Abstract 333P
Background
The Incidence of EOCRC rose abruptly starting in the mid-1990s. Inherited genes and inflammatory bowel disease (IBD), known risk factors, are not the causes of most EOCRC. We hypothesized that the increasing incidence may be an off-target effect of a medication (meds) not previously widely used in a genetically susceptible subgroup of young adults. To identify potential pharmaceutical agents as risk factors in EOCRC, we employed novel machine learning methods using a large Israeli electronic medical records (EMR) database, with digitized pharmacy records.
Methods
Using EMR from Maccabi Healthcare Services, an Israeli HMO with 2.6 million members, we identified 941 EOCRC cases (<50 years of age). Cases were density matched with up to 10 controls based on sociodemographic factors. IBD patients (Pts) were excluded. Complete meds dispensing history (OTC and prescription meds) was obtained for all participants, excluding drugs started within 2-years of EOCRC diagnosis to minimize risk of reverse causation. Gradient boosted decision trees coupled with Bayesian model optimization and a nested cross-validation design were used to sort through the very high-dimensional drug dispensing data to identify specific meds groups that were consistently linked with outcome while allowing for synergistic or antagonistic interactions between drugs.
Results
Of nearly 800 binary predictors, we identified several medication classes that were consistently (>50%) associated with EOCRC risk across independently trained models. Interactions between meds groups did not seem to substantially affect the risk. In our initial analysis, drug groups that were consistently associated with EOCRC, with increased odds ratio after covariate adjustment included: a beta blocker (OR 1.82 (CI 1.26-2.63, p<0.01), a hypoglycemic (OR 2.85, CI 1.04-7.82, P-0.04) and an anti-emetic (OR 3.37, CI 1.53-7.42 p<0.01); antibiotics were not consistently associated with increased risks.
Conclusions
Our initial analysis implies that EOCRC may be correlated with prior use of specific medications. Additional study of pharmaceutical agents as a driver in EOCRC and potential mechanisms for this off-target effect, are underway.
Clinical trial identification
Editorial acknowledgement
Legal entity responsible for the study
The authors.
Funding
Applebaum Foundation.
Disclosure
All authors have declared no conflicts of interest.