Study design and data sources

We conducted an observational study of data from two nationwide Swedish registries with the aim of identifying risk factors for development of Charcot foot in individuals with diabetes mellitus. The National Diabetes Register (NDR) was initiated in 1996, with a continuously improved coverage reaching up to 87.7% in 2020 throughout Sweden [11] for individuals aged ≥18 years who have been diagnosed with diabetes mellitus. It includes demographic, clinical and laboratory data and lists type of medication, as recorded by a healthcare provider. All included individuals have agreed to inclusion in the register through informed consent. The National Patient Register (NPR) contains data in the form of ICD-10 codes (https://icd.who.int/browse10/2019/en), which are associated with diagnoses related to hospitalisations and outpatient visits. Cross-links between these two registries were established with the help of the personal identification number, which is unique for every Swedish resident. Data was anonymised before delivery to the researchers.

The study was approved by the Regional Ethics Review Board in Stockholm (diary approval numbers 2016/652–31/1, 2016/1429–32, 2016/2141–32 and 2017/410–32).

Study participants

All individuals with diabetes mellitus registered in the NDR, aged ≥18 years, were included in the study. Diabetes mellitus was defined using an inpatient or outpatient visit with the use of an appropriate reported ICD-10 code (E10–E14). Type 1 diabetes was defined using the ICD-10 code E10 and type 2 diabetes was defined using the ICD-10 code E11. The exclusive use of insulin did not classify a participant as having type 1 diabetes. If there was a discrepancy regarding the recorded diagnostic codes for the types of diabetes, the most recent record was assumed to be the most accurate. Charcot foot was defined using an inpatient or outpatient diagnosis of ICD-10 code M14.2, M14.6 or M90.8 in the NPR. The date of diagnosis of diabetes was recorded as the date of the initial diagnosis.

Risk factors

Individuals with diabetes mellitus but no Charcot foot who were registered in the NDR (during the study period) were matched with the highest available number of control individuals (7:1 for those with type 1 diabetes and Charcot foot; 10:1 for those with type 2 diabetes and Charcot foot), with respect to the length of time (similar number of years of diabetes duration) they had had the disease at time of initial diagnosis of Charcot foot (index date). Potential risk factors for Charcot foot were identified from NDR and NPR data. Data regarding demographics, clinical characteristics, biochemical variables and therapies for blood glucose, lipid and BP regulation were identified in the NDR (from January 2001 until December 2016). Data collected for the 12 months preceding the index date was used.

In the NDR, HbA1c values are reported dually as mmol/mol (International Federation of Clinical Chemistry and Laboratory Medicine) and % (National Glycohemoglobin Standardization Program). Microalbuminuria was defined using an albumin/creatinine ratio of 3–30 mg/mmol or a urine albumin concentration of 20–200 µg/min; macroalbuminuria was defined using an albumin/creatinine ratio of >30 mg/mmol or a urine albumin concentration of >200 µg/min. Two positive tests obtained from at least three different samples within one calendar year were needed to confirm this finding, otherwise individuals were defined as not having albuminuria. The presence of retinopathy was reported according to the existing national guidelines [21] and defined using the presence of alterations or damage to the retina caused by diabetes mellitus. BMI was defined as low/normal (≤25 kg/m2) or high (>25 kg/m2). Hypertension was defined as a BP of ≥140/85 mmHg. While both main and secondary diagnoses of peripheral vascular disease (PVD) (ICD-9 code [http://www.icd9data.com/2007/Volume1/default.htm] 443.9; ICD-10 code I73), atherosclerosis (ICD-9 code 440; ICD-10 code I70) and osteoporosis (ICD-9 code 733; ICD-10 code M81) were identified in the NPR, diagnoses of prior ischaemic heart disease or stroke, along with the use of acetylsalicylic acid and BP- and lipid-lowering medication were identified in the NDR.

Statistical analysis Descriptive data

The cohorts were divided into two independent groups according to the type of diabetes. Data are described using means ± SDs for normally distributed numeric data, and medians (IQRs) for data with skewed distributions. Counts and percentages are presented for categorical data. Two-sample t tests were used to compare mean values, Pearson’s χ2 test was used to test the difference between proportions and Wilcoxon rank-sum test was used to compare distributions of skewed distributed variables.

Potential causal links

To facilitate the evaluation of potential causal links between HbA1c, duration of diabetes, macroalbuminuria, microalbuminuria, atherosclerosis and retinopathy and Charcot foot, we created direct acyclic graphs (DAGs) to identify possible confounders for inclusion in the statistical analysis.

Applying d-separation criteria on the DAG, the following terms were indicated as confounders: for HbA1c, we used age, macroalbuminuria, microalbuminuria (type 1 diabetes and type 2 diabetes) and atherosclerosis (type 2 diabetes) (ESM Figs 1, 2) [22, 23]; for duration of diabetes, we used age (type 1 diabetes and type 2 diabetes) (ESM Fig. 3); for macroalbuminuria, we used age, HbA1c and BP (type 1 diabetes and type 2 diabetes); for microalbuminuria, we used age, HbA1c and BP (type 1 diabetes and type 2 diabetes) (ESM Fig. 4); for retinopathy, we used age, LDL-cholesterol, plasma triglycerides, HbA1c and BP (type 1 diabetes and type 2 diabetes) (ESM Fig. 5); and for atherosclerosis, we used age, HbA1c, LDL-cholesterol, HDL-cholesterol, triglycerides, duration of diabetes, BP, BMI and smoking (type 1 diabetes and type 2 diabetes) (ESM Fig. 6).

In addition, a sensitivity analysis was performed by reproducing the logistic regression models with the category ‘unknown’ included for BMI, BP, creatinine and diabetic retinopathy. This category was introduced to increase statistical power.

The robust sandwich estimator was used to estimate SEs and two-sided p values are reported. A p value of <0.05 was considered to represent statistical significance. The analyses were performed using Stata version 15 (StataCorp, College Station, TX, USA) and were conducted at the Biostatistics Core Facility of the Karolinska Institutet in Stockholm.