DKA frequency or average blood glucose/A1c levels) can serve as proxies for delayed diagnosis. Such analyses would help untangle whether we are seeing more T1DM occurring or simply more T1DM being detected late. Importantly, the observed approximately 25% relative increase in DKA at diagnosis during the pandemic does not preclude a concurrent rise in true T1DM incidence; rather, it underscores that two non-exclusive mechanisms - diagnostic delay and potential etiologic triggers - could both be contributing, highlighting the need for multivariable time-series or record-linkage analyses to disentangle their effects[9].

Viral surveillance data biases

Carmon et al[1] correlated T1DM patterns with the circulation of common respiratory viruses, using a national polymerase chain reaction testing database to assert that non-COVID respiratory infections declined sharply during lockdowns. It is true that NPIs (masking, distancing, school closures) in 2020 led to dramatic reductions in many infectious diseases - for example, influenza and respiratory syncytial virus (RSV) cases plummeted worldwide in 2020. However, we must be cautious in interpreting the testing data as a precise proxy for actual community viral exposure. Changes in testing practices and health-seeking behavior during the pandemic could introduce bias. Routine surveillance testing for viruses like influenza, RSV, etc., was often de-emphasized or limited to hospitalized patients when health systems were strained. Many mild infections in the community likely went untested as people avoided clinics for minor symptoms[11]. Thus, the recorded drop in positive tests may overestimate the decline in true infections. (Conversely, SARS-CoV-2 testing was widely implemented, but even there, asymptomatic or mild cases could be missed.) In short, the denominator of total infections is unknown - the surveillance data may not fully represent community viral circulation. One study comparing pre-pandemic and pandemic viral testing in emergency rooms found that despite a fivefold increase in tests, positivity rates for influenza A, B, and RSV fell to approximately 0% in 2020/21, effectively eliminating the usual flu/RSV season[12]. This was a striking finding, but it reflects both real reduction and the focus on testing only the sickest patients. Therefore, concluding that viral prevalence was “reduced” purely from test counts needs nuance - it’s very likely true that circulation of common respiratory viruses dropped in 2020, but the exact extent is hard to quantify without broader measures (such as serology or community sampling). Carmon et al[1] did not explicitly address potential limitations of their virological data, such as changes in testing coverage or case definitions. We believe this warrants mention, as the strength of their conclusions about environmental viral triggers depends on the quality and representativeness of the virus surveillance. Future research could improve this by using multiple measures of viral activity - for example, combining lab test data with syndromic illness surveillance, wastewater viral monitoring, or seroprevalence studies - to ensure a more robust assessment of population exposure to viruses. Additionally, analyzing test positivity rates (not just absolute counts) and comparing them to historical baselines can help confirm that lower observed virus circulation was genuine and not an artifact of reduced testing. In summary, while the hypothesis that reduced non-COVID viral exposure influenced T1DM incidence is intriguing, it should be handled carefully given the potential biases in pandemic-era infection data.

Biological plausibility

Notably, plausible biological pathways exist by which SARS-CoV-2 might influence T1DM risk - for instance, the virus can infect pancreatic β-cells via the angiotensin-converting enzyme 2 receptor, and viral antigens might trigger autoimmunity through molecular mimicry[13,14]. These hypotheses, although unconfirmed, suggest a mechanistic basis for a causal relationship.

CONCLUSION

The observations by Carmon et al[1] highlight a potentially important link between the COVID-19 pandemic and T1DM incidence, but the above limitations temper the strength of the causal claims. Notably, plausible biological pathways exist by which SARS-CoV-2 might influence T1DM risk - for instance, the virus can infect pancreatic β-cells via the angiotensin-converting enzyme 2 receptor, and viral antigens might trigger autoimmunity through molecular mimicry. These hypotheses, although unconfirmed, suggest a mechanistic basis for a causal relationship. To truly disentangle the effects of the pandemic, future research in this field should strive to address these gaps. In particular, studies should utilize individual-level data to establish causality (for example, record-linkage cohorts or case-control designs with prior infection data) and include robust adjustment for socioeconomic and healthcare disruptions to isolate viral effects. Researchers must also differentiate true increases in incidence from diagnostic delays; examining markers of disease severity at diagnosis (such as DKA rates) can indicate if surges partly reflect late presentations. Stratified analyses comparing outcomes in COVID-positive vs COVID-negative individuals are urgently needed to directly evaluate SARS-CoV-2’s role. Finally, assessing viral exposure - potentially integrating routine testing data with serological surveys and other epidemiologic indicators - will yield a more reliable picture of environmental triggers. Insights from such refined studies could also inform public health strategies, for example, by identifying high-risk children for closer monitoring after infections or ensuring continuity of pediatric diabetes care during future pandemics. The authors appreciate Carmon et al’s valuable contribution and hope that recognizing these methodological limitations will help inform more refined research and interpretation in future studies on this important topic[1].