We used data from the Epidemiological Study of Risk Factors for LADA and Type 2 diabetes (ESTRID), a population-based case–control study with incident cases (www.ki.se/IMM/ESTRID). The details of ESTRID have been described previously . In short, ESTRID was initiated in 2010, recruiting incident cases of diabetes through diabetes registries in the Swedish counties of Skåne (All New Diabetics in Skåne [ANDIS], http://andis.ludc.med.lu.se/) and Uppsala (All New Diabetics in Uppsala [ANDiU], www.andiu.se/). The aim of these registers is to characterise all incident cases of diabetes according to type, genetic factors and clinical features.
All individuals with incident LADA registered in Skåne (2010–) and Uppsala (2012–) were invited to take part in ESTRID, together with a random sample of people with type 2 diabetes (four per LADA case) and control participants free from diabetes (six per LADA case). The controls were randomly selected via the National Population Registry, which covers the whole population of Skåne and Uppsala (~1.6 million inhabitants). They had to be ≥35 years of age (corresponding to the criteria used to define LADA) and were matched on geographical area (county) and date of participation, i.e. incidence density sampling . All participants in ESTRID between 1 September 2010 and 1 July 2014 with self-reported information on birthweight were eligible for the analyses presented here.
Of the original sample, 44.6% (Skåne) and 45.3% (Uppsala) of participants could recall their birthweight information, including 134 patients with LADA, 350 patients with type 2 diabetes and 603 controls. Those who filled in the questionnaire but did not remember their birthweight (answered ‘I do not know’ to the birthweight question) were excluded from the analysis. The overall response rates for ESTRID were 80% for cases and 66% for controls. Participants gave written informed consent and the study was approved by the Ethical Review Board in Stockholm (reference number 2010/336-31/2).
Assessment of exposure and covariates
A comprehensive questionnaire soliciting a broad range of items on lifestyle, family history and health was sent out by mail to all cases and their matched controls. Patients received the questionnaire close to diagnosis, with careful instructions to report lifestyle habits as they had been before diagnosis.
Information on birthweight was based on self-reports. To validate the self-reported information on birthweight, ESTRID was linked to the Swedish Medical Birth Register. This registry contains information on prenatal and neonatal care for nearly all births in Sweden from 1973 onwards; information on birthweight was collected for all ESTRID participants born in 1973 onwards (n = 112). Of these individuals, 76 (68%), including 16 cases and 60 controls, also had self-reported information on birthweight from the ESTRID questionnaire.
The BMI was calculated by self-reported weight and height as kg/m2 and participants were categorised based on WHO criteria as normal weight (BMI < 25 kg/m2), overweight (BMI ≥ 25 kg/m2) and obese (BMI ≥ 30 kg/m2). Detailed information on family history of diabetes was gathered through questions on diabetes in first-degree relatives (mother, father, sister, brother and children) and second-degree relatives (maternal and paternal grandparents, aunts and uncles). Relatives were classed as having type 1 diabetes if they had onset <age 40 and insulin treatment, but were otherwise deemed to have type 2 diabetes. Responders were also asked to report their highest level of education, with the categories elementary, high school and university. Smoking status was collected by questions on smoking habits and respondents were divided into current, former or never smokers. Alcohol intake was assessed through questions on amount and frequency of wine, beer and liquor intake, with participants classed as abstainers, ex-drinkers, or low, moderate or high consumers based on this information. Information on physical activity was obtained by questions on leisure time activity during the preceding year, with four response options ranging from sedentary to very active.
Clinical data and case definition
Patients were diagnosed within the healthcare system of Skåne and Uppsala and registered in the ANDIS (Skåne) and ANDiU (Uppsala) registries. At the time of diagnosis, blood samples from all cases were collected and analysed for glutamic acid decarboxylase antibody (GADA), C-peptide and fasting blood glucose levels at the central laboratory of the university hospital of each county.
Positivity for GADA, which is the most common autoantibody, being present in 90% of all individuals with adult autoimmune diabetes , was measured by ELISA, with cut-off ≥10 U/ml, according to the manufacturer’s directions . The values are reported as an index value in relation to standard serum. The assay gives a maximum value of 250 with a cut-off level 10.7 U/ml; when GADA was measured in Ca2+ treated plasma, sensitivity and specificity were estimated as 84% and 98%, respectively .
Assessment of C-peptide level was performed using a Cobas e 601 analyser, (Roche Diagnostics, Mannheim, Germany) or IMMULITE 2000 (Siemens Healthcare Diagnostics Products, Llanberis, UK) .
For this study, LADA was defined as onset ≥35 years of age, GADA positivity (≥10 U/ml) and C-peptide levels ≥0.3 nmol/l (Cobas e 601) or ≥0.2 nmol/l (IMMULITE). Individuals with onset ≥35 years but who were GADA negative (<10 U/ml) and with C-peptide levels >0.72 nmol/l (Cobas e 601) or >0.6 nmol/l (IMMULITE) were classed as having type 2 diabetes. There is no universal definition of LADA, but this definition is in line with previous reports, except for C-peptide which was used as an indicator of remaining insulin production and a slow ‘latent’ onset. This replaced the commonly used insulin criterion, which is subjective. Via linkage to the ANDIS and ANDiU registries, clinical data from the time of diagnosis were available for all patients who gave consent for this (98.8% of all cases; two LADA cases and four type 2 diabetes cases are missing in these analyses). The HOMA model was used to estimate insulin resistance (HOMA-IR) and beta cell function (HOMA-β) based on fasting plasma glucose and C-peptide . Information from blood samples was not available for the controls in the present study.
By means of logistic regression we calculated ORs with 95% CIs for the associations for birthweight, LADA and type 2 diabetes. Conditional logistic regression analysis, matched for date of participation and residential area, was also conducted, but we present only the results from the unmatched analyses as these were in close agreement with those from the matched analyses, but allowed us to use all the available information. The ORs were interpreted as incidence rate ratios, as this case–control study was based on incident cases and controls sampled by incidence density sampling .
The ORs were adjusted for age (continuous) and sex (model 1), age, sex and BMI (continuous) (model 2) and age, sex, BMI and family history of diabetes (yes vs no) (model 3). Results from model 3 are presented unless otherwise stated. Information on these covariates was complete for all participants with information on birthweight. Additional adjustment for education, smoking, alcohol and physical activity did not affect the ORs and therefore these variables were not included in the final model. Birthweight was modelled both as a continuous variable (per kg reduction in birthweight) and as a categorical variable divided into low (<3 kg), normal (3–<4 kg) and high (≥4 kg) birthweight. We also modelled birthweight in five categories with reference to the WHO definition of low birthweight (<2.5 kg, 2.5–<3 kg, 3–<3.5 kg, 3.5–<4 kg and ≥4.0 kg) . To determine possible interactions between low birthweight and overweight we calculated the relative excess risk due to interaction (RERI) . In these analyses, birthweight and BMI were dichotomised as low (<3 kg) and high (≥3 kg) birthweight and normal weight (BMI < 25 kg/m2) vs overweight/obesity (BMI ≥ 25 kg/m2). Sensitivity analyses were performed using a higher cut-off for the GADA assay (≥20 U/ml rather than ≥10 U/ml). We also performed separate analyses in which: participants reporting diabetes in the mother were excluded to account for maternal diabetes during pregnancy; and participants born outside Sweden or with immigrant parents were excluded, to check if ethnicity could affect the results. Analyses were performed separately for individuals older than 60 years as we could not validate self-reported birthweight in higher age groups and participants’ recall may have been less reliable.
Two-sided p values were calculated using χ
2 (proportion), Student’s t (mean) and Kruskal–Wallis (median) tests. All statistical calculations were performed using the statistical software program SAS version 9.4 (SAS Institute, Cary, NC, USA). Where ORs are reported, the 95% CI are shown in parentheses.