We have thoroughly engaged with the article “Computed tomography Hounsfield unit values as a treatment response indicator for spinal metastatic lesions in patients with non-small-cell lung cancer: a retrospective study in Japan” by Taniwaki et al. [1]. We congratulate the team and commend the efforts they put into this important and understudied area—quantitative evaluation of treatment response in spinal metastases using Hounsfield units (HU) in patients with non-small-cell lung cancer. However, despite the promising findings, we would like to suggest important methodological pointers that can further strengthen the findings of future studies with similar data.

First, the exclusion of 61 patients—33 of whom died within two months with rest lacking sufficient followup imaging—introduces survivorship bias. These patients likely had more aggressive disease and may have influenced HU trends, reducing generalizability of the results. Including them through imputation or sensitivity analysis can help mitigate bias [2].

Second, HU measurements were taken from single-slice manual regions of interests (ROIs) without assessing inter- or intra-observer variability. Given known HU variability due to ROI size, positioning, and scanner calibration, reproducibility is uncertain [3,4]. Future work should adopt reliability metrics such as intraclass correlation coefficients and consider automated volumetric analysis [5].

Third, defining HU “responders” using a median cutoff of 124 HU lacks biological justification. Dichotomizing a continuous variable without validation (e.g., receiver operating characteristic curve analysis) risks misclassification and limits interpretability. Future studies should seek to find and confirm best HU thresholds for survival prediction in separate groups [6].

Fourth, the multivariate model included only age, sex, and the modified Katagiri score, omitting critical prognostic factors such as visceral metastases, epidermal growth factor receptor/anaplastic lymphoma kinase status, programmed death-ligand 1 expression, and Eastern Cooperative Oncology Group performance status, which may confound results [7,8]. Their absence reduces the validity of the inferences made.

Finally, the retrospective, single-center design, small sample (n=85), and varied treatments limit generalizability. Prospective, multicenter studies with standardized imaging and treatment protocols would improve external validity and statistical power [9].

In conclusion, although the study identifies a potentially useful imaging biomarker, methodological improvements, more extensive prognostic modeling, and outside validation would support its findings. The treatment of spinal metastases may benefit from quantitative imaging like HU, but strict standardization and contextual interpretation are necessary.