Leveraging patient-reported outcomes and serum markers to predict active ulcerative colitis: a multicenter cohort.

Background and aim: This study aims to develop a score system via noninvasive and reliable clinical tools for individuals to distinguish remission and active Ulcerative Colitis (UC). Methods: We performed a retrospective multicenter study collecting 173 patients in the training cohort and 124 patients in the validation cohort with UC. Then we assessed the relationship between patient-reported outcomes (PROs) and serum indicators with endoscopic disease activity (defined as UCEIS ≥1). Univariate and multivariate logistic regression analyses were performed, with a stepwise backward selection approach used to select significant variables. Two predictive models were ultimately developed based on PROs and serum biomarkers. The performance of the models was evaluated through ROC curves, and calibration was assessed using Spiegelhalter's Z-test. Results: A total of 173 and 124 patients were enrolled in the training and validation groups, respectively. Univariate and multivariate analyses revealed that stool frequency (SF), rectal bleeding (RB), CRP/TB, and PDW were significantly associated with endoscopic active UC. Two predictive models were developed, with SF (model A) and RB (model B) combined with CRP/TB and PDW, respectively. Both models demonstrated excellent discriminative ability for endoscopic activity, with area under the ROC curve (AUC) values of 0.906 (95% CI 0.863-0.949) and 0.899 (95% CI 0.855-0.943) in the training cohort. In the external validation cohort, the AUC values were 0.793 and 0.794, showing similar strong discriminative ability. In the Mayo Endoscopic subscore (MES) system, model A and model B exhibited AUC values of 0.894 and 0.884 in the training cohort, and 0.769 and 0.758 in the validation cohort. Subgroup analysis based on disease severity further validated the models' stability and reliability. Conclusion: The predictive models based on SF and RB developed in this study demonstrated good discriminative ability in predicting endoscopic activity in patients with UC. Both models performed well in the internal and external validation. Additional validation utilizing the MES and disease severity provided further evidence supporting the reliability and effectiveness of these models. These findings underscored the potential clinical utility of the SF- and RB-based models as valuable tools for predicting endoscopic disease activity in UC patients, which could facilitate more informed clinical decision-making and improve patient outcomes in the management of UC.