Restless Legs Syndrome — Network Meta-Analysis
Comparing all randomized RLS treatments on one scale (IRLS (0-40), lower = better) by combining direct and indirect evidence. Reference = placebo; random-effects (DL) via R netmeta 3.6.1.
Evidence network
Core treatments (≥2 trials) + placebo. Node size ∝ number of trials; line thickness ∝ number of direct head-to-head studies.
Effect vs placebo (core drugs)
Ranking (core drugs, ≥2 trials)
P-score is the frequentist analogue of SUCRA (higher = better), ranked here over the full network of all interventions. For a like-for-like comparison to Zhou's drug-only SUCRA, see the validation section below.
| Treatment | Trials | MD vs placebo (95% CI) | P-score |
|---|---|---|---|
| cabergoline | 2 | -11.93 (-16.80, -7.06) | 0.88 |
| dipyridamole | 2 | -7.09 (-10.45, -3.73) | 0.65 |
| pramipexole | 12 | -6.39 (-7.77, -5.01) | 0.61 |
| gabapentin | 2 | -6.49 (-11.31, -1.67) | 0.59 |
| gabapentin enacarbil | 8 | -5.98 (-7.46, -4.51) | 0.57 |
| iron | 8 | -4.91 (-6.85, -2.97) | 0.47 |
| rotigotine | 11 | -4.63 (-6.05, -3.20) | 0.45 |
| levodopa/benserazide | 2 | -4.43 (-8.75, -0.11) | 0.44 |
| pregabalin | 5 | -4.44 (-6.37, -2.50) | 0.43 |
| ropinirole | 11 | -3.67 (-5.09, -2.25) | 0.36 |
League table (core treatments)
| Placebo | Cabergoline | Dipyridamole | Gabapentin | Gabapentin Enacarbil | Iron | Levodopa/Benserazide | Pramipexole | Pregabalin | Ropinirole | Rotigotine | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Placebo | — | +11.93 | +7.09 | +6.49 | +5.98 | +4.91 | +4.43 | +6.39 | +4.44 | +3.67 | +4.63 |
| Cabergoline | -11.93 | — | -4.84 | -5.44 | -5.95 | -7.02 | -7.50 | -5.54 | -7.49 | -8.26 | -7.30 |
| Dipyridamole | -7.09 | +4.84 | — | -0.60 | -1.11 | -2.18 | -2.66 | -0.70 | -2.65 | -3.42 | -2.46 |
| Gabapentin | -6.49 | +5.44 | +0.60 | — | -0.51 | -1.58 | -2.06 | -0.10 | -2.05 | -2.82 | -1.87 |
| Gabapentin Enacarbil | -5.98 | +5.95 | +1.11 | +0.51 | — | -1.07 | -1.55 | +0.41 | -1.54 | -2.31 | -1.36 |
| Iron | -4.91 | +7.02 | +2.18 | +1.58 | +1.07 | — | -0.48 | +1.48 | -0.47 | -1.24 | -0.28 |
| Levodopa/Benserazide | -4.43 | +7.50 | +2.66 | +2.06 | +1.55 | +0.48 | — | +1.96 | +0.01 | -0.76 | +0.20 |
| Pramipexole | -6.39 | +5.54 | +0.70 | +0.10 | -0.41 | -1.48 | -1.96 | — | -1.95 | -2.72 | -1.77 |
| Pregabalin | -4.44 | +7.49 | +2.65 | +2.05 | +1.54 | +0.47 | -0.01 | +1.95 | — | -0.77 | +0.19 |
| Ropinirole | -3.67 | +8.26 | +3.42 | +2.82 | +2.31 | +1.24 | +0.76 | +2.72 | +0.77 | — | +0.96 |
| Rotigotine | -4.63 | +7.30 | +2.46 | +1.87 | +1.36 | +0.28 | -0.20 | +1.77 | -0.19 | -0.96 | — |
Cell = row treatment vs column treatment, IRLS mean difference (green/negative favors the row; red/positive favors the column). Bold = 95% CI excludes 0; hover any cell for the interval.
Validation vs Zhou et al. 2021
Re-run on only the 9 pharmacological drugs Zhou analysed, so the P-score ranks over the same node set as Zhou's SUCRA (the full 57-node landscape shown above ranks placebo mid-pack and is not directly comparable). Zhou MD/CI are primary-RLS estimates where available; SUCRA is Zhou's all-drug ranking.
| Drug | Trials | Project MD (95% CI) | P-score | Zhou MD | SUCRA | ΔMD | Sig. |
|---|---|---|---|---|---|---|---|
| cabergoline | 2 | -11.95 (-16.84, -7.05) | 0.99 | -12.05 | 98.7 | +0.10 | ✓ |
| pramipexole | 12 | -6.40 (-7.79, -5.01) | 0.75 | -5.44 | 57.2 | -0.96 | ✓ |
| gabapentin enacarbil | 8 | -5.99 (-7.47, -4.50) | 0.68 | -3.69 | 32.6 | -2.30 | ✓ |
| gabapentin | 1 | -6.50 (-11.35, -1.65) | 0.67 | -8.25 | 80.1 | +1.75 | ✓ |
| iron | 8 | -4.91 (-6.86, -2.96) | 0.47 | -5.65 | 51.0 | +0.74 | ✓ |
| rotigotine | 11 | -4.63 (-6.07, -3.20) | 0.41 | -5.12 | 54.7 | +0.49 | ✓ |
| levodopa/benserazide | 2 | -4.44 (-8.78, -0.10) | 0.41 | -4.33 | 41.5 | -0.11 | ✗ |
| pregabalin | 5 | -4.43 (-6.39, -2.48) | 0.38 | -5.34 | 56.4 | +0.91 | ✓ |
| ropinirole | 10 | -3.67 (-5.11, -2.24) | 0.24 | -2.50 | 16.8 | -1.17 | ✓ |
Coverage gap: oxycodone-naloxone — analysed by Zhou but absent from the project's extracted network.
Cabergoline ranks first and ropinirole near-last in both; most MDs agree within ~1 IRLS point. The one significance disagreement is levodopa (the project CI barely excludes 0, while Zhou reports it no better than placebo). Largest MD gap: gabapentin enacarbil.
Single-trial treatments (low evidence)
These can top the P-score ranking but rest on one small trial each (46 total) — shown for completeness, not as recommendations.
Caveats
- Automated extraction. Effect signs, SDs (often derived from SE/CI), arm drug identity and timepoints were machine-extracted; an NMA propagates such errors through indirect links.
- Heterogeneity is high (I² = 59.1%): a broad 2004–present pull mixes severities, primary/secondary RLS, doses and trial lengths, weakening transitivity.
- Scope. One IRLS 0–40 scale only; endpoint scores preferred; crossover handled approximately; non-randomized designs excluded.
- Sparse nodes. Most compounds have a single trial; the credible signal is the ≥2-trial core above.