This observational study is conducted within a register-based framework, utilizing data from several Swedish national health and administrative registers. The linking of data is facilitated through the personal identification number, a unique identifier assigned to each Swedish citizen. The registers incorporated in this study comprise the Swedish National Diabetes Register (NDR) [17], the Total Population Register (TPR) (administered by the government agency Statistics Sweden) [18], the National Patient Register (NPR) [19], the Prescribed Drug Register (PDR) [20], the Swedish Cause of Death Register [21], and the Longitudinal Integration Database for Health Insurance and Labor Market Studies (LISA) [22].

The NDR has been thoroughly documented in previous literature [4, 10, 17]. In summary, established in 1996, the NDR boasts nationwide coverage and includes nearly all individuals with T2D in Sweden. On an annual basis, data encompassing patient characteristics, risk factors, treatment, and complications related to diabetes mellitus are collected from patient encounters at healthcare facilities specializing in the management of people with T2D, and reported to the NDR.

The epidemiological definition of T2D (dietary treatment with or without the use of oral antihyperglycemic medication; and additionally, individuals diagnosed with diabetes mellitus at the age of 40 years or older and treated with insulin, either exclusively or in combination with other oral antihyperglycemic agents, are also categorized as having T2D) was used to identify persons with T2D from the NDR in this study.

The study cohort consisted of all persons with T2D who had at least one registration in the NDR between January 1, 2005, and December 31, 2008. At the initial registration in the NDR, which also served as the index date for the study, each person with T2D was matched for sex, age, and county with two control participants without T2D. The controls were randomly sampled with replacement from the general population (Total Population Register, Statistics Sweden), and the primary objective of the matching process was to secure controls without T2D for use as matched comparators. The control participants were initially free of T2D at the time of their selection, but a very small subset of these individuals developed T2D during the wash-in period (2005–2008) before follow-up. Instead of censoring them, these cases were matched with two new controls, resulting in a non-integer case-control ratio. Controls who received a T2D diagnosis in the NPR during follow-up were censored at the time of diagnosis. Study participants with manifest PAD or amputation at baseline were excluded from the present analyses.

The subset of individuals with T2D and controls who experienced onset of incident PAD diagnosis during the study period had their follow-up initiated at the time of the registered incident PAD diagnosis in the NPR. Their baseline data were then set as the last pre-PAD observation, encompassing sex, age, and all relevant comorbidities and socioeconomic factors listed below. Study participants were followed until death, covering both all-cause mortality and ASCVD mortality, or until the conclusion of the study period on December 31, 2020.

Baseline comorbidity data and the identification of incident PAD during the study period were acquired from the NPR, which employs Swedish revisions of International Classification of Diagnoses (ICD) version 10 codes, for diagnosing and classifying medical conditions. Administered by the Swedish National Board of Health and Welfare, the NPR serves as the source for this health-related data [19]. A detailed description and definition of all comorbidities, along with their corresponding ICD-10 codes considered in this study, are provided in Supplementary Table 1, available in the online data supplement. The definition of ASCVD in this study encompasses a combination of myocardial infarction, ischemic heart disease, and stroke. Socioeconomic factors such as marital status, educational level, region of origin, and income level were gathered from the LISA, administered by Statistics Sweden [22]. The PDR facilitated the collection of data on filled prescriptions for statins, antihypertensive, antiplatelet, and anticoagulant medications over a one-year period preceding the index date [20]. The methodology of defining PAD, other comorbidities and outcomes using corresponding ICD-10 codes in the NPR, along with the collection of additional baseline and outcome data from various Swedish national health and administrative registers, has been employed in prior national and international collaborations [4, 8, 10, 23, 24].

Clinical characteristics and cardiovascular risk factors for persons with T2D were gathered from the NDR. In the analyses exploring the associations of risk factors with all-cause and ASCVD mortality in individuals with T2D after the onset incident PAD, the following baseline variables were considered: glycated haemoglobin (HbA1c), blood pressure levels (systolic and diastolic), total cholesterol, low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), triglycerides, body mass index (BMI), smoking status (defined as a current smoker at study entry), presence of micro- and macroalbuminuria, and estimated glomerular filtration rate (eGFR). The eGFR was calculated using the MDRD equation [25]. Microalbuminuria was defined as two separate samples of urine albumin/creatinine ratio of 3–30 mg/mmol (30–300 mg/g) within a year, while macroalbuminuria followed a similar definition but with higher ratios than 30 mg/mmol. Risk factor data were not available for control subjects in this study.

The two pre-defined main outcomes in this study were incident all-cause mortality and ASCVD mortality (defined with ICD-codes I00-I99, R570, R960, R961). Information on these outcomes were obtained from the Swedish Cause of Death Register, administered by the Swedish National Board of Health and Welfare. This register contains comprehensive details on all deaths, including causes and timing of death, for individuals registered in Sweden. The underlying cause of death is coded based on ICD codes, and reporting to this register is mandatory [21]. Both people with T2D and controls had their outcomes equally ascertained.

Descriptive statistics are reported using means with standard deviations (SD) or, when appropriate, medians with interquartile range (IQR) values for continuous variables. Categorical variables are presented as counts (n) with corresponding percentages (%) for categorical variables. The magnitude of differences at baseline across variables between study groups, were calculated with standardized mean differences (SMD), where values below 0.1 were considered non-significant.

Event rates were calculated as the number of incident all-cause mortality and ASCVD mortality events per person-years for individuals with T2D and controls. Event rates in individuals without PAD are reported per PAD-free year. At time of PAD-diagnosis, the individuals are censored. For individuals with PAD, event rates are reported per PAD-exposed year. Stratification was performed based on whether onset of incident PAD occurred during the study period or not, and sex aswell. These rates were presented with 95% exact Poisson confidence intervals (CI) and visualized using Kaplan-Meier and cumulative incidence curves.

The risk of incident all-cause mortality and ASCVD mortality was further assessed by employing Cox proportional hazards regression. A competing risk approach was applied for ASCVD mortality using the Fine-Gray method. These models were also stratified on whether onset of incident PAD diagnosis occurred or not in a multi-state design. The comparison of risk was conducted between individuals with T2D and controls. Specifically, the risk comparison was divided into two sub analyses: one sub analyses where the risk was compared between individuals with T2D and controls who had experienced onset of incident PAD diagnosis during the study period, and the other sub analyses where the risk was compared between those individuals with T2D and controls who did not experience the onset of incident PAD. This approach allowed for a more nuanced evaluation of risk differences in these distinct subgroups. The Cox proportional hazards models underwent a stepwise adjustment process. The final model encompassed adjustments for age, sex, income level, educational level, marital status, region of origin, selected medications (antihypertensive medication, statins, anticoagulant medication, antiplatelet therapy), and comorbidities as defined in Supplementary Table 1. The associations are summarized as hazard ratios (HR) presented with 95% confidence intervals (CIs).

The secondary objective, focused on evaluating T2D-related ASCVD risk factors and their associations with all-cause and ASCVD mortality among persons with T2D after the onset of incident PAD, was analysed using Cox proportional hazards testing. To assess the relative importance of T2D-related cardiovascular risk factors in the models estimating the risk of all-cause and ASCVD mortality, a Gradient-Boosting Model (GBM) with a proportional hazards loss function was employed [26]. The GBM-model was fitted with a shrinkage of 0.05 for computational efficiency and an interaction depth of 3. The number of trees, optimized at 1167, was determined through 5-fold cross-validation. The relative importance, also known as the partial graded effect, of each predictor to the model for the outcome risk is presented as an estimate. This estimate, termed relative importance, indicates the predictive significance of each risk factor to the model for outcomes risk and was determined using GBM, a machine learning technique.

Given the register-based nature of this study, missing data on T2D-related cardiovascular risk factors in the NDR were acknowledged. Presumed to be missing at random, a multiple imputation model—multiple imputations by chained equations (MICE)—was applied to handle missing data, mainly clinical biomarkers from the NDR. This iterative imputation method predicted missing values using observed data, and the imputed datasets were combined to create a final dataset for subsequent analyses. The imputed data was used the assessment of mortality risk as well as the evaluation of risk factor importance to the models and association of risk factors with outcomes. Supplementary Table 2 provides a comprehensive list of variables considered during the imputation process, ensuring transparency in the imputation model.

The assessment of Schoenfeld’s and Martingale residuals indicated that the proportional hazard assumption was met. All statistical testing adhered to a two-sided significance level of 5%, and 95% confidence intervals (CIs) were employed. Given the exploratory nature of the study, no correction for multiple comparisons was applied. The statistical analyses were conducted using R version 4.0.3.