Cohen 2015.

Methods	Multicentre, randomised, double‐blind (double‐dummy), active‐controlled, parallel groups, partial enrichment. Titration to maximum tolerated dose (1800 mg to 3600 mg daily) over 15‐24 days (3 months total)
Participants	Radicular leg pain for ≥ 6 weeks and < 4 years, PI ≥ 4/10 or ≥ 3/10 if leg pain ≥ back pain, symptoms of lumbosacral radicular pain. Findings of herniated disc or spinal stenosis on MRI concordant with presentation. Age ≥ 17 years Excluded: neuropathic pain > 4 years, previous failed trial with gabapentin or pregabalin, steroid injections ≤ 3 years, cauda equina syndrome, planned surgery N = 145 Mean age 43 years, 26% women 18% had pain ≤ 3 months, 25% taking opioids Initial PI: worst leg 7.8/10; average leg 5.4/10
Interventions	Gabapentin capsule + saline injection, n = 72 Depomethylprednisolone 60 mg injection + 1 ml 0.25% bupivacaine + placebo capsule, n = 73 Gabapentin titrated to 1800 mg to 3600 mg/day (3 divided doses) over 15‐24 days, but ≥ 5 days before follow‐up Steroid injected into epidural space (interlaminar or transforaminal), saline injected into posterior ligaments Tramadol and NSAIDs "as needed" for rescue medication, or opioids increased by ≥ 20% for those taking them. No other co‐interventions
Outcomes	Mean PI, for average and worst leg and back pain Global evaluation (non‐standard scales) Adverse events Withdrawals
Notes	Oxford Quality Score: R = 2, DB = 2, W = 1, Total = 5 Congressional grant from the Center for Rehabilitation Sciences Research, Bethesda
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Computer‐generated randomisation tables, stratified by site. Central pharmacy
Allocation concealment (selection bias)	Low risk	"sequentially numbered opaque sealed envelope"
Blinding (performance bias and detection bias) All outcomes	Low risk	"A central research pharmacy over‐capsulated 300 mg gabapentin and placebo capsules to appear identical". Participants "visually shielded from the image screen" during injections", had no further contact with physician
Incomplete outcome data (attrition bias) Efficacy	Unclear risk	LOCF imputation
Size Efficacy	Unclear risk	50‐200 participants per treatment arm