The Comparative Efficacy of Multiple Acupuncture for Alzheimer's Disease: A Bayesian Network Meta-Analysis

Background Alzheimer's disease (AD) is a progressive neurodegenerative disease. Numerous cases have illustrated that the acupuncture method could improve AD patients' cognitive function and daily living ability. However, the optimal acupuncture treatments remain controversial. Therefore, we aimed to conduct a systematic review to compare the efficacy of multiple acupuncture therapies for AD and identify the optimal acupuncture intervention for delaying AD progression. Methods To select potentially concerned randomized controlled trials (RCTs), we searched four English databases, four Chinese databases, and additional sources from 1 May 2021. Two independent reviewers conducted study screening, data extraction, and methodological quality assessment. The primary outcome was global cognitive function improvement. Pairwise and Bayesian network meta-analyses were performed using STATA v15.0 and ADDIS v1.16.8. The Grading of Recommendations Assessment, Development, and Evaluation (GRADE) tool was used to assess the quality of evidence. Results This study included 34 RCTs with 2,071 participants. Regarding global cognitive function improvement, the pairwise meta-analysis confirmed that electronic acupuncture (EA) plus conventional medicine (CM) and manual acupuncture (MA) plus CM were statistically significantly different from CM, and EA plus CM was ranked as the best combination in the network meta-analysis. In terms of response rate, MA outperformed CM statistically significantly; warm acupuncture (WA) was ranked as the best in the network meta-analysis. Regarding activity of daily living improvement, EA plus CM, MA plus CM, and fire acupuncture plus CM, MA, and scalp acupuncture were statistically significantly different from CM, and EA plus CM was ranked as the best combination in the network meta-analysis. However, the evidences were ranked as low to critically low. Conclusions Acupuncture, as a monotherapy or an adjuvant therapy, may have a beneficial effect on efficacy for AD. EA plus CM may be the optimal acupuncture therapy for AD and should be administered to AD patients. It may aid and support patient, operative, and societal decision-making. Due to the dearth of high-quality evidence, additional high-quality studies should be conducted to ensure these findings in the future. This study is registered with PROSPERO (CRD42021252305).


Introduction
Alzheimer's disease (AD), the most prevalent disorder of dementia, is characterized by tau and amyloid β (Aβ) accumulation [1,2]. e disorder is manifested by a progressive decline of cognitive function [3]. Moreover, AD may negatively affect the activity of daily living (ADL) and psychological and behavioral conditions [4]. It is estimated that approximately 44 million people suffer from AD [5]. As the aging population proliferates worldwide, AD prevalence increases [6,7], particularly in China, which has experienced exponential growth [8,9]. In addition, it leads to a huge financial burden associated with AD control [9,10].
Besides, because AD's mechanism remains unclear, there is no specific remedy for the whole AD process [11]. Current interventions for AD included several conventional medicines (CM), which have been approved by the US Food and Drug Administration (FDA), such as acetylcholinesterase inhibitors (AChEIs), N-methyl-D-aspartate (NMDA) receptor, and monoclonal antibody [12]. However, some evidences illustrated that their effect was negligible [13,14]. Consequently, it is indispensable to seek new effective treatments for AD.
Notably, acupuncture has a long history of managing dementia [15,16]. Several systematic reviews and metaanalyses demonstrated that acupuncture was not inferior to pharmacotherapy in treating AD [17][18][19][20]. Furthermore, numerous articles have indicated that acupuncture may improve cholinergic neurotransmission, decrease Aβ protein concentration, and stimulate exciter motor-related brain regions associated with cognitive function [21][22][23]. However, several acupuncture treatments have been applied for AD, including manual acupuncture (MA), electronic acupuncture (EA), fire acupuncture (FA), warm acupuncture (WA), scalp acupuncture (SA), and so on. Because the most effective acupuncture technique is unknown, seeking the optimal acupuncture intervention for AD is critical.
is study aimed to conduct a Bayesian network metaanalysis (NMA) [24] to drastically compare and rank various acupuncture therapies for AD in improving global cognitive function and activity of daily living. Meanwhile, our findings provided a new reference for clinical decision-making regarding acupuncture for AD.

Methods
e NMA is registered on the PROSPERO platform (number: CRD42021252305) and reported following the preferred reporting items for systematic reviews and metaanalyses (PRISMA-NMA) checklist [25].

Types of Studies.
is study included all randomized controlled parallel trials published in English/Chinese, regardless of region or publication restriction. By the way, the randomized cross-over trials' first period would be covered. On the other hand, randomized controlled cluster trials, case reports, experts' experience, and so on were excluded.

Types of Participants.
is study included all participants with definite AD diagnostic criteria, regardless of their gender, country, ethnic origin, or severity. Participants with dementia who did not have a precise AD diagnosis were excluded.

Types of Intervention.
MA, EA, WA, FA, and SA are high-frequency acupuncture therapies. ese acupuncture treatments were regarded as either monotherapies or integrative therapies. Moreover, integrative treatments combining acupuncture and CM would be covered. e Chinese herb, exercise, music therapies, and so on would be excluded.

Types of Control Groups.
Numerous acupuncture treatments and CM (AChEIs, such as donepezil, huperzine, and rivastigmine) were used to form the basis.

Types of Outcome Measures.
We included studies that addressed one or more of the below-highlighted outcomes. Our primary outcome measure for NMA is global cognitive function improvement, as determined by Mini-Mental State Examination (MMSE) and Alzheimer's Disease Assessment Scale-Cognitive (ADAS-cog). Secondary outcomes included response rate and improvement in activity of daily living (ADL) as measured by ADL scales. Meanwhile, adverse events (AEs) would be included to measure intervention safety. Other outcomes of AD would be eliminated.

Study Selection and Data Extraction.
Two reviewers (ZY and LW) were trained on a professional course on NMA. ZY and LW independently screened titles, abstracts, and keywords to identify duplicate trials and clearly ineligible studies and then excluded them. Following that, the full text of the studies was examined to ensure that they met inclusion criteria. If no ideal solution exists, the referee (LZ or FL) would make the final decision.
Two independent investigators (ZY and LW) extracted data using a six-part standardization extraction form: (1) identification information (publication year and first author), (2) general information (language, sample size, allocation ratio, diagnostic criteria, age, gender, course of disease, and severity of disease), (3) details of the acupuncture group (according to Revised Standards for Reporting Interventions in Clinical Trials of Acupuncture (STRICTA) [26], (4) details of the control group, (5) outcomes, and (6) main results. e selection procedure is depicted using a PRISMA flow graph.

Study Quality Assessment.
e risk of bias (ROB) of each study was independently evaluated by two assessors using Cochrane Handbook [27]. is Cochrane ROB Tool comprised seven parts (random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessors, incomplete outcome data, selective reporting, and other bias) and ranked the methodological quality as unclear, low, or high. A third party (LZ or FL) was consulted and aided in the final decision-making process. e ROB plot was generated using ReviewManager (RevMan) version 5.4 software (Cochrane, London, UK).

Statistical Analysis
e pairwise meta-analysis was conducted using STATA software version 15.0 (Stata Corp LP, College Station, Texas, USA). e pre-post differences or end-point scores were calculated as outcomes. For the metaanalysis, three-arm trials were divided into two two-arm trials.
e Mantel-Haenszel method used a fixed-effects model, whereas Der Simonian-Laired method utilized a randomeffects model. e statistical heterogeneity was identified and measured by I 2 statistics and p-value. e risk ratios (RR) were used for dichotomous data with a 95% confidence interval (CI). For continuous with 95% CI, weighted mean differences (WMD)/mean differences (MD) were applied. Based on guidelines from Cochrane Handbook 5.4, we deemed no statistical heterogeneity when I 2 <50% and p > 0.05.  [28]. Meanwhile, ADDIS V.1.16.8 was employed to generate indirect and direct comparisons. Using node-splitting analysis, ADDIS models were separated into consistency and inconsistency models. All nodes had p-values ≥0.05, indicating no statistically significant difference between indirect and direct comparisons, and we may employ the consistency model. e consistency model was used to determine the probability ranking of the best treatment for each outcome. e model's convergence is indicated by the potential scale reduced factor (PSRF). If the PSRF value was less than 1.2, it would be considered acceptable. For each acupuncture method, the ranking probabilities were generated in each outcome.

Network
2.6. Publication Bias. As the analysis included over 10 RCTs, we used a comparison-adjusted funnel graph to assess reporting bias. If the included studies were symmetrically distributed on either side of the midline, there is a low risk of reporting bias.

Quality of Evidence.
Using Grades of Recommendations, Assessment, Development, and Evaluation (GRADE) [29,30], the overall quality of evidence was assessed and ranked as high, moderate, low, and critically low.

Study Selection.
Following a comprehensive search, 5,647 potential trials were identified. After removing duplicate trials, 2,689 studies remained. After initial screening, 51 articles remained. Finally, after reading the full text of the articles, 17 articles were excluded (8 non-RCT, 8 ineligible  intervention groups, and 1 ineligible control group), and 34 RCTs remained . e selection process is displayed in Figure 1, and the excluded full-text studies with reasons are listed in Table S2.

Study Characteristics.
All included studies were implemented in China. e 34 studies were published between 2002 and 2021, with 2,071 patients; 30 trials were reported in Chinese, and 4 [44][45][46]51] were published in English. Most studies had sample sizes of less than 100, and only 2 RCTs [31,59] were equal to or greater than 100. e treatments of included studies consisted of MA, EA, WA, FA, SA, and CM, as well as integrations between these acupuncture therapies or with CM. Donepezil was the most frequently used medication in control groups. e allocation ratio was 1:1 or nearly 1:1. e frequently used diagnostic criteria were the Diagnostic and Statistical Manual of Mental Disorders (DSM) criteria. However, in the past 5 years, the National Institute on Aging and Alzheimer's Association (NIA-AA) criteria were commonly used. Besides, the mean age of AD participants was 60 to 80 years, and the number of males was less than that of females. e course of the disease mainly ranged from 3 to 5 years. Only 7 studies [41,44,46,[56][57][58]62] reported on the severity of AD. e treatment duration in the included studies ranged from 28 to 84 days. MMSE score improvement was the most mentioned outcome. Table 1 summarizes the major characteristics of all included RCTs.
3.3. Acupuncture Details. As determined by the STRICTA tool, the details of acupuncture methods were extracted and displayed in Table 2. All included trials referred to acupuncture rationale. In needling details, the number of needle insertions per subject per session of the 34 trials mainly ranged from 8 to 10; the frequently used acupoints for AD were Baihui (DU 20), Zu San Li (ST 36), and Sishencong (EX-HN1); the acupuncture insertion depth varied widely due to different acupoints; a total of 13 studies [36, 38, 40, 41, 49, 51-55, 60, 62, 63] were lacking in response sought; the commonly used needle stimulation was manual acupuncture; the generally frequently used acupuncture brand was Hwato, as well as the diameter and length of acupuncture were 0.35 and 40 mm, respectively. e number of treatment sessions was various in the treatment regimen, and the frequency of treatment sessions was 5 to 6 times per week. In other components, only 2 trials [33,56] covered the details of other interventions. In practitioners, only 3 studies [32,34,44] covered acupuncturists' details. In comparator interventions, more than half of included articles reported rationale for control/comparator, while two studies [37,38] did not clearly illustrate the control/comparator.

Quality
Assessment. ROB of included trials was evaluated using Cochrane ROB Assessment Tool v.5.4. Although all 34 RCTs were reported using a random method, 12 trials Evidence-Based Complementary and Alternative Medicine were unclearly reported in random sequence generation, and 1 study [50] was grouped by date. Only 4 studies [34,39,43,44] produced low risk in allocation concealment. Due to acupuncture's exceptionality, a high risk of performance bias existed. In 2 trials, the method of blinding outcome assessors was successfully implemented [43,44]. All included trials exhibited a low ROB in other parts. Figure 2 illustrates the ROB results.

Secondary Outcome.
(1) Response Rate. We performed 10 classic pairwise meta-analyses using a random-effects model to compare the effectiveness of various acupuncture  Evidence-Based Complementary and Alternative Medicine Evidence-Based Complementary and Alternative Medicine           (2) Improvement in ADL. Herein, we generated 9 classic pairwise meta-analyses using a random-effects model to compare the effectiveness of various acupuncture therapies with CM.

Network Plot for Different
Interventions. We conducted 4 network plots using STATA 15.0. e line thickness is proportional to the 2 therapies, and the point size is positively correlated with the treatment sample size in Figure 3. MMSE improvement was reported in 31 studies involving 10 therapies and 1,874 subjects (Figure 3(a)), whereas ADAS-cog reduction was reported in 9 RCTs involving 708 patients and 6 interventions (Figure 3(b)). e response rate was revealed in 19 studies with 1,206 participants and 10 methods (Figure 3(c)). ADL improvement was reported in 21 RCTs with 1,366 patients and 9 interventions (Figure 3(d)).

Evaluating Statistical Inconsistency.
e node-splitting method was used to test the local inconsistency in MMSE improvement (Table S3(a)) and ADL (Table S3(b)). We found that p ≥ 0.05, demonstrating no significant difference between direct and indirect evidences. Due to no indirect evidence of a reduction in ADAS-cog and response rate, we performed a model of consistency.

Evaluating Convergence of Consistency Model.
According to PSRF results (all PSRF-value ≤1.2) in MMSE improvement (Table S4(a)), ADAS-cog reduction (Table S4(b)), response rate (Table S4(c)), and ADL improvement (Table S4(d)), the consistency model's convergence was acceptable.     Figure 4(a), EA + CM was proved the optimal acupuncture intervention in improving the MMSE score of 10 therapies in this NMA.

Secondary Outcome. (1) Response rate.
e NMA response rate is displayed in Table Figure 4(c), WA was proved as the optimal acupuncture intervention in response rate of 10 methods in this NMA.
(2) Improvement in ADL. e NMA for ADL improvement is displayed in Table   3.8. Heterogeneity. Acupuncture techniques, methods, acupuncture points, treatment duration, and other factors were different, resulting in high clinical heterogeneity. erefore, we performed a sensitivity analysis using STATA 15.0 to assess the stability of the results, and we found that most combined effects were comparatively minor, and the results were reliable.

Publication Bias.
e reporting bias was assessed by comparing the symmetry of the comparison-adjusted funnel graph. Based on funnel plots regarding MMSE improvement ( Figure  5(a)), response rate ( Figure 5(b)), and ADL improvement ( Figure 5(c)), most included studies were symmetrically distributed on either side of the midline, demonstrating that the likelihood of small sample effects was reduced. In ADAS-cog reduction, since the number of included RCTs did not exceed 10, funnel plots were not used to evaluate publication bias.

Quality of Evidence.
According to the GRADE tool, the quality of the four outcomes (improvement in MMSE, ADAS-cog, ADL, and response rate) was low to critically low. Due to ROB, inconsistency, and imprecision, most evidence was rated critically low. Tables S5-S8 contain details about the evidence's quality.

Discussion
AD presents a remarkable public health problem, but FDA has approved only a few medical therapies, which could not affect the disease process [65,66]. Numerous studies [17][18][19][20] proved that the acupuncture method effectively improved AD cognitive function and daily life ability. Some reviews [67,68] demonstrated that acupuncture could induce neural plasticity, cell communication, regeneration, and gene expression in AD. Meanwhile, these studies have provided a mechanistic basis for acupuncture's efficacy in AD treatment. While various acupuncture therapies are applied nowadays, these have not been normalized or standardized. Doctors are compelled to use diversified acupuncture interventions, which imposes significant manpower and high economic burdens. erefore, this NMA aims to identify optimal acupuncture therapy for AD using the most comprehensive information.
is meta-analysis aims to determine the efficacy of multiple acupuncture methods for AD treatment. e primary outcomes were shown as follows: (1) for MMSE improvement, EA + CM, MA + CM, and FA + CM efficacies were statistically different compared with CM efficacy; EA + CM was regarded as the optimal acupuncture method for MMSE improvement. (2) Meanwhile, for ADAS-cog reduction, when EA or MA was combined with CM, a high reduction in ADAS-cog was observed compared with CM alone; EA plus CM was deemed the most efficient acupuncture treatment. e results of secondary outcomes were summarized as follows: (1) regarding response rate, we discovered remarkable differences between MA and CM; WA was considered the optimal acupuncture therapy. (2) In improving ADL, EA + CM, MA + CM, FA + CM, MA, and SA efficacies were statistically different compared with CM efficacy; EA + CM was regarded as the optimal acupuncture method for ADL improvement. Moreover, six trials (16.22%) reported the safety of acupuncture-related AEs (pain, local      AD treatment was identified, and (2) this NMA was strictly accomplished according to international standards. For instance, the report followed PRISMA-NMA guidelines, and acupuncture details met STRICTA criteria. (3) While numerous previous studies have examined efficacy outcomes, they lack objective and uniform standards. In this systematic review, the internationally acknowledged and commonly used measurement tools for cognitive function, MMSE and ADAS-cog, were regarded as the primary outcome indicators. ADL scale was applied as a secondary outcome indicator to assess improvement in activity function. In addition, the clinical efficacy of AD was further illustrated using response rate and adverse events. (4) To ensure the robustness of the results, we conducted a sensitivity analysis. (5) e quality of evidence is critical for clinical decision-making, which can be assessed using GRADE. Meanwhile, this systematic review has some limitations. First, all included trials were conducted in China, which may introduce regional bias. In addition, the sample size of included trials was small, which may cause insufficient statistical efficiency. ird, because numerous acupuncture articles did not adhere to the STRICTA statement, they may leave many important details. Besides, the methodological and evidence quality of included studies was low, which could impair the findings' reliability and efficiency. Fifth, as is known to all, AD can be divided into mild, moderate, and severe periods. Since most of the RCTs did not provide sufficient AD stage information, we did not accomplish the analyses to explore the various stage differences in the curative effect of acupuncture interventions. Next, although this study had limited intervention methods in detail, nonstandard factors contributed to clinical heterogeneity in acupuncture and CM. Additionally, while acupuncture is known for its long-term effects, this study focused exclusively; this study only concentrated on short-term efficacy and lacked long-term efficacy.
Numerous recommendations for future research exist. (1) In terms of methodological quality of the included trial, the influencing factors leading to poor quality are randomization, allocation concealment, and blinding; ROB,   inconsistency, and imprecision resulted in degradation in evidence quality. erefore, future studies should strictly adhere to the latest edition of the Cochrane Handbook for Systematic Reviews and the GRADE tool. (2) Besides, numerous investigations failed to report acupuncture details in a standardized approach in acupuncture trials. us, Consolidated Standards of Reporting Trials (CONSORT) [69,70] and STRICTA were proposed to govern the reporting.
(3) Clinical heterogeneity was observed due to various factors, including acupoint selection, treatment duration, CM dose, and so on. erefore, the acupuncture industry should not only seek the optimal acupuncture methods but also address the optimal acupoints, treatment time, and frequency for AD treatment. (4) Additionally, future acupuncture studies for AD should focus on both short-and long-term efficacy. Besides, attention should be paid to acupuncture prevention in AD. (5) Finally, potential mechanisms (markers in neuroimage, biochemical, and gene) of acupuncture for AD should be studied more closely.

Conclusion
According to our findings, acupuncture therapy has been demonstrated to be effective against AD in terms of improving cognitive function, the ability of daily living, and response rate. EA + CM may be the optimal acupuncture method for improving AD cognitive function and ADL. Meanwhile, WA therapy was deemed the most effective treatment in terms of response rate. However, the overall quality of evidences was ranked as low to critically low. erefore, well-designed and high-quality trials are expected to validate and re-evaluate scientific discoveries.

Data Availability
No additional data are available.

Conflicts of Interest
e authors declare that there are no conflicts of interest.

Authors' Contributions
Zihan Yin and Xiang Li contributed equally to this work. Zihan Yin, Ling Zhao, and Fanrong Liang conceived this study. Zihan Yin developed and implemented the systematic review, under the supervision of Xiang Li. Zihan Yin provided the statistical analysis plan for the study and performed data analysis. Zihan Yin and Linjia Wang performed a study search, screening, and extraction of data, whereas Mingsheng Sun reviewed the work. Zihan Yin and Xiang Li wrote the first draft of the current review, with Ling Zhao and Fanrong Liang providing input to the final draft. All authors read and approved the final manuscript.