Study design

The Atherosclerosis Risk in Communities (ARIC) Study is a prospective community-based cohort which began in 1987–1989 with 15,792 participants [13]. Participants were recruited from four communities in the USA: Forsyth County, North Carolina; suburban Minneapolis, Minnesota; Washington County, Maryland; Jackson, Mississippi. Visit 2 (1990–1992) serves as baseline for this analysis because it was the first ARIC visit with HbA1c measurements. We used data from six clinic visits in ARIC participants (visit 2, 1990–1992; visit 3, 1993–1994; visit 4, 1996–1998; visit 5, 2011–2013; visit 6, 2016–2017; and visit 7, 2018–2019) and from ongoing surveillance to identify outcomes for this study.

Of the 14,348 participants who attended visit 2, we excluded participants who were missing measurement of HbA1c (n=247), had a non-fasting blood draw (n=197), had missing data for adjustment covariates (n=88), had prevalent diabetes (glucose-lowering medication use, HbA1c ≥6.5%, or FG ≥7.0 mmol/l) (n=2401), had prevalent outcomes of interest (n=1001) or were missing information on outcomes during follow-up (n=47). Due to small numbers, we excluded participants who were neither Black nor White (n=34), and Black participants at the Maryland and Minnesota centres (n=32). Our analytic sample for all outcomes included 10,310 participants.

Definitions of baseline prediabetes and incident diabetes

HbA1c was measured in whole blood using HPLC methods (Tosoh 2.2 Plus in 2003–2004 and the Tosoh G7 in 2007–2008; Tosoh Corporation, Tokyo, Japan) aligned to the DCCT [14]. FG was measured in serum (or plasma at visits 3–5) using the hexokinase method.

We defined prediabetes at baseline as an HbA1c 39–47 mmol/mol (5.7–6.4%) and/or FG 5.6–6.9 mmol/l, according to ADA criteria [1]. We conducted secondary analyses of four other prediabetes definitions: HbA1c 39–47 mmol/mol (5.7–6.4%) only; FG 5.6–6.9 mmol/l only; the WHO’s definition for FG (6.1–6.9 mmol/l) only; and the National Institute for Health and Care Excellence (NICE) [12] and International Expert Committee (IEC) [15] definition for HbA1c (42–47 mmol/mol [6.0–6.4%]) only.

Participants were asked to bring their medication containers to the clinic visit and the containers were transcribed and coded. We identified cases of diabetes at subsequent visits using information on glucose-lowering medication use (self-report or medication inventory), biochemical measures of hyperglycaemia (FG ≥7.0 mmol/l or non-FG ≥11.1 mmol/l, HbA1c ≥48 mmol/mol [≥6.5%] at visits 5, 6, 7, or OGTT 2 h glucose ≥7.8 mmol/l at visit 4; see ESM Methods), or based on glucose-lowering medication use self-reported during annual follow-up calls (semi-annual calls after 2012 through 2020).

Incident complications

Participants were followed through 2020 for incidence of any complications (ASCVD, heart failure, chronic kidney disease [CKD] or all-cause mortality). We analysed the composite endpoint (any complication) and we also examined the individual components (ASCVD, heart failure, CKD, all-cause mortality). ASCVD was based on adjudicated CHD [16] and ischaemic stroke [17] events. CHD included definite fatal CHD or definite or probable myocardial infarction as previously described [16]. Ischaemic stroke was adjudicated based on physician review with National Survey of Stroke criteria and a computer algorithm [17]. Heart failure was identified based on hospitalisations or deaths with ICD-9 code 428 and ICD-10 code 150 (http://www.icd9data.com/2007/Volume1/default.htm and https://icd.who.int/browse10/2019/en, respectively) [18, 19]. CKD was based on CKD-related hospitalisations or death, creatinine-based eGFR [20] <60 ml/min per 1.73 m2 (stage G3a or higher) at baseline or eGFR decline ≥25% from baseline, or end-stage kidney disease identified by the US Renal Data System registry [21]. All-cause mortality was identified based on telephone contact with proxies, hospital records, obituaries, death certificates and linkage to state and national death indexes.

Covariates

Participants self-reported their physical activity, smoking status (current, former, never) and alcohol consumption (current, former, never). Physical activity was based on the Baecke questionnaire (measured at visit 1) [22] and used standard categories of recommended, intermediate or poor based on American Heart Association guidelines [23]. BMI was calculated based on measured height and weight. HDL-cholesterol and total cholesterol were measured using enzymatic methods. Hypertension was defined as systolic BP ≥140 mmHg, diastolic BP ≥90 mmHg, or antihypertensive medication use. BP measurements were based on the mean of the second and third BP reading following 5 min rest. All measurements were conducted by trained clinical staff.

Statistical analysis

We report descriptive statistics of the participants at baseline according to prediabetes status. For each outcome, person-time accrued from baseline up to the date of the complication, loss-to-follow-up, death or administrative censoring (31 Dec 2020), whichever came first. In models with time-varying diabetes as a covariate, person-time accrued before and after the date of diabetes incidence. We conducted Kaplan–Meier survival analyses for the composite outcome (any complication) and all-cause mortality by baseline prediabetes status. We generated cumulative incidence functions for ASCVD, heart failure and CKD by baseline prediabetes status and considered all-cause mortality as a competing event. We used Cox regression to assess associations of prediabetes with incidence of outcomes before and after adjusting for time-varying diabetes status occurring after visit 2 (baseline) and before the complication of interest (see Fig. 1). We calculated the remaining excess risk associated with prediabetes not explained by progression to diabetes using the following equation: 1 – (HRprediabetes – HRprediabetes allowing progression to diabetes) / (HRprediabetes − 1). We bootstrapped the CIs for these estimates (1000 samples). We calculated the median time to diabetes diagnosis and the percentage of participants who progressed from prediabetes to diabetes prior to each outcome of interest.

Fig. 1

Illustration (swimmer plot) of baseline prediabetes (HbA1c 39–47 mmol/mol [5.7–6.4%] or FG 5.6–6.9 mmol/l) and time-varying diabetes status with ~30 year risks of any complication. Each bar represents one hypothetical individual in the study. For each outcome, person-time accrued from baseline (1990–1992) up to the date of the complication, loss-to-follow-up, death or administrative censoring (31 December 2020), whichever came first. When time-varying diabetes was included as a covariate, person-time also accrued before (medium or light blue) and after (purple) the date of diabetes incidence

In Model 1a, we adjusted for age (continuous) and sex (male, female). In Model 1b, we adjusted for the variables in Model 1a plus time-varying diabetes (i.e. incident diabetes prior to outcome of interest). Our primary analyses were based on the comparison of Models 1a and 1b, which addresses the clinical question of whether people who screen positive for prediabetes in clinical practice (regardless of other cardiometabolic risk factors) have an elevated risk of complications before and after adjusting for time-varying diabetes status. In secondary analyses, we adjusted for additional covariates: Model 2a included age, sex, race-centre (Maryland-White, Minnesota-White, Mississippi-Black, North Carolina-White, North Carolina-Black), smoking status, alcohol consumption (current, former, never), physical activity level (recommended, intermediate, poor), BMI, total cholesterol, HDL-cholesterol, lipid-lowering medication use and hypertension. Model 2b included all variables in Model 2a plus time-varying diabetes. We used Stata version 17.0 (StataCorp, USA) and R version 4.3.0 (R Foundation for Statistical Computing, Austria) for all analyses. Two-tailed p values less than 0.05 were considered statistically significant.