DOP58 predictors of complicated disease course in CD in an administrative database: a nationwide study from the epi-IIRN

Abstract Background Several studies have proposed models to predict disease outcomes in Crohn’s disease (CD), but with limited accuracy, often due to small sample size. We aimed to use a large nationwide cohort to explore predictors of disease course in CD. Methods Data of patients diagnosed with CD in the epi-IIRN cohort 2005-2020 were retrieved from the four Israeli Health-Maintenance-Organizations covering 98% of the population. The following potential predictors were explored: demographic data, laboratory results, induction medications, change in medications during the induction period, extra-intestinal manifestations, perianal disease, and diagnostic delay. The primary outcome was complicated disease course defined as CD-related surgery, steroid-dependency, or need for more than one biologic class. Hierarchical clustering categorized disease severity at diagnosis based on available laboratory results into five groups of disease severity (minimal, mild, moderate, severe, and extreme). A Cox proportional hazard model with Bonferroni correction to adjust for multiple comparisons was sought to assess the relationship between predictors and time to complicated outcome. A nomogram was created by conversion of the hazard ratios from the model to points on a scale between 0 to 100. Results A total of 19,263 patients with 140,354 person-years of follow-up were included, of whom 6,874 (36%) had complicated disease course. In a Cox multivariable regression model, complicated disease course was predicted by diagnostic delay (HR 1.2 [95%CI 1.09-1.33]), induction therapy with biologics (HR 2.8 [2.21-3.64]), severity of laboratory tests prior to diagnosis (HR 1.6 [1.28-1.99]) and perianal disease (HR 1.5 [95%CI 1.4-1.6]; Figure 1). Specifically in the laboratory tests, there was a gradual increase in rates of patients who had complicated disease course amongst the disease severity clusters (p<0.001; Figure 2). These variables formed a model to predict time to complicated disease, with fair accuracy during the first year from diagnosis (AUC 0.69 [95%CI 0.65-0.73]; Figure 3). Conclusion In this nationwide cohort, complicated disease course was apparent in a third of patients and predicted by diagnostic delay, serum laboratory tests, perianal disease and type of induction therapy. Our results also highlight the ability of clustering analyses to define clinically meaningful categories in administrative database of CD.