Abstract / Summary
Background: Real-world evidence on food-as-medicine programs for irritable bowel syndrome (IBS) is scarce, and single-arm evaluations overstate improvement because of regression to the mean (RTM). We estimated change in the IBS Symptom Severity Scale (IBS-SSS, 0-500) in a multi-omics personalized digital health food-as-medicine program, unadjusted and RTM-corrected. Methods: Retrospective single-arm cohort. Of 15,933 members with a valid baseline, 2,254 had a follow-up reading and 1,683 one at least 14 days after baseline. Change at 4, 12 (primary) and 26 weeks was estimated by baseline severity stratum, unadjusted and RTM-corrected by analysis of covariance. A responder fell by at least 50 points (the minimal clinically important difference benchmark). Sensitivity analyses covered equity, inverse-probability weighting and tipping-point analyses for members without a reading in each landmark window, a negative-control stratum and medication. The association of weight loss and dietary nutrient density with symptom change was evaluated within and between members. Results: At 12 weeks (n = 430) IBS-SSS fell by 43.88 points unadjusted (95% confidence interval [CI] -52.44 to -35.32; p < 0.001) and by 47.73 RTM-corrected (-55.48 to -39.98; p < 0.001); 720 of 1,339 symptomatic members (53.8%; 95% CI 51.1 to 56.4) fell by at least 50 points. Larger improvement in more severe members persisted after RTM correction (severe stratum, n = 37: unadjusted -127.43, RTM-corrected -60.71, 95% CI -97.75 to -23.68, p = 0.001). A higher meal nutrient density was associated with lower IBS-SSS within a member's own trajectory (p = 0.018) while weight change was not (p = 0.361). Improvement was similar across groups in spite of baseline differences by age and sex: RTM-corrected 12-week change did not differ by age (p = 0.458), body-mass index (p = 0.328), sex (p = 0.749), income (p = 0.486) or ethnicity (p = 0.170). Estimates were robust to weighting for members without a reading in the landmark window (RTM-corrected estimates moved by less than 2 points), and the change of those members would have had to be 39 to 73 points worse than observed before the mean change reached zero. Baseline medication use and starting a GLP-1 receptor agonist did not explain symptom changes. Conclusions: IBS-SSS fell significantly among program participants, with steeper improvement in the more severe group, even after RTM correction. The improvement was comparable across socioeconomic characteristics, age and sex and robust to the confounders and missing-data assumptions evaluated. These associations from a self-selected single-arm cohort warrant replication in independent studies.