Clarification and sensitivity analysis of peri-operative metabolomic alterations in colorectal cancer: a focused query.
Dear Editor, We read with interest the Publish Ahead of Print article by Liu and colleagues examining whether peri-operative serum metabolomic alterations are associated with post-operative complications and hospital stay in colorectal cancer (CRC) surgery[1]. In this observational cohort of 243 patients enrolled from January to December 2024, paired serum samples were obtained 1 day before surgery and again on the first day after surgery, quantified using an advance IVDr high-throughput nuclear magnetic resonance platform, and analyzed alongside peri-operative clinical indicators. The authors evaluated both pre-operative metabolite levels and post-/pre-operative metabolite ratios. They report that integrating metabolic variables with clinical indicators improved discrimination for in-hospital major complications: the AUC increased from 0.749 (clinical indicators alone) to 0.776 when pre-operative metabolites were added, and from 0.727 to 0.805 when metabolite ratios were added. For hospital stay, the combined model increased the explained variance to R2 0.137 (vs 0.102 for clinical indicators alone). The discussion also highlights that recovery-related physiological alterations may be detectable as early as 1 day after surgery, suggesting a potential window for early post-operative risk stratification. These results are encouraging and clinically thought-provoking. One time-related point, however, could affect how readers interpret what the post-operative metabolomic signal represents. Because the post-operative blood draw was performed on the first day after surgery, it is plausible that a subset of major complications – or the earliest clinical changes that prompt diagnostic work-up and treatment – may have emerged before the sampling time. If so, post-operative day 1 metabolite ratios could partly act as early markers of complications already evolving, rather than predictors of events that occur later in the hospital course. This distinction matters clinically: an “early warning” marker measured during immediate recovery can be valuable, but it is conceptually different from a model that is framed as predicting future complications from a baseline time point. To keep this issue answerable within the existing dataset, we would be grateful if the authors could clarify two closely linked items, consistent with guidance for reporting prediction model studies[2]. First, what was the typical interval between the end of surgery and the post-operative blood sampling (for example, median [IQR] hours), and was sampling generally standardized (e.g., morning draws) or dependent on workflow and patient stability? Second, among patients who developed major complications, on which post-operative day – or at what time from surgery – was the first complication first documented or treated? Even a brief summary of the proportion of events first recognized before versus after the post-operative blood draw, and the median time to first event, would help readers judge the effective lead time provided by the metabolomic ratios. In addition, if these timestamps are available in the routinely collected records already analyzed, a timing-restricted sensitivity analysis could strengthen interpretability without requiring any new measurements – for example, repeating the ROC analysis after excluding complications first documented before the post-operative sample, or restricting the endpoint to complications first identified after a prespecified window such as 24 hours. Reporting discrimination separately for “early” versus “later” events could also clarify whether the model’s strength lies in capturing immediate post-operative physiology or in forecasting subsequent deterioration. Such evaluation may facilitate comparison with other peri-operative metabolomics studies in CRC surgery that likewise leveraged early post-operative sampling to identify patients at risk for severe complications[3]. Notwithstanding this query, we commend Liu and colleagues for assembling one of the larger peri-operative metabolomics datasets in CRC surgery and for using metabolite ratios as a pragmatic way to capture individual metabolic responses to surgical stress. We hope that clarification of event timing – and, if feasible, a timing-restricted sensitivity analysis – will further strengthen clinical interpretability and support translation of this promising approach.
Read more