Understanding Fatigue in Inflammatory Bowel Disease through Machine Learning
Fatigue is a debilitating symptom that significantly impacts the lives of individuals with inflammatory bowel disease (IBD). Affecting over 50% of those diagnosed, the prevalence of fatigue in IBD mirrors that of many common immune-mediated inflammatory diseases (IMIDs). Despite its widespread nature, the mechanistic basis of fatigue remains elusive, largely due to the clinical heterogeneity and multifactorial origins of this complex symptom.
Methods and Analysis
To tackle this challenge, researchers have employed a machine learning (ML) conceptual framework, utilizing one of the largest sets of prospectively collected, real-world patient-reported outcomes (PROs) from three concurrent cohorts spanning from 2020 to the present. This dataset includes 2970 responses from 2290 participants across the UK and internationally, with a subset of non-IBD controls and 100 rows of clinical metadata. Notably, a patient engagement group conducted a thematic analysis of this dataset, identifying fatigue as a critical research priority (www.musicstudy.uk).
Key Findings
Researchers have systematically defined the fatigue threshold as a primary endpoint, specifically looking at patients experiencing ≥10/14 fatigue days. Among 1604 patients (1151 responses during active disease and 1061 during remission), the mean fatigue days were 14 and 7, respectively, with statistical significance (p < 0.001). This data informed the development of the ML approach, incorporating routinely available clinical data for population-level analysis.
Seven different ML methods were employed, with external validation across three cohorts in the United Kingdom, Spain, and Australia (n=252). The use of Shapley Additive Explanations (SHAP) analysis allowed for the disaggregation of clinical heterogeneity and examination of clinical predictive factors at the individual level. Through this process, researchers identified five distinct clusters of fatigue patients, including a subgroup with lower fatigue levels.
Conclusions
The study provides a robust ML “roadmap” for predicting and deconstructing fatigue in IBD and potentially other IMIDs. This approach enables a patient-level analysis that goes beyond symptom-based classification, integrating deep molecular data. Such advancements represent a significant step towards developing clinical scientific artificial intelligence models with immediate applications in stratifying patients for human experimental studies. This could lead to improved identification of patterns associated with fatigue at an individual level, ultimately enhancing patient care and outcomes.
For further details, you can access the full study Here.
“`
