Studies that met all of the following criteria were included in this systematic review:

Participants: Patients diagnosed with any type of MCI using standardized diagnostic criteria or based on the clinicians’ assessment. Patients with coexisting conditions, such as cerebrovascular disease, were not excluded from consideration in this review.

Interventions: Any orally administered CHM. Studies evaluated single compounds extracted from certain herbs, such as the standardized extract of Ginkgo biloba L., were not included in this review since these herbs were not classified as traditional CHM (DeFeudis, 2003). Usual care for underlying diseases was allowed if the same treatments were applied to both the CHM and placebo groups, except for any other types of Chinese medicine therapies, anti-dementia drugs or other therapies aimed at improving cognitive function (e.g., cognitive training).

Controls: Only placebo-controlled trials were included. Co-interventions were allowed if they were the same as those used in the intervention group.

Outcomes: Studies reporting at least one of the following outcomes at the end of treatment were included: scores of Mini-Mental State Examination (MMSE) and Montreal Cognitive Assessment (MoCA). This review also analyzed adverse events (AEs) if the original RCTs reported this outcome.

Study design: Only RCTs were included.

Two reviewers (LL and CSZ) independently searched nine databases: PubMed, Excerpta Medica Database (Embase), Cochrane Central Register of Controlled Trials (CENTRAL) (including the Cochrane Library), Cumulative Index of Nursing and Allied Health Literature (CINAHL), the Allied and Complementary Medicine Database (AMED), China Biomedical Literature (CBM), China National Knowledge Infrastructure (CNKI), Wanfang and Chongqing VIP (CQVIP) databases, from their respective inceptions to January 2023. No restrictions were placed on the language of publication. The search terms were the keywords and their synonyms of MCI, CHM, and RCT. Details of the search strategy employed on searching nine databases are presented in Supplementary Text S1. In addition, references from published systematic reviews on Chinese medicine for MCI were hand-searched.

Two reviewers (LL and CSZ) independently screened the articles’ titles and abstracts against the pre-defined selection criteria, excluding irrelevant studies and duplicates. The full-text articles of potential studies were then retrieved for further screening. Any discrepancy between these two reviewers was resolved through discussion with a third reviewer (ALZ).

For data extraction, two independent reviewers (LL and CSZ) extracted information from each eligible study, including sample size, characteristics of participants, details of intervention and control, duration of treatment and follow-up, and clinical outcomes data.

The methodological quality of each study was independently evaluated by two reviewers (LL and CSZ) using the Version 2 of the Cochrane risk-of-bias tool for randomized trials (RoB 2) tool (Higgins et al., 2022). Any disagreement between these two reviewers was resolved by discussing with a third reviewer (ALZ). The RoB 2 assesses different aspects of trial design, conduct, and reporting compared with the original RoB tool. Judgements were summarized as “low” or “high” risk of bias or “some concerns” (Higgins et al., 2022).

Review Manager 5.4 and Stata 15 were used for data analyses in this review. For continuous data (scores of MMSE and MoCA), mean difference (MD) with 95% confidence intervals (CIs) was calculated. For categorical data (adverse event), risk ratio (RR) and 95% CIs were used to present the effect size. A random-effect meta-analysis model was utilized to calculate the pooled effect size of all outcomes. Meta-analyses were conducted on the end-of-treatment data of the primary outcome measures, MMSE and MoCA, to evaluate the efficacy of CHM on cognitive function among individuals with MCI. Heterogeneity between trials was assessed using the I² test, which was incorporated into the forest plots. Subgroup analyses were conducted based on treatment duration, subtypes of MCI and herbal ingredients, to explore the source of heterogeneity. Several sensitivity analyses, including the use of different effect models, exclusion of studies that were considered to be at a “high risk of bias” in the overall judgment or each domain of the ROB 2, and a one-by-one exclusion approach, were conducted to assess the robustness of the finding. The potential presence of publication bias in the primary outcomes was evaluated by constructing a funnel plot and using Egger’s test when the number of included studies exceeded ten (Sterne et al., 2011).

The certainty of the evidence for the primary outcome, cognitive function evaluated using validated tools (MMSE and MoCA), was assessed using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) method. The certainty was categorized as “high”, “moderate”, “low”, or “very low” considering the risk of bias, inconsistency of results, indirectness of evidence, imprecision, and publication bias (Schünemann et al., 2013).

A total of 7,931 records were initially identified by searching the nine databases, of which 7,905 were retrieved for screening; and an additional 26 records were identified by searching reference lists of previously published systematic reviews. After screening, 13 studies were included in this review (Zhou et al., 2007; Wu et al., 2010; Dai et al., 2011; Huang et al., 2013; Su, 2013; Liu, 2014; Shi et al., 2015; Wu, 2016; Zhang et al., 2016; Park et al., 2019; Lu et al., 2020; Chen et al., 2021; Shin et al., 2021). The study search and selection procedure is shown in Figure 1.

The 13 studies included in this review were published between 2007 and 2021. Five studies were published in English (Zhang et al., 2016; Park et al., 2019; Lu et al., 2020; Chen et al., 2021; Shin et al., 2021) and the other eight studies were published in Chinese (Zhou et al., 2007; Wu et al., 2010; Dai et al., 2011; Huang et al., 2013; Su, 2013; Liu, 2014; Shi et al., 2015; Wu, 2016). Ten trials were conducted in China (Zhou et al., 2007; Wu et al., 2010; Dai et al., 2011; Huang et al., 2013; Su, 2013; Liu, 2014; Shi et al., 2015; Wu, 2016; Zhang et al., 2016; Lu et al., 2020), two in South Korea (Park et al., 2019; Shin et al., 2021), and one involved three centers in Singapore, Vietnam, and the Philippine (Chen et al., 2021). Detailed characteristics of each study are presented in Table 1.

A total of 1,043 participants were enrolled in the 13 studies. The number of participants per trial ranged from 33 to 128, with 80 participants on average. Nine studies reported a total of 51 dropouts, of which 22 were from the CHM groups and 29 were from the placebo groups. Ten trials recruited patients diagnosed with MCI without indicating a specified subtype, and three trials recruited only vascular MCI (i.e., vascular cognitive impairment no dementia) (Liu, 2014; Lu et al., 2020; Chen et al., 2021). Eleven studies reported the baseline scores of MMSE: the mean MMSE score in ten studies ranged from 23 to 27 (Zhou et al., 2007; Wu et al., 2010; Dai et al., 2011; Huang et al., 2013; Su, 2013; Liu, 2014; Wu, 2016; Zhang et al., 2016; Park et al., 2019; Lu et al., 2020; Shin et al., 2021), the other one reported an extremely low mean MMSE score (Shi et al., 2015). Four studies reported the baseline scores of MoCA (Huang et al., 2013; Su, 2013; Liu, 2014; Chen et al., 2021), with the mean scores ranging from 18 to 21. Table 2 presents detailed information on the participants’ baseline characteristics, including age, gender, education level and cognitive score.

All 13 included studies compared CHM with placebo. The treatment duration varied from two to 24 months, with eight out of 13 studies having a duration of 6 months or more. Among the 13 studies, four conducted follow-up assessments after the treatment phase, ranging from 1 month to 1 year (Zhou et al., 2007; Wu et al., 2010; Dai et al., 2011; Shin et al., 2021). One study mentioned that participants were followed-up after the treatment phase, but did not specify the duration (Park et al., 2019). The remaining eight studies did not include a follow-up phase after the treatment.

Twelve studies evaluated the treatment effects using MMSE (Zhou et al., 2007; Wu et al., 2010; Dai et al., 2011; Huang et al., 2013; Su, 2013; Liu, 2014; Shi et al., 2015; Wu, 2016; Zhang et al., 2016; Park et al., 2019; Lu et al., 2020; Shin et al., 2021), and three studies reported data on MoCA (Huang et al., 2013; Su, 2013; Liu, 2014). Eleven studies clearly outlined in the methods section that AEs would be monitored during the study period, and reported the outcome of AEs in the published articles (Zhou et al., 2007; Wu et al., 2010; Huang et al., 2013; Su, 2013; Liu, 2014; Shi et al., 2015; Wu, 2016; Zhang et al., 2016; Park et al., 2019; Chen et al., 2021; Shin et al., 2021). The remaining two studies reported information on AEs in the results/discussion sections without stating this in the methods section (Dai et al., 2011; Lu et al., 2020).

Except four studies (Huang et al., 2013; Liu, 2014; Shi et al., 2015; Zhang et al., 2016), the other nine studies identified the pharmaceutical manufacturer. Nonetheless, only one study explicitly stated that both the CHM preparation and the placebo were produced by the same manufacturer, utilizing standardized methods in accordance with Good Manufacturing Practice guidelines (Shin et al., 2021). The remaining studies, however, did not specify the quality control methods implemented for the CHM or placebo preparations (Supplementary Table S2).

Out of the 13 included studies, two used the form of decoction for CHM administration (Huang et al., 2013; Liu, 2014). The other 11 studies utilized more convenient format of CHM preparation, they are: capsules (Wu et al., 2010; Zhang et al., 2016; Lu et al., 2020; Chen et al., 2021), granules (Dai et al., 2011; Su, 2013; Shi et al., 2015; Shin et al., 2021), powder (Park et al., 2019), pills (Wu, 2016) and oral solution (Zhou et al., 2007). Thirteen unique CHM formulae were identified from the 13 included studies. Twelve studies detailed the herbal ingredients of the CHM formulae, and the other one study did not provide such information (Wu, 2016). A total of 50 herbs were used in these trials. Although the formulae were diverse across trials, some herbs were frequently used by most of the studies, with the most common herbs being Panax ginseng C.A.Mey. (Ren shen), Acorus calamus var. angustatus Besser (Shi chang pu), Polygala tenuifolia Willd. (Yuan zhi). Supplementary Table S3 presents the top 15 most frequently used herbs.

All 13 included studies used a placebo CHM as the control, but the ingredients used to prepare the placebo varied across the trials. Seven trials prepared placebo only containing inactive substances such as starch, food coloring and bittering agent (Zhou et al., 2007; Wu et al., 2010; Shi et al., 2015; Park et al., 2019; Lu et al., 2020; Chen et al., 2021; Shin et al., 2021). In one study, the placebo comprised a bittering agent, food coloring, and stir-fried Medicated leaven (Shen qu) (Huang et al., 2013). It is worth noting that Medicated leaven is a fermented product by mixing flour with other Chinese medicines (including Artemisia annua Pall., Xanthium sibiricum Patrin ex Widder, Polygonum hydropiper L., Vigna angularis (Willd.) Ohwi & H. Ohashi, Prunus armeniaca L.), which has been commonly used to treat gastrointestinal diseases in Chinese medicine (Fu et al., 2020). Another study stated that the placebo consisted of a 10% dose of the CHM decoction used in the experimental group (Liu, 2014). Four studies did not provide detailed information about the preparation of the placebo. However, three out of these four studies indicated that the appearance, smell and taste of the placebos were the same as the CHM in the experimental groups (Dai et al., 2011; Su, 2013; Zhang et al., 2016), while the remaining one study mentioned that the control group received a mimetic agent with a corresponding dosage to the CHM (Wu, 2016). Detailed information on quality control and ingredients of CHM/placebo preparations are presented in Supplementary Table S2.

We used the Cochrane’s Risk of Bias (RoB) 2 tool to assess the risk of bias of included studies based on two outcome measures: MMSE and MoCA. The results are merged in Supplementary Figure S1 since there is no difference between the assessments based on these two outcomes. In our assessment, an intention-to-treat analysis model was used. In addition to the published results articles, the trial registry records related to four trials (Zhang et al., 2016; Lu et al., 2020; Chen et al., 2021; Shin et al., 2021) and one published protocol related to one trial (Chen et al., 2021) were also checked to inform the RoB assessment.

For overall RoB, 10 studies (76.9%) were assessed as “some concerns”, one study (Shin et al., 2021) is “low risk of bias” and two studies are “high risk” (Lu et al., 2020; Chen et al., 2021). In terms of randomization process, six studies were judged as “some concerns” because they did not provide information on the generation of allocation sequence and sequence concealment (Zhou et al., 2007; Wu et al., 2010; Dai et al., 2011; Huang et al., 2013; Shi et al., 2015; Park et al., 2019), the others were assessed as “low risk of bias” for this domain. For the domain of “deviations from the intended interventions”, five studies were assessed as “some concerns” because they did not apply an intention-to-treat analysis to deal with missing data (Huang et al., 2013; Su, 2013; Liu, 2014; Wu, 2016; Chen et al., 2021), one study was “high risk of bias” for this domain due to a high dropout rate (18%) without appropriate analysis method (Lu et al., 2020); the remaining seven studies were “low risk of bias” since there was no dropout cases. The Lu 2020 study (Lu et al., 2020) was also given a “high risk of bias” judgement for “missing outcome data”. In terms of the outcome measurements, all 13 studies were assessed as “low risk of bias” since both MMSE and MoCA consist of a series of questions with clearly defined scoring criteria, which were unlikely to be influenced by knowing which intervention was received. For the “selection of the reported result”, one study was assessed as “high risk of bias” because detailed data on a pre-specified outcome (MMSE) was not reported (Chen et al., 2021); 10 studies were rated as “some concerns”, due to a lack of pre-registered trial protocols or insufficient information on outcome measurements and statistical analyses plans provided in the trial registration; the remaining two studies were rated as “low risk of bias” (Park et al., 2019; Shin et al., 2021).

Three studies reported MoCA scores at the end of treatment (Huang et al., 2013; Su, 2013; Liu, 2014). Meta-analysis shows that the CHM group achieved superior effects in improving the MoCA score compared to the placebo group (MD: 2.88 [1.69, 4.06], I² = 81%, 3 studies, 241 participants) (Figure 6). Subgroup analysis, sensitivity analysis, and publication bias evaluation were not feasible due to the small number of studies reporting this outcome.

Twelve studies mentioned AE information: four studies reported no AEs during the treatment phase (Wu et al., 2010; Dai et al., 2011; Su, 2013; Shi et al., 2015) and eight studies reported AEs occurred (Zhou et al., 2007; Huang et al., 2013; Liu, 2014; Wu, 2016; Zhang et al., 2016; Park et al., 2019; Chen et al., 2021; Shin et al., 2021). Among the eight studies reporting AEs, one study mentioned that no statistical difference was detected between the CHM and placebo group in terms of the Treatment Emergent Symptom Scale score (Huang et al., 2013). However, the exact number of participants who experienced the AEs was not reported (Huang et al., 2013). Therefore, the remaining 11 studies were included in the meta-analysis for AEs, with varied adverse events observed in 76 participants (9.06%). There is no statistically significant difference between the CHM and placebo groups in terms of the number of participants who reported AEs (RR: 1.31 [0.92, 1.87], I² = 0%, 11 studies, 839 participants) (Supplementary Figure S3). Notably, gastrointestinal symptoms like constipation, nausea, and vomiting emerged as the predominant manifestations among all the reported AEs. One study reported that one case in the treatment group dropped out from the study due to gastrointestinal discomfort (Liu, 2014). One study did not report the safety outcome but mentioned that of the 18 patients (CHM and placebo groups) who dropped out of the study, none were due to AEs (Lu et al., 2020).

As for the severity of the AE, two studies mentioned that no serious adverse events (SAEs) were observed in either the CHM groups or the placebo groups (Zhou et al., 2007; Liu, 2014). One study reported that, according to the criteria for SAEs which encompassed factors such as death, life-threatening situations, prolonged hospitalization, incapacity, or important medical events, 13 participants (22.8%) experienced SAEs in the CHM group while 6 participants (13.0%) experienced SAEs in the placebo group (Chen et al., 2021). However, there was no significant difference between the CHM and placebo group in terms of the proportion of participants experiencing SAE (Chen et al., 2021).

Regarding the association of AEs with the study interventions, three studies found no identified connection between these AEs and the administered interventions (Su, 2013; Park et al., 2019; Shin et al., 2021). One study noted that a total of six AEs were possibly linked to the administration of the CHM intervention, including dry throat, dizziness, hematoma in the right thigh, vomiting, neuropathic pain, and sleepiness (Chen et al., 2021). Supplementary Table S4 details the AEs reported in the included studies.

The GRADE assessments for MMSE and MoCA were downgraded by one level due to inconsistency. Additionally, due to the limited sample size, the certainty of the evidence for MoCA was further downgraded by one level for imprecision. Overall, the evidence certainty for MMSE was evaluated as “moderate”, while it was “low” for MoCA (Supplementary Table S5).

This systematic review synthesized 13 RCTs that compared CHM with placebo to evaluate the efficacy and safety of CHM for the treatment of MCI. The meta-analyses on post-treatment MMSE and MoCA scores indicated that, the use of CHM led to significant improvement in MCI patients’ cognitive function compared to placebo.

All except one study reported the outcome of MMSE. The overall effect estimate was found to be robust, and no significant publication bias was detected. However, substantial heterogeneity was observed among the studies. The evidence regarding the outcome on MMSE was assessed with moderate certainty, while it was “low” for MoCA. Subgroup analyses were conducted according to the subtypes of MCI, treatment duration and different CHM herbal ingredients. The results indicated that CHM is beneficial for MCI, either vascular MCI or unclassified MCI. Subgroup analyses also revealed a trend suggesting that the treatment effects of CHM diminish with longer treatment durations. Additionally, herbs include P. ginseng C.A.Mey., A. calamus var. angustatus Besser, P. tenuifolia Willd. May enhance the therapeutic effects of CHM. Regarding the safety of CHM, gastrointestinal symptoms have been identified as the most commonly reported AEs. None of the included studies identified an association between the administration of CHM and a higher risk of AEs. The comprehensive meta-analysis on AEs revealed no significant difference between the CHM and placebo groups. While the available evidence suggests that CHM is generally safe, caution is advised due to the lack of AE reporting or detailed information in some instances.

Previous reviews have shown that CHM may improve cognitive functions in patients with MCI (Dong et al., 2016; Dong et al., 2019; Wang et al., 2021; Liang et al., 2022), which is consistent with the results of our study. However, there were some differences in the inclusion criteria between our review and previous reviews.

In our study, we aimed to reflect the real-world clinical practice by including participants with all types of MCI, while previous systematic reviews had a narrow focus on only one specific subtype (Dong et al., 2016; Dong et al., 2019). It is important to note that the etiology and pathogenesis of MCI are highly complex (DeCarli, 2003; Mufson et al., 2012; Jongsiriyanyong and Limpawattana, 2018). Furthermore, MCI typically occurs in older adults (Petersen et al., 2018), and there is a high prevalence of overlapping neuropathology in this population (Brenowitz et al., 2017). Therefore, it can be challenging to distinguish a single or pure pathological subtype of cognitive impairment, even with autopsy (Brenowitz et al., 2017). Based on this background, there is a need to evaluate the effects of CHM on all types of MCI.

As for the intervention, in contrast to the latest systematic review which focused solely on Chinese patent medicine (Liang et al., 2022), our review included all types of CHM interventions to provide more comprehensive and representative evidence for real-world clinical practice.

In addition, we prioritize randomized placebo-controlled trial as the optimal study design to assess the efficacy of CHM, as they provide the highest level of evidence (Burns et al., 2011), and the use of indistinguishable placebos is the most effective method for determining the pure biological effect of an intervention in experimental settings (Cummings et al., 2013). Whilst previous reviews explored the effects of CHM using various types of comparisons (Dong et al., 2016; Dong et al., 2019; Wang et al., 2021; Liang et al., 2022). In comparison to previous reviews, our review includes a larger number of original studies from a more diverse range of countries, thereby providing more comprehensive evidence.

Panax ginseng C.A.Mey., A. calamus var. angustatus Besser, and P. tenuifolia Willd. Were the three most commonly used herbs in the included studies. Moreover, the inclusion of P. ginseng C.A.Mey., A. calamus var. angustatus Besser, and P. tenuifolia Willd. appears to enhance the therapeutic effects. These three herbs have been traditionally used in East Asia for its potential cognitive improving effects.

Preclinical studies suggest that the active compounds or extracts from these three herbs—P. ginseng C.A.Mey., A. calamus var. angustatus Besser, and P. tenuifolia Willd. Can enhance cognitive function across different experimental models of cognitive impairment. The potential underlying mechanisms include antioxidant, anti-apoptosis, anti-neurotoxicity, anti-cytotoxicity and anti-inflammatory effects, mitigation of Alzheimer’s disease-related pathology, synaptic protection, and the upregulation of neuronal cells through various signaling pathways (Li et al., 2020; Liang et al., 2021; Zhang et al., 2023).

Additionally, these three herbs are traditionally valued for their common function in soothing the heart and calming the mind, indicated for treating symptoms like forgetfulness and sleep disturbances (Chinese Pharmacopoeia Commission, 2020). Emerging research has demonstrated a strong association between sleep disturbances and cognitive decline (Naismith et al., 2010; Suh et al., 2018; Ma et al., 2020). Experimental studies have indicated that ginsenosides, the active chemical compounds in P. ginseng C.A.Mey. May reverse memory deficits in animal models suffering from sleep deprivation-induced memory impairment (Lu et al., 2017; Lu et al., 2018). Additionally, active compounds from these three herbs have demonstrated sedative and hypnotic effects, improving sleep in animal studies, which aligns with their traditional use (Cao et al., 2016; Radhakrishnan et al., 2017; Xu et al., 2019; Shao et al., 2020). Details information on the pharmacological effects and potential mechanisms of action is presented in Table 5.

Although efforts were made to achieve an impartial conclusion, the present study has inherent limitations. First, the presence of substantial heterogeneity among the included studies is one limitation of our meta-analysis. One potential source of heterogeneity could be the variations in participant characteristics, such as age, gender, education level or the underlying etiology of MCI. Unfortunately, due to the limited availability of individual patient data in the included studies, we were unable to perform further analysis based on these factors. Moreover, differences in the treatment protocols used in the individual trials, such as duration or different CHM ingredients, might have contributed to the observed heterogeneity.

Second, although we conducted comprehensive research to collect as many studies as possible, the number of eligible randomized placebo-controlled trials on CHM for MCI is limited. Additionally, the sample sizes for most included studies were small, and all of the studies were conducted in Asia. Therefore, it remains unclear whether the findings can be generalized to other ethnic populations. More large-scale and multi-cantered trials with detailed patient-level data are still needed to provide more robust evidence in the field of CHM for MCI.