Response to comments on ‘Adverse in‐hospital outcomes in patients with paraplegia who undergo radical prostatectomy’

We thank Ma et al. [1] for their thoughtful remarks on our article [2]. We agree that 30-day outcomes represent a contemporary benchmark; however, as acknowledged in our manuscript, the National Inpatient Sample (NIS) does not capture post-discharge events. For this reason, 30-day metrics should be viewed as a complementary direction for future registry-based work rather than a requirement for NIS-based analyses. Commentators should recognise that this limitation reflects the structure of the NIS database rather than an analytical choice. We also appreciate the remarks regarding the exposure definition. In our study, paraplegia was identified using the International Statistical Classification of Diseases and Related Health Problems, ninth revision-Clinical Modification (ICD-9-CM) and ICD 10th revision-CM (ICD-10-CM) diagnostic code clusters that have been widely employed in prior NIS-based and other administrative-database evaluations of surgical outcomes among patients with spinal cord-related paralysis (e.g., Di Bello et al. [3]). These codes were selected to ensure comparability with the existing literature and to maximise case capture across both coding eras. We acknowledge that the paraplegia cluster includes several non-spinal cord injury (SCI) neurological entities; however, because the NIS spans both ICD-9 and ICD-10 periods, restricting the definition to SCI-specific ICD-10 codes would only apply to data from 2016 onward, leading to substantial reductions in cohort size and limiting comparability across years. For these reasons, we used the broader paraplegia definition commonly adopted in national inpatient studies, while recognising that future sensitivity analyses focusing on SCI-specific codes—where feasible in datasets with uniform ICD-10 coding—may further refine specificity. Finally, we agree that for outcomes of moderate or high frequency, adjusted odds ratios (ORs) may overstate the magnitude of association relative to adjusted risk ratios (aRRs). In line with this suggestion, we acknowledge that providing complementary effect measures—such as aRRs using a modified Poisson approach with robust variance, or absolute risk differences derived through marginal standardisation—can enhance bedside interpretability. In the present analysis, we used logistic regression models clustered by hospital because this approach remains the most common and reproducible method within the NIS framework, particularly following propensity score matching. Nonetheless, we concur that future studies using datasets with individual-level longitudinal follow-up could incorporate modified Poisson models or g-computation to provide adjusted absolute risks alongside ORs. We appreciate this valuable recommendation, which aims to improve the clinical readability of effect estimates. The authors report no conflict of interest.