1. Overall assessment Ethics approval, study registration, and funding disclosure are complete, and the study generally complies with international ethical standards for medical research. The work is based on a 499-patient multi-centre retrospective cohort and uses OS together with 3- and 5-year cumulative survival rates as primary endpoints; Kaplan-Meier survival analysis and Cox proportional-hazards modelling are used to evaluate prognostic factors, and the overall analytical framework is consistent with common analytical approaches in oncology prognostic research. Statistical procedures, including survival time definition, censoring handling, and baseline comparison, are reported relatively completely, and the figures and baseline characteristics are presented clearly. In the context of relatively limited real-world long-term survival evidence on ICWM in CRLM, this study provides a useful foundation for future research. Although the research question is clinically valuable and the cohort size is reasonable, several methodological limitations may affect causal interpretation; the findings are therefore more appropriately regarded as hypothesis-generating evidence. 2. Major methodological concerns (a) Immortal time bias The study defines cohort A by "continuous TCM >=3 months" and uses the date of initial diagnosis as the OS time origin. This combination of exposure definition and survival time origin may introduce immortal time bias, because patients have to survive long enough to qualify for cohort A. The reclassification of some patients from cohort B to cohort A during follow-up also points to a potential time-dependent exposure issue. A landmark analysis (e.g., a 3- or 6-month landmark) or modelling TCM as a time-dependent covariate would help reduce this risk of bias. (b) Residual confounding The grouping is not randomized, and the choice of TCM may be related to factors such as adherence, socioeconomic status, family support, and treatment tolerability. These unmeasured variables may also influence survival, and Cox regression alone may not fully address residual confounding. Propensity-score-based approaches (e.g., PSM or IPTW) could be considered to improve covariate balance. If these unmeasured data cannot be retrospectively extracted, this should be explicitly listed as a study limitation in the Discussion. (c) Exposure heterogeneity and treatment-intensity imbalance The TCM exposure in cohort A includes individualized decoctions together with several patent medicines (Anteke, Pingxiao, Compound Cantharidin, Xihuang, Weimaining). These patent medicines may differ in mechanism and potential efficacy, and combining them under a single “TCM exposure” may weaken interpretability. Additional variation across treatment stages may further increase within-exposure heterogeneity. In addition, the manuscript states only that patients received "best available treatment according to guidelines," without reporting treatment-intensity indicators such as chemotherapy lines, biologic agent use rate, or relative dose intensity (RDI). Stratified sensitivity analyses by patent-medicine category or by decoction-only exposure, together with reporting of systemic treatment intensity, would be useful additions. 3. Statistical and reporting considerations Covariate adjustment in the Cox model The multivariable Cox model includes only four covariates: primary tumour site, RAS/BRAF status, local treatment of liver metastases, and cohort assignment. For CRLM survival analysis, prognostic models generally need to cover key dimensions such as patient status, tumour biology, disease burden, and treatment intensity in a relatively comprehensive way. The variables currently included are concentrated on selected treatment factors and molecular features, while several well-recognized prognostic factors have not yet been entered into the model. For example, age differs significantly between the two groups (62.47 vs 58.51 years, P<0.01) but is not adjusted for; ECOG/KPS is used only as an inclusion criterion (0-2) rather than as a covariate; and liver metastatic burden indicators (such as number of metastases, maximum lesion diameter, synchronous vs metachronous status, and extrahepatic disease) are also not included. Treatment-intensity indicators such as chemotherapy lines and RDI are not reported. Multiplicity and subgroup analyses Figure 2 presents multiple subgroup analyses (age, sex, primary site, gene status, local treatment, treatment lines, and metastasis location, among others), but the manuscript does not address the multiple-comparisons issue. A large number of subgroup tests may increase the risk of false-positive findings; we suggest that the authors interpret subgroup results cautiously and consider reporting adjusted P values for key subgroups in order to reduce the likelihood of false-positive findings. Proportional-hazards assumption Follow-up in this study is relatively long (up to 132 months), and time-varying treatment effects are plausible. Formal assessment of the Cox proportional-hazards assumption (e.g., Schoenfeld residuals) would strengthen confidence in the reported hazard ratios. Safety reporting The methods section notes that major adverse reactions would be tracked during follow-up, but the manuscript does not report a safety comparison between the two cohorts, nor does it present indicators such as HILI, myelosuppression, or hepatic/renal function. Given that ICWM is evaluated as a long-term treatment exposure, additional safety data would help support a more comprehensive assessment of its clinical value. If these data cannot be systematically obtained, this limitation should be further clarified in the Discussion. References The reference base broadly covers CRLM epidemiology, local liver-directed therapies, and TCM-related mechanisms, and is reasonably current and on-topic. Incorporating the most recent systematic reviews or meta-analyses on integrative medicine in advanced colorectal cancer, where available, would further strengthen the literature base. 4. Language and interpretation The manuscript is clearly structured and well organized, and on the whole follows standard reporting conventions for observational cohort studies. However, causal and associative language are mixed in places. For example, the title uses verbs with a more causal connotation such as "prolong," and the Discussion contains expressions such as "reduced the risk of death by 35%" and "protective factors," whereas the Abstract Conclusion is phrased associatively as "is associated with longer OS." Maintaining greater consistency in the use of associative rather than causal language throughout the manuscript would better match the evidence level of a retrospective observational design. 5. Conclusion Overall, this study addresses a clinically relevant question and contributes useful real-world evidence regarding ICWM in CRLM. The findings are best interpreted as hypothesis-generating. Addressing the methodological limitations identified above, particularly immortal time bias, residual confounding, and limited covariate adjustment, in future studies would strengthen causal inference and improve the interpretability of the results.