Clarifying Composite Outcomes and Baseline Prognostic Analysis in Ulcerative Colitis

To the Editor,

1.

We thank Ahmed for raising important methodological considerations regarding our prospective multicenter inception cohort study of newly diagnosed moderate‐to‐severe ulcerative colitis (UC) [1]. We wish to clarify the intended interpretation of the composite outcome and the purpose of the baseline prognostic analysis.

The disabling disease course was prespecified as a pragmatic measure of real‐world clinical burden; it was not intended to imply that each component represented equivalent biological severity. As further clarified in our recent response to related correspondence, the original composite was intended to capture the broad clinical burden experienced by patients rather than serve as a direct measure of intrinsic disease severity [2]. We agree that treatment‐related components can be influenced by practice patterns and healthcare‐system factors. As stated in our Discussion, Korean reimbursement criteria influenced access to advanced therapies but were not independent of disease activity because reimbursement required an inadequate response to or intolerance of conventional therapies. Although these requirements formalized the treatment sequence, they broadly reflected the step‐up treatment paradigm that prevailed internationally when the cohort was established and should not be interpreted as uniquely Korean restrictions unrelated to disease activity [3]. Treatment escalation therefore reflected disease activity and treatment response within a defined access framework, not reimbursement policy in isolation.

This issue was specifically examined in the published article using prespecified sensitivity analyses that excluded treatment‐related components [1]. Excluding step‐up to thiopurines, step‐up to advanced therapies, and switching or swapping of advanced therapies yielded a 5‐year cumulative incidence of 39.1%. After additionally excluding repeated systemic corticosteroid treatment, the most stringent definition, which was least dependent on treatment‐related components, yielded a 5‐year cumulative incidence of 18.8%, driven predominantly by UC‐related hospitalization. The association between trajectory clusters and a disabling disease course remained significant under both alternative definitions. Thus, the original composite should be interpreted as a broad measure of clinical burden, although the results under alternative definitions clarify its interpretation and potential applicability across healthcare settings.

The proposed time‐varying treatment analysis would address a different question from that examined in our baseline Cox model. Our Cox analysis assessed associations between baseline characteristics and subsequent outcomes; it was not designed to estimate causal treatment effects, the direct effect of baseline white blood cell count independent of subsequent treatment decisions, or the performance of a clinical prediction model. Treatment initiated during follow‐up is a post‐baseline response to evolving disease activity and may lie on the pathway between baseline severity and later outcomes. Adding treatment as a conventional time‐dependent covariate would therefore change the estimand and would not necessarily reduce bias, because evolving disease activity may both influence subsequent treatment and predict outcomes. A causal analysis of disease modification would require a clearly defined causal estimand, longitudinal data on disease activity and treatment exposure, and methods appropriate for time‐varying confounding, such as marginal structural models [4].

We agree that a baseline‐only model cannot fully capture the evolving disease course. This limitation and the corresponding future directions were already acknowledged in the published article: Limited early biologic use precluded formal evaluation of its potential disease‐modifying effect, and no early partial Mayo score cutoff or prediction rule was proposed because the study was not designed for prediction‐model development or validation [1]. The Cox analysis was intended to explore baseline factors measurable at diagnosis that might inform future risk‐stratification research. The association between baseline white blood cell count and a disabling disease course was modest and exploratory and was not presented as an immediately actionable prognostic marker. Dynamic prognostic modeling in this setting would require serial disease‐activity measures and time‐updated treatment data. Any resulting model would also require formal assessment of predictive performance and external validation [5]. By contrast, our trajectory analysis used repeated partial Mayo score assessments to characterize longitudinal clinical phenotypes and was not presented as a dynamic prediction model.

We appreciate the opportunity to clarify these distinctions. Future work incorporating component‐specific outcomes, serial disease‐activity measures, and appropriately designed analyses of evolving treatment strategies should help refine these findings.

Funding

The original MOSAIK cohort study was funded and supported by Johnson & Johnson Korea.

Ethics Statement

The author has nothing to report.

Conflicts of Interest

The author declares no conflicts of interest.

Data Availability Statement

Data sharing is not applicable to this article because no datasets were generated or analyzed for this correspondence.