(142) Diabetes and IBD: Utilization of High Dimensional Propensity Score Methods for Screening and Balancing Risk Factors Between Type 2 Diabetes and Non-Diabetes Population in Risk of IBD Assessment

Publication Number: 473

Background: Recent findings on high gene expression rate of the insulin sensitivity regulator in adipose tissue in Chinese Ulcerative Colitis(UC) patients suggested an association between diabetes and Inflammatory bowel disease (IBD).

Objectives: We tested if T2DM is a major risk factor in developing Crohn’s Disease (CD) or Ulcerative Colitis (UC) when balancing other identified risk factors by high dimensional-propensity score (HD-PS) method.

Methods: A retrospective negative-control cohort of new T2DM and non-diabetes (non-Diab) patients were created using Optum claims data during 2010- Q3, 2020. The index dates were the first T2DM diagnosis date in T2DM or the first of any medical records date in non-Diab patients. Besides demographics or predefined covariates were assessed, the top confounders to comparison groups associating the outcome were selected through Schelesselman bias score algorithm. Hazard ratio (HR) with 95% confidence intervals(CI) on the risk of CD or UC were obtained using Cox regression either adjusting or balancing covariates by PS matching(PSM), PS weightings(PSW) methods, which included average treated effect(ATE), average treated effect on the treated(ATT), or overlapped(OVLP) and with or without trimming options.

Results: There were 206,036 and 1,821,862 patients in T2DM vs. non-Diab cohorts. Unbalanced frequencies in demographics and disease covariates were found in prebalanced cohorts. Mean and standard deviation of Index age were 60.31 (17.07) vs. 35.62 (20.65). All disease covariates, including known or HD newly identified IBD conditions, were more frequent in T2DM than non-Diab patients. Some newly identified conditions were fatigue & malaise, thyroid, lipid metabolism, chronic obstructive pulmonary disease.
By balancing methods, the standardized differences(STD) in prebalanced, matched, or weighted cohorts indicated the OVLP PSW method demonstrated the best-balance with STD=0 for all covariates.
The HR(95% CI) of CD estimated by OVLP PSW: 0.9 (0.8, 1.0) was close to adjusted HR: 1.0 (0.9, 1.1), and 1.0 ( 0.9 ,1.1) by PSM. For UC, HRs are 1.4 (1.2,1.5), 1.4 (1.3,1.5), 1.5 (1.4, 1.5), respectively.

Conclusions: HD Schelesselman bias score algorithm could powerfully screen possible risk factors in comparative study. OVLP PSW is the best balancing method. Cox model with OVLP PSW provides reliable and similar result as adjustment or PSM. T2DM is found to be a major confounder to UC, but not CD.