Letter: Enhancing Predictive Models for Paediatric Ulcerative Colitis—Addressing Socioeconomic, Environmental and Clinical Factors. Authors' Reply

Abstract

Dr. Luo from Liangshan (China) has provided valuable comments on improving predictive models for poor outcomes in paediatric ulcerative colitis (pUC) [1].

We included predictors identified by the Paediatric Inflammatory Bowel Disease (PIBD) Ahead Program [2] when available in the CEDATA registry of the German-speaking Society for Paediatric Gastroenterology and Nutrition (GPGE). Although all baseline variables were collected according to the Porto criteria [3], not all criteria of interest for predicting poor outcomes in pUC are included in the registry [4].

Sex, family history and initial disease severity were included in our study [5]. Another study has shown that initial treatment regimens (intensified infliximab induction) may improve disease outcome in pUC [6]. We also included time since diagnosis as a possible predictor [5], as access to healthcare had no significant effect on any of the outcomes assessed. However, socioeconomic factors such as socioeconomic status are not assessed in the CEDATA registry [4]. Nevertheless, a meta-analysis on the association of education, income differences and ethnicity with hospitalisation in IBD were heterogeneous and pointed in different directions [7]. Problems with treatment adherence was reported in 61 (8.2%) of our patients; hence, overall adherence in our cohort was high. We performed Cox regressions as described in the original publication [5] and a reduced treatment adherence was associated with an increased need for systemic steroids, when controlling for age and sex (b = 0.643; SE = 0.185; HR = 1.905 95% CI 1.325–2.734; p < 0.001). None of the other outcomes showed an association with treatment adherence. Many of the other important factors mentioned by Dr. Luo (e.g., information on genetics and gut microbiome) are not (yet) included in the CEDATA registry [4]. Regional or institutional differences as well as environmental factors (e.g., diet, air pollution, urbanisation) are also not available in the fully anonymised CEDATA data set [4].

Multiple imputation is an important approach to deal with missing data and has been shown to correct for underestimation of remission rates in a large registry study of paediatric IBD [8]. However, to perform robust multiple imputation, missing data must be completely at random, and factors predictive of missing data must be included in the imputation model [9]. We did not assume missing completely at random: for example, since 2013 data can be submitted online, and some new clinical variables were added [10]. We chose a conservative statistical approach to improve robustness, with a minimum of 15 events per predictor in Cox regression and an additional alpha correction method [5].

Clearly, the clinical course of many paediatric patients with UC is far from optimal. Therefore, the trend towards personalised medicine to optimise treatment requires further studies with multiple subgroups as well as large sample sizes to provide models with independent predictors to facilitate decision-making. This gap could be filled by artificial intelligence-based analyses as an advanced data handling technique.

Merle Claßen: writing – original draft, methodology. Benjamin Schiller: writing – review and editing. Jan Däbritz: writing – review and editing, conceptualization, supervision.

The authors’ declarations of personal and financial interests are unchanged from those in the original article [5].

This article is linked to Claßen et al papers. To view these articles, visit https://doi.org/10.1111/apt.18262 and https://doi.org/10.1111/apt.18327.