Six databases, Embase, PubMed, Web of Science, CNKI, VIP Database, and Wanfang Database, were searched from their inception to 1 September 2025. Articles were limited to English and Chinese. The search strategy used Medical Subject Headings (MeSH) terms related to acupuncture and insomnia. Table 1 summarizes the complete PubMed search strategy. The same terminologies were used in the English and Chinese databases (see Supplementary material 1). Additionally, the Cochrane Central Register of Controlled Trials (CENTRAL), SinoMed, PROSPERO, and INPLASY databases were searched to identify relevant literature. Protocols, reviews, pilot studies, case reports, case series, editorials, surveys, animal experiments, notes, letters, and conference reports were excluded.

Inclusion criteria: (1) The study was a randomized controlled trial (RCT) that exclusively used either manual acupuncture or electroacupuncture as an experimental intervention, and used sham acupuncture, wait list control, or drugs as control intervention, (2) the participants were diagnosed with chronic insomnia according to published guidelines, (3) outcomes measured in the study should include at least either the PSQI or Insomnia Severity Index (ISI) score, and (4) the article was in either English or Chinese.

Exclusion criteria include the following: (1) non-comparative studies, (2) duplicate studies (3) studies focusing on secondary insomnia, and studies in which patients were diagnosed with cancer, anxiety, depression, or other psychological illnesses that may lead to secondary insomnia, (4) studies that compared the effects of different acupoints, or used other acupoints on the meridians without the effect of treating insomnia as a control, (5) studies that did not report the selected acupoints, (6) studies without available data on preoperative and postoperative outcomes or those outcomes that could not be calculated using a specific formula, and (7) studies that did not report the randomization process, and studies that did not report ethical statements or registration numbers.

There were no restrictions on sex, needle material, treatment course, or needle retention time.

After a systematic search, two authors (Peiqi Li and Qianwen Yang) conducted a preliminary screening of the titles, abstracts, and keywords using the Citation Management Software EndNote X9. Subsequently, potentially eligible full texts were obtained and read for article re-screening, and the data were cross-checked. Any disputes were settled by a third reviewer (Qinhui Fu) or by consensus between the two reviewers. When the studies needed to be excluded, the exclusion reasons were recorded. Subsequently, the two reviewers screened the text and extracted data from the eligible studies independently in duplicate, which are detailed in Table 2. The extracted information is listed as follows: year of publication, author, sample size, sex, interventions, outcomes, acupoints selected, treatment frequency, and adverse effects. The missing data were obtained by the reviewers by inquiring with the authors of the original study via email.

This assessment was performed by the two reviewers (Peiqi Li and Qianwen Yang) independently using version 2 of the Cochrane tool for assessing the Risk of Bias (RoB2) in randomized trials. The randomization process, deviations from intended interventions, missing outcome data, outcome measurement, and selection of reported results were the five factors used to assess the risk of bias.

All the studies were classified into “low risk of bias,” or “high risk of bias,” or “unclear.” Possible disagreements were settled by the third reviewer (Qinhui Fu).

Review Manager Software (RevMan, version 5.4) was utilized for all statistical analyses. The primary outcomes were the PSQI and ISI scores. To adhere to the Cochrane unit-of-analysis guidance, we used only one pair of comparisons to conduct the pairwise meta-analysis. The preliminary input rules were as follows: (1) when there were two kinds of experiment intervention, we would choose single acupuncture or usual acupuncture rather than acupuncture with special manipulations or acupoints, introducing less heterogeneity, and (2) when there were two kinds of control intervention, we would primarily choose drugs rather than sham acupuncture or waiting list, in order to better present the difference between acupuncture and the conventional pharmacological therapy. As for secondary outcomes, four neurobehavioral assessment tools were selected to evaluate different daytime symptoms: (1) Epworth Sleepiness Scale (ESS) is used to quantifying drowsiness tendencies in different daytime situations, (2) Auditory Verbal Memory Test (AVMT) could measure four hippocampal-dependent memory consolidation critical for daytime cognitive performance, (3) Montreal Cognitive Assessment (MoCA), recognized for its high diagnostic sensitivity and clinical feasibility, serves as a gold standard brief cognitive screening tool across for overall cognitive function, and (4) Fatigue Scale-14 (FS-14) is chosen to quantify the severity of fatigue, indirectly reflecting the quality and recovery effect of sleep at night. Other outcomes, such as cortisol (CORT) level and total effective rate, were also recorded if reported. Quantitative data, such as sample size, were extracted from all the eligible RCTs, as well as means and standard deviation values of outcome measurements (PSQI and ISI scores) after treatment in each group. We exclusively extracted data from the earliest post-treatment time-point (usually immediately after the completion of the treatment intervention). If a study reported results of multiple post-treatment or follow-up time-points, data from the earliest one were prioritized. Standardized mean difference (SMD), mean difference (MD), and 95% confidence intervals (CIs) were used for the efficacy analysis of continuous variables.

Note that when there were no direct data on mean value and standard deviation, we calculated them through the size, first quartile, median, and third quartile of the sample using a statistical formula (15, 16).

The statistical heterogeneity of the results was evaluated using the I² test, and then a fixed or random-effect model would be applied accordingly. Additional analyses (e.g., sensitivity analysis and subgroup analysis) were executed to identify potential sources of heterogeneity and assess the suitability of using a random-effects model. Furthermore, evaluations of subgroups were carried out to look into heterogeneity pertaining to the intervention, treatment duration, number of acupoints used, treatment frequency, diagnostic criteria, age range, and article quality. Two methods were used to test the interactions between the factors: (1) qualitative assessment: an interaction was probably absent if each subgroup’s 95% CI overlapped; and (2) quantitative assessment: an interaction was probably absent if the subgroup difference was greater than 0.05.

In addition, funnel plots were employed to evaluate potential bias in the RCTs encompassed in this review when the total number of trials included was more than or equal to 10. When heterogeneity larger than 50% was referred to a random-effects model application, the RevMan software was unable to generate funnel plots with 95% CI lines. Therefore, Stata 13.1 software was utilized to generate the corresponding funnel plot. The publication bias was deemed low if both sides of the funnel plots were balanced.

The quality of evidence was assessed and classified into four levels: high, moderate, low, and very low quality, according to the Grading of Recommendations Assessment, Development, and Evaluation (GRADE). The evidence quality was assessed separately by two authors using the Guideline Development Tool (GRADEpro GDT). Disagreements were settled through additional discussion.

In Figure 1, the diagram illustrates that 5,037 studies were found across the six databases. Ninety-one replications were removed, and no grey literature was added. Based on titles, keywords, and abstracts, 4,560 articles were excluded, including reviews, case reports and series, protocols, animal experiments, letters, surveys, and commentaries. The articles that did not focus on insomnia were excluded (n = 712). In the remaining articles, participants with insomnia caused or complicated by cancer or psychological problems (n = 77), studies that did not use acupuncture as the exclusive treatment for the intervention group (n = 956), and articles not published in English or Chinese (n = 12) were excluded. The complete texts of the remaining 386 studies were acquired and subsequently evaluated; 368 studies were eliminated due to the specified reasons: no ISI or PSQI measurements (n = 78), did not report acupoints used (n = 203), and no ethical statement or registration was reported (n = 87). Finally, 18 randomized controlled trials that satisfied the criteria for inclusion were ultimately incorporated into the meta-analysis.

Totally, 1,767 patients were included from the 18 RCTs (17–35). Four of the studies were published in English, and 14 in Chinese. Participants were identified as having insomnia based on either the third edition of the International Classification of Sleep Disorders (ICSD-3) or the fifth edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-5). In each study, the participants’ characteristics (age, sex, and other parameters) at baseline were balanced between the treatment and control groups (Table 2). The interventions were manual acupuncture in 11 studies and electroacupuncture in the other 7 studies. The comparators included no treatment (waiting list), sham (electro-) acupuncture, and medication. In these trials, the course of treatment varied from 2 to 5 weeks, and each treatment lasted approximately 30 min. The treatment frequency of acupuncture and electroacupuncture was either once a day or once every other day, while that of drug therapy was generally every night.

Randomization was applied in all 18 included studies. The results showed a low risk of deviation from the intended interventions. The lack of missing outcome data degrades the risk of bias. Regarding outcome measurements, five studies raised concerns about missing outcome data owing to dropouts. Almost all trials blinded the outcome assessors to the participants’ interventions. The risk of selection of the reported results was low. The results of the risk of bias assessment for all included studies are illustrated in Figures 2, 3. In general, 13 studies were of low risk, whereas 5 exhibited some concern.

When enough data of the same measurement were reported in more than one study, they were extracted and are shown in Table 4. Meta-calculations were then performed using RevMan 5.4 software.

The synthesized data showed that acupuncture therapy had significant curative effects when measured using four subscales of AVMT (instant memory, delayed recognition, short delayed recall, and long delayed recall), FS-14, total effective rate, and MoCA. Nevertheless, the heterogeneity of some results was generally high, so these results should be interpreted cautiously.

Original data of CORT and ESS scores were not reported in the included articles, preventing a meta-analysis. Accordingly, we extracted the reported p-values and summarized them in the table above to provide a general overview.

Among the included studies, seven studies reported negative events and the exact number of patients who experienced them (see Table 5). The most common event was a subcutaneous hematoma, predominantly in the acupuncture intervention groups. Drug-related effects emerged, such as thirst and lethargy, in the medication control groups. The majority of events were mild to moderate in severity, indicating a favorable overall safety profile for the interventions studied. In these studies, there were no withdrawals from the study owing to adverse effects.

Other than the above six studies, Chung et al. (17) also reported the adverse events, but without the exact number of each incident. The study reported that the total number of adverse events was 260, of which 243 were mild, 14 moderate, and 3 severe, with the incidence rate of any negative event being 62%. The top three adverse events were acupoint bruising (26.6%), headache and other pain events (21.4%), and needle site pain (17.7%). Moreover, 2.1 and 3.6% of the participants in the acupuncture treatment group and combination therapy groups, respectively, dropped out because of adverse events. The authors of that study emphasized that all reported events were mild and the treatments were deemed safe and acceptable to patients.

The evidence for the quality of the synthesized results was assessed by the GRADE process (Figure 7). Evidence of moderate quality supported that acupuncture could improve ISI, AVMT, and MoCA scores. In contrast, the evidence of acupuncture improving PSQI and FS-14 scores was of low quality.