blood-based transcriptomic signature might further improve the LNM predictive efficacy of our risk-assessment model in future studies.Second, the authors suggested that in addition to receiver operating characteristic curve analysis, inclusion of calibration curves that reflect the agreement between the actual and predicted probabilities might be useful. In such analysis, a perfect calibration would be where the observed versus predicted probability would be equal. Once again, we appreciate this suggestion very much. Accordingly, we performed such calibration analyses in our validation cohort patients using the RMS package. The flexible calibration curve was based on local regression. In this regard, when we interrogated the performance of our transcriptomic panel, we noted that patients at high risk tended to get underestimated risk predictions, whereas a good calibration was observed when patients were at low risk. We noticed that compared with the transcriptomic panel, our final riskstratification model (which included lymphatic and venous invasion, tumor budding grade, and depth of tumor invasion) exhibited a superior calibration performance (data not shown); once again highlighting the clinical significance of our reported risk-assessment model for predicting LNM in patients with T1 CRC.Third, this correspondence also recommended inclusion of decision curve analysis (DCA), which potentially offers a better measure of net benefit of any predictive biomarkers in clinical settings. The authors are correct in suggestion that DCA is a widely used method to evaluate the alternative diagnostic strategy based on "net benefit" of using any molecular assay, by itself or as an adjunct to other clinicpathological tools used in the clinic. 6 As suggested, we undertook these analyses, and observed that across most of the threshold probabilities, both the transcriptomic panel and risk-stratification model exhibited higher net benefit than the strategy for treating all the patients or none of the patients. Not surprisingly, the risk-assessment model on its own was superior to the transcriptomic panel. The DCA analysis further proved that the risk-assessment model could limit the probability of potential overtreatment in patients with T1 CRC. However, the current calibration analysis and DCA were somewhat limited by number of patients with LNM; hence, future studies with a larger number of such patient populations are needed to better appreciate the clinical significance of such analytical approaches.Last, this letter proposed enrollment of patients with T1 CRC from multiple centers to further improve the predictive accuracy of our signature. We agree with this important suggestion, because in our published study, all patients had a similar demographic profile and were enrolled at 2 institutions in Japan. Given the importance of this suggestion, we currently have several collaborations under way in which we are prospectively enrolling patients with T1 CRC at different institutions in the United States, Europe, and Asia. We ...
