Response to Letter to the Editor for “Impact of Immunosuppression Type on 30-Day Postoperative Outcomes Following Anterior Cervical Discectomy and Fusion: A Retrospective National Cohort Study of 66,162 Patients”

We thank Kaplan et al. for their close reading of our study and for two points that allow us to clarify both our findings and our intent.^1,2

The propensity-matched mortality estimate. The correspondents are correct that the matched 30-day mortality result is reported inconsistently, and we are grateful for the opportunity to resolve it. ¹ The estimate that should stand is the one given in the Results text and Table 2, an adjusted odds ratio (aOR) of 1.50 (95% CI 0.79-2.86; p = 0.22), which did not reach statistical significance. ² The value of 2.10 (95% CI 1.09-4.03; p = 0.03) appearing in the Abstract, the Figure 1 legend, and the Table 3 footnote reflects a transcription error introduced during revision. ² Critically, this correction does not alter any primary conclusion. The matched mortality analysis rested on only 16 deaths in the steroid group (0.6% vs 0.2%); as we cautioned in the original text, an estimate this sparse is unstable and hypothesis-generating, whichever point estimate is adopted. ² The findings that carry our conclusions are the adequately powered composite and readmission endpoints, which rest on 225 and 134 events and are unchanged by this correction: composite adverse events (matched aOR 1.47; 95% CI 1.24-1.75) and unplanned readmission (matched aOR 1.53; 95% CI 1.23-1.90). ² We would add, in the same spirit, that the individual infectious subtypes rest on similarly modest event counts and were not themselves propensity-matched; we regard them as directionally concordant with the composite result rather than as independent confirmation of it. ² The non-significant primary multivariable mortality estimate (aOR 1.69; 95% CI 0.97-2.94) is fully concordant with the corrected matched value. ²

The management implication. We agree that our data cannot establish a comparative hierarchy between corticosteroid minimization and withholding of biologic or synthetic DMARDs, and we did not intend our underpowered medication-class analysis (biologic n = 71, synthetic n = 129; < 10% power) to carry that claim; as stated, those null results reflect inadequate power, not demonstrated safety. ² Our recommendation to prioritize corticosteroid burden is directional, not comparative, and rests on two better-supported pillars. First, the corticosteroid signal was statistically robust in our adequately powered primary analysis (multivariable aOR 1.35; 95% CI 1.16-1.56) and was reproduced, on a far smaller sample, in the 2021 granular category cohort (aOR 1.95; n = 181 corticosteroid-only patients). ² It converges with a large external literature: the posterior lumbar spine surgery NSQIP analysis of Kebaish and colleagues in 140,519 patients, ³ the Medicare lumbar-fusion analysis of Singla and colleagues, ⁴ the perioperative review of Modha and Whinney, ⁵ and the Danish nationwide cohort of 143,782 appendectomy patients reported by Orgun and colleagues. ⁶ Together these identify chronic glucocorticoid exposure as a consistent and modifiable pharmacologic risk factor for postoperative infection, and two of them address exposure intensity directly: Orgun and colleagues report elevated 90-day infection risk above a threshold of 5 mg prednisone-equivalent daily, ⁶ while Modha and Whinney conclude that chronic glucocorticoid dose should be lowered to the safest level that controls disease activity. ⁵ By contrast, a single perioperative dexamethasone dose was noninferior to placebo for surgical-site infection in the large randomized PADDI trial reported by Corcoran and colleagues in the New England Journal of Medicine. ⁷ That contrast is precisely the dose-and-duration distinction our binary NSQIP variable cannot resolve, and which we acknowledged as a central limitation. ² Second, reflexive biologic cessation carries its own hazard: the systematic review of van Duren and colleagues found significantly fewer disease flares with continuation (OR 0.22) and no significant excess of surgical-site infection (OR 1.11). ⁸ The 2026 Cochrane review of Ponkilainen and colleagues reaches a parallel conclusion, with the important qualification that its included trials enrolled only conventional synthetic DMARD patients and that its authors judge the corresponding risks for biologic and targeted synthetic agents to be less clear. ⁹ The nationwide Taiwanese arthroplasty cohort of Chen and colleagues, which did study biologics, reported the same pattern (flares 6.1% vs 22.3%; surgical-site infection 2.9% vs 2.5%; p = 0.39). ¹⁰ Guideline positions follow similar logic without being uniform: the 2022 ACR/AAHKS guideline conditionally favors continuing conventional synthetic DMARDs and withholding biologics for a single dosing interval rather than by blanket cessation, ¹¹ whereas the 2025 ASRA infection-control guideline continues conventional DMARDs but advises considering biologic interruption before deep, neuraxial, and implanted-device procedures. ¹² We fully agree that these sources derive from arthroplasty, general orthopaedic, and interventional pain populations rather than ACDF, and we would add that their divergence on biologics is itself an argument against extrapolation.^8,11,12 We flagged this limitation in the original article and reiterate our call for spine-specific perioperative immunosuppression guidance and an adequately powered, procedure-specific study. ²

Risk stratification versus a treatment hierarchy. Finally, we share the correspondents’ view that these data are best positioned to support corticosteroid-associated risk stratification rather than a hierarchy of immunosuppression strategies. ¹ Operationalizing that association is where we see the most immediate clinical value, and any instrument built from it must be judged on external validation and calibration rather than on incremental discrimination, the standard we set out in our scoping review of risk prediction in spine surgery. ¹³ Bedside tools built only from variables available before incision are feasible: our FORD score predicts non-home discharge after orthopaedic trauma from admission data alone, and an equivalent instrument could flag the corticosteroid-exposed ACDF patient for intensified infection surveillance and targeted preoperative optimization. ¹⁴ The same evidentiary standard applies to the machine-learning and large-language-model approaches now entering spine care. ¹⁵ Where pharmacologic risk cannot be modified, reducing soft-tissue disruption is a complementary lever. ¹⁶ We have defended these standards in subsequent correspondence on risk prediction, and we hold the present association to them. ¹⁷ Generating the procedure-specific, adequately powered evidence needed to convert this association into practice remains the shared goal, and we thank the correspondents for advancing it.

ORCID iDs

Rohan A. Phadke https://orcid.org/0000-0002-8611-6711

Samer G. Salman https://orcid.org/0009-0007-9897-4071

Connor Welsh https://orcid.org/0009-0002-4340-813X

Ethical Considerations

This correspondence presents no new patient data; the original study analyzed the American College of Surgeons National Surgical Quality Improvement Program Participant Use Files, a deidentified registry available to participating hospitals.

Data Availability Statement

NSQIP ACS participant use file data is privately available to participating hospitals; however, data used may be made available to the journal and editors upon reasonable request.*