Statistical analysis

We used unconditional logistic regression to calculate odds ratios (ORs) and 95% confidence intervals (CIs) for associations between autoimmune conditions and PC. Models were adjusted for age category (66–69, 70–74, 75–79, 80–84, 85–89, 90–99 years), sex, calendar year category of PC diagnosis/selection (1992–2000, 2001–2005, 2006–2009, 2010–2015), race/ethnicity (non-Hispanic White, non-Hispanic Black, Asian/Pacific Islander, Hispanic, other/unknown), geographic region (SEER registry), Medicaid eligibility (ever, never), average number of physician visits per 6 months in the 5 years before diagnosis/selection (quintiles in controls), average duration of Medicare part A, B, non-HMO coverage (quintiles in controls), Medicare low-income subsidy (ever, never, unknown), overweight/obesity (ever, never diagnosed), smoking behavior-related diagnoses (ever, never diagnosed), COPD (ever, never diagnosed), alcohol-related diagnoses (ever, never diagnosed), type 2 diabetes mellitus (ever, never diagnosed), chronic or acute pancreatitis (ever, never diagnosed).

We conducted exploratory stratified analyses by PC subtypes (PDAC, PNETs for the most significant autoimmune conditions), sex, age (< 75 years, ≥ 75 years), and race (non-Hispanic White, non-Hispanic Black). Race of other categories were not included in the analysis due to limited sample size. We fitted a two-stage hierarchical logistic regression model (online supplementary methods) separately to three groups (“primary gastrointestinal”, “secondary gastrointestinal-associated”, and “not gastrointestinal-associated”) to assess the combined effects of conditions within the same group.²⁵ The hierarchical models also account for correlations due to multiple autoimmune conditions occurring in the same individual and thus improve the accuracy of risk estimates. To account for the long duration of pancreatic cancer development and surveillance bias due to autoimmune disease diagnosis, we also conducted a sensitivity analysis to examine the associations of autoimmune conditions with all primary PC among participants whose cancer diagnosis/control selection was at least five years after autoimmune condition diagnosis. All statistical tests were two-sided. We considered a false discovery rate (FDR)-adjusted P (P_FDR) value < 0.10 statistically significant.²⁶ Analyses were conducted using SAS version 9.4 (SAS Institute, Inc., Cary, North Carolina).