Neurosurgery

Double negative controls reduce bias from non‑compliance in surgical RCT simulations

In simulation scenarios informed by lumbar discectomy trials, double negative control (DNC) methods produced low bias and near‑nominal coverage across crossover rates up to 40%. Traditional estimators (ITT, PP, AT) showed severe bias with coverage probabilities ≤ 0.1%, and IV estimators had high variability (MSE reported as 25.612 at 40% crossover). Difference‑in‑differences performed well only when confounding was equal across groups, whereas DNC remained stable across confounding structures. These findings apply to surgical RCTs with notable non‑compliance but derive from simulations rather than empirical trial analyses. 1

Pivotal device trials report basic demographics but rarely embed equity frameworks

Scoping review of 74 pivotal device studies found universal age and sex reporting but limited analytic integration of equity frameworks. Only 18.9% performed age‑based subgroup analyses and 14.8% performed sex‑based analyses; race/ethnicity was reported in 35.1% of studies. No study used CONSORT‑Equity and only 2.7% explicitly framed trials around equity considerations. These practices limit assessment of external validity and generalisability to underrepresented patients in device approvals. 2

Validated models reasonably predict 12‑month disability after lumbar spinal stenosis surgery

Using three Scandinavian registries (development n=31,908; external validations n=30,700 and n=4,063), models predicted 12‑month ODI with pooled MAE 12.4 after internal‑external cross‑validation. Binary ODI models achieved C‑statistics ~0.75–0.78, with calibration slopes near 1 and modest systematic underprediction in some cohorts. Pain prediction performed less well (MAE 2.2–2.6; C 0.64–0.73), and XGBoost offered no clear advantage over regression. Applicable to elective lumbar stenosis in Scandinavian registry settings; local recalibration recommended before clinical use. 3