We analyzed data using R version 4.0.4 (18). Categorical data are presented as numbers and percentages. Continuous data are summarized as medians with interquartile ranges (IQRs). Time from type 2 diabetes diagnosis to sepsis was assessed using Cox regression analysis and the following models: unadjusted, adjusted for age and sex, and fully adjusted. In the fully adjusted model, we included the following potential confounders: age, sex, smoking, country of birth, socioeconomic factors, diabetes treatment, other medications, BMI, blood pressure, eGFR, albuminuria, and coexisting conditions (Supplementary Table 1). Based on previous data showing a J-shaped relationship between HbA1c and infections (7), we also used restricted cubic splines with three knots at the 25th, 50th, and 75th percentiles to flexibly model the association of HbA1c with sepsis before and after full adjustment for covariates. In the spline analyses, we used HbA1c of 53 mmol/mol (7.0%) as reference. We tested for potential nonlinearity by using a likelihood ratio test comparing the model with only a linear term against the model with linear and cubic spline terms. Because the association between HbA1c and sepsis was approximately log linear below and above the reference value, we used a log linear model to calculate hazard ratios per SD increase in HbA1c.

Because immunosuppressed patients are particularly prone to infections, we conducted two sensitivity analyses excluding such patients. In the first analysis, we did a complete case exclusion of individuals with an ICD-10 diagnosis associated with immunosuppression (cancer, renal dialysis and transplantation, and immunological deficiency) and/or individuals receiving treatment with immunosuppressive drugs (systemic glucocorticoids and/or other immunosuppressive drugs) (Supplementary Table 1 lists ICD-10 and Anatomical Therapeutic Chemical codes). In the second analysis, individuals were censored at the time point when an immunosuppressive disease was diagnosed or when immunosuppressive therapy commenced.

The association between sepsis and death was assessed using Cox regression analysis and covariates included in the fully adjusted model (Supplementary Table 1). We further explored the association between HbA1c and mortality among individuals with type 2 diabetes who developed sepsis using Cox regression analysis before and after adjustment.

All tests were two tailed and conducted at a significance level of 0.05.