Study population

We used data from the Epidemiological Study of Risk Factors for LADA and Type 2 Diabetes (ESTRID), nested within the ANDIS (All New Diabetics in Scania) biobank and incidence register covering the Swedish region of Scania [19]. Since 2010, all incident cases of LADA (termed ‘slowly evolving, immune-mediated diabetes of adults’ by WHO [20]) in ANDIS have been invited to participate in ESTRID, together with randomly sampled type 2 diabetes patients (ratio 1:4) from the ANDIS biobank and diabetes-free control participants (ratio 1:6) randomly sampled from the population of Scania and matched by participation date and residential area [21]. All cases and control participants recruited between 2010 and 2019 were eligible for the present study (LADA, n=586; type 2 diabetes, n=2003; control participants, n=2355). The Swedish Ethical Review Board approved the study (numbers 2010/336-31/1 and 2018/1036-32).

Diabetes classification

Cases of diabetes were diagnosed in the healthcare system of Scania. Patients aged ≥35 years at diagnosis were classified as having LADA if they were GADA-positive (≥10 U/ml) with C-peptide ≥0.3 nmol/l (measured using a Cobas e601 analyser; Roche Diagnostics, Germany) or ≥0.2 nmol/l (measured using an IMMULITE 2000 assay system; Siemens Healthcare, UK) and as having type 2 diabetes if they were GADA-negative with C-peptide ≥0.72 nmol/l (Cobas) or ≥0.60 nmol/l (IMMULITE). LADA patients were further categorised into LADAhigh (GADA ≥250 U/ml) and LADAlow (GADA <250 U/ml) based on the median GADA level. The sensitivity and specificity of GADA measurement using ELISA (RSR, UK) were 0.84 and 0.98, respectively. HOMA-IR and HOMA-B values were calculated from fasting glucose and C-peptide levels. Genotyping was conducted using iPlex (Sequenom, USA) or TaqMan assays (Thermo Fisher Scientific, USA) [19]. We used three SNPs (rs3104413, rs2854275 and rs9273363) to predict HLA-DRB1 (DR3/DR4) and HLA-DQB1 (DQ2/DQ8) genotypes. An overall accuracy of 99.3% has been demonstrated for this approach [22].

Autoimmune diseases

Information on ADs was obtained through the National Patient Register (NPR) (1997–2019) and the Scania Healthcare Register (2004–2019) (see electronic supplementary material [ESM] Fig. 1). The NPR provides nationwide coverage of inpatient care since 1987 and outpatient care since 2001 [23] and the Scania register provides almost complete coverage of primary care since 2004 [24]. We retrieved information on diagnoses of 33 ADs (ESM Table 1) that have been shown to be relatively prevalent in Sweden [25]. In addition, we used self-reported information on diagnoses of coeliac disease, thyroid dysfunction, Sjögren’s syndrome, systemic lupus erythematosus, ulcerative colitis or Crohn’s disease, vitiligo, psoriasis, multiple sclerosis or rheumatoid arthritis and year of diagnosis. In the analyses, we integrated self-reported information with the register data and combined diagnoses of Graves’ disease and Hashimoto’s disease into thyroid dysfunction, and diagnoses of ulcerative colitis and Crohn’s disease into inflammatory bowel disease. There were also questions on family history of these ADs in the patient’s mother, father, siblings or other relatives.

Diabetic retinopathy

Information on diabetic retinopathy (hereafter referred to as ‘retinopathy’) was retrieved from the NPR, the Cause-of-Death Register (CDR) and the National Diabetes Register (NDR) [26]. Retinopathy was assessed in the patients with diabetes, and was defined as the first occurrence in NPR or NDR of severe non-proliferative retinopathy, pre-proliferative diabetic retinopathy, proliferative diabetic retinopathy, diabetes with advanced eye disease, other proliferative retinopathy, diabetic cataract, retinal haemorrhage, visual impairment or vitreous haemorrhage, or death attributable to diabetic retinopathy in the CDR. The ICD-10 codes (https://icd.who.int/browse10/2019/en) for these forms of retinopathy are given in ESM Table 2.

Covariates

Information on covariates was collected by questionnaire at enrolment. BMI was based on self-reported weight and height. Smoking history was used to categorise individuals into never, former or current smokers. Based on validated questions [27], individuals were categorised as sedentary or displaying light, moderate or high physical activity. Family history of type 1 diabetes in first-degree relatives (mother, father, siblings or children) or grandparents was defined as diagnosis at age <40 years with insulin treatment; otherwise, the diagnosis was recorded as type 2 diabetes. Educational level was categorised into primary, secondary or tertiary. We used median values to impute missing data on lifestyle covariates (missing rate 1.5%), and created a dummy variable to indicate missingness, which we incorporated into regression models. Information on HbA1c levels and diabetes medications was retrieved from NDR and the prescribed drug register (2005–2019), respectively.

Statistical analyses

We assessed differences in the distribution of baseline characteristics using the χ2 test (for proportions with expected frequency ≥5), Fisher’s exact test (for proportions with expected frequency <5), Student’s t test (unpaired) for means and the Wilcoxon test for medians. Values were considered significant at p<0.05. R version 4.3.3 (R Foundation for Statistical Computing, Austria), SAS 9.4 (IBM, USA) and STATA 17.0 (StataCorp, USA) were used.

Case–control analyses

Conditional logistic regression was used to estimate ORs and 95% CIs for LADA/type 2 diabetes in relation to ADs and family history of ADs in first-degree relatives. We used information on ADs occurring at least one year prior to the index date (date of diagnosis for patients or date of enrolment for control participants). ORs were calculated in relation to any vs no AD, number of ADs, and individual ADs if there were five or more affected LADA patients. Regarding family history, ORs were estimated in relation to any AD, number of ADs, number of affected relatives and individual ADs. Model 1 was adjusted for age and sex; model 2 was additionally adjusted for education, smoking, physical activity and BMI; and model 3 was additionally adjusted for family history of type 1 diabetes, type 2 diabetes and any AD. For analyses of individual ADs, the models were mutually adjusted for other ADs. The results of model 3 are presented below unless otherwise stated. We calculated the attributable proportion due to interaction [28] to assess additive interaction between having an AD and having first-degree relatives with AD.

Cohort analyses

Cox proportional hazard regression models, with age as the time scale, were used to estimate HRs and 95% CIs for retinopathy in LADA with and without AD, compared with type 2 diabetes. Person-years were counted from baseline date until date of the first event or end of follow-up (31 December 2019). The main model was adjusted for sex, calendar year at diabetes diagnosis, diabetes duration, education, smoking, alcohol, physical activity, BMI, HbA1c, BP, lipids and eGFR. HbA1c trajectories for LADA with and without AD were estimated using generalised linear models adjusted for age, sex and calendar year.

Sensitivity analysis

We performed analyses restricted to register information on ADs, and adjusted the analyses of individual ADs and family history of ADs for multiple testing. As ADs were more common in female participants than male participants, we stratified the analyses by sex. As a post hoc analysis, we estimated the risk of type 2 diabetes in relation to ADs while excluding psoriasis. We analysed retinopathy in LADA vs type 2 diabetes while adjusting for glucose-lowering drugs, statins and antihypertensive treatment, with non-diabetic retinopathy death as a competing event using the Fine–Gray method [29].

Biological pathway analysis

For ADs associated with LADA, we explored the underlying biology through pathway analysis using FUMA, a web-based platform that incorporates various biological resources to functionally annotate GWAS results and prioritise genes [30]. First, we used the SNP2GENE function to apply positional mapping (restricted to exonic and splicing SNPs with a combined annotation-dependent depletion [CADD] score ≥12.37), expression quantitative trait loci (eQTL) mapping (performed using GTEx version 8 eQTLs with false discovery rate [FDR] <0.05), and chromatin interaction mapping (performed using Hi-C data, with interactions filtered by FDR <1×10−6) to GWAS summary statistics for LADA [31], Crohn’s disease [32], ulcerative colitis [32], Graves’ disease [32], hypothyroidism [33] and vitiligo [34]. Subsequently, we chose genes present in both LADA and the respective ADs. Finally, we used the GENE2FUNC function to annotate the shared genes for biological mechanisms. Details of the mechanisms are shown in ESM Fig. 2.