We searched the following databases on October 7, 2020 for systematic reviews of randomised controlled clinical trials: EMBASE via OVID (from 1947), MEDLINE via OVID (from 1946), CINAHL (Cumulated Index to Nursing and Allied Health Literature from 1977) and the Cochrane Library including the Cochrane Central Register of Controlled Trials (CENTRAL). Each search strategy was adapted to take into account differences in controlled vocabulary and syntax rules. An example search strategy is given in Supplementary Material. We also contacted experts in the field to identify any relevant studies which had not been found by the search.

Two reviewers independently screened titles and abstracts to select articles for full-text screening. Two reviewers then independently screened the full-text articles. We selected articles which met the following inclusion criteria:

- Study type: Systematic reviews of randomised controlled clinical trials with a reported systematic search strategy and with the intention to perform a meta-analysis.

We excluded reviews which only presented results in a narrative format and did not attempt to meta-analyse the outcomes. We did however include systematic reviews which found only a single relevant trial and presented its results in the correct format–where a meta-analysis had been intended but included only a single trial. Some reviews included trials both on T2DM and also on prediabetes. If results for T2DM were presented separately, and/or if trials in T2DM were the majority of included trials, we included these. We excluded reviews where the majority of included trials were not on patients with T2DM and where it was not possible to separate out the results for T2DM patients. We also excluded reviews where results for type 1 diabetes (T1DM) were not presented separately. We excluded reviews of multiple different herbal remedies and of pure compounds extracted from herbs, because none of these presented meta-analyses of individual medicinal plants. We did not apply any language restrictions.

Two reviewers independently extracted relevant data using a data extraction form created on Microsoft Excel, and any discrepancies were checked by a third reviewer. Where a review reported several patient groups and/or outcomes, we extracted the number of trials and participants which matched our inclusion criteria (type 2 diabetes) and which reported each relevant outcome (HbA1c and FPG). When results were presented separately for different types of control, we preferentially chose the comparison against placebo (rather than comparison against standard treatment), in order to gauge the effect size of the medicinal plant itself. Where HbA1c results were reported in mmol/mol, they were multiplied by the conversion factor 0.09148 to give the equivalent as a percentage (National Glycohemoglobin Standardization Program, 2010). Where FPG results were presented as mg/dL, they were divided by 18 to convert to mmol/L. For each review we extracted the number of trials which had reported on adverse effects, and the number of these which reported any specific adverse effects.

Two reviewers independently appraised the quality of the studies using the AMSTAR-2 tool (Shea et al., 2017) and discrepancies were resolved by discussion with a third reviewer.

Results from meta-analyses of HbA1c and FPG were ranked in order of effect size and presented on a Forest plot. A clinically significant reduction in HbA1c has been defined by clinicians as a reduction of at least ≥0.5% (Lenters-Westra et al., 2014); we defined a clinically significant reduction in FPG as a change of 0.5 mmol/l or more. We conducted a narrative synthesis of the other results. We calculated the Spearman’s rank correlation coefficient for the correlation between rank of effect on HbA1c and FPG. In this analysis we only included remedies for which both measures were reported. Where a remedy had differing results from several reviews, we took the rank of the best result for each of HbA1c and FPG.

Our initial search identified 2,363 articles after removing duplicates (Figure 1). Forty-nine full texts were screened and of these, 25 met all our inclusion criteria (Davis and Yokoyama, 2011; Kim et al., 2011; Leach and Kumar, 2012; Ooi et al., 2012; Allen et al., 2013; Ooi and Loke, 2013; Neelakantan et al., 2014; Gibb et al., 2015; Gui et al., 2016; Li et al., 2016; Shin et al., 2016; Suksomboon et al., 2016; Tian et al., 2016; Zhang et al., 2016; Daryabeygi-Khotbehsara et al., 2017; Poolsup et al., 2017; Schwingshackl et al., 2017; Gu et al., 2018; Deyno et al., 2019; Gao et al., 2019; Huang et al., 2019; Namazi et al., 2019; Peter et al., 2019; Yang et al., 2019; Ziaei et al., 2020). The commonest reason for exclusion was that the review did not attempt a quantitative meta-analysis of randomised controlled trials. One of the meta-analyses was excluded because it had incorrectly reported underlying data from included studies and its results were inaccurate (Gong et al., 2016).

There were reviews on 18 different medicinal plants (Table 1). Some herbal remedies had more than one review: cinnamon (Davis and Yokoyama, 2011; Leach and Kumar, 2012; Allen et al., 2013; Deyno et al., 2019; Namazi et al., 2019), ginseng (Kim et al., 2011; Gui et al., 2016), Aloe vera (Suksomboon et al., 2016; Zhang et al., 2016) and karela (Momordica charantia) (Ooi et al., 2012; Peter et al., 2019). Three reviews evaluated the effect of a standard traditional Chinese herbal formula which contained a mixture of several herbs. Gegen formulae contained Pueraria lobata root as their main constituent alongside other ingredients such as Salvia miltiorrhiza root, liquorice root and Dioscorea opposita rhizome (Yang et al., 2019). Jinqi Jiangtang contains Astragalus membranaceus root, Coptis spp rhizome and lonicera japonica (Gao et al., 2019). Tianmai Xiaoke contains Trichosanthes root, Ophiopogon japonicus root, Schisandra chinensis fruit and chromium picolinate (Gu et al., 2018). Some reviews studied the effect of specific plant products which are also used as foods: olive oil (Schwingshackl et al., 2017), sweet potato (Ooi and Loke, 2013), dragon fruit (Poolsup et al., 2017), and fenugreek powder incorporated into chapatis (Neelakantan et al., 2014).

All the reviews included mainly clinical trials in patients with T2DM (see Table 1) but four also included a few trials in patients with pre-diabetes. One included a single trial in patients with T1DM, but its results were presented separately and excluded from this review. Five reviews included only trials of patients with diet-controlled diabetes, not taking any conventional antidiabetic medications. Fourteen reviews included trials in which both intervention and control groups received concomitant conventional treatment with oral hypoglycaemic agents (OHA). Five reviews did not specify whether concomitant treatment was given. In 19 reviews, the control groups received a placebo, in four reviews they received only the conventional care (diet and/or medications) and in one, some control groups received a fish oil supplement. In three reviews, some studies gave a conventional OHA to the control group only, not to the treatment group (Ooi et al., 2012; Shin et al., 2016; Peter et al., 2019) but for this review we only extracted the outcomes from the studies using a placebo control.

Duration of follow-up was most often 4–12 weeks, but there was a wide range with a few included studies following up for as little as 1 week or for as long as 4 years. All the reviews included randomised controlled trials but two also included a few non-randomised controlled trials. The reviews included a median of eight trials and 390 participants but the smallest included only a single trial and the largest review included 25 studies (1724 participants).

The AMSTAR-2 scores for each study are shown in Supplementary Material. Only three reviews scored “yes” on all the criteria–all of them Cochrane reviews (Leach and Kumar, 2012; Ooi et al., 2012; Ooi and Loke, 2013). Several quality issues were identified with the other reviews. Most did not report that there was a pre-established published protocol. Most did not have a fully comprehensive search strategy including the grey literature. Most did not list all excluded studies and most did not report on the sources of funding for the studies included in the review. Seven did not adequately investigate publication bias. Six did not report conflicts of interest, including the review on Psyllium which was led by a company marketing a Psyllium product (Gibb et al., 2015)—this review is at high risk of bias.

Twenty-one studies on 16 remedies attempted to conduct a meta-analysis quantifying the reduction in HbA1c (Figure 2). The most effective remedy appeared to be Aloe vera (freshly extracted juice) (Suksomboon et al., 2016). Psyllium fibre (Gibb et al., 2015) and Fenugreek seeds (Neelakantan et al., 2014) also led to similar reductions in HbA1c although with wider confidence intervals. Nigella sativa seeds (Daryabeygi-Khotbehsara et al., 2017), Astragalus membranaceus root (Tian et al., 2016), and two complex traditional Chinese formulae (Gegen Qinlian (Yang et al., 2019) and Jinqi Jiantang (Gao et al., 2019)) also led to clinically and statistically significant reductions in HbA1c. Nettle (Urtica dioica) appeared to lead to a clinically significant reduction but this was not statistically significant because of very wide confidence intervals (Ziaei et al., 2020).

Several remedies produced a statistically significant reduction in HbA1c but the standard mean difference fell below the pre-determined threshold of 0.5%. These were the patent traditional Chinese formula Tianmai Xiaoke (Gu et al., 2018), ginger (Huang et al., 2019), sweet potato tablets (Ooi and Loke, 2013), olive oil (Schwingshackl et al., 2017), karela (Momordica charantia) (Peter et al., 2019) and cinnamon (Namazi et al., 2019). Momordica charantia was studied by two reviews which came to differing conclusions; an early Cochrane review found only a single small RCT with 40 participants, which concluded that Karela dried powder in capsules appeared to be ineffective (Ooi et al., 2012). However, a more recent and comprehensive review including five RCTs (243 participants) found that there was a statistically significant reduction in HbA1c by 0.26% (Peter et al., 2019). Similarly, the four reviews on cinnamon which reported HbA1c came to slightly different conclusions; only one found a statistically significant reduction and none of them reported a clinically significant reduction.

Two meta-analyses of ginseng (Kim et al., 2011; Gui et al., 2016) and one of tea (Camellia sinensis) and tea extracts (Li et al., 2016) all showed that these remedies had no clinically or statistically significant effect on HbA1c.

Twenty-five reviews meta-analysed the reduction in FPG (Figure 3). All the remedies which produced clinically significant reductions in HbA1c also produced clinically and statistically significant reductions in FPG, with the exception of Astragalus membranaceus, which reduced FPG slightly less than the predetermined clinically significant threshold of 0.5 mmol/l. Nettle (Urtica dioica), Momordica charantia and sweet potato also all produced clinically significant reductions in FPG.

There were varying results in the five meta-analyses of cinnamon, but the largest and most recent (which only included patients with T2DM) showed a clinically significant reduction in FPG of −1.07 mmol/l (95% CI-1.56 to −0.58) (Namazi et al., 2019). Other reviews also included patients with T1DM (Leach and Kumar, 2012) or pre-diabetes (Davis and Yokoyama, 2011; Deyno et al., 2019).

For several remedies, there was a wide degree of uncertainty regarding their effectiveness in reducing FPG. Dragon fruit appeared to have a large effect but this was not statistically significant as there were very wide confidence intervals (Poolsup et al., 2017). There was also a large degree of uncertainty about the effect of Mulberry leaf–there was a wide confidence interval, and a second trial (not included in the meta-analysis) reported that it was more effective than glibenclamide (Shin et al., 2016). Of the two reviews on Ginseng, that by Kim et al. (2011) was the only one to focus purely on T2DM; it showed a non-significant reduction in FPG. Another meta-analysis did report a significant reduction in FPG but also included pre-diabetic patients (Gui et al., 2016).

It can be stated with some certainty that ginger and tea (Camellia sinensis) extracts were ineffective for reducing FPG. Neither had a significant effect, and confidence intervals were tight.

Spearman’s rank correlation coefficient was 0.70, indicating that there was a moderate correlation between effect on HbA1c and FPG.

None of the included reviews reported any serious adverse events. In most cases there was no significant difference in the incidence of adverse events between the treatment and control groups (Table 2). Mild gastrointestinal symptoms such as diarrhoea, vomiting and abdominal discomfort were reported in a few cases for certain herbal remedies, in particular Momordica charantia (three participants) and Fenugreek seeds (three participants). There was no specific mention of drug interactions although 14 of the reviews included trials in which the herbal medicine was given in addition to conventional oral hypoglycaemic agents. Only three of these reviews mentioned cases of hypoglycaemia, including only one reported case of a hypoglycaemic seizure in a clinical trial of cinnamon given to adolescent T1DM patients on insulin (Leach and Kumar, 2012).

There have been many RCTs on different phytomedicines and herbal medicines for T2DM, and 25 published meta-analyses on 18 different medicinal plants. Of these, seven have a clinically and statistically significant effect on HbA1c and 12 on FPG (Figures 2, 3). The most effective on both measures appear to be Aloe vera, Psyllium fibre, Fenugreek seeds, Nigella sativa seeds, and the complex traditional Chinese formula Jinqi Jiangtang. Tea and tea extracts were ineffective. The 12 other remedies showed some degree of effectiveness on either HbA1c or FPG, but in some cases with a wide degree of uncertainty. All of the medicinal plants evaluated in this review appeared to be safe, with no serious adverse effects reported. However, some were associated with minor side-effects, in particular gastrointestinal disturbances.

This the first study to provide a systematic, evidence-based overview of meta-analyses of the effectiveness of medicinal plants for glycaemic control in type 2 diabetes. Our systematic approach with broad search terms ensured that we probably found most relevant articles. One limitation is that we did not have the time to search the grey literature or databases in foreign languages such as Chinese. Another limitation is that we were not able to include medicinal plants for which there had been no systematic review with a meta-analysis. For example there was a systematic review of the Ayurvedic remedy Gymnema sylvestre (Leach, 2007) but this found no clinical trials which met its inclusion criteria. It is also likely that there are other potentially effective medicinal plants which have been evaluated in RCTs but not reviewed in a meta-analysis, and others which have not been evaluated in an RCT although lower-level evidence suggests they could be effective (Sissoko et al., 2020).

Our results are also limited by the quality of the trials included in the meta-analyses. Although most only included RCTs, in some cases the preparation or dosage of the phytomedicine may have been suboptimal; in some reviews both herbal remedies and standardised phytomedicines were included. The clinical condition of the patients may have been different between trials where patients were taking concomitant oral antidiabetics and those who were purely diet controlled. In some trials, the duration of follow-up was insufficient to measure the effect on HbA1c, which should be measured at least 3 months after the start of treatment to reveal its full effect. Follow-up duration was generally short: only three reviews included studies with follow-up of 1 year or more, so there is little information on long-term adherence to herbal remedies.

The effect of the most promising medicinal plants was similar to that of standard oral hypoglycaemic agents. In a meta-analysis, metformin monotherapy lowered HbA1c by 1.12% (95% CI 0.92–1.32) versus placebo. Metformin added to oral therapy lowered HbA1c by 0.95% (0.77–1.13) versus placebo added to oral therapy (Hirst et al., 2012). In another meta-analysis, metformin reduced FPG by −2.0 mmol/l (95% CI: −2.4, −1.7) (Johansen, 1999). Other conventional hypoglycaemic medications have a smaller effect, for example sitagliptin lowers HbA1c by −0.94% and FPG by 1.2 mmol/l (Aschner et al., 2006).

Several mechanisms of action explain the effect of medicinal plants. Firstly, many plant products contain gel-forming fibres which delay gastric emptying and interfere with glucose absorption from the intestines–for example Aloe vera (Suksomboon et al., 2016), Fenugreek (Madar and Shomer, 1990) and Psyllium (Gibb et al., 2015). Secondly, some medicinal plants contain substances which inhibit enzymes involved in digestion of carbohydrates (eg α-amylase, α-glucosidase), such as nettle (Ziaei et al., 2020) and the Chinese formula Jinqui Jiangtan (Gao et al., 2019). Third, others stimulate release of insulin; these include Fenugreek seeds (Neelakantan et al., 2014) and Nigella sativa seeds (Daryabeygi-Khotbehsara et al., 2017). Fourth, some medicinal plants inhibit gluconeogenesis, including Nigella sativa (Daryabeygi-Khotbehsara et al., 2017). Fifth, some, such as nettle (Ziaei et al., 2020), mimic the effect of insulin by increasing peripheral uptake of glucose, while others such as Nigella sativa induce insulin sensitivity (Daryabeygi-Khotbehsara et al., 2017).

Dietary and lifestyle advice for patients with diabetes rarely includes information on natural remedies, herbs and spices that can help with glycaemic control. The results presented here can guide patients who wish to try herbal supplements and foods as part of their self-care and diet, and clinicians who wish to advise them. Several of the remedies tested are effective and safe. Many of these herbs and spices with clinically assessed hypoglycemic properties are common food products, and as such generally considered very safe. Some can easily be incorporated into the diet–for example in some studies fenugreek seed powder was mixed with flour for baking chapatis, to reach a total daily dose of 100 g (Neelakantan et al., 2014); but the most effective preparation appeared to be a standardised extract of Fenugreek seed total saponins given in six capsules three times daily after meals (Lu et al., 2008). Other herbs can easily be purchased without a prescription (for example Aloe vera, Psyllium fibre, and Nigella seeds). However, it would be necessary to ensure that an adequate dosage is taken of the most effective preparations. The most effective preparation of Aloe vera appeared to be freshly extracted juice, followed by powdered gel in capsules (Suksomboon et al., 2016). In the case of Psyllium, the most effective preparation appeared to be the seed husk of Plantago ovata Forssk (Ziai et al., 2005). For Nigella sativa, the seed powder (at a dose of 2 g daily) was more effective than the oil (Daryabeygi-Khotbehsara et al., 2017). It is equally important to inform patients and clinicians about remedies which appear to be ineffective–such as tea extracts–and those for which there is insufficient evidence of effectiveness–for example cinnamon and ginseng.

Firstly, some of the meta-analyses were performed more than 5 years ago and need to be updated to include the most recent trials. Some reviews were not performed to the highest standards and could be improved. In particular we recommend that the meta-analysis on Fenugreek should be updated because this appears to be one of the most effective remedies but the systematic review was done in 2014 (Neelakantan et al., 2014). A later systematic review suggested an even greater effect but incorrectly reported some of the underlying data (Gong et al., 2016). It would also be useful to perform a network meta-analysis to estimate the relative effects between the different herbal interventions.

Secondly, it would be interesting to evaluate the impact on glycaemic control of including information on effective medicinal plants and herbal remedies within dietary and lifestyle advice for patients with type 2 diabetes. These may have an additional benefit, and for some patients may be more acceptable, so may be a useful addition to the “menu” of options. This information would need to include clear instructions on the most effective preparations and dosages, and to warn patients about potential side-effects.

Thirdly, this review found a large number of potentially effective medicinal plants for which there is insufficient evidence of effectiveness. For example, Nettle (Urtica dioica) appears to have a significant effect on HbA1c and FPG (Ziaei et al., 2020) but the confidence intervals are very wide. Larger trials are needed to provide a more precise estimate of efficacy. Although it appears effective, the results on Psyllium were at high risk of bias because the review was undertaken by a company selling it–a higher quality review, with low risk of bias, would be helpful. In some studies, cinnamon appears to significantly reduce FPG, but not HbA1c. However, there is a wide variety of cinnamon species, preparations and doses–it is likely that some are more effective than others. Further research is needed to identify the most effective preparations and dosages, and to conduct high-quality clinical trials of these.

Fourth, for the majority of the 1,200 remedies which have been traditionally used in the treatment of diabetes (Simmonds et al., 2006), no meta-analyses and/or no RCTs have been conducted. Some of these have preliminary evidence of effectiveness, for example on post-prandial glucose; these include the Ayurvedic remedy Gymnema sylvestre (Leach, 2007) and the West African tree Moringa oleifera (Sissoko et al., 2020). It is important to conduct high-quality clinical trials of these (at low risk of bias, using a standardised, replicable dosage and preparation, and measuring HbA1c after at least 12 weeks).