Joint Longitudinal-Survival Modelling of Patient-Reported Gastrointestinal Symptom Trajectories and Treatment Discontinuation in Irritable Bowel Syndrome: A Prospective Cohort Study from the Canadian Gut Project

Background: Irritable bowel syndrome (IBS) is characterized by heterogeneous symptom trajectories and high treatment discontinuation rates. Traditional analyses examine longitudinal outcomes and time-to-event endpoints separately, potentially missing informative dropout and the association between symptom dynamics and treatment persistence. Objective: To jointly model patient-reported IBS symptom trajectories and time-to-treatment discontinuation using shared random effects, characterizing the association between individual symptom dynamics and treatment persistence in a large Canadian prospective cohort. Methods: We analyzed 2,847 adults with Rome IV diagnosed IBS enrolled in the Canadian Gut Project (2018 to 2024) across 14 gastroenterology centres in Alberta, British Columbia, and Ontario. The longitudinal submodel used linear mixed-effects regression for the IBS Severity Scoring System (IBS-SSS) measured at baseline and months 3, 6, 12, 18, and 24. The survival submodel used a Weibull proportional hazards model for time-to-treatment discontinuation. The joint model linked both processes through shared random effects (random intercept and slope), estimated via maximum likelihood with adaptive Gauss-Hermite quadrature (15 nodes). We conducted sensitivity analyses using Bayesian estimation, alternative association structures (current value, time-dependent slopes), and multiple imputation for intermittent missingness. Results: Mean baseline IBS-SSS was 298.4 (SD 72.1). Over 24 months, 1,042 participants (36.6%) discontinued treatment. The longitudinal submodel revealed a mean IBS-SSS decline of -8.7 points/month (95% CI: -10.2, -7.1) with substantial between-person heterogeneity in both intercepts (STD = 4,218.3) and slopes (STD = 12.4). The association parameter linking the shared random intercept to the hazard of discontinuation was = 0.0034 (95% CI: 0.0021, 0.0047; p < 0.001), indicating that each 10-point increase in individual-specific baseline severity increased the hazard of discontinuation by 3.5%. The shared slope association parameter was 2 = -0.187 (95% CI: -0.264, -0.110; p < 0.001), demonstrating that individuals with steeper symptom improvement had lower discontinuation hazards. IBS-D subtype (HR = 1.41; 95% CI: 1.18, 1.69), concurrent anxiety (HR = 1.28; 95% CI: 1.09, 1.50), and social media health information use (HR = 0.82; 95% CI: 0.71, 0.95) were significant predictors in the survival submodel. Conclusion: Joint longitudinal-survival modelling reveals that IBS symptom trajectories and treatment discontinuation are dynamically linked through individual-level latent processes. Higher baseline severity and slower improvement trajectories significantly predict earlier discontinuation. These findings support personalized treatment monitoring approaches that use real-time symptom trajectory data to identify patients at risk of discontinuation.