We analysed data from the ARIC study, an ongoing community-based cohort designed to examine the aetiology of atherosclerotic disease. The study was comprised of 15,792 adults aged 45–64 years from four US communities: Forsyth County, NC; Jackson, MS; suburbs of Minneapolis, MN; and Washington County, MD. Participants were recruited in 1987–1989 for clinical examinations, medical interviews and laboratory tests (visit 1). Subsequent study visits were conducted in 1990–1992 (visit 2), 1993–1995 (visit 3), 1996–1998 (visit 4), 2011–2013 (visit 5), 2016–2017 (visit 6) and 2018–2019 (visit 7). All participants provided written informed consent and the study was approved by institutional review boards at all research sites. Further details about the ARIC study are available elsewhere .
Visit 1 was the baseline for our current study. We excluded respondents with missing information for diabetes status and covariates. Because of small sample size, we also excluded those with a race other than Black or White and excluded Black adults in Minneapolis, MN, or Washington County, MD. These restrictions yielded an analytical sample of 12,379 respondents.
During study visits, participants self-reported if they had ever been diagnosed with diabetes by a physician. They brought in all prescription medications used over the prior 2 weeks and had plasma glucose measured using the hexokinase method. We defined diabetes at baseline as fasting glucose ≥7 mmol/l, non-fasting glucose ≥11.1 mmol/l, self-report of a diagnosis of diabetes by a physician, or use of glucose-lowering medication at study visit 1. Because no antibody testing was performed, we could not distinguish between type 1 and type 2 diabetes in our analyses.
Hospitalisation for infection and infection mortality
The ARIC study conducted continuous active surveillance of hospitalisations for all participants. We defined first hospitalisation for infection as an infection ICD-9-CM/ICD-10-CM code (www.icd9data.com/2007/Volume1/default.htm and http://apps.who.int/classifications/icd10/browse/2016/en, respectively) in the first diagnostic position from hospital discharge records (ESM Table 1). Diagnostic codes were derived from prior ARIC studies focused on infection-related hospitalisation , along with research specifically focused on infection in individuals with diabetes . We considered first hospitalisation for specific diabetes-related infections (respiratory, urinary, foot, gastrointestinal, sepsis and postoperative wound) as secondary outcomes (ESM Table 1) .
Vital status was determined through linkage to the National Death Index, telephone interviews with participant proxies and review of state records. We defined infection mortality as an infection ICD-9-CM/ICD-10-CM code (ESM Table 1) listed as the underlying cause of death in death certificates.
We calculated person-time from study visit 1 until the first hospitalisation for infection (for hospitalisation for infection analysis only), death due to infection (for infection mortality analysis only), loss to follow-up, or administrative censoring, whichever came first. The final day of follow-up was 31 December 2017 for participants at one research site (Jackson, MS) and 31 December 2019 for all other participants.
Information on covariates
Structured questionnaires were administered at baseline to collect information on respondents’ demographic characteristics (age, sex, race, study centre, health insurance status, household income, education level) and health behaviours (smoking, alcohol consumption). Race and study centre were combined into a single measure (race-centre) to account for the uneven distribution of Black and White participants across research sites. Area deprivation was derived by combining 17 different neighbourhood socioeconomic status (SES) measures (e.g. median family income of a neighbourhood) into an index based on 2000 Census data . Categories of participants’ annual household income (0, >$50,000; 1, $25,000–$49,999; 2, $12,000–$24,999; 3, <$12,000), education level (0, graduate/professional school; 1, college with or without completion; 2, high school/general educational development/vocational school; 3, <high school) and area deprivation (0, lowest area deprivation quartile, lowest deprivation; 3, highest quartile, highest deprivation) were summed to create a SES score (high, 0–2; medium, 3–6; low, ≥7).
BMI was calculated as weight (kg) divided by height (m) squared. Obesity status was defined as either being obese (BMI ≥30 kg/m2) or non-obese (BMI <30 kg/m2). HDL-cholesterol and triacylglycerol levels were measured using enzymatic methods. LDL-cholesterol was calculated using the Friedewald equation. Hypertension was defined as BP ≥140/90 mmHg or self-reported use of BP-lowering medication. eGFR was determined using the Chronic Kidney Disease Epidemiology Collaboration formula . Chronic kidney disease was defined as eGFR <60 ml min−1 [1.73 m]−2. Prevalent CHD was defined by self-reported prior diagnosis or cardiovascular revascularisation, or evidence of a myocardial infarction on a study ECG.
We examined differences in participant characteristics at baseline by diabetes status using χ2 or t tests to assess group differences. We estimated the age-adjusted rate of hospitalisation for infection and infection mortality using predictive margins from Poisson models with robust error variance , using follow-up time as an offset. We used Cox proportional hazard models to examine the associations between diabetes and risk of hospitalisation for infection and infection mortality. We evaluated three models: model 1 included age, sex, race-centre; model 2 included all variables in model 1 plus SES score, health insurance status, smoking and alcohol consumption; model 3 included all variables in model 2 plus BMI, hypertension, chronic kidney disease, prevalent CHD status, HDL-cholesterol, triacylglycerols and LDL-cholesterol. In exploratory analyses, we examined potential effect modification by age, sex, race, SES, health insurance status and obesity status using Wald tests.
In sensitivity analyses, we examined the risk for hospitalisation for infection with diabetes measured as a time-varying exposure. We assessed the association between diabetes and incident hospitalisation for infection while accounting for the competing risk of death using the Fine–Gray model. We estimated the rate of total hospitalisation for infection (including recurrent events) using Poisson models.
We performed all analyses using Stata 16.0 (StataCorp, College Station, TX, USA). A two-sided p value of <0.05 was considered statistically significant.