This prospective, nested case–control study included twins from the nationwide Swedish Twin Registry (STR), which was started in the 1960s . In 1998–2002, all living twins in the registry who were born in 1958 or earlier were invited to participate in the Screening Across the Lifespan Twin study (SALT), a full-scale screening that gathered data on an extended set of variables via computer-assisted telephone interviews. Out of 44,919 twin individuals eligible for the telephone interview, we excluded 3184 with heart disease before screening and 272 with type 1 diabetes, resulting in 41,463 individuals with data for the current analyses (Fig. 1).
Information on age, sex, educational attainment, marital status and zygosity was obtained from the SALT survey . All twins were categorised as monozygotic, dizygotic or of undetermined zygosity. Education was defined as the maximum years of formal schooling attained, and dichotomised into <8 vs ≥8 years. Marital status was defined as married/cohabiting vs single (including divorced and widows/widowers).
Information on history of type 2 diabetes and heart disease was derived from the National Patient Registry (NPR), which covers all inpatient diagnoses in Sweden from the 1960s to the end of 2014, and outpatient (specialist clinic) diagnoses since 2001. Each medical record in the NPR included up to eight discharge diagnoses according to the ICD. The seventh revision (ICD-7) was used up to 1968, the eighth revision (ICD-8) from 1969 to 1986, the ninth revision (ICD-9) from 1987 to 1996 and the tenth revision (ICD-10) since 1997.
All participants provided informed consent. The data collection procedures were approved by the Regional Ethics Committee at Karolinska Institutet, Stockholm, Sweden, and by the Institutional Review Board of the University of Southern California, USA.
Ascertainment of type 2 diabetes
Type 2 diabetes was ascertained based on self- and informant-reported history of diabetes, glucose-lowering medication use or the NPR (ICD-7 code 260; ICD-8 and -9 code 250; and ICD-10 codes E10–E14). The age at type 2 diabetes onset was estimated according to the earliest recorded date of type 2 diabetes in the NPR or the date of type 2 diabetes onset available in SALT.
Assessment of heart disease
Information on heart disease diagnoses (ICD-7 codes 420, 433 and 434; ICD-8 and -9 codes 410–414, 427 and 428; and ICD-10 codes I20–I25 and I48–I50) was obtained from the NPR. According to the ICD codes, the major subtypes of heart disease included: (1) coronary heart disease: angina pectoris, acute myocardial infarction, chronic ischaemic heart disease and other coronary heart disease (such as coronary thrombosis and Dressler’s syndrome); (2) cardiac arrhythmias: atrial fibrillation and flutter, and other cardiac arrhythmias (such as ventricular fibrillation and flutter, atrial premature depolarisation and junctional premature depolarisation); and (3) heart failure: congestive heart failure, left ventricular failure and unspecified heart failure. The age of heart disease onset was estimated as the earliest date that a heart disease diagnosis was recorded in the NPR.
Assessment of lifestyle-related factors
Information on smoking status, alcohol consumption, physical activity and BMI was obtained from the SALT survey. Smoking status was dichotomised as never vs ever being a smoker. Data on alcohol consumption were collected by a question on drinking habits, ‘Think about your use of alcohol over your entire life. Has there ever been a period in your life when you drank too much?’, with two response options: (1) ‘no’; and (2) ‘yes’. We defined ‘no’ as ‘no/mild drinking’ and ‘yes’ as ‘heavy drinking’. Data on physical activity were collected by a question on average exercise, with seven response options: (1) ‘almost never’; (2) ‘much less than average’; (3) ‘less than average’; (4) ‘average’; (5) ‘more than average’; (6) ‘much more than average’; and (7) ‘maximum’ . For the analyses, we combined these categories into two groups and defined ‘low’ as exercise ‘almost never’ to ‘much less than average’, and ‘regular’ physical activity as ‘less than average’ to ‘maximum’. BMI was calculated as weight (kg) divided by height squared (m2), and was categorised as non-overweight (BMI <25) and overweight (BMI ≥25).
In the current study, we considered four healthy lifestyle-related factors: being a non-smoker, no/mild alcohol consumption, regular physical activity and being non-overweight. Participants were divided into three groups according to the number of lifestyle-related factors: (1) unfavourable: participants who had no or only one healthy lifestyle factor; (2) intermediate: those who had any two or three healthy lifestyle factors; and (3) favourable: those who had four healthy lifestyle factors.
The characteristics of participants in different groups were compared using χ2 tests, t test and Mann–Whitney test. Generalised estimating equation (GEE) models were used to analyse the unmatched case–control data while controlling for the clustering of twins within a pair. To examine the associations between type 2 diabetes and risk of heart disease independently, we looked at the first onset of one specific subtype of heart disease with no others. Data for the co-twin control study were analysed by using conditional logistic regression, in which twin pairs were discordant for outcome; thus, cases and control participants were comparable with respect to early-life familial environmental factors (such as shared childhood socioeconomic status and adolescent environment) and genetic background (monozygotic twins shared 100% of their genetic background and dizygotic twins shared only 50%) . In both GEE and conditional logistic regression, the ORs and 95% CIs were estimated for the association between type 2 diabetes and heart disease.
Logistic regression was used to test the difference in ORs from GEE models and conditional logistic regression by examining the difference between the proportions of type 2 diabetes in unmatched control participants and in co-twin control participants [21,22,23,24]. Absence of a statistically significant difference in ORs from the GEE and conditional logistic regression analyses suggests that genetic and early-life familial environmental factors might not account for the observed associations. In contrast, a statistically significant difference in ORs from the GEE and conditional logistic regression analyses indicates that genetic and/or shared environmental factors likely play a role in the observed associations [15, 21,22,23,24,25].
The combined effect of the type 2 diabetes and lifestyle on heart disease risk was assessed by creating dummy variables based on the joint exposures to both factors. The presence of an additive interaction was examined by estimating the relative excess risk due to interaction (RERI), the attributable proportion (AP) and the synergy index (SI). Additionally, we examined multiplicative interaction by incorporating the two variables and their cross-product term in the same model.
Age, sex, education, BMI, smoking, alcohol consumption, marital status and physical activity were considered as potential confounders in the type 2 diabetes–heart disease association. Missing values on education (n = 1217), smoking (n = 1167), alcohol consumption (n = 1261), BMI (n = 1918), marital status (n = 755) and physical activity (n = 5938) were imputed by chained equation to obtain valid statistical inferences with five completed datasets generated. All statistical analyses were performed using SAS statistical software version 9.4 (SAS Institute, Cary, NC, USA) and IBM SPSS Statistics 24.0 (IBM Corp, New York, NY, USA).