Study population

We leveraged the large Utah Population Database (UPDB) to investigate the association between endometriosis and type 2 diabetes. The UPDB is a comprehensive, population-based data resource that includes information on more than 11 million individuals [16]. It uses probabilistic record linking based on multiple identifiers to link vital records (birth, death, marriage), health records from statewide inpatient, ambulatory surgery and emergency department facilities, and University of Utah and Intermountain Health electronic health records, in addition to other records including census information and driver licences (see electronic supplementary material [ESM] Fig. 1) [17]. Our study protocol was approved by the Resource for Genetic and Epidemiologic Research, the University of Utah Institutional Review Board, and the Intermountain Health Institutional Review Board. All research was conducted under a waiver of informed consent approved by the University of Utah Institutional Review Board.

Our UPDB‐based, dynamic cohort comprised all individuals assigned female at birth who were born on 1 January 1935 or later, and who had documented Utah residency on or after 1 January 1996. Cohort entry (‘time zero’) was defined as 1 January 1996 or date of first Utah residency for women who were not residents at the study start date. This approach ensured that all individuals were available for exposure and outcome ascertainment from the start of follow-up, while avoiding inclusion of unobservable person-years prior to Utah residency [18].

Exposure

Endometriosis exposure was identified using validated [19] ICD-9 and ICD-10 diagnosis codes drawn from all linked health facility sources. In line with prior research [20], we defined endometriosis subtypes by 617* or N80* ICD-9/10 codes: ICD-9 codes 617, 617.2 and 617.3 and ICD-10 codes N80.1 and N80.2 for SE; ICD-9 code 617.1 and ICD-10 code N80.1 for OE; ICD-9 codes 617.4 and 617.5 and ICD-10 codes N80.4 and N80.5 for DE; and ICD-9 codes 617.6, 617.8 and 617.9 and ICD-10 codes N80.6, N80.8 and N80.9, respectively, for other endometriosis including endometriosis in a cutaneous scar, other specified sites (such as bladder, lung or umbilicus), and ‘other site unspecified’. Adenomyosis was defined by ICD-9 code 617.0 and ICD-10 code N80.0. Additionally, in sensitivity analyses, we used laparoscopy codes (ICD-9, 54.21; CPT4 codes 49320–58674; Current Procedural Terminology, 4th edition) to identify women who had been diagnosed with endometriosis using this gold standard of visualised disease [21].

Outcome

Diagnoses of type 2 diabetes were identified using ICD-9 codes 250.x0 and 250.x2 and ICD-10 codes E11*. In ICD-9, the fifth digits ‘0’ and ‘2’ denote type 2 or unspecified diabetes without and with complications, respectively, thus excluding type 1 diabetes diagnoses. ICD-10 code E11* corresponds specifically to type 2 diabetes mellitus, including related subcodes.

Covariates

Demographic data, obtained from the UPDB, included birth month and year, birth state (Utah/other state, or born outside the USA), race (American Indian or Alaska Native, Asian, Black or African American, Native Hawaiian or other Pacific Islander, White, multiple races), ethnicity (Hispanic/non-Hispanic) baseline education (less than high school, high school graduate, some college, college graduate and post college), BMI (derived from linked driver licence records), and residence (urban, rural, frontier) [22]. Specifically, frontier areas were defined as highly rural and geographically isolated locations characterised by low population density and substantial travel time to urban centres and essential services, consistent with national and rural health definitions). The Utah Department of Health and Human Services facility provided information on infertility diagnoses and gynaecological surgery (e.g. laparoscopy, laparotomy, hysterectomy, oophorectomy). Birth and foetal death records were used to obtain information on captured parity and pregnancy complications. Death month and year, and last month and year known to be a resident of Utah, were captured from UPDB vital records for censoring purposes.

Statistical analysis

Women contributed unexposed person-years from cohort entry until their first qualifying endometriosis diagnosis, after which all subsequent person-years were classified as exposed. This time-varying approach avoids the immortal time bias that would arise if women were classified based solely on ever/never endometriosis status at baseline [20]. Follow-up duration accrued from cohort entry until the date of type 2 diabetes diagnosis, death, permanent out-migration from Utah (loss to follow-up), or the end of follow-up (31 December 2021). For time-to-event models, women without type 2 diabetes were administratively censored at their last known date in Utah or 31 December 2021, whichever came first.

Descriptive statistics were used to report characteristics of populations, comparing women with and without endometriosis. Cox proportional hazard models, with calendar time as the time scale, were used to estimate HRs and 95% CIs to assess the risk of developing type 2 diabetes in women with endometriosis compared with those without. We selected calendar time as the time scale to allow us to account for temporal trends in both endometriosis ascertainment and type 2 diabetes diagnosis, which is particularly important in the UPDB given changes in diagnostic practices and healthcare utilisation over time. Consistent with current methodological guidance, departures from the proportional hazards model were not considered a violation of the validity of our model, as Cox model HRs represent weighted averages of time-varying effect after follow-up [23].

Separate models were fitted for SE, DE, OE, DE and OE combined (as both include deep lesions), adenomyosis and ‘other site’ endometriosis, with no endometriosis as the reference group. Models were adjusted a priori for potential confounders including birth year, birth state (born in Utah and not born in Utah), race (American Indian or Alaska Native, Asian, Native Hawaiian or other Pacific Islander, Black or African American, White, multiple races), ethnicity (Hispanic or non-Hispanic), and age and BMI (kg/m2) at time of cohort entry (<18.5, 18.5–24.9, 30–34.9, 35–39.9, >40).

Based on prior findings, we tested for effect modification by mean menopausal age (<50 vs ≥50 years), BMI (<30 vs ≥30 kg/m2) and history of GDM (yes or no), using stratified models and Wald tests for interaction.

Given the heterogeneity of ICD coding used to define ‘other site’ endometriosis and the potential for diagnostic or healthcare utilisation-related bias, we conducted sensitivity analyses that further refined the ‘other site’ subtype into unspecified and specified anatomical categories. ‘Unspecified other site’ endometriosis was defined using non-specific ICD codes (ICD-9 code 617.9 and ICD-10 code N80.9 or N80.9x), which may reflect incomplete anatomical documentation rather than true disease heterogeneity. ‘Specified other site’ endometriosis was defined by the presence of at least one diagnosis associated with an anatomically specified ICD code (ICD-9 codes 617.6, 617.8 or 617.95 and ICD-10 codes N80.6, N80.8 or N80.Cx); approximately 20% of all cases of ‘other site’ endometriosis were classified as occurring at a specified site.

We conducted several sensitivity analyses to evaluate the robustness of our findings. To assess the impact of the time scale, models were refitted using age as the underlying time scale, with additional adjustment for study start year, with all other covariates unchanged. To address potential delays in endometriosis diagnosis, we conducted analyses assuming that diagnosis occurred 6, 8 and 10 years earlier. We further evaluated the influence of prevalent type 2 diabetes and exposure classification at a time point close to cohort entry by restricting follow-up to 1 January 1998 or later, excluding individuals with type 2 diabetes diagnosed prior to this date while retaining those with endometriosis diagnoses before 1998 and classifying them as exposed.

To assess the potential role of reproductive factors and missing data, we conducted additional analyses adjusting for parity and infertility, performed complete-case analyses restricted to individuals with non-missing covariate data, and incorporated updated electronic health record-derived measures of education, race/ethnicity, smoking and geographic residence. Finally, to evaluate the impact of exposure classification based on surgical confirmation by laparoscopy, we conducted a sensitivity analysis restricted to women with a recorded laparoscopy, for whom endometriosis diagnoses were more likely to reflect surgically evaluated disease; models were otherwise specified identically to the primary analysis.