Toward a more robust predictive model in esophageal cancer: methodological considerations for the combined use of immunoscore and pan-immune inflammation value

Dear Editor, We read with interest the article by Huang et al1, entitled “Combined immunoscore and pan-immune inflammation value associated with pathological response and survival outcomes in esophageal squamous cell carcinoma receiving neoadjuvant immunotherapy,” published in your journal. The authors’ effort to integrate the local immunoscore (IS) with the systemic pan-immune inflammation value (PIV) offers a promising dual-parameter approach for prognostic assessment in locally advanced ESCC treated with neoadjuvant chemoimmunotherapy. While the reported predictive performance (AUC 0.78–0.82) is notable, we have several methodological considerations regarding model development, validation, and reporting that we wish to highlight for further scholarly discussion. All statements in this commentary regarding analytical methodology follow the TITAN guidelines2. First, the study completed variable processing, model construction, and performance evaluation within the same cohort. Although the results are significant, this represents a single-sample development and validation process, which may lead to an overestimation of model performance. Employing techniques such as splitting the data into training and validation sets or using internal resampling methods (e.g., bootstrapping) in future work would allow for a more robust assessment of the model’s generalizability3. Moreover, dichotomizing the inherently continuous variables IS and PIV at the cohort median, while analytically convenient, results in a loss of information granularity. The biological rationale for this cutoff and its consistency across different studies remain to be established. Future efforts could explore incorporating the raw continuous values into the modeling framework or determining optimized cut-points based on clinical outcomes (e.g., using the Youden index) to enhance the model’s precision and stability4. Second, a notable feature of the study design is the temporal discrepancy between its key parameters: the IS is derived from post-treatment surgical specimens, while the PIV is measured pre-treatment. This suggests that the current model is more aligned with providing a comprehensive postoperative prognostic stratification. Its ability to serve as a pure pre-treatment predictor of response to neoadjuvant therapy requires further prospective validation. Additionally, the subgroup analysis for Stage II patients, as presented in the supplementary material, is based on a limited sample size (n ≈ 25). The reported AUC values show considerable variability. To prevent potential misinterpretation by readers, it would be prudent to add a note in the relevant figure or text highlighting the exploratory nature and inherent uncertainty of such small-sample subgroup analyses, clarifying that any conclusions require validation in larger cohorts5. Finally, the study demonstrates impressive survival differences. To more fully adhere to reporting guidelines for prognostic studies (e.g., the REMARK guidelines) and enhance transparency and reproducibility, we recommend supplementing the analysis with the total number of events and censored cases for each comparison group (e.g., Groups A–D)6. This would allow readers to better evaluate the precision of the long-term survival estimates. In summary, these suggestions are intended to contribute to the ongoing refinement of this promising biomarker framework, viewed through the lenses of predictive model science, statistical rigor, and clinical translation. The study has laid a solid foundation. We look forward to seeing its further evolution into a standardized, individualized clinical decision-making tool through future validation in multicenter, prospective studies.