Data analysis

We used the NDR registrations in the period 1/12006 to 31/122017. The baseline measurements for a participant were based on the three last NDR-registrations prior to the start of follow up. Missing data was imputed using last observation carried forward, i.e., if the last observation in NDR was missing then information was taken from the second last, if that was also missing it was retrieved from the third last NDR registration. If all the last three NDR observations were missing, then data was classified as missing.

Time to event was defined as the time from start of follow up to the first date of worsening diabetes control or last observation in NDR, whichever came first. Hazard ratios (HR) and 95% confidence interval (CI) for worsening of diabetes control as defined above were obtained using Cox proportional hazards regression models. All models were adjusted for age at PCa onset, duration of T2DM, education level, civil status, the Charlson Comorbidity Index (CCI), smoking habits, physical activity and body mass index (BMI). Cumulative incidence of worsening T2DM control was presented using Kaplan-Meier curves.

All data management was performed with Statistical Analysis Systems release 9.4 (SAS Institute, Cary, NC) and R 3.5.2 (R Foundation for Statistical Computing). The study has been approved by The Research Ethics Board at Uppsala University, Sweden.