Predictors of Treatment Success Following Pulsed Radiofrequency of the Lumbar Dorsal Root Ganglion - Leoni, Varrassi et al
A Multivariable Prediction Model with Clinical Nomogram
Leoni et al. (Current Pain and Headache Reports, 2026) retrospectively studied 308 patients who underwent pulsed radiofrequency of the lumbar dorsal root ganglion (DRG-PRF) for chronic radicular pain at a single Italian centre. At 6 months, 43.5% achieved ≥50% NRS pain reduction. LASSO regression identified four independent predictors — daily morphine milligram equivalents (OR 0.981), pain duration (OR 0.816), baseline NRS (OR 1.558), and prior fusion/decompression surgery (OR 0.315) — which were built into a multivariable logistic model and bedside nomogram. Performance was moderate (AUC 0.734, corrected 0.742 on bootstrap validation) with good calibration.
Key points
- Opioid burden is the strongest modifiable negative predictor. Each 1 mg/day increase in MME cut the odds of success by ~2%; responders averaged 17.0 mg/day vs 40.4 mg/day in non-responders. This supports pre-procedural opioid tapering as a way to improve outcomes, and fits a wider literature on opioids blunting interventional pain treatments.
- Surgical subtype matters more than surgery per se. Prior fusion or decompression roughly tripled the odds of failure (OR 0.315), plausibly via epidural fibrosis, altered biomechanics and central sensitisation (FBSS/PSPS-T2), while prior discectomy was prognostically neutral — so a discectomy history should not exclude a patient from DRG-PRF, whereas fusion history warrants careful counselling or alternative approaches.
- The model is promising. Higher baseline pain paradoxically predicted better response (OR 1.558), a finding the authors flag as needing external validation. More broadly: AUC 0.734 means roughly one in three patients is misclassified; the study is retrospective, single-centre, limited to a pain-only endpoint at 6 months and omits psychological predictors (depression, catastrophising) and procedural parameters. External validation in independent cohorts is required.