Efficacy of acupuncture for generalized anxiety disorder: a systematic review

Abstract

Background

This systematic review aimed to assess the efficacy and safety of acupuncture for generalized anxiety disorder (GAD), with a particular focus on acupuncture alone and acupuncture combined with medication, compared with medication or sham acupuncture.

Methods

We searched MEDLINE/PubMed, Web of Science, Embase, Cochrane Central Register of Controlled Trials (CENTRAL), China National Knowledge Infrastructure (CNKI), Wanfang Data, VIP Database for Chinese Technical Periodicals, Chinese Biomedical Literature Database (CBM), and ClinicalTrials.gov up to August 31, 2025. Randomized controlled trials (RCTs) of acupuncture for GAD were independently screened and extracted by two reviewers following strict quality assessment. Risk of bias was evaluated using the RoB 2.0 tool, evidence certainty was assessed with the GRADE approach, and meta-analyses were conducted with RevMan 5.3. Subgroup analyses were performed according to comparator type and acupuncture parameters (treatment modality, duration, frequency, and sessions).

Results

A total of 41 RCTs involving 3,209 participants were included. Meta-analysis showed that acupuncture alone (MD -1.79, 95% CI -2.69 to -0.88; 19 studies) and acupuncture combined with medication (MD -2.26, 95% CI -3.05 to -1.48; 20 studies) were superior to medication alone in reducing Hamilton Anxiety Rating Scale (HAMA) scores. Acupuncture also outperformed sham acupuncture (MD -3.46, 95% CI -4.76 to -2.16; 3 studies). Significant improvements were also observed in total effective rate, Self-Rating Anxiety Scale (SAS) scores, and Clinical Global Impression-Efficacy Index (CGI-EI), with lower rates of adverse events in the acupuncture groups. Subgroup analyses suggested that a treatment duration of 4–6 weeks was associated with better outcomes; for stand-alone acupuncture, a frequency of seven sessions per week was more effective than medication, while for combined therapy, a frequency of 2–4 sessions per week was more effective than medication.

Discussions

This meta-analysis indicates that acupuncture, either as a monotherapy or in combination with medication, may provide superior efficacy to medication and sham acupuncture, with fewer adverse events. However, most included studies were rated as having “some concerns” in risk of bias by RoB 2.0, and GRADE assessment suggested low or very low certainty of evidence for the primary outcome (HAMA). Egger’s test indicated no significant publication bias in HAMA. Further high-quality, large-scale RCTs are required to confirm these findings.

Registration

The protocol of the review was registered in PROSPERO (CRD42015016033).

Supplementary Information

The online version contains supplementary material available at 10.1186/s12991-025-00614-5.

Introduction

Generalized anxiety disorder (GAD) is the most common anxiety disorder in primary care, characterized by excessive and uncontrollable worry lasting at least six months [1, 2]. GAD is associated with an increased risk of suicide, with a comorbidity prevalence of 10.4%, and severely impairs patients’ quality of life and social functioning [3, 4]. Nearly one-third of patients miss an average of 6.3 workdays per month, leading to increased healthcare and societal burden [5].

First-line treatments for GAD include psychological interventions such as cognitive-behavioral therapy (CBT), as well as pharmacological options such as selective serotonin reuptake inhibitors (SSRIs) and serotonin–norepinephrine reuptake inhibitors (SNRIs) [2]. Although these approaches are effective, approximately one-third of patients show little or no response [6]. Moreover, medication is often associated with serious adverse effects, including cognitive and motor impairment, as well as risk of misuse and dependence [7]. Acupuncture, as a complementary and alternative therapy, has been shown in several trials to alleviate anxiety symptoms with fewer side effects [8, 9].

A previous meta-analysis reported that acupuncture improved anxiety-related outcomes and was generally well tolerated and safe [10]. However, that study only included 20 RCTs published before 2021, lacked standardized control group selection, and assessed treatment duration without considering treatment frequency, a key determinant of efficacy. In addition, the meta-analysis pooled equivalence and superiority trial designs, resulting in substantial heterogeneity, and did not apply the Grading of Recommendations Assessment, Development and Evaluation (GRADE), which reduced the strength of its conclusions. Another meta-analysis suffered from similar methodological limitations [11].

Therefore, our objective was to conduct a more comprehensive and rigorous systematic review and meta-analysis of randomized controlled trials (RCTs) of acupuncture for GAD, with particular emphasis on standardized control settings, placebo effects of acupuncture, and treatment parameters such as frequency and sessions.

Methods

This systematic review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement [12] (Supplementary Material 1). The protocol of the review was registered in PROSPERO (CRD42015016033) (Available from: https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD42015016033).

Eligibility and inclusion criteria

Studies were included if they met the following criteria: (1) randomized controlled trials (RCTs); (2) participants diagnosed with generalized anxiety disorder according to DSM, ICD, or CCMD criteria; (3) interventions consisting of acupuncture alone, or acupuncture combined with medication (with identical medication in both groups); and (4) control interventions consisting of medication or sham acupuncture. Different forms of acupuncture were allowed, including manual acupuncture, scalp acupuncture, auricular acupuncture, electroacupuncture, transcutaneous acupoint electrical stimulation, pricking-blood therapy, cupping, or moxibustion, without restrictions on treatment duration or sessions. Studies involving other complementary and alternative therapies (e.g., relaxation, massage, psychotherapy, music therapy, or other traditional Chinese medicine approaches) were excluded. No language restrictions were applied.

We categorized studies into stand-alone designs (acupuncture vs. medication), add-on designs (acupuncture plus medication vs. medication), and sham-controlled designs (acupuncture vs. sham acupuncture).

Search strategy

A comprehensive search strategy was developed. The following electronic databases were searched up to August 31, 2025: MEDLINE/PubMed, Web of Science, Embase, Cochrane Central Register of Controlled Trials (CENTRAL), China National Knowledge Infrastructure (CNKI), Wanfang Data, VIP Database for Chinese Technical Periodicals, Chinese Biomedical Literature Database (CBM), and ClinicalTrials.gov. Both subject terms and free-text terms were applied, without restrictions on language or publication type. Reference lists of included studies were also screened to identify additional eligible trials (Supplementary Material 4).

Screening and study selection process

Two reviewers independently screened studies in two stages: title/abstract screening and full-text review. Discrepancies were resolved through discussion or adjudication by a third reviewer. Inter-rater agreement was assessed using the kappa statistic.

Data extraction

Data were extracted on study characteristics (Study ID, funding, diagnostic criteria), participant characteristics (age, sex, sample size, dropout), intervention details (type, frequency, duration, sessions, acupoints, needle retention times and combined medication), control interventions (drug type, dose, frequency, or sham procedures), and outcome measures. Outcomes included the Hamilton Anxiety Rating Scale (HAMA), total effective rate (as defined in the original studies), Self-Rating Anxiety Scale (SAS), Clinical Global Impression (CGI; including Severity of Illness [SI], Global Improvement [GI], and Efficacy Index [EI]), Treatment Emergent Symptom Scale (TESS), and adverse events. Data extraction was performed independently by two reviewers and verified by a third.

Quality assessment

The methodological quality of all included randomized controlled trials (RCTs) was evaluated using the Cochrane Risk of Bias 2.0 tool (RoB 2.0) [13]. Risk of bias assessments were conducted independently by two reviewers, with disagreements resolved through discussion or consultation with a senior reviewer.

Data synthesis

To capture the breadth and variability of reported outcomes, we collected continuous data on anxiety-related scales (HAMA, SAS, and CGI), categorical data on total effective rate, as well as information on TESS scores, adverse events, and dropout rates. Dropout was defined as all randomized participants who did not complete treatment, regardless of reason.

If the original trial did not provide the mean and standard deviation (SD), we first attempted to contact the study authors to obtain the missing information. If the data could not be obtained, we applied the standard imputation methods recommended in the Cochrane Handbook for Systematic Reviews of Interventions [14], such as estimating SDs from standard errors, confidence intervals, or P values when available. Trials without sufficient information for calculation were excluded from the quantitative synthesis. Intention-to-treat (ITT) data were prioritized when available. For total effective rate, the effect size was calculated according to the ITT principle, defined as the ratio of patients reported as effective to the total number randomized to each group.

Meta-analyses were performed using RevMan 5.3 (Cochrane Collaboration). Categorical data were summarized as risk ratios (RR) with 95% confidence intervals (CI), and continuous data as mean differences (MD) with 95% CI. The main characteristics and results of the included studies were summarized in tables. For quantitative syntheses, forest plots were used to visually present the effect estimates with their 95% CI for individual studies as well as the pooled results. Given the inherent clinical and methodological heterogeneity in acupuncture trials (e.g., different acupoint combinations, needling techniques, and treatment parameters), a random-effects model was applied regardless of the degree of statistical heterogeneity.

Heterogeneity was assessed using the I² statistic. When substantial heterogeneity was detected, prespecified subgroup analyses were performed based on acupuncture type (e.g., manual acupuncture, electroacupuncture), treatment frequency, sessions, duration, and type of control medication (SSRIs, SNRIs, 5-HT1A receptor partial agonists, benzodiazepines, or others). Sensitivity analyses were conducted by excluding individual studies one at a time to evaluate the robustness of the pooled results and to explore potential sources of heterogeneity. If the overall effect estimates remained consistent after sequential exclusion, the results were considered stable.

Publication bias was assessed for the primary outcome (HAMA score) only when at least ten studies were included. Funnel plots were generated using R software, and Egger’s regression test was performed to statistically examine potential small-study effects.

Certainty of evidence

The certainty of evidence for all outcomes was assessed using the GRADE approach [15]. Five domains were evaluated: risk of bias, inconsistency, indirectness, imprecision, and publication bias. Evidence quality was downgraded when concerns were identified and upgraded when large effect sizes, dose–response relationships, or residual confounding suggested underestimation of the true effect. The certainty of evidence for each outcome was classified as high, moderate, low, or very low.

Results

Search results

A total of 4,471 records were identified from nine electronic databases, and 20 additional records were obtained from other sources. After removing duplicates, 2,691 records remained for screening. Following title and abstract screening, 124 full-text articles were assessed for eligibility (Supplementary Material 5). In the end, 41 studies were included [7, 16–55], of which 39 were published in Chinese and 2 in English. The kappa statistic for agreement between the two independent reviewers was 0.605, indicating moderate-to-good agreement. Detailed information on the search process is presented in the PRISMA 2020 flow diagram (Fig. 1).

Fig. 1.

PRISMA 2020 flow diagram of study selection

Characteristics of included studies

The characteristics of the 41 included studies are summarized in Table 1. Sample sizes ranged from 30 to 220 participants, with a total of 3,209 participants (1,685 in the intervention groups and 1,524 in the control groups). The mean age of participants ranged from 16 to 60 years. Regarding study design, 22 trials compared acupuncture with medication (stand-alone design), 18 trials compared acupuncture plus medication with medication (add-on design), and 4 trials compared acupuncture with sham acupuncture (sham-controlled design). Treatment duration varied from 2 to 12 weeks, with the number of sessions ranging from 6 to 60 and frequency ranging from once to seven times per week.

Table 1.

Characteristics of included studies

Study ID	Funding	Diagnostic criteria	Age (years)	Gender (male/female)	Sample (I/C)	Frequency, duration, and session	Intervention (acupoints and needle retention times)	Comparison	Outcome measures
Wang YN 2010 [16]	NR	CCMD-3	I:48.9 ± 10.1 C:50.2 ± 10.4	I:11/19 C:9/21	30/30	6/w, 3w, 18 times.	Acupuncture BL15 (Xinshu), BL18 (Ganshu), BL19 (Danshu), BL17 (Geshu), BL23 (Shenshu); DU20 (Baihui), HT7 (Shenmen), PC6 (Neiguan), KI3 (Taixi), LR3 (Taichong), BL20 (Pishu), BL21 (Weishu) etc. 40 min	Medication: Alprazolam 0.4 ~ 0.8 mg tid	HAMA, SAS, total effective rate
Li XJ 2019 [17]	NR	CCMD-3	I:53.41 ± 8.53 C:51.24 ± 8.37	I:8/21 C:9/20	30/30 dropout: I:1, C:1	Qd, 30 d, 30 times.	Acupuncture DU20 (Baihui), DU20 (Baihui) left and right 1 unit of length, DU24 (Shenting), DU24 (Shenting) left and right 1 unit of length, EX-HN3 (Yintang), EX-HN3 (Yintang) left and right 1 unit of length, LI4 (Hegu), LR3 (Taichong), LI4 (Hegu), LR3 (Taichong). 30 min	Medication: Buspirone 5 mg tid	HAMA, SAS, total effective rate
Shi CP 2010 [18]	NR	CCMD-3	I:47.66 ± 10.74 C:43.46 ± 11.29	I:9/20 C:11/17	30/30 dropout: I:1, C:1	Qd, 4w, 28 times.	Acupuncture DU20 (Baihui), DU23 (Shangxing), DU24 (Shenting), DU26 (Shuigou), RN17 (Shanzhong), RN14 (Juque), RN12 (Zhongwan). 30 min	Medication: Flupentixol and Melitracen Tablets 10.5 mg bid	HAMA, SAS, total effective rate, TESS
Che LN 2015 [19]	NR	CCMD-3	43.2 ± 7.8	32/48	40/40	6/w, 6w, 36 times.	Acupuncture DU20 (Baihui), DU23 (Shangxing), DU24 (Shenting), DU26 (Shuigou), RN17 (Shanzhong), RN14 (Juque), RN12 (Zhongwan); others were selected by syndrome differentiation. 30 min	Medication: Paroxetine 20 mg qd	HAMA, SAS, total effective rate, TESS
Sun XW 2014 [20]	NR	CCMD-3	I:47.57 ± 12.16 C:46.87 ± 10.78	I:12/18 C:11/19	30/30	Qd, 4w, 28 times.	Acupuncture DU20 (Baihui), DU24 (Shenting), EX-HN3 (Yintang), ST8 (Touwei), DU26 (Shuigou), RN17 (Shanzhong), LR3 (Taichong), BL18 (Ganshu). 50 min	Medication: Flupentixol and Melitracen Tablets 21 mg qd	HAMA, total effective rate
Xu Y 2021 [21]	NR	CCMD-3	I:43.52 ± 10.20 C:44.32 ± 11.08	I:22/20 C:19/23	42/42	5/w, 4w, 20 times.	Acupuncture DU20 (Baihui), DU26 (Shuigou), HT7 (Shenmen), PC6 (Neiguan), LR3 (Taichong). 30 min	Medication: Paroxetine 20 mg ~ 40 mg/d	HAMA, SAS, total effective rate
Liu EJ 2020 [22]	Y	CCMD-3	I:38 ± 3 C:38 ± 3	I:20/30 C:22/28	50/50	Qd, 60 d, 60 times.	Acupuncture DU20 (Baihui), EX-HN1 (Sishencong), EX-HN3 (Yintang), DU24 (Shenting), EX-HN5 (Taiyang), HT7 (Shenmen), HT4 (Lingdao), PC6 (Neiguan), PC9 (Zhongchong), PC7 (Daling), SP6 (Sanyinjiao), LR3 (Taichong). 30 min	Medication: Tandospirone 10 mg tid	HAMA, total effective rate, TESS
Wang GM 2007 [23]	NR	CCMD-3	I:16 ~ 63 C:16 ~ 64	I:9/12 C:8/12	21/20	Qd, 30 d, 30 times.	Acupuncture DU20 (Baihui), EX-HN3 (Yintang), HT7 (Shenmen), SP6 (Sanyinjiao). 45 min	Medication: Flupentixol and Melitracen Tablets 10.5 mg bid	HAMA, total effective rate
Zhao Y 2014 [24]	NR	CCMD-3	I:42.31 ± 8.44 C:41.59 ± 7.89	I:11/19 C:13/17	30/30	NR, 6w, NR.	Acupuncture EX-HN1 (Sishencong) to DU20 (Baihui), DU24 (Shenting), DU26 (Shuigou), GB13 (Benshen), GB20 (Fengchi), PC6 (Neiguan), HT7 (Shenmen). 50 min	Medication: Paroxetine 20 mg qd	HAMA, SAS, total effective rate, TESS
Wang ZL 2005 [25]	NR	CCMD-3	NR	NR	35/30	Qd, 30 d, 30 times.	Acupuncture EX-HN3 (Yintang), DU20 (Baihui), PC6 (Neiguan), HT7 (Shenmen), RN17 (Shanzhong), SP6 (Sanyinjiao); others were selected by syndrome differentiation. 30 min	Medication: clomazone 0.5 mg ~ 2 mg bid/tid + oryzanol 20 mg tid or propranolol 10 mg ~ 20 mg tid	SAS, total effective rate
Zhou XF 2013 [26]	Y	CCMD-3	I:38 ± 8 C:38 ± 10	I:16/24 C:18/22	40/40	Qd, 6w, 42 times.	Acupuncture LU7 (Lieque), LI4 (Hegu), HT7 (Shenmen), SI3 (Houxi), SJ5 (Waiguan), PC8 (Laogong), KI1 (Yongquan), BL54 (Zhibian), SP9 (Yinlingquan), ST36 (Zusanli), LR3 (Taichong), GB34 (Yanglingquan). Penetration needling: LI4 (Hegu) through PC8 (Laogong), SJ5 (Waiguan) through PC6 (Neiguan), GB34 (Yanglingquan) through SP9 (Yinlingquan), LR3 (Taichong) through KI1 (Yongquan)	Medication: Clonazepam 2 ~ 8 mg/d	HAMA
Gu X 2019 [27]	NR	CCMD-3	I:46.97 ± 9.35 C:48.83 ± 9.64	I:12/18 C:10/20	30/30	5/w, 4w, 20 times.	Acupuncture RN12 (Zhongwan), ST25 (Tianshu), RN6 (Qihai), ST36 (Zusanli), ST37 (Shangjuxu), ST39 (Xiajuxu); GB34 (Yanglingquan), GB39 (Xuanzhong), GB13 (Benshen), LR3 (Taichong), DU20 (Baihui). 30 min	Medication: Flupentixol and Melitracen Tablets 10.5 mg ~ 21 mg/d	HAMA, SAS, total effective rate
Xiong YQ 2013 [28]	NR	CCMD-3	16 ~ 55	32/39	36/35	Qd ~ q2d, 60 d, 30 ~ 60 times.	Acupuncture RN4 (Guanyuan), RN12 (Zhongwan), RN15 (Jiuwei), RN17 (Shanzhong), RN24 (Chengjiang), DU20 (Baihui), DU24 (Shenting), DU16 (Fengfu), DU14 (Dazhui), DU11 (Shendao), DU9 (Zhiyang), DU4 (Mingmen); others were selected by syndrome differentiation. 30 ~ 60 min	Medication: Flupentixol and Melitracen Tablets 10.5 mg bid + oryzanol 10 mg tid	SAS, total effective rate
Liu C 2023 [29]	NR	DSM-5	I:39.16 ± 10.66 C:39.93 ± 9.43	I:15/16 C:16/14	33/33 dropout: I:2, C:3	Qd, 4w, 28 times.	Acupuncture BL52 (Zhishi), BL49 (Yishe), SP3 (Taibai), KI3 (Taixi), ST36 (Zusanli), RN4 (Guanyuan), RN4 (Guanyuan), EX-HN3 (Yintang), DU20 (Baihui).Moxibustion applied on SP3, KI3, ST36. 30 min	Medication: Buspirone 5 mg tid for the first week, then increase to 10 mg tid for the second week.	HAMA, SAS, total effective rate, TESS
Zhao RZ 2018 [30]	Y	CCMD-3	I:42.59 ± 8.17 C:43.31 ± 7.96	I:20/35 C:23/36	60/60 dropout: I:7, C:1	NR, 6w, NR.	Electroacupuncture EX-HN3 (Yintang), DU20 (Baihui), GB20 (Fengchi), PC6 (Neiguan), SP6 (Sanyinjiao), LR3 (Taichong). Connect EA to EX-HN3, DU20, and GB20. 30 min	Medication: Buspirone 15 mg/d	HAMA, total effective rate, TESS
Wang CY 2003 [31]	NR	CCMD-3	I:36.4 ± 7.4 C:36.5 ± 8.6	I:12/18 C:13/19	30/32 dropout: I:0, C:3	5/w, 6w, 30 times.	Electroacupuncture EX-HN3 (Yintang), DU20 (Baihui), GB5 (Xuanlu), GB20 (Fengchi). 6–12 Hz, 3–10 V. 45 min	Medication: Trazodone 50 mg ~ 150 mg/d	HAMA, SAS, total effective rate
Chen Y 2013 [32]	NR	CCMD-3	I:45.33 ± 11.04 C:45.09 ± 10.74	I:11/22 C:10/23	33/33 dropout: I:1, C:2	5/w, 6w, 30 times.	Electroacupuncture + ear acupuncture DU20 (Baihui), EX-HN1 (Sishencong). EA on EX-HN3 (Yintang), DU20 (Baihui), 2–15 Hz. Auricular points: Jiaogan, Shenmen, Gan, Dan, Pi, Wei, Nao, Xin, Neifenmi	Medication: Paroxetine 10 mg ~ 20 mg/d	HAMA, total effective rate
Sheng GB 2015 [33]	NR	CCMD-3	Ia:24 ~ 64 Ib:18 ~ 62 C:22 ~ 60	Ia:9/21 Ib:7/23 C:12/18	30/30/30	6/w, 6w, 36 times.	a) Electroacupuncture: Jiaji points T3–L2 (EX-B2 Jiajixue), emotional scalp area. 30 min. b) Acupuncture: PC6 (Neiguan), HT7 (Shenmen), ST36 (Zusanli), SP6 (Sanyinjiao), scalp area. 30 min	Medication: Paroxetine 20 mg bid	total effective rate
Li JM 2011 [34]	NR	CCMD-3	I:41.00 ± 10.72 C:43.87 ± 12.56	I:11/19 C:6/24	30/30 dropout: I:2, C:2	Qd (first 3 days), 1/w. 4w, 6 times.	Pricking blood therapy BL17 (Geshu), BL19 (Danshu).Pricking blood on the points, pull out the glass cupping cup at the acupoint, and the bleeding volume within 1 ~ 2mL each time	Medication: Alprazolam 0.4 mg qd	HAMA, SAS, total effective rate
Zhou BH 2015 [35]	NR	CCMD-3	I:33.97 ± 10.18 C:36.67 ± 10.54	I:10/20 C:8/22	30/30	Q2d, 8w, 28 times.	Acupuncture + medication DU19 (Houding), DU21 (Qianding), BL8 (Luoque), DU24 (Shenting), EX-HN3 (Yintang), HT7 (Shenmen), SP6 (Sanyinjiao), LI4 (Hegu), LR3 (Taichong); others were selected by syndrome differentiation. 30 min. Medication: same with comparison	Medication: Paroxetine 20 mg ~ 40 mg qd	HAMA, total effective rate
Deng XY 2009 [36]	NR	CCMD-3	I:46.25 ± 11.06 C:44.63 ± 9.54	I:9/19 C:9/18	30/30 dropout: I:2, C:3	Q2d, 30 d, 15 times.	Acupuncture + medication DU20 (Baihui), EX-HN3 (Yintang), EX-HN1 (Sishencong), GB13 (Benshen), HT7 (Shenmen), PC6 (Neiguan), LR3 (Taichong), ST36 (Zusanli), SP6 (Sanyinjiao). 30 min. Medication: same with comparison	Medication: Flupentixol and Melitracen Tablets 10.5 mg bid	HAMA, total effective rate, CGI, TESS
Xu MM 2016 [37]	NR	CCMD-3	I:38(25 ~ 60) C:38(23 ~ 61)	I:14/16 C:18/12	30/30	Qd, 4w, 28 times.	Acupuncture + medication Scalp “emotional area”. 120 min. Medication: same with comparison	Medication: Buspirone 10 ~ 40 mg/d	HAMA, total effective rate
Liu K 2021 [38]	NR	DSM-5	I:39.1 ± 10.4 C:38.6 ± 8.3	I:10/20 C:11/19	30/30	5/w, 12w, 60 times.	Acupuncture + medication PC6 (Neiguan), RN17 (Shanzhong), DU26 (Shuigou), SP6 (Sanyinjiao), LR3 (Taichong), DU20 (Baihui), HT7 (Shenmen); others were selected by syndrome differentiation. 20 min. Medication: same with comparison	Medication: Chlorpromazine 12.5 mg ~ 50.0 mg bid	HAMA, total effective rate
Sui AM 2010 [39]	NR	CCMD-3	I:39.55 ± 11.5 C:37.24 ± 10.1	I:15/23 C:14/24	38/38	6/w, 6w, 36 times.	Acupuncture + medication Sishencong, Dingshen, DU26 (Shuigou), PC6 (Neiguan), HT7 (Shenmen), SP6 (Sanyinjiao); others were selected by syndrome differentiation. Medication: Paroxetine 20 mg qd; alprazolam 0.4 ~ 0.8 mg qd or Lola 0.5 ~ 1 mg qd	Medication: Paroxetine 20 ~ 40 mg qd; alprazolam 0.4 ~ 1.2 mg qd or Lola 0.5 ~ 2 mg qd	HAMA, SAS, total effective rate, TESS
Zhang HX 2019 [40]	NR	ICD-10	I:51.89 ± 9.40 C:48.22 ± 9.38	I:0/35 C:0/37	35/37	Qd, 2w, 14 times.	Acupuncture + medication LR3 (Taichong), LI4 (Hegu), CV12 = RN12 (Zhongwan), CV10 = RN10 (Xiawan), CV6 = RN6 (Qihai), CV4 = RN4 (Guanyuan), ST25 (Tianshu), SP15 (Daheng). 30 min. Medication: same with comparison	Medication: Escitalopram Oxalate Tablets 10 mg qd	SAS, total effective rate
Guan J 2021 [41]	NR	DSM-3	I:44.91 ± 16.69 C:43.42 ± 16.01	I:10/24 C:12/21	35/35 dropout: I:1, C:2	3/w, 6w, 18 times.	Acupuncture + medication Forehead midline, vertex midline, RN12 (Zhongwan), HT7 (Shenmen), SP6 (Sanyinjiao), ST36 (Zusanli), LR3 (Taichong). 30 min. Medication: same with comparison	Medication: Paroxetine 20 mg qd increase by 10 mg per week, maximum dose of 40 mg qd	HAMA, SAS, total effective rate, TESS
Lu H 2022 [42]	Y	CCMD-3	I:47.32 ± 2.54 C:47.82 ± 2.74	I:34/26 C:32/28	60/60	Qd, 20 days, 20 times.	Acupuncture + medication GV29 = DU29 (Suliu), HT7 (Shenmen), KI6 (Zhaohai), BL62 (Shenmai), LR2 (Xingjian), GB43 (Xiaxi). Press needles. 24 h. Medication: same with comparison	Medication: Escitalopram Oxalate Tablets 10 mg qd	HAMA, SAS, total effective rate
Sun QY 2024 [43]	Y	DSM-5	I:35.24 ± 8.19 C:34.71 ± 8.24	I:24/35 C:21/38	59/59	Q2d, 4w, 14 times	Acupuncture + medication DU20 (Baihui), HT7 (Shenmen), KI3 (Taixi, bilateral), BL15 (Xinshu), BL18 (Ganshu), BL20 (Pishu), BL23 (Shenshu), DU24 (Shenting), LR3 (Taichong). 30 min. Medication: same with comparison	Medication: Tandospirone Citrate Capsules 10 mg tid	HAMA, SAS, total effective rate
Xu WC 2021 [44]	NR	DSM-5; CCMD-3	I:47.23 ± 8.842 C:48(31.75 ~ 55.25)	I:7/23 C:11/19	30/30	Q3d, 4w, 10 times.	Acupuncture + moxibustion + medication “Navel inner ring”, “Faxuan”, “Anmian three points”, PC6 (Neiguan), LR3 (Taichong), KI7 (Fuliu), SP6 (Sanyinjiao), ST36 (Zusanli). 30 min. Moxibustion: PC6, LR3, KI7, SP6, ST36. 2 ~ 3 times. Medication: same with comparison	Medication: Flupentixol and Melitracen Tablets 10.5 mg ~ 21 mg/d	HAMA, total effective rate
Gong Y 2012 [45]	NR	CCMD-3	I:30.85 ± 8.12 C:31.17 ± 7.24	I:29/27 C:32/25	56/57	5/w, 6w, 30 times.	Electroacupuncture + medication DU20 (Baihui), DU24 (Shenting), EX-HN1 (Sishencong), EX-HN3 (Yintang), PC6 (Neiguan), HT7 (Shenmen), KI6 (Zhaohai), BL62 (Shenmai), SP6 (Sanyinjiao), BL15 (Xinshu), BL18 (Ganshu), BL23 (Shenshu). 30 min. Medication: same with comparison	Medication: Duloxetine 20 mg ~ 60 mg/d, average 40 mg/d	HAMA, total effective rate, CGI, TESS
Fan CH 2014 [46]	Y	DSM-4; CCMD-3	I:44.58 ± 14.99 C:44.36 ± 15.23	I:19/21 C:17/24	40/41	5/w, 6w, 30 times.	Electroacupuncture + medication DU20 (Baihui), EX-HN1 (Sishencong), HT7 (Shenmen), PC6 (Neiguan), SP6 (Sanyinjiao). 30 min. Medication: same with comparison	Medication: Paroxetine 20–40 mg/d, briefly given sedative-hypnotic drugs if necessary	HAMA, total effective rate
Liu LY 2013 [47]	Y	CCMD-3	I:35.6 ± 8.5 C:37.2 ± 9.0	I:38/72 C:30/80	110/110 dropout: I:5, C:12	Qd, 6w, 42 times.	Acupoint transcutaneous electrical nerve stimulation + medication PC6 (Neiguan), bilateral. 30 min. Medication: same with comparison	Medication: Sertraline 50 mg ~ 100 mg/d qd	HAMA, total effective rate, TESS
Zheng XY 2020 [48]	NR	CCMD-3	I:45.50 ± 9.77 C:43.31 ± 9.56	I:19/13 C:21/11	32/32	Q3d, 6w, 14 times.	Cupping + medication RN8 (Shenque)–RN10 (Xiawan), RN8 (Shenque)–ST25 (Tianshu), RN8–RN6 (Qihai); RN10–RN12, ST25–SP15, RN6–RN4, RN12–RN14, SP15–GB26, RN4–RN2 (Qugu). Flash the cup 3 ~ 4 times, and run the cup for 5 ~ 8 min at a speed of 10 ~ 25 cm/s, until the local skin becomes ruddy or hyperemia. Medication: same with comparison	Medication: Sertraline 50 mg qd	HAMA, total effective rate, CGI
Liang HH 2020 [49]	NR	CCMD-3	Ia:50.36 ± 11.09 Ib:49.85 ± 11.35 C:51.03 ± 12.74	Ia:14/19 Ib:13/20 C:13/19	38/38/38 dropout: Ia:5, Ib:5, C:6	4/w, 6w, 24 times.	a) Acupuncture + medication: LR3 (Taichong), SP6 (Sanyinjiao), PC6 (Neiguan), HT7 (Shenmen), BL15 (Xinshu) b) Acupuncture + medication: LR3 (Taichong), SP6 (Sanyinjiao), DU20 (Baihui), EX-HN3 (Yintang), DU24 (Shenting), EX-HN1 (Sishencong), PC6 (Neiguan), HT7 (Shenmen), BL15 (Xinshu) 30 min. Medication: same with comparison	Medication: Paroxetine 20 mg ~ 50 mg/d	HAMA, SAS, total effective rate, TESS
Wei D 2017 [50]	NR	CCMD-3; ICD-10	18 ~ 60	Ia:11/11 Ib:11/11 Ic:11/11 C:11/11	22/22/22/22	Qd, 4w, 28 times.	a) Acupuncture: PC8 (Laogong), HT7 (Shenmen), PC6 (Neiguan), SP6 (Sanyinjiao), etc b) Shallow acupuncture + medication: same points, with 2 ~ 3 mm shallow needling c) Acupuncture + medication: same points 30 min. Medication: same with comparison	Medication: Flupentixol and Melitracen Tablets 10.5 mg qd	HAMA, total effective rate, TESS
Fu AN 2008 [51]	NR	CCMD-3	34.9 ± 11.1	23/37	20/20/20 dropout: Ia:2, Ib:1, C:2	6/w, 6w, 36 times.	(a) Acupuncture; (b) Acupuncture + medication BL15 (Xinshu), BL14 (Jueyinshu), BL18 (Ganshu), BL19 (Danshu), BL20 (Pishu), BL21 (Weishu), BL23 (Shenshu), and others were selected by syndrome differentiation. 45 min Medication: same with comparison	Medication: Paroxetine 20 mg qd or/and alprazolam 0.4 ~ 1.6 mg qd	HAMA, SAS, total effective rate, CGI, TESS
Liu HJ 2007 [52]	NR	CCMD-3	38.48 ± 15.80	30/56	29/28/29 dropout: Ia:2, Ib:2, C:3	6/w, 6w, 36 times.	(a) Acupuncture; (b) Acupuncture + medication Sishencong, Dingshen, PC6 (Neiguan), HT7 (Shenmen), SP6 (Sanyinjiao). 45 min. Medication: same with comparison	Medication: Fluoxetine or paroxetine 20 mg qd; and/or alprazolam 0.4 mg-1.6 mg qd	HAMA, SAS, total effective rate, CGI, TESS
Liu X 2019 [53]	NR	DSM-5	I:26.40 ± 3.80 C:28.40 ± 4.29	I:9/6 C:7/8	15/15	3/w, 8w, 24 times.	Acupuncture DU20 (Baihui), DU16 (Fengfu), BL15 (Xinshu), HT7 (Shenmen). 20 min	Sham acupuncture: non-acupoint surface points beside the acupoints in the acupuncture group, with the same operation.	HAMA, total effective rate, TESS
Mak 2019 [7]	Y	DSM-5	I:50.85 ± 11.57 C:50.83 ± 14.15	I:20/20 C:18/22	40/40 dropout: I:0, C:2	1/w, 10w, 10 times.	Electroacupuncture PC6 (Neiguan), HT7 (Shenmen); ST36 (Zusanli), ST37 (Shangjuxu); SP6 (Sanyinjiao), LR3 (Taichong); DU20 (Baihui), EX-HN3 (Yintang). 2-Hz continuous wave in the range of 0.5–1.5 mA	Sham EA: adopted non-penetrating sham acupuncture without electrical stimulation.	total effective rate
Huang HB 2023 [54]	Y	DSM-5, ICD-10	I:42.40 ± 10.72 C:39.07 ± 9.86	I:11/19 C:13/17	30/30	4/w, 2w, 8 times.	Acupuncture DU26 (Shuigou), PC7 (Daling), BL62 (Shenmai), DU16 (Fengfu), ST6 (Jiache), RN24 (Chengjiang), LI11 (Quchi). 30 min	Sham acupuncture: non-acupoint surface points 0.5 ~ 1.0 cm away from the acupoints in the acupuncture group, with the same operation.	HAMA
Liu X 2023 [55]	NR	DSM-5	I:49.00 (47.00, 52.00) C: 50.00 (48.00, 52.00) *	I:0/56 C:0/56	56/56 dropout: I:12, C:12	3/w, 4w, 12 times.	Acupuncture DU24 (Shenting), DU29 (Suliu), HT7 (Shenmen), SP6 (Sanyinjiao). 30 min	Sham acupuncture: blunt tipped sham needles that were not inserted into the skin. Acupoints were the same.	HAMA

Y, the study had funding; NR, not reference; I, intervention group; C, control group; qd, once a day; q2d, every 2 days; q3d, every 3 days; bid, twice a day; tid, three times a day; w, week; EA, electroacupuncture; HAMA, Hamilton Anxiety Rating Scale; SAS, Self-Rating Anxiety Scale; CGI, Clinical Global Impression; TESS, Treatment Emergent Symptom Scale; * Median (Q1, Q3)

Risk of bias of included studies

The results of the RoB 2.0 assessment are shown in Table 2. Overall, 39 studies were judged to raise “some concerns” and 2 studies were judged at “high risk” of bias. Most studies were assessed as low risk in the domains of bias due to deviations from intended interventions and bias due to missing outcome data. However, concerns were common in the domains of bias arising from the randomization process, bias in measurement of the outcome, and bias in selection of the reported result. One study [33] was rated at high risk for selective reporting because it reported only HAMA-defined response rates without providing HAMA scale scores. Another study [30] was judged at high risk due to a high number of post-randomization exclusions and dropouts without reported reasons, affecting the domains of randomization process and deviations from intended interventions.

Table 2.

Risk of bias assessment using Cochrane RoB 2.0 tool

Study ID	Bias arising from the randomization process	Bias due to deviations from intended interventions	Bias due to missing outcome data	Bias in measurement of the outcome	Bias in selection of the reported result	Total risk of bias
Che LN 2015 [19]	S	L	L	S	S	S
Chen Y 2013 [32]	S	L	L	L	S	S
Deng XY 2009 [36]	S	S	L	S	S	S
Fan CH 2014 [46]	S	L	L	L	S	S
Fu AN 2008 [51]	S	L	L	S	S	S
Gong Y 2012 [45]	S	L	L	S	S	S
Gu X 2019[27]	S	L	L	S	S	S
Guan J 2021 [41]	L	L	L	S	S	S
Huang HB 2023 [54]	S	L	L	S	S	S
Li JM 2011 [34]	L	L	L	S	S	S
Li XJ 2019 [17]	S	L	L	S	S	S
Liang HH 2020 [49]	L	L	L	L	S	S
Liu C 2023 [29]	S	L	L	S	S	S
Liu EJ 2020 [22]	S	L	L	S	S	S
Liu HJ 2007 [52]	L	L	L	S	S	S
Liu K 2021 [38]	S	L	L	S	S	S
Liu LY 2013 [47]	S	S	S	S	S	S
Liu X 2019 [53]	S	L	L	L	S	S
Liu X 2023 [55]	L	L	L	S	L	S
Lu H 2022 [42]	S	L	L	S	S	S
Mak 2019 [7]	L	L	L	S	L	S
Sheng GB 2015 [33]	S	L	L	S	H	H
Shi CP 2010 [18]	S	L	L	S	S	S
Sui AM 2010 [39]	S	L	L	S	S	S
Sun QY 2024 [43]	S	L	L	S	S	S
Sun XW 2014 [20]	S	L	L	S	S	S
Wang CY 2003 [31]	S	L	S	S	S	S
Wang GM 2007 [23]	S	L	L	S	S	S
Wang YN 2010 [16]	S	L	L	S	S	S
Wang ZL 2005 [25]	S	L	L	S	S	S
Wei D 2017 [50]	S	L	L	S	S	S
Xiong YQ 2013 [28]	S	L	L	S	S	S
Xu MM 2016 [37]	S	L	L	S	S	S
Xu WC 2021 [44]	L	L	L	S	S	S
Xu Y 2021 [21]	S	L	L	S	S	S
Zhang HX 2019 [40]	L	L	L	S	S	S
Zhao RZ 2018 [30]	H	S	H	S	S	H
Zhao Y 2014 [24]	S	L	L	S	S	S
Zheng XY 2020 [48]	S	L	L	S	S	S
Zhou BH 2015 [35]	S	L	L	S	S	S
Zhou XF 2013 [26]	S	L	L	S	S	S

L, low risk; S, some concerns; H, high risk

Clinical data

Stand-alone design

In the stand-alone design, nearly all outcomes showed significant effects (Table 3). Meta-analysis indicated that acupuncture was more effective than medication in reducing HAMA scores (MD −1.79, 95% CI −2.69 to −0.88; 1,224 participants; I² = 84%) and SAS scores (MD −3.66, 95% CI −5.71 to −1.61; 860 participants; I² = 88%), and in improving total effective rate (RR 1.10, 95% CI 1.06 to 1.15; 1,407 participants; I² = 0%). For CGI outcomes, no significant differences were observed in SI (MD 0.24, 95% CI −0.23 to 0.71; 89 participants; I² = 0%) or GI (MD 0.00, 95% CI −0.43 to 0.43; 89 participants; I² = 0%), whereas the EI was significantly higher in the acupuncture group (MD 1.62, 95% CI 1.12 to 2.11; 89 participants; I² = 0%). Subgroup analysis suggested that manual acupuncture was superior to medication, while electroacupuncture and pricking-blood therapy showed comparable efficacy to medication. Further subgroup analyses based on duration, number of sessions, frequency, and control medication demonstrated that manual acupuncture significantly improved HAMA scores (MD −2.31, 95% CI −3.38 to −1.24); both short courses (3–4 weeks) and long courses (8 weeks) were more effective than medication alone; acupuncture treatment with 18–30 sessions produced the most stable and significant effects; and a frequency of seven sessions per week was superior to lower-frequency interventions (Table 4). Sensitivity analysis showed that after excluding Che LN 2015 [19] and Xu Y 2021 [21], heterogeneity decreased and acupuncture remained superior in reducing HAMA scores (MD −1.06, 95% CI −1.63 to −0.48; 1,060 participants; I² = 45%).

Table 3.

Meta-analysis outcomes for acupuncture in generalized anxiety disorder

Variable	No. of trials	Participants	Heterogeneity (I2) a	Test for overall effect	95% CI
Stand-alone designs
HAMA score	19	1224	84%, P < 0.00001	Z = 3.88, P = 0.0001*	MD=−1.79 [−2.69,−0.88]
SAS score	14	860	88%, P < 0.00001	Z = 3.50, P = 0.0005*	MD=−3.66 [−5.71,−1.61]
Total effective rate	22	1407	0%, P = 0.88	Z = 4.51, P < 0.00001*	RR = 1.10 [1.06,1.15]
CGI score
SI	2	89	0%, P = 0.97	Z = 0.98, P = 0.33	MD = 0.24 [−0.23,0.71]
GI	2	89	0%, P = 0.62	Z = 0.01, P = 0.99	MD = 0.00 [−0.43,0.43]
EI	2	89	0%, P = 0.97	Z = 6.39, P < 0.00001*	MD = 1.62 [1.12,2.11]
Add-on Designs
HAMA score	19	1390	74%, P < 0.00001	Z = 5.68, P < 0.00001*	MD=−2.26 [−3.05,−1.48]
SAS score	8	568	95%, P < 0.00001	Z = 2.17, P = 0.03*	MD=−4.18 [−7.95,−0.41]
Total effective rate	20	1511	0%, P = 0.64	Z = 5.57, P < 0.00001*	RR = 1.11 [1.07,1.16]
CGI score
SI	3	144	27%, P = 0.26	Z = 0.90, P = 0.37	MD = 0.19 [−0.23,0.61]
GI	3	144	45%, P = 0.16	Z = 0.63, P = 0.53	MD=−0.15 [−0.61,0.31]
EI	5	333	84%, P < 0.0001	Z = 2.64, P = 0.008*	MD = 0.68 [0.17,1.18]
Sham Control Designs
HAMA score	3	178	0%, P = 0.95	Z = 5.21, P < 0.00001*	MD=−3.46 [−4.76,−2.16]
Total effective rate	2	110	89%, P = 0.003	Z = 0.45, P = 0.65	RR = 1.15 [0.62,2.14]

CI, confidence interval; MD, mean difference; RR, risk ratio; HAMA, Hamilton Anxiety Rating Scale; SAS, Self-Rating Anxiety Scale; CGI, Clinical Global Impression; SI, Severity of Illness; GI, Global Improvement; EI, Efficacy Index; TESS, Treatment Emergent Symptom Scale

All analyses were conducted using random-effects models. A negative effect size indicates superiority of the experimental treatments over the control treatments in HAMA, SAS, and CGI-SI/GI scores, whereas the opposite interpretation applies to total effective rate and CGI-EI scores

* indicates a statistically significant difference between groups

^a P values indicate whether statistical heterogeneity across the included trials is significant

Table 4.

Subgroup analysis of acupuncture efficacy

Variable	No. of trials	Participants	Heterogeneity (I2) a	Test for overall effect	95% CI	Test for subgroup differences (I2) b
Stand-alone designs
HAMA score	19	1224	84%, P < 0.00001	Z = 3.88, P = 0.0001 *	MD=−1.79 [−2.69,−0.88]
Intervention						81.7%, P = 0.004
Manual acupuncture	15	934	86%, P < 0.00001	Z = 4.22, P < 0.0001 *	MD=−2.31 [−3.38,−1.24]
Electroacupuncture	3	234	0%, P = 0.84	Z = 0.01, P = 0.99	MD=−0.00 [−0.87,0.86]
Pricking blood therapy	1	56	NA	Z = 0.27, P = 0.79	MD=−0.36 [−2.97,2.25]
Duration						57.3%, P = 0.07
3w	1	60	NA	Z = 2.42, P = 0.02 *	MD=−3.10 [−5.61,−0.59]
4w	9	521	77%, P < 0.0001	Z = 3.48, P = 0.0005 *	MD=−2.82 [−4.40,−1.23]
6w	8	543	84%, P < 0.00001	Z = 1.06, P = 0.29	MD=−0.74 [−2.10,0.63]
8w	1	100	NA	Z = 2.93, P = 0.003 *	MD=−0.98 [−1.64,−0.32]
Session						35.2%, P = 0.16
6 times	1	56	NA	Z = 0.27, P = 0.79	MD=−0.36 [−2.97,2.25]
18 ~ 20 times	3	204	89%, P = 0.0001	Z = 2.52, P = 0.01 *	MD=−3.64 [−6.46,−0.81]
28 ~ 30 times	8	443	57%, P = 0.02	Z = 2.55, P = 0.01 *	MD=−1.84 [−3.26,−0.43]
36 times	3	169	85%, P = 0.001	Z = 0.80, P = 0.42	MD=−1.83 [−6.29,2.64]
42 times	1	80	NA	Z = 3.42, P = 0.0006 *	MD=−0.83 [−1.31,−0.35]
60 times	1	100	NA	Z = 2.93, P = 0.003 *	MD=−0.98 [−1.64,−0.32]
NR (duration 6w)	2	172	0%, P = 0.70	Z = 0.12, P = 0.90	MD = 0.06 [−0.94,1.07]
Frequency						47.0%, P = 0.11
3 times qd + 1/w	1	56	NA	Z = 0.27, P = 0.79	MD=−0.36 [−2.97,2.25]
5/w	4	266	92%, P < 0.00001	Z = 1.28, P = 0.20	MD=−1.91 [−4.84,1.02]
6/w	4	229	79%, P = 0.003	Z = 1.57, P = 0.12	MD=−2.29 [−5.14,0.56]
7/w	8	501	54%, P = 0.03	Z = 3.74, P = 0.0002 *	MD=−1.58 [−2.41,−0.75]
NR	2	172	0%, P = 0.70	Z = 0.12, P = 0.90	MD = 0.06 [−0.94,1.07]
Control medication						13.3%, P = 0.33
Antidepressants	9	677	91%, P < 0.00001	Z = 2.23, P = 0.03 *	MD=−1.86 [−3.49,−0.23]
SSRIs combined with Benzodiazepines	2	89	13%, P = 0.28	Z = 0.34, P = 0.73	MD = 0.52 [−2.44,3.48]
Benzodiazepines	3	196	38%, P = 0.20	Z = 1.85, P = 0.06	MD=−1.16 [−2.38,0.07]
Flupentixol and Melitracen Tablets	5	262	53%, P = 0.08	Z = 3.03, P = 0.002 *	MD=−2.34 [−3.85,−0.83]
SAS score	14	860	88%, P < 0.00001	Z = 3.50, P = 0.0005 *	MD=−3.66 [−5.71,−1.61]
Intervention						46.7%, P = 0.15
Manual acupuncture	12	745	89%, P < 0.00001	Z = 3.62, P = 0.0003 *	MD=−4.06 [−6.26,−1.86]
Electroacupuncture	1	59	NA	Z = 0.24, P = 0.81	MD=−0.39 [−3.52,2.74]
Pricking blood therapy	1	56	NA	Z = 0.27, P = 0.79	MD=−1.04 [−8.55,6.47]
Duration						83.6%, P = 0.0004
3w	1	60	NA	Z = 2.56, P = 0.01 *	MD = 6.30 [1.48,11.12]
4w	7	441	84%, P < 0.00001	Z = 4.17, P < 0.0001 *	MD=−5.73 [−8.42,−3.04]
6w	5	288	87%, P < 0.00001	Z = 1.74, P = 0.08	MD=−2.68 [−5.71,0.34]
8w	1	71	NA	Z = 1.31, P = 0.19	MD=−3.20 [−7.99,1.59]
Session						0%, P = 0.43
6 times	1	56	NA	Z = 0.27, P = 0.79	MD=−1.04 [−8.55,6.47]
18 ~ 20 times	3	204	96%, P < 0.00001	Z = 0.53, P = 0.59	MD=−1.81 [−8.43,4.82]
28 ~ 30 times	5	300	79%, P = 0.0007	Z = 3.07, P = 0.002 *	MD=−5.30 [−8.68,−1.91]
36 times	3	169	79%, P = 0.008	Z = 1.53, P = 0.13	MD=−3.72 [−8.50,1.06]
NR	2	131	0%, P = 0.61	Z = 3.79, P = 0.0002 *	MD=−1.98 [−3.00,−0.95]
Frequency						57.6%, P = 0.05
3 times qd + 1/w	1	56	NA	Z = 0.27, P = 0.79	MD=−1.04 [−8.55,6.47]
5/w	4	274	92%, P < 0.00001	Z = 1.64, P = 0.10	MD=−3.65 [−8.00,0.70]
6/w	4	229	91%, P < 0.00001	Z = 0.34, P = 0.74	MD=−1.08 [−7.37,5.21]
7/w	4	241	62%, P = 0.05	Z = 4.45, P < 0.00001*	MD=−6.63 [−9.55,−3.71]
NR	1	60	NA	Z = 3.59, P = 0.0003*	MD=−1.92 [−2.97,−0.87]
Control medication						17.8%, P = 0.30
Antidepressants	6	402	94%, P < 0.00001	Z = 3.25, P = 0.001 *	MD=−5.52 [−8.84,−2.19]
SSRIs combined with Benzodiazepines	2	89	0%, P = 0.95	Z = 0.73, P = 0.46	MD=−1.26 [−4.63,2.11]
Benzodiazepines	3	181	91%, P < 0.0001	Z = 0.09, P = 0.93	MD=−0.41 [−9.01,8.18]
Flupentixol and Melitracen Tablets	3	188	2%, P = 0.36	Z = 3.23, P = 0.001*	MD=−2.73 [−4.38,−1.07]
Total effective rate	22	1407	0%, P = 0.88	Z = 4.51, P < 0.00001*	RR = 1.10 [1.06,1.15]
Intervention						0%, P = 0.91
Manual acupuncture	17	1054	0%, P = 0.83	Z = 4.17, P < 0.0001*	RR = 1.11 [1.06,1.16]
Electroacupuncture	4	293	13%, P = 0.33	Z = 1.47, P = 0.14	RR = 1.09 [0.97, 1.22]
Pricking blood therapy	1	60	NA	Z = 0.77, P = 0.44	RR = 1.20 [0.76,1.90]
Duration						0%, P = 0.59
3w	1	60	NA	Z = 0.76, P = 0.45	RR = 1.08 [0.88, 1.32]
4w	10	600	0%, P = 0.87	Z = 3.67, P = 0.0002 *	RR = 1.13 [1.06,1.21]
6w	9	576	0%, P = 0.51	Z = 2.78, P = 0.005*	RR = 1.11[1.03,1.19]
8w	2	171	0%, P = 0.87	Z = 0.70, P = 0.48	RR = 1.04 [0.93,1.15]
Session						0%, P = 0.54
6 times	1	60	NA	Z = 0.77, P = 0.44	RR = 1.20 [0.76, 1.90]
18 ~ 20 times	3	204	0%, P = 0.46	Z = 2.08, P = 0.04*	RR = 1.13[1.01,1.28]
28 ~ 30 times	9	524	0%, P = 0.82	Z = 3.58, P = 0.0003*	RR = 1.13 [1.06,1.21]
36 times	5	268	0%, P = 0.47	Z = 2.50, P = 0.01*	RR = 1.14 [1.03, 1.27]
60 times	1	100	NA	Z = 0.52, P = 0.60	RR = 1.05 [0.87, 1.26]
NR (duration 6w)	3	251	0%, P = 0.87	Z = 0.39, P = 0.70	RR = 1.02 [0.93, 1.12]
Frequency						0%, P = 0.72
3 times qd + 1/w	1	60	NA	Z = 0.77, P = 0.44	RR = 1.20 [0.76, 1.90]
5/w	5	343	12%, P = 0.34	Z = 2.42, P = 0.02*	RR = 1.11[1.02,1.21]
6/w	6	328	0%, P = 0.58	Z = 2.57, P = 0.01*	RR = 1.13[1.03,1.24]
7/w	8	496	0%, P = 0.85	Z = 2.97, P = 0.003*	RR = 1.11 [1.04, 1.19]
NR	2	180	0%, P = 0.69	Z = 0.02, P = 0.98	RR = 1.00 [0.87,1.16]
Control medication						0%, P = 0.56
5HT1A receptor partial agonist	4	346	0%, P = 0.58	Z = 1.12, P = 0.26	RR = 1.06 [0.96,1.18]
SARI (Trazodone)	1	62	NA	Z = 2.28, P = 0.02*	RR = 1.23[1.03, 1.46]
SSRIs	6	380	0%, P = 0.57	Z = 3.11, P = 0.002*	RR = 1.17 [1.06, 1.29]
SSRIs combined with Benzodiazepines	2	98	0%, P = 0.60	Z = 0.77 P = 0.44	RR = 1.06 [0.92,1.22]
Benzodiazepines	3	185	0%, P = 0.88	Z = 1.66 P = 0.10	RR = 1.11 [0.98, 1.27]
Flupentixol and Melitracen Tablets	6	336	0%, P = 0.63	Z = 2.18, P = 0.03*	RR = 1.08[1.01,1.16]
Standard of total effective rate						0%, P = 0.68
HAMA reduction rate > 20%	2	112	0%, P = 0.83	Z = 0.77 P = 0.44	RR = 1.04 [0.94,1.16]
HAMA reduction rate > 25%	13	816	0%, P = 0.84	Z = 3.82, P = 0.00031*	RR = 1.12[1.06,1.18]
HAMA reduction rate > 30%	6	414	0%, P = 0.47	Z = 2.19, P = 0.03*	RR = 1.12[1.01,1.24]
Clinical symptoms improve	1	65	NA	Z = 1.35, P = 0.18	RR = 1.13[0.95,1.35]
Add-on Designs
HAMA score	19	1390	74%, P < 0.00001	Z = 5.68, P < 0.00001*	MD=−2.26 [−3.05,−1.48]
Intervention						69.5%, P = 0.02
Manual acupuncture	15	929	66%, P = 0.0002	Z = 6.07, P < 0.00001 *	MD=−2.73 [−3.62,−1.85]
Electroacupuncture	2	194	12%, P = 0.29	Z = 1.40, P = 0.16	MD=−0.72 [−1.73,0.29]
TENS	1	203	NA	Z = 2.97, P = 0.003 *	MD=−1.73 [−2.87,−0.59]
Cupping	1	64	NA	Z = 2.28, P = 0.02 *	MD=−1.19 [−2.21,−0.17]
Duration						89.6%, P < 0.00001
20 days	1	120	NA	Z = 7.71, P < 0.00001 *	MD=−5.15 [−6.46,−3.84]
4w	6	359	0%, P = 0.59	Z = 6.81, P < 0.00001 *	MD=−2.36 [−3.04,−1.68]
6w	10	791	51%, P = 0.03	Z = 3.45, P = 0.0006 *	MD=−1.42 [−2.23,−0.62]
8w	1	60	NA	Z = 2.45, P = 0.01 *	MD=−2.24 [−4.03,−0.45]
12w	1	60	NA	Z = 8.05, P < 0.00001 *	MD=−5.10 [−6.34,−3.86]
Session						85.3%, P < 0.00001
10 ~ 15 times	4	297	23%, P = 0.27	Z = 5.39, P < 0.00001 *	MD=−1.96 [−2.68,−1.25]
18 ~ 24 times	4	285	51%, P = 0.11	Z = 5.40, P < 0.00001 *	MD=−3.81 [−5.19,−2.43]
28 ~ 30 times	6	380	41%, P = 0.13	Z = 2.85, P = 0.004*	MD=−1.53 [−2.59,−0.48]
36 times	3	165	0%, P = 0.68	Z = 0.13, P = 0.90	MD = 0.13 [−1.79,2.05]
42 times	1	203	NA	Z = 2.97, P = 0.003 *	MD=−1.73 [−2.87,−0.59]
60 times	1	60	NA	Z = 8.05, P < 0.00001 *	MD=−5.10 [−6.34,−3.86]
Frequency						40.4%, P = 0.15
2 ~ 4/w	5	369	47%, P = 0.11	Z = 5.89, P < 0.00001 *	MD=−2.22 [−2.96, −1.48]
4/w	3	153	0%, P = 0.70	Z = 3.08, P = 0.002 *	MD=−2.86 [−4.68,−1.04]
5/w	3	254	94%, P < 0.00001	Z = 1.44, P = 0.15	MD=−2.20 [−5.19,0.79]
6/w	3	165	0%, P = 0.68	Z = 0.13, P = 0.90	MD = 0.13 [−1.79,2.05]
7/w	5	449	79%, P = 0.0009	Z = 3.05, P = 0.002 *	MD=−2.96 [−4.87,−1.06]
Control medication						89.0%, P < 0.00001
SSRIs	2	187	75%, P = 0.04	Z = 4.64, P < 0.00001 *	MD=−4.22 [−6.00,−2.44]
Antidepressants	9	797	19%, P = 0.28	Z = 5.81, P < 0.00001 *	MD=−1.60 [−2.15, −1.06]
SSRIs combined with Benzodiazepines	3	165	0%, P = 0.68	Z = 0.13, P = 0.90	MD = 0.13 [−1.79,2.05]
Flupentixol and Melitracen Tablets	4	181	17%, P = 0.31	Z = 3.76, P = 0.0002 *	MD=−2.47 [−3.75,−1.18]
Chlorpromazine	1	60	NA	Z = 8.05, P < 0.00001 *	MD=−5.10 [−6.34,−3.86]
SAS score	8	568	95%, P < 0.00001	Z = 2.17, P = 0.03*	MD=−4.18 [−7.95,−0.41]
Intervention						99.3%, P < 0.00001
Manual acupuncture	7	448	0%, P = 0.44	Z = 7.10, P < 0.00001 *	MD=−3.49 [−4.45, −2.53]
Press needle	1	120	NA	Z = 22.34, P < 0.00001 *	MD=−12.16 [−13.23,−11.09]
Total effective rate	20	1511	0%, P = 0.88	Z = 4.51, P < 0.00001*	RR = 1.10 [1.06,1.15]
Intervention						8.0%, P = 0.35
Manual acupuncture	16	1033	0%, P = 0.92	Z = 4.50, P < 0.00001*	RR = 1.11 [1.06,1.16]
Electroacupuncture	2	194	81%, P = 0.02	Z = 0.82, P = 0.42	RR = 1.23 [0.74,2.05]
TENS	1	220	NA	Z = 3.40, P = 0.0007 *	RR = 1.25 [1.10, 1.42]
Cupping	1	64	NA	Z = 1.19, P = 0.23	RR = 1.11 [0.93, 1.32]
Duration						0%, P = 0.56
2w	1	72	NA	Z = 0.40, P = 0.69	RR = 1.03 [0.90,1.16]
20days	1	120	NA	Z = 2.37, P = 0.02*	RR = 1.16 [1.03,1.31]
4w	6	364	0%, P = 0.99	Z = 3.04, P = 0.002*	RR = 1.12 [1.04, 1.20]
6w	10	835	25%, P = 0.21	Z = 2.98, P = 0.003*	RR = 1.11 [1.04,1.19]
8w	1	60	NA	Z = 2.00, P = 0.05*	RR = 1.27 [1.01,1.61]
12w	1	60	NA	Z = 1.94, P = 0.02*	RR = 1.21 [1.00, 1.46]
Session						27.5%, P = 0.23
10 ~ 15 times	5	374	0%, P = 0.80	Z = 2.67, P = 0.008*	RR = 1.10 [1.02,1.17]
18 ~ 24 times	4	304	0%, P = 0.98	Z = 2.96, P = 0.003*	RR = 1.16 [1.05,1.29]
28 ~ 30 times	6	380	25%, P = 0.25	Z = 2.27, P = 0.02*	RR = 1.12 [1.02,1.23]
36 times	3	173	0%, P = 0.82	Z = 0.83, P = 0.41	RR = 1.04 [0.95,1.14]
42 times	1	220	NA	Z = 3.40, P = 0.0007*	RR = 1.25 [1.10,1.42]
60 times	1	60	NA	Z = 1.94, P = 0.05*	RR = 1.21 [1.00,1.46]
Frequency						0%, P = 0.53
2 ~ 3/w	5	372	0%, P = 0.88	Z = 3.55, P = 0.0004*	RR = 1.15 [1.06,1.24]
4/w	3	174	0%, P = 0.92	Z = 1.44, P = 0.15	RR = 1.13 [0.96,1.34]
5/w	3	254	65%, P = 0.06	Z = 1.53, P = 0.13	RR = 1.18 [0.95,1.45]
6/w	3	173	0%, P = 0.82	Z = 0.83, P = 0.41	RR = 1.04 [0.95,1.14]
7/w	6	538	17%, P = 0.30	Z = 3.39, P = 0.0007*	RR = 1.13 [1.05,1.21]
Control medication						10.1%, P = 0.35
5-HT1A receptor partial agonist	2	178	0%, P = 0.86	Z = 2.34, P = 0.02*	RR = 1.14 [1.02,1.26]
SNRIs	1	113	NA	Z = 0.58, P = 0.56	RR = 1.04 [0.91,1.18]
SSRIs	9	801	14%, P = 0.32	Z = 4.29, P < 0.0001*	RR = 1.16 [1.08,1.24]
SSRIs combined with Benzodiazepines	3	173	0%, P = 0.82	Z = 0.83, P = 0.41	RR = 1.04 [0.95,1.14]
Flupentixol and Melitracen Tablets	4	186	0%, P = 0.96	Z = 1.99, P = 0.05*	RR = 1.10 [1.00,1.22]
Chlorpromazine	1	60	NA	Z = 1.94, P = 0.05*	RR = 1.21 [1.00,1.46]
Standard of total effective rate						0%, P = 0.40
HAMA reduction rate > 25%	17	1250	0%, P = 0.71	Z = 4.76, P < 0.00001*	RR = 1.11 [1.06,1.15]
HAMA reduction rate > 50%	1	81	NA	Z = 2.05, P = 0.04*	RR = 1.57 [1.02,2.41]
SAS reduction rate > 25%	1	120	NA	Z = 2.37, P = 0.02*	RR = 1.16 [1.03,1.31]
Clinical symptoms reduction rate > 30%	1	60	NA	Z = 1.19, P = 0.23	RR = 1.12 [0.93,1.35]

NA, not applicable; CI, confidence interval; MD, mean difference; RR, risk ratio; HAMA, Hamilton Anxiety Rating Scale; SAS, Self-Rating Anxiety Scale; CGI, Clinical Global Impression; SI, Severity of Illness; GI, Global Improvement; EI, Efficacy Index; TESS, Treatment Emergent Symptom Scale

All analyses were conducted using random-effects models. A negative effect size indicates superiority of the experimental treatments over the control treatments in HAMA and SAS scores, whereas the opposite interpretation applies to total effective rate

* indicates a statistically significant difference between groups

^a P values for heterogeneity indicate whether the statistical heterogeneity among trials is significant (P ≥ 0.05 indicates no significant heterogeneity within the group)

^bP values for subgroup differences indicate whether the difference between subgroups is significant (P < 0.05 indicates significant subgroup differences)

Add-on design

In add-on designs, significant improvements were observed in HAMA scores (MD −2.26, 95% CI −3.05 to −1.48; 1,390 participants; I² = 74%), SAS scores (MD −4.18, 95% CI −7.95 to −0.41; 568 participants; I² = 95%), total effective rate (RR 1.11, 95% CI 1.07 to 1.16; 1,511 participants; I² = 0%), and CGI-EI scores (MD 0.68, 95% CI 0.17 to 1.18; 333 participants; I² = 84%) (Table 3). Although heterogeneity was high for HAMA and SAS, subgroup analyses stratified by intervention type, duration, number of sessions, frequency, and control medication indicated that acupuncture plus medication produced consistently better outcomes across different treatment courses, with significant benefits for ≤ 30 sessions or ≥ 42 sessions, and for frequencies of ≤ 4 or 7 sessions per week (Table 4). Sensitivity analyses showed that excluding Liu K 2021 [38] and Lu H 2022 [42] reduced heterogeneity, while acupuncture plus medication remained superior in reducing HAMA scores (MD −1.81, 95% CI −2.37 to −1.25; 1,210 participants; I² = 39%). For SAS scores, excluding Lu H 2022 [42] markedly reduced heterogeneity (MD −3.49, 95% CI −4.45 to −2.53; 448 participants; I² = 0%), possibly related to the use of press-needle intervention in that trial.

Sham-controlled design

In sham-controlled designs, three studies reported HAMA outcomes and two studies reported total effective rate. Acupuncture was more effective than sham acupuncture in reducing HAMA scores (MD −3.46, 95% CI −4.76 to −2.16; 178 participants), but no significant difference was found for total effective rate (RR 1.15, 95% CI 0.62 to 2.14; 110 participants) (Table 3).

Dropout rates

Dropout rates were reported in 15 trials. A total of 51 participants in the intervention groups and 55 in the control groups dropped out, with dropout rates of 3.14% and 3.75%, respectively. The odds ratio for dropout was 0.83 (95% CI 0.56 to 1.22), indicating no significant difference.

TESS and adverse events

Meta-analysis showed that compared with medication alone, both acupuncture (MD −3.31, 95% CI −5.12 to −1.50; 6 studies; 386 participants) and acupuncture plus medication (MD −1.39, 95% CI −2.02 to −0.76; 9 studies; 701 participants; I² = 0%) were associated with significantly lower TESS scores. Similarly, adverse event rates were lower in the acupuncture group (RR 0.33, 95% CI 0.22 to 0.49; 10 studies; 650 participants) and the acupuncture plus medication group (RR 0.67, 95% CI 0.52 to 0.86; 8 studies; 600 participants; I² = 0%). Detailed records of adverse events are presented in Table 5.

Table 5.

Adverse events reported across intervention designs

Study ID	Adverse events
Stand-alone designs
Shi CP 2010 [18]	I: 5 cases, including fainting (n = 1), mild hematoma (n = 4); C: 10 cases, including dizziness (n = 4), insomnia (n = 1), fatigue (n = 2), sweating (n = 1), thirst (n = 4), decreased appetite (n = 4), constipation (n = 3), diarrhea (n = 1)
Che LN 2015 [19]	I: 6 cases, including mild hematoma (n = 4), dizziness (n = 2); C: 12 cases, including insomnia (n = 4), dizziness (n = 4), diarrhea (n = 1), constipation (n = 1)
Zhao Y 2014 [24]	I: 2 cases, including fainting (n = 1), mild hematoma (n = 1); C: 8 cases, including dizziness (n = 2), insomnia (n = 1), sweating (n = 2), thirst (n = 2), decreased appetite (n = 1)
Gu X 2019 [27]	I: no side effects; C: 6 cases, including thirst and constipation (n = 3), abdominal distension and loss of appetite (n = 2), dizziness (n = 1)
Zhao RZ 2018 [30]	I: 3 cases, including fainting (n = 1), mild hematoma (n = 1), pain persisting to the next day (n = 1); C: 9 cases, including dizziness (n = 2), nausea (n = 1), insomnia (n = 1), sweating (n = 2), thirst (n = 2), decreased appetite (n = 1)
Wang CY 2003 [31]	I: no side effects; C: 18 cases, including dizziness, fatigue, thirst, etc
Li JM 2011 [34]	I: no side effects; C: gastrointestinal discomfort (n = 2), 1 of whom discontinued treatment
Liu C 2023 [29]	I: no side effects; C: 1 case, including dizziness, nausea and thirst
Add-on designs
Deng XY 2009 [36]	I: 2 cases, including thirst (n = 1), dizziness (n = 1), akathisia (n = 1); C: 8 cases, including headache (n = 2), loss of appetite (n = 1), dizziness (n = 4), hypoactivity (n = 1), thirst (n = 3), tremor (n = 2), sweating (n = 1), 1 diarrhea (n = 1)
Gong Y 2012 [45]	I: 13 cases, including nausea and vomiting (n = 4), thirst (n = 5), dizziness (n = 2), hand tremor (n = 1), fatigue (n = 1); C: 16 cases, including nausea and vomiting (n = 5), thirst (n = 4), dizziness (n = 2), hand tremor (n = 2), electrocardiogram changes (n = 2), elevated transaminase (n = 1)
Liu LY 2013 [47]	The main side effects of the two groups were nausea, diarrhea, insomnia, thirst, fatigue, etc
Zheng XY 2020 [48]	I: 7 cases, including mild insomnia (n = 3), mild thirst (n = 2), constipation (n = 1), mild diarrhea (n = 1); C: 8 cases, including mild insomnia (n = 2), mild thirst (n = 1), constipation (n = 3), mild diarrhea (n = 2)
Liang HH 2020 [49]	Ia: 4 cases, including hematoma (n = 1), palpitation (n = 1), pain and local numbness at the acupoints (n = 2); Ib: 6 cases, including hematoma (n = 4), palpitation (n = 1), pain and local numbness at the acupuncture site (n = 1); C: no side effects
Guan J 2021 [41]	I: 8 cases, including constipation (n = 1), somnolence (n = 1), thirst (n = 3), fatigue(n = 3); C: 15 cases, including constipation (n = 6), somnolence (n = 1), thirst (n = 4), fatigue(n = 4)
Sun QY 2024 [43]	I: 4 cases, including abdominal distension(n = 2), decreased appetite (n = 2); C:8 cases, including nausea(n = 2), emesis(n = 1), abdominal distension(n = 3), decreased appetite (n = 2)
Both stand-alone designs and add-on designs in a trial
Wei D 2017 [50]	Ia: 2 cases, including thirst (n = 1), dizziness (n = 1); Ib: no side effects; Ic: 5 cases, including thirst (n = 1), dizziness (n = 2), insomnia (n = 1), tremor (n = 1); C: 6 cases, including thirst (n = 2), dizziness (n = 1), insomnia (n = 1), tremor (n = 1), akathisia (n = 1)
Fu AN 2008 [51]	Ia: 5 cases, including fainting (n = 2), mild hematoma (n = 3); Ib: 11 cases, including dizziness (n = 3), fatigue (n = 2), thirst (n = 4), decreased appetite (n = 4), nausea (n = 5), constipation (n = 4); C: 18 cases, including dizziness (n = 6), fatigue (n = 4), thirst (n = 5), decreased appetite (n = 4), nausea (n = 5), constipation (n = 4), excessive sedation (n = 5)
Placebo-controlled designs
Liu X 2019 [53]	No side effects
Liu X 2023 [55]	A total of 16 participants reported bleeding after acupuncture

Publication bias

Funnel plots appeared symmetrical. Egger’s regression test showed an intercept of −0.91 (SE = 0.94, t = −0.97, df = 17, P = 0.35) for stand-alone designs, and 0.51 (SE = 1.19, t = 0.43, df = 17, P = 0.68) for add-on designs, indicating no evidence of publication bias (Fig. 2).

Fig. 2.

Funnel plots of HAMA scores: (a) stand-alone designs; (b) add-on designs

GRADE assessment results

According to the GRADE evaluation (Tables 6, 7 and 8), in the stand-alone design, evidence certainty was rated as low for HAMA and SAS, moderate for total effective rate, and very low to low for CGI-EI. In the add-on design, evidence certainty was moderate for HAMA and total effective rate, and low for SAS and CGI-EI. In the sham-controlled design, evidence certainty was moderate for HAMA and very low for total effective rate.

Table 6.

GRADE assessment of stand-alone designs

Table 7.

GRADE assessment of add-on designs

Table 8.

GRADE assessment of sham-controlled designs

Discussion

This meta-analysis suggests that, compared with medication, acupuncture alone or in combination with medication provides additional benefits in improving HAMA, SAS, and CGI-EI scores. In sham-controlled studies, acupuncture was more effective than sham acupuncture in reducing anxiety. Both acupuncture and acupuncture plus medication were associated with fewer adverse events than controls, indicating good tolerability and safety. These findings suggest that acupuncture may be an effective treatment for generalized anxiety disorder (GAD), although the overall certainty of evidence remains low and the results should be interpreted with caution.

Given the characteristics of acupuncture, treatment type (manual acupuncture, electroacupuncture, TENS, etc.) and treatment parameters (duration, frequency, number of sessions) have a direct impact on outcomes, warranting subgroup analyses. (1) Treatment type: In stand-alone designs, manual acupuncture was superior to medication, whereas electroacupuncture and pricking-blood therapy showed comparable efficacy. In add-on designs, combining manual acupuncture or TENS with medication produced better outcomes, while electroacupuncture did not show clear benefits, possibly due to small sample sizes. (2) Treatment parameters: In stand-alone designs, all evaluation time points (3, 4, 6, and 8 weeks) demonstrated efficacy. Short courses (3–4 weeks) produced rapid improvements, and 8-week courses showed sustained benefits, suggesting that acupuncture may act more quickly and maintain efficacy better than medication; however, due to the limited number of studies, these results should be interpreted cautiously. Treatment with 18–30 sessions appeared most stable, and even once-weekly acupuncture produced benefits, while daily acupuncture (7 sessions/week) showed the greatest effect superior to the control group, suggesting a possible dose–response relationship. In add-on designs, both short (3 weeks) and long (12 weeks) treatment courses showed greater effects, indicating potential short- and long-term advantages of combined therapy. Conventional 4–6 week regimens also maintained stable improvements. Regarding the number of sessions, 18–30 treatments were consistently effective, whereas 36 sessions did not show significant additional benefit but indicated a potential trend, which may be related to the wide data range of the included studies. Regarding treatment frequency, ≤ 4 sessions per week were effective, and high-frequency treatment (7 sessions/week) produced even greater benefits.

The type of control medication also influenced relative efficacy. When control groups received a single agent (e.g., SSRIs, 5-HT1A partial agonists, trazodone, or Flupentixol and Melitracen), acupuncture (alone or combined with medication) produced greater symptom relief. However, when benzodiazepines were used (either alone or in combination with antidepressants) in control group acupuncture showed no additional benefit compared with controls. This may reflect the strong and rapid anxiolytic effect of benzodiazepines, which makes the additional benefit of acupuncture less evident; conversely, when control treatments were less effective, the advantage of acupuncture was more apparent.

Compared with previous systematic reviews of acupuncture for GAD, this study has multiple advantages: (1) stricter inclusion and exclusion criteria, with a broader and more comprehensive literature search, resulting in a larger number of included trials; (2) more rigorous study design, incorporating stand-alone, add-on, and sham-controlled trials for more detailed and objective evaluation; (3) use of standardized pharmacotherapy in control groups without additional complementary therapies, strengthening the validity of comparisons; (4) subgroup analyses based on acupuncture modalities and treatment parameters, providing more targeted and clinically relevant findings; (5) systematic risk-of-bias assessment and evidence grading, enhancing the reliability of results; and (6) multidimensional outcome measures, including clinician-rated scales (HAMA, CGI), patient-reported scales (SAS), total effective rate, and adverse events, allowing for a more comprehensive evaluation of efficacy and safety.

Nevertheless, several limitations should be noted. First, most included RCTs were published in Chinese and were of generally low methodological quality, with frequent concerns regarding randomization, blinding, and outcome reporting. Nearly all studies were rated as having “some concerns” or “high risk” in multiple bias domains, raising the possibility of overestimation of treatment effects. Second, the number of sham-controlled trials was limited (only four), leaving the specific placebo effect of acupuncture insufficiently established. Third, statistical heterogeneity was high in several analyses. While subgroup and sensitivity analyses explained some variability, methodological and clinical heterogeneity inherent to acupuncture (e.g., variability in acupoints and needling techniques) could not be fully avoided. Inconsistencies across subgroups may also reflect the limited number and uneven quality of included studies, adding uncertainty to interpretation. Fourth, some trials had relatively short treatment courses, small numbers of sessions, or short follow-up periods, limiting assessment of long-term efficacy and stability. Finally, the search strategy did not employ controlled vocabulary comprehensively (i.e., Emtree terms in EMBASE and MeSH terms in the Cochrane Library), which may have resulted in missing some potentially relevant studies.

Future studies should address these limitations. More high-quality sham-controlled RCTs with adequate blinding are needed to confirm the specific effects of acupuncture. Future trials should adhere strictly to reporting standards, improving methodological quality in randomization, allocation concealment, and outcome assessment to minimize bias. Further research should also explore optimal acupuncture parameters, including treatment frequency, number of sessions, and follow-up duration, to establish dose–response relationships and best practice protocols. Large-scale trials with longer follow-up are warranted to assess long-term efficacy and safety, as well as cost-effectiveness, to better define the clinical value of acupuncture in the treatment of GAD.

Conclusion

This study suggests that acupuncture, either alone or combined with medication, is likely effective for generalized anxiety disorder and may be superior to medication alone. Acupuncture was also associated with fewer adverse events and better tolerability. However, given the limited number of sham-controlled trials, challenges in blinding, and the overall low certainty of evidence, these findings should be interpreted with caution. Further high-quality clinical research is warranted.

Abbreviations

CBM

Chinese biomedical literature database

CCMD

Chinese classification of mental disorders

CENTRA

Cochrane central register of controlled trials

CGI

Clinical global impression

CGI-EI

Clinical global impression-efficacy index

CGI-GI

Clinical global impression-global improvement

CGI-SI

Clinical global impression-severity of illness

CI

Confidence interval

CNKI

China national knowledge infrastructure

DSM

Diagnostic and statistical manual of mental disorders

EA

Electroacupuncture

GAD

Generalized anxiety disorder

GRADE

Grading of recommendations assessment, development and evaluation

HAMA

Hamilton anxiety rating scale

ICD

International classification of diseases

ITT

Intention-to-treat

MD

Mean difference

NA

Not applicable

NR

Not reported

PRISMA

Preferred reporting items for systematic reviews and meta-analyses

RCT

Randomized controlled trial

RoB

Cochrane risk of bias tool

RR

Risk ratio

SAS

Self-rating anxiety scale

SNRIs

Serotonin-norepinephrine reuptake inhibitors

SSRIs

Selective serotonin reuptake inhibitors

TENS

Transcutaneous electrical nerve stimulation

TESS

Treatment emergent symptom scale

VIP

VIP database for chinese technical periodicals

Author contributions

J. Lai identified and selected studies, conducted the meta-analysis, and drafted the manuscript; Y. Wang identified and selected studies, extracted data, and assessed risk of bias; X. Yao and J. Yu designed the study and registered the protocol; S. Lu extracted data and assessed risk of bias; J. Lu revised and approved the manuscript; J. Liu conceptualized the study, designed the research, and revised the manuscript. All authors reviewed the manuscript.

Funding

This study was supported by the High-Level Chinese Medicine Hospital Promotion Project (Grant No. HLCMHPP2023125).

Data availability

The search strategies for each database, the list of excluded studies with reasons, and all forest plots generated from the analyses have been provided as supplementary materials. The template data collection forms, extracted data from the included studies, and other materials used in this review are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.