Risk of New-Onset Type 2 Diabetes Mellitus in 600,055 Persons after COVID-19: a cohort study

Background

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection may increase the risk for new-onset type 2 diabetes mellitus (T2DM). SARS-CoV-2 can infect pancreatic islet cells [1–3] and metabolic complications can occur in persons with COVID-19 who have no prior history of diabetes [4–6]. Currently, there is limited epidemiological evidence to suggest a temporal association between SARS-CoV-2 infection and new-onset T2DM.

Methods

We used the TriNetx Analytics Network, a federated health research network that aggregates health records from 63 health-care organizations (HCOs) compromising 70 million patients; 93% of the study population derives from HCOs across all 4 US census regions, while 7% of patients were included from HCOs outside of the US. All types of insurances including self-pay are represented.

Our study population was persons diagnosed between January 20, 2020-January 20, 2021 with COVID-19 (ICD-10 U07.1) whereas our control population were persons diagnosed with influenza (ICD-10 J09-J11) between January 20, 2018-January 20, 2021. Multiple years for diagnoses were included in the control population to increase size for robust propensity-matching given influenza burden was substantially less during the COVID-19 pandemic [7]. Persons with COVID-19 or a positive test for SARS-CoV-2 were excluded from the control group. Data collection was performed after July 20, 2021 to ensure all persons had an opportunity for 6 months of follow up. Study and control groups were divided into mild (requiring outpatient services only) vs moderate/severe disease (requiring inpatient/intensive unit care) (supplementary). To evaluate the contribution of steroid-induced T2DM a subgroup analysis was performed whereby persons who received dexamethasone, hydrocortisone, methylprednisolone, or prednisone within 30 days after diagnosis of COVID-19 or influenza were excluded. We investigated first-ever T2DM diagnosis (ICD-10 E11) occurring 1–180 days after index events (i.e date of diagnosis of COVID-19 for cases and influenza for controls). The use of ≥ 1 ICD-10 E11 to identify persons with T2DM demonstrated 96% specificity and 78% sensitivity in the Veterans Affairs Diabetes Epidemiology Cohort [8]. Persons <18 years of age or with recorded death were excluded from analysis.

We balanced cohorts either by severity of disease or lack of steroid use using 1:1 greedy nearest-neighbor propensity score matching by age, sex, race, ethnicity, obesity, hypertension, hyperlipidemia, nicotine dependence, substance use, socioeconomic deprivation, and family history of diabetes mellitus (supplementary). Any characteristic with a standardized mean difference between cohorts lower than 0.1 was considered well matched. We calculated relative risk (RR) of new T2DM with 95% CIs in the 1–180 days after index events for each group. Statistical significance was set at two-sided p-value <0.05 using the TriNetX Analytics Function. All data within TriNetX is in aggregate and de-identified.

Results

We identified 600,055 persons diagnosed with COVID-19 (546, 819 mild and 53,236 moderate/severe disease) and 394,667 persons diagnosed with influenza (377,903 mild and 16, 764 moderate/severe disease). After 1:1 propensity matching, demographic and clinical characteristics were balanced in all groups (standardized mean difference <0.1). The risk for new-onset T2DM within 180 days of either mild and moderate/severe COVID-19 disease was 1.1% (3,510/313,924) and 4.1% (424/10,436), respectively. The estimated rate per 1000 person years was 23 for the mild COVID-19 cohort and 83 for the moderate/severe cohort. Persons with mild COVID-19 disease had 1.54 (95% CI 1.46–1.62) times higher risk for new onset diabetes than mild influenza controls. A similar higher risk was observed among those with moderate/severe COVID-19 compared to moderate/severe influenza [(RR 1.46 (95% CI 1.26–1.69)]. After excluding those with steroid use in subgroup analysis, risk for new-onset T2DM was less among those with mild COVID-19 vs mild influenza [RR 1.22 (95% CI 1.14–1.29)], but remained similar in magnitude among those with moderate/severe COVID-19 relative to moderate/severe influenza [RR 1.42 (95% CI 1.13–1.80)].

Conclusion

The study found that COVID-19 was associated with an increased risk for new-onset T2DM compared to influenza in a multi-institutional research network. Further, steroid use contributed to the association for stronger risk for new-onset T2DM among persons with mild COVID-19. Moreover, the rate of new-onset T2DM after COVID-19 was higher among persons with moderate/severe compared to mild disease.

The finding of steroid use being associated with higher risk for T2DM in the mild COVID-19 cohort is concerning. The NIH COVID-19 treatment guidelines currently do not recommend steroid use for persons non-hospitalized with COVID-19 given the lack of data of proven efficacy and for fear of adverse effects such as steroid-induced hyperglycemia [9]. Reassuringly, our mild COVID-19 cohort include persons diagnosed early in the pandemic, a time of which the utility of steroids was unclear and guideline recommendations were absent; therefore, potentially explaining a treatment practice that would be controversial today.

In contrast to the mild COVID-19 cohort, removal of those with steroidal treatment did not affect the strength of association between new T2DM and moderate/severe COVID-19 (Steroids vs no steroids, RR 1.42 vs RR 1.46). Further, moderate/severe COVID-19 had a higher rate per 1000 person years for new-T2DM than mild COVID-19 (83 vs 23 per 1000 person years). The association of increased risk with more severe COVID-19 disease even in the absence of steroid use may suggest higher viral load or greater immune dysregulation underlie risk for T2DM.

The findings of an association of increased risk for new-T2DM against an influenza comparator offers potent insight. For example, an association of higher risk for T2DM after COVID-19 compared to influenza can suggest new T2DM burden may be attributed to SARS-CoV-2-mediated mechanisms rather than general morbidity after viral illness. Greater inflammation after COVID-19 relative to influenza is likely implicated, but given there was increased risk among persons not requiring hospitalization, for whom the inflammatory stage may be minimal, other mechanisms such as directly viral mediated should be considered. The entry receptor for SARS-CoV-2, ACE2, has been shown to have higher expression in the pancreas in comparison to the lungs [2]. Further, ACE2-knockout mice relative to wild-type mice have been demonstrated to more susceptible to pancreatic ß-cell dysfunction after high-fat diets [10]. Lastly, other coronaviruses such as SARS-CoV also uses ACE2 as its cellular entry point and has been shown to be associated with new-onset diabetes that persists even 3 years after recovery [11].

Our study has limitations that bears consideration. SARS-CoV-2 infection may cause acute hyperglycemia that resolves with acute illness, but is mistakenly diagnosed as T2DM. Completeness of electronic medical records cannot be ascertained, particularly rates of persons with socioeconomic deprivation or family history of diabetes. Laboratory values were not included in analysis and may exclude important covariates such as prediabetes, inflammatory markers, and degree of insulin resistance. Propensity score matching is subject to imperfect balance of covariates. Further, misclassification bias is inherent to the use of diagnostic codes and can impact the accuracy of T2DM as outcome. The use of >1 ICD-10 to identify persons with T2DM although validated in other databases, has not been assessed for its accuracy within TriNetX. However, it should be considered the use of an influenza comparator group can help alleviate biases inherent to retrospective cohort analyses, as persons infected with influenza or SARS-CoV-2 likely exhibit similar health-seeking behaviors. Therefore, limitations such as misclassification of outcome or preexisting undiagnosed diabetes, may be less likely to affect data as any bias may presumably impact those with COVID-19 or influenza the same. Lastly, we could not capture patients who did not seek care at any of the healthcare organizations included in the database.

In conclusion, the findings in this report suggest COVID-19 is associated with risk of new-onset T2DM. The CDC estimates the rate of T2DM to be 6.9 per 1,000 years among the general adult population [12]. While a direct comparison cannot be made as 7% of our population is non-US and represents a health-seeking population, our data estimates a 3–12 fold increase in rate of T2DM after mild and moderate/severe COVID-19, respectively, compared to the general US population. Given, there remains opposing evidence whether acute glycemia may persist after SARS-CoV-2 infection, long-term follow-up investigation, including high-quality prospective studies, are warranted assessing the relationship between COVID-19 and T2DM [13–15]. As evidence accumulates, physicians caring for survivors of COVID-19 should be aware of the association of increased risk and monitor for signs and symptoms of T2DM, which include excessive urination and thirst, nausea/vomiting, fatigue and weight loss.

Table 1: Risk of Type 2 Diabetes Mellitus Among Persons after COVID-19.

Representation of risk of T2DM among persons with SARS-CoV-2 infection compared to influenza. The relative risk compares the risk of T2DM between persons infected with SARS-CoV-2 against persons infected with influenza after matching for age, sex, race, ethnicity, obesity, hypertension, hyperlipidemia, nicotine dependence, substance use, socioeconomic deprivation, and family history of diabetes mellitus

Cohort	Persons in Cohorta	Persons with new-onset T2DM	Rate (per 1000 person years)b	Risk Ratio, 95% CI
Mild COVID-19 VS Mild Influenza	313,924	3,510	23	1.54 (1.46–1.62)
	319,783	2,328	15
Moderate/Severe COVID-19 VS Moderate/Severe Influenza	10,436	424	83	1.46 (1.26–1.69)
	10,951	304	56
Mild COVID-19 (no steroids) VS Mild Influenza (no steroids)	276,748	2,154	16	1.22 (1.14–1.29)
	280,851	1,798	13
Moderate/Severe COVID-19 (no steroids) VS Moderate/Severe Influenza (no steroids)	5,357	166	63	1.42 (1.13–1.80)
	5,424	118	44

^aPersons with outcome prior to the time window were excluded from results

^bRate per 1000 person years were calculated as followed: (Persons with new T2DM) / [(Persons in cohort) × (180/365)] × 1000

Abbreviations – T2DM; Type 2 Diabetes Mellitus