To the Editor: We read with interest the article by Xu et al in the International Journal of Surgery, “Plasma Proteomic Markers Predict Risk for Bowel Resection in Inflammatory Bowel Disease: A Retrospective Cohort Study”1. Using plasma proteomics from the UK Biobank Pharma Proteomics Project and a random survival forest (RSF) approach, the authors report that an inflammation-related protein signature predicts subsequent bowel resection in inflammatory bowel disease (IBD) better than conventional clinical markers. We commend the authors for tackling long-term surgical risk stratification with a scalable proteomics-based framework. To further strengthen interpretability, reproducibility, and clinical translation, we offer several constructive comments. First, the definition and directionality of the “predicted risk score” require clearer specification. The Methods describe transforming RSF-derived survival probabilities into a continuous score, with higher values indicating higher predicted resection risk and using the median to define high/low-risk groups for Kaplan–Meier analyses. However, the baseline table refers to the score as the “probability of remaining surgery-free,” while the reported medians (e.g., 2.70 vs 7.58) do not align with a 0–1 probability scale. In addition, the adjusted hazard ratio for the score in multivariable Cox regression is <1, which may appear inconsistent with the “higher score = higher risk” interpretation. These discrepancies may reflect the specific transformation (e.g., 1–S(t) vs alternative links), coding direction, or labeling. We suggest explicitly reporting the mathematical definition (transformation, range, and any standardization/log or sign changes) and verifying consistency across Kaplan–Meier grouping, table text, figure legends, and Cox model coding; harmonized terminology would prevent misinterpretation of the score’s clinical meaning. Second, the comparator clinical model may be overly parsimonious, which could overstate the incremental value of proteomics. The “traditional clinical marker model (Model 2)” includes only C-reactive protein, albumin, platelets, hemoglobin, and WBC, yielding a C-index of 0.543 versus 0.784 for the proteomic model. The authors also note that UK Biobank lacks key IBD variables (e.g., disease severity, location, and biologic therapy), suggesting residual confounding. In this context, a simplified clinical baseline may not represent the best-available model in routine practice. We therefore recommend constructing an expanded baseline model using available covariates beyond laboratory markers – such as age, sex, IBD subtype, disease duration, smoking status, and medication history (at least as proxies or sensitivity analyses) – and then reassessing proteomic incremental utility. Reporting ΔC-index, calibration across clinically relevant horizons, and net benefit via decision curve analysis (DCA) would further address concerns about “straw-man” comparators and better align performance gains with decision-making. Third, greater transparency in endpoint operationalization for bowel resection would strengthen interpretability. While the exclusion criteria reference prior ileal resection or colectomy, details on incident “bowel resection” ascertainment during follow-up (coding sources, event-date definition, elective vs emergency procedures, and differentiation between total colectomy in UC and segmental resections in CD) are not fully delineated in the main text. Given procedure heterogeneity and the potential influence of competing events, we suggest providing a code list and censoring rules, clearly defining time zero (e.g., blood draw/baseline assessment) and follow-up end, and considering competing-risk sensitivity analyses where feasible. Finally, decision-level reporting would improve clinical usability. Beyond discrimination and NRI/IDI, adoption requires clarity on actionable thresholds. We encourage reporting clinically interpretable cut-points (or stratification schemes) with sensitivity, specificity, and predictive values, alongside DCA to quantify net benefit and clarify how proteomic testing might inform surveillance intensity, treatment escalation, or trial enrichment. Overall, Xu et al provide promising evidence that plasma proteomics may enhance long-range prediction of bowel resection risk in IBD. Addressing the points above would further improve clarity, robustness, and transportability, supporting future external validation and prospective evaluation. Ethical approval Not applicable. Consent Not applicacble.
Zheng et al. (Mon,) studied this question.