Deep phenotyping of patient lived experience in functional bowel disorders using machine learning

Area of Science:

• Gastroenterology and Computational Biology
• Utilizing advanced machine learning and Bayesian generative graph frameworks for complex biological systems.

Background:

• Functional bowel disorders (FBDs) present significant heterogeneity, lacking definitive diagnostic markers or universally effective treatments.
• Current clinical management often relies on diagnostic labels that do not fully capture individual patient experiences.
• Understanding the intricate interplay of factors influencing FBD patient experiences requires sophisticated analytical approaches.

Purpose of the Study:

• To develop and apply a machine learning and Bayesian generative graph framework to elucidate the complex lived experiences of patients with functional bowel disorders.
• To identify key predictors of patient-reported health, quality of life, and treatment response within a large FBD cohort.
• To challenge traditional diagnostic-centric approaches in FBD management by exploring broader patient characteristics.

Main Methods:

• Employed machine learning models to assess the predictive power of 59 clinical factors on patient outcomes.
• Utilized Bayesian stochastic block models to map the network community structure of FBD patient heterogeneity.
• Analyzed a large cohort (n=1175) including demographics, diagnoses, symptoms, life impact, mental health, healthcare access, and treatment effectiveness.

Main Results:

• Machine models identified life impact, mental well-being, employment status, and age as primary predictors of patient-reported health and quality of life, outperforming diagnostic group or symptom severity.
• Predictive accuracies included: personal health rating (R² 0.35), anxiety/depression severity (R² 0.54), employment status (balanced accuracy 96%), healthcare attendance (R² 0.71), and treatment effectiveness (R² 0.08-0.41).
• Observed a stratification in treatment response, where patients responding to one treatment were more likely to respond to others, indicating distinct patient subgroups.

Conclusions:

• Clinical assessment for FBDs should prioritize a holistic view, focusing on the broader life impact, mental health, and employment status over strict diagnostic classification.
• The identified predictors have significant implications for refining clinical practice and designing more effective clinical trials for FBDs.
• Further research is warranted to explore the stratification of treatment response and resistance in functional bowel disorders.