Methods
Study setting
Clalit Health Services (Clalit) is the largest healthcare organization (insurer and provider) in Israel where health insurance is universal and mandatory for all citizens. Clalit provides primary, secondary, and tertiary care to 4.8 million individuals (52% of the population) and owns and operates a network of hospitals which account for 30% of all general hospital beds in the country. Clalit’s information systems are fully digitized and feed into a central data warehouse including both administrative and clinical data.
Study design
This retrospective, population-based cohort study included all Clalit members who were diagnosed with COVID-19 between March 01, 2020 and August 31, 2021 and matched individuals who were not infected by COVID-19. The date of COVID-19 diagnosis was considered as the index date. Baseline variables including sociodemographic clinical and treatment characteristics were defined based on the most recently available data prior to index date.
Five landmark times were defined, at index date (T0) and at 1 (T1), 2 (T2), 3 (T3), and 4 (T4) months post-index date. At each landmark date, individuals with COVID-19 and their matched controls were followed until an occurrence of one of the following events: new documentation of DM, occurrence of COVID-19, pregnancy, corticosteroid use in the outpatient setting, disruption in Clalit membership, all-cause mortality or end of study (December 31, 2021). This landmark design enabled to assess the association of COVID-19 with incident DM during the acute and the post-acute phase in varying time frames, as well as to reduce the possibility of reverse causality and surveillance bias (figure 1).
Figure 1Study design. aPersonal index date was defined for each participant based on date diagnosis of COVID-19, with a maximum potential follow-up period of 22 months (from March 01, 2020 to December 31, 2021).
Study population
All Clalit members aged ≥25 years with documented positive PCR test for COVID-19 during March 01, 2020–August 31, 2021 and with continuous membership at Clalit during the 2 years prior to the index date were included in the study. The minimum inclusion age of 25 years was determined to avoid a lack of data during the period of mandatory service in the Israel Defense Forces. For each individual diagnosed with COVID-19, one individual who was not infected by COVID-19 (before index date) was matched based on age group (stratified by 5-year intervals), sex and primary care clinic/neighborhood characteristic which represents different ethnic and religious groups in Israel (Arabs, Orthodox Jewish, traditional/secular Jewish). Matching for primary care clinic/neighborhood characteristics was performed due to statistically significant differences in DM27 28 and COVID-1929 incidence between the varied communities living in different neighborhoods in Israel. Individuals were excluded if they met one or more of the following criteria: documentation of DM before the index date; positive antibody test for COVID-19 during the study period without prior vaccination or positive PCR test (to remove individuals for whom there was probably an undocumented COVID-19 event); pregnancy at index date; corticosteroid use during the 3 months prior to index date (based on Anatomical Therapeutic Chemical (ATC) code H02, at least one filled prescription in the outpatient setting); long-term care facility housing (home nursing or psychiatric geriatric or rehabilitative hospital); and home confinement for medical reasons. The last two exclusion criteria were set due to absence of continuous medical reporting on the long-term care facility residents, and accordingly, substantial missing information in Clalit data warehouse related to this subpopulation. Non-COVID-19 matches met the same inclusion and exclusion criteria as the patients with COVID-19 except the positive PCR test at index date. Potential matches were selected irrespective of their future COVID-19 status; therefore, there were patients in the matched non-COVID-19 group who were diagnosed with COVID-19 during the follow-up period time. Those individuals and their matched pair were censored at the time of their COVID-19 diagnosis. Patients with COVID-19 for whom suitable matches were not found were excluded from the analysis (figure 2). To evaluate whether the association of COVID-19 with new-onset DM varies by disease severity, three subgroups were defined including non-hospitalized patients, hospitalized patients, and those hospitalized with severe COVID-19.
Figure 2Selection of study population. *Those individuals and their matched pair were included in the analysis but were censored at the time of their COVID-19 diagnosis. T2DM, type 2 diabetes mellitus.
Baseline measurements
Incident COVID-19 was assessed based on documentation of a positive PCR test for SARS-CoV-2 in the hospitals or community laboratory records. Information regarding laboratory-confirmed COVID-19 infections was also received from the Israeli Ministry of Health which established and maintains a national database with mandatory daily reporting of PCR results and disease status from all hospitals in Israel. Three levels of COVID-19 severity were defined including not hospitalized, hospitalized, and severe hospitalized individuals (SpO2 <94% on room air at sea level, a ratio of arterial partial pressure of oxygen to fraction of inspired oxygen <300 mm Hg, a respiratory rate >30 breaths/min, or lung infiltrates >50%).30–33 DM as of index date was determined based on documentation of one or more of the following criteria in hospitals or community records: International Classification of Diseases Ninth Revision (ICD-9) code 250xx, HbA1c ≥6.5%, and glucose-lowering medication use (ATC classification system code A10), defined as ≥1 dispensed prescription any time prior to index date. Individuals with documented fasting glucose levels of ≥126 mg/dL in the community records only and individuals listed with DM in the Clalit chronic disease registry were also considered as patients with DM at index date. The demographic variables assessed at the index date included age, sex, primary care clinic location (according to the main ethnic and religious group living in the clinic area), and socioeconomic status (low, medium, or high based on the primary care clinic level). Smoking status (non-smoker, former, or current) was evaluated based on last documentation in the community records as reported by the patient. Body mass index (BMI) (kg/m2), systolic and diastolic blood pressures (mm Hg), and laboratory data were gathered from the community clinic records based on last documented level prior to index date. Laboratory data included fasting plasma glucose concentration (mg/dL), HbA1c concentration (%), total cholesterol (mg/dL), high-density lipoprotein (HDL) cholesterol (mg/dL), low-density lipoprotein (LDL) cholesterol (mg/dL), and triglycerides (mg/dL). Medical diagnoses were defined based on ICD-9 codes documented in the hospital or the community clinic records anytime prior to the index date (online supplemental etable 1) and included pre-diabetes, hypertension, cardiovascular disease (CVD), congestive heart failure, hyperlipidemia, chronic kidney disease, and polycystic ovary syndrome (PCOS)/polycystic ovarian disease. CVD was defined by having one or more of the following diagnoses: myocardial infarction, unstable angina pectoris, stable angina pectoris, ischemic heart disease, ischemic stroke (cerebrovascular accident), coronary artery bypass graft, and percutaneous transluminal coronary angioplasty. Pharmaceutical treatment was defined according to the ATC classification system based on at least one filled prescription during the 2 years prior to index date. Medications included beta-blocking agents (ATC code C07), calcium channel blockers (ATC code C08), agents acting on the renin–angiotensin system (RAS) (ATC code C09), and lipid-modifying agents (ATC code C10).
Primary outcome
New diagnosis of DM (yes/no) during the follow-up period was defined based on fulfillment of one of the following criteria: (a) new documentation of DM diagnosis—ICD-9 code 250xx in hospitals and community records; (b) HbA1c ≥6.5% based on hospital and community records; (c) at least two documentations of fasting glucose levels of 126 mg/dL or higher based on community records only; and (d) dispensed prescription of glucose-lowering medication (ATC code A10) based on hospital and community records. Since metformin or liraglutide is also used for treating conditions other than diabetes, individuals who started only those medications during the follow-up period without any other criteria were not considered as new DM cases. The earliest of these criteria was defined as the date of diagnosis.
Statistical analysis
Main characteristics are presented for COVID-19 and non-COVID-19 as means with SDs or medians with IQRs for quantitative variables that were normally or non-normally distributed, respectively. Categorical variables are presented by percentages. The association of COVID-19 with DM was evaluated by Kaplan-Meier non-parametric method and the log-rank test was used to compare the survival curves of COVID-19 and non-COVID-19 groups. HR and 95% CI for the association between COVID-19 and occurrence of new DM were assessed by stratified Cox proportional hazards model using 1:1 matched pair. Models were adjusted for age (years), socioeconomic status, pre-diabetes diagnosis of hypertension, diagnosis of dyslipidemia, diagnosis of PCOS, smoking status, BMI (kg/m2), glucose concentration (mg/dL), LDL cholesterol (mg/dL), HDL cholesterol (mg/dL), triglycerides (mg/dL), and lipid-modifying agents. To assess the possibility of period effect due to change in the therapy and severity of the infection over time, the follow-up was divided into three periods (January 20–August 31, 2020, September 01, 2020–March 31, 2021, and April 01, 2021–August 31, 2021), and the significance of the main effect of the period and the interaction period×COVID-19 were evaluated. To evaluate the possibility of surveillance bias,34 that is, capturing diagnosis of diabetes more in patients with COVID-19 than among non-COVID-19 matches, the percentage of individuals who had any interaction with the primary care clinics (that is, virtual visits, in-person visits, and blood glucose tests) was measured and compared between the two groups.
Individuals were censored from the analysis based on new diagnosis of DM, all-cause mortality, disruption in Clalit membership, COVID-19 diagnosis among controls, pregnancy, corticosteroid use in the outpatient setting, or end of study (December 31, 2021), whichever came first. The association of COVID-19 with DM was also evaluated for each of the COVID-19 severity groups. Five landmark models35 were performed including baseline and 1, 2, 3 and 4 months after COVID-19 diagnosis, evaluating the occurrence of new DM in different time frames in relation to COVID-19 infection with continued follow-up until the end of the study (December 31, 2021). Analyses at landmark 1, 2, 3 and 4 excluded patients with DM diagnosed from index date until the landmark date. Finally, five HRs were calculated, each of them representing the average HR for the specified period (from landmark date until the end of the study), and a time-varying association was assessed based on changes between these HRs. All statistical analyses were performed using R Statistical Software V.4.1.1. R package MICE V.3.1336 was used to impute missing baseline characteristics of the study population.
