Machine learning identifies key comorbidities predicting premature death in IBD patients

By Dr. Priyom Bose, Ph.D.Reviewed by Benedette Cuffari, M.Sc.Mar 28 2025

New machine learning study reveals how early-life chronic conditions like arthritis, mood disorders, and hypertension may drive premature death in people with IBD—highlighting critical opportunities for earlier intervention.

Study: Machine learning prediction of premature death from multimorbidity among people with inflammatory bowel disease: a population-based retrospective cohort study. Image Credit: Apichatn / Shutterstock.com

A recent Canadian Medical Association Journal study uses machine learning models to investigate patterns between multimorbidity and premature deaths among decedents with irritable bowel disease (IBD).

IBD and premature death

IBD is an umbrella term for a group of chronic inflammatory disorders, including Crohn's disease and ulcerative colitis, that affect the gastrointestinal tract. Researchers predict that approximately 470,000 Canadian people will develop IBD by 2035.

Individuals diagnosed with IBD are more likely to develop chronic health conditions and, as a result, experience premature death as compared to other individuals. Thus, it is crucial to identify which comorbidities are responsible for the increased risk of premature death among IBD patients.

About the study

The current population-based, retrospective cohort study used Ontario administrative health data to predict premature death rates among IBD patients using machine learning (ML) techniques based on three tasks. Researchers also identified patterns between non-IBD chronic conditions and premature death among decedents with IBD.

Task 1 predicted premature death without considering chronic conditions that developed later in life, such as congestive heart failure, dementia, and chronic coronary syndrome. Comparatively, tasks 2 and 3 considered the association between the presence of non-IBD chronic conditions and premature death. Task 3 considered young age at diagnosis for mood disorders, male sex, hypertension, osteo- and other arthritis types, and mental health disorders.

For tasks 1 and 2, logistic regression, random forest, and Extreme Gradient Boosting (XGBoost) models were developed. The XGBoost model (XGB3) was used for task 3, which comprises a total of seven models.

The current study considered data from individuals who resided in Ontario, were diagnosed with IBD, and died between January 1, 2010, and January 31, 2020. The Ontario Crohn’s and Colitis Cohort study data was used to identify patients diagnosed with IBD.

Using validated algorithms for health administrative data, individuals with a history of chronic conditions like congestive heart failure, asthma, diabetes, chronic obstructive pulmonary disease, hypertension, cardiac arrhythmia, rheumatoid arthritis, and mental health disorders were identified.

Study findings

A total of 9,278 decedents with IBD were included in the current study, 49.3% of whom were female and 47.2% of whom experienced premature death. The most prevalent comorbidities at sixty years of age were osteo- and other arthritis types, mood disorders, and hypertension. At death, frequently occurring conditions included osteo- and other arthritis types, hypertension, mood disorders, renal failure, and cancer.

All seven machine learning models demonstrated strong performance and calibration on testing data. Superior model performance was observed in tasks 2 and 3, both of which included individuals who were diagnosed with comorbid conditions diagnosed before 60 years of age were only included.

The strongest feature considered for predicting premature death varied across and within tasks. Thus, although all models exhibited similar prediction capabilities, their outcomes were based on different relationships within the data.

Models used in task 3 exhibited fewer prediction errors at a rate of 11%. The false positive prediction was associated with specific conditions including osteo- and other arthritis types (58%), hypertension (56%), and mood disorders (53%), whereas false negative errors occurred in individuals with fewer comorbidities.

Similar predictions were obtained across IBD subtypes and genders. For example, as compared to all models, one that incorporated age at diagnosis for each chronic condition developed at or before age 60 years exhibited the best performance.

Conclusions

ML models have the potential to accurately predict premature death associated with non-IBD comorbidities, particularly when these models were trained with early-life conditions. Furthermore, younger ages of diagnosis for mood disorders, hypertension, osteo and other arthritis types, and mental health disorders, as well as male sex, were also important features that can be used to predict premature death.

Our model helps dissect and capture patient heterogeneity, identifying areas where more targeted follow-up is needed to better understand their clinical importance and relation to IBD severity.”

To develop effective preventive care at the population level, additional multidisciplinary research is needed to elucidate how multimorbidity in IBD causes premature death.