Machine Learning Uncovers Key Comorbidities Linked to Premature Death in IBD Patients
IBD and Premature Mortality
Inflammatory Bowel Disease (IBD) encompasses chronic gastrointestinal disorders such as Crohn’s disease and ulcerative colitis. By 2035, an estimated 470,000 Canadians are expected to be diagnosed with IBD.
Individuals living with IBD face a higher likelihood of developing chronic health conditions, which in turn increases their risk of premature death compared to the general population. Understanding which comorbidities contribute most to this elevated risk is essential for improving patient outcomes.
Study Overview
This population-based, retrospective cohort study leveraged Ontario’s administrative health data to examine premature mortality risks among IBD patients through machine learning (ML) techniques. Researchers explored associations between IBD, coexisting chronic conditions, and early mortality across three distinct analytical tasks.
-
Task 1 predicted premature mortality without factoring in chronic conditions that emerged later in life, such as dementia, chronic coronary syndrome, and congestive heart failure.
-
Tasks 2 and 3 examined the influence of chronic conditions on early death. Task 3 specifically focused on young age at diagnosis for conditions like mood disorders, hypertension, arthritis, and mental health disorders.
Three ML models—logistic regression, random forest, and Extreme Gradient Boosting (XGBoost)—were implemented for tasks 1 and 2. Task 3 utilized the XGBoost model (XGB3), comprising a total of seven predictive models.
The study included individuals diagnosed with IBD who resided in Ontario and passed away between January 2010 and January 2020. The Ontario Crohn’s and Colitis Cohort provided the patient data. Researchers identified chronic conditions such as diabetes, asthma, chronic obstructive pulmonary disease (COPD), hypertension, cardiac arrhythmia, mental health disorders, and rheumatoid arthritis using validated health data algorithms.
Key Findings
The study analyzed data from 9,278 IBD patients, of whom 49.3% were female. Premature mortality was observed in 47.2% of cases. The most common comorbidities at age 60 included arthritis, hypertension, and mood disorders, while conditions frequently observed at death were cancer, renal failure, hypertension, and arthritis.
All seven machine learning models performed well in predicting premature mortality, with the best results observed in tasks 2 and 3. These tasks included individuals diagnosed with chronic conditions before turning 60.
The most influential predictors of premature death varied depending on the model used. Although the models exhibited comparable predictive accuracy, their interpretations of data patterns differed.
Task 3 models had the lowest prediction error rate (11%). False positives were most commonly associated with hypertension (56%), mood disorders (53%), and arthritis (58%). Conversely, false negatives were more likely in patients with fewer chronic conditions.
Across IBD subtypes and genders, the models delivered consistent predictions. The highest-performing model was one that incorporated the age at which each chronic condition was diagnosed, specifically for conditions emerging before age 60.