We read with interest the systematic review and meta-analysis by Dr. Vatvani et al. [1] suggesting that low-dose naltrexone (LDN) is more effective than a placebo in lowering pain in patients diagnosed with fibromyalgia. The authors’ conclusion was based on two meta-analyses with pooled data from two and four randomized controlled trials (RCTs), respectively, including data from a study published by our research group in 2024 [2].

The meta-analysis on “at least 30% improvement in pain symptoms” particularly caught our attention, as the odds ratio reported from our study (OR = 2.82, 95% confidence interval [CI] 1.25 to 6.39) [1] was higher than the data presented in our original publication (RR = 1.57, 95% CI 0.88 to 2.79) [2], which when recalculated, corresponds to an OR = 1.96 (95% CI 0.84 to 4.60). Upon closer examination of the data extracted from our study, we noticed that the response rates of 32/49 in the LDN group and 20/50 in the placebo group were inflated by the addition of 30% and 50% pain responders. However, in our publication, the 50% pain responders were already included in the 30% pain response group, and the actual 30% pain response rate is 20/49 in the LDN group and 13/50 in the placebo group [2]. We verified the accuracy of the data extracted from the study by Younger et al. [3], and conducted a new meta-analysis using the correct numbers from our study. This yielded a substantially lower pooled odds ratio (OR = 2.47, 95% CI 1.10 to 5.54) (Fig. 1) compared to the estimate reported by Vatvani et al. (OR = 3.32, 95% CI 1.59 to 6.92).

Regarding the meta-analysis by Vatvani et al. investigating between-group changes in pain scores, we were concerned that the pooled effect size of –0.86 (–1.20 to –0.51) may be overestimated for several reasons. Most notably, the authors calculated a pooled mean difference despite the included studies using different outcome measures, such as the Summed Pain Intensity Rating (SPIR), an 11-point Numeric Rating Scale (NRS), and a 0–100 Visual Analog Scale (VAS). When outcome measures differ across studies, a Standardized Mean Difference (SMD) should be used [1]. Secondly, data from the two crossover trials were not reported consistently. In the study by Bested et al. [4], only data from period 2 were extracted, while for the study by Younger et al. [3], the overall percentage change was used. Regarding the handling of data from cross-over trials, The Cochrane Handbook recommends using data from the first treatment period, as these represent a parallel-group trial and avoid potential carryover effects [5]. Estimates of percentage change from baseline to follow-up cannot be pooled in a meta-analysis, as measures of variation (e.g., standard deviation) cannot be derived. We contacted the authors from the Younger et al. study and obtained raw change scores with standard deviation, which yielded markedly different results from those reported in the meta-analysis by Vatvani et al. Finally, in the study by Paula et al. [6], which was a 4-arm study, where patients received LDN or a placebo for the first 21 days, and then transcranial direct current stimulation (tDCS) or sham tDCS was added from day 21 to day 26, Vatvani et al. only included data from arm 2 (LDN + sham tDCS) and arm 4 (placebo + sham tDCS) at the 26-day follow-up. We argue that sham tDCS could potentially bias the estimated efficacy of LDN. Thus, we suggest using data from all four arms at the 21-day follow-up.

Therefore, we chose to extract data from the included studies and re-do the meta-analysis. From the Paula et al. study we extracted data at the 21-day follow-up comparing arm 1 with arm 3 (named Paula et al. A), and comparing arm 2 with arm 4 (named Paula et al. B). From the Younger et al. study, mean change scores with standard deviation from baseline to end of the first treatment arm (i.e., before cross-over) were obtained, using the data from the “placebo first” group and the “LDN first” group, respectively. For the Bested et al. study, data from the first treatment period was extracted. We then applied a random-effects meta-analysis fitted to a restricted maximum likelihood (REML) model to calculate the pooled treatment effect estimate expressed as an SMD. The results show no significant difference between LDN and placebo treatment (Fig. 2). However, a very high heterogeneity in the pooled estimate was observed, which may be explained by differences in the treatment duration. Thus, we performed a sensitivity analysis pooling data from the studies with a treatment duration of 3 weeks and 12 weeks, respectively (Fig. 3). The sensitivity analysis showed a non-significant SMD of –0.25 (–0.59 to 0.09) in favor of LDN following 12 weeks of treatment with no observed heterogeneity. For the studies with 3 weeks treatment the SMD was –0.11 (–2.18 to 1.95), and very high heterogeneity was observed.

Based on our updated data-extraction and re-analysis of data from four RCTs investigating the effect of LDN on pain in patients with fibromyalgia, we conclude that there is no evidence of an effect from 3 or 12 weeks of treatment with LDN. In agreement with the findings by Vatvani et al., we demonstrated a significantly higher pain response rate among patients treated with LDN compared to a placebo; however, the calculated OR was substantially lower than the estimate reported by Vatvani et al.