This systematic review and meta-analysis was performed following the guidelines outlined in the Cochrane Handbook and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA). A systematic search was carried out across three electronic databases: PubMed, CINAHL, and Google Scholar to identify relevant publications on synthetic pharmaceutical adulterants in weight-loss natural products, covering studies from their inception to July 2024. Additionally, a manual search of reference lists from eligible publications was undertaken to identify relevant articles not indexed in the primary databases. The search strategy employed keywords and their synonyms across three main domains: “weight loss,” “adulterants,” and “natural products.” Initially, search terms were applied individually and subsequently combined using Boolean operators such as “AND” and “OR.”

The detailed search string used was: (weight loss OR slimming) AND (herbal products OR herbal supplements OR herbal remedies OR dietary supplements OR natural products OR food supplements OR herbal medicine) AND (adulteration OR synthetic pharmaceuticals OR contaminants OR chemical adulterants).

Primary studies were eligible for inclusion if they were chemical-laboratory studies reporting the detection of synthetic pharmaceutical adulterants in weight-loss ‘herbal’ preparations, including herbal supplements, dietary supplements, herbal medicines, or food supplements, without language restrictions. Studies were excluded if they were review articles, case reports and conference papers. Additionally, studies with inadequate data to calculate the prevalence rate of adulteration were excluded from the quantitative analysis. Publications were managed using Endnote™ 21 (Clarivate Analytics, PA), which facilitated the importation of references and the removal of duplicate publications.

Titles and abstracts of the identified studies were independently screened by two reviewers (DTAP and THV) to verify compliance with the selection criteria. Following this, a thorough review of the full texts was conducted for potentially eligible articles. For non-English articles, translations were conducted using Google Translate to facilitate evaluation (Jackson et al., 2019). Each researcher independently entered data into spreadsheets, capturing key attributes, including the author, year of publication, country, product information, formulation type, sample collection method, adulterants of interest, analytical methods, and the number of samples adulterated with synthetic pharmaceutical substances.

The adulteration prevalence rate was determined by dividing the number of adulterated samples by the total number of collected samples. Likewise, the prevalence of sibutramine adulteration was calculated as the proportion of samples containing sibutramine relative to the overall sample count.

Two reviewers (DTAP and TCMD) independently evaluated the quality of the included studies using the Toxicological Data Reliability Assessment Tool (ToxRTool), a standardized instrument for assessing the reliability of toxicological research (Schneider et al., 2009). The ToxRTool^® consists of two sections tailored for in vivo and in vitro studies. For this analysis, the in vitro section, comprising 18 criteria, was utilized as it was applicable to all included studies. Each criterion was assigned a score of 1 (“criterion met”) or 0 (“criterion not met”). The evaluation covered key aspects such as (i) identification of the test substance, (ii) characterization of the test system, (iii) description of the study design, (iv) documentation of study results, and (v) the plausibility of both the study design and findings. The total score determined the study’s quality category: category 1 (15–18 points) indicated good quality, category 2 (11–14 points) indicated moderate quality, and category 3 (<11 points) indicated low quality.

Any differences in data extraction or quality assessment were reconciled through agreement between the two reviewers (DTAP and TCMD) with additional investigators (CK and MH).

The overall prevalence of adulteration was estimated across the included studies using the DerSimonian-Laird random-effects model with corresponding 95% confidence intervals (CI) displayed in forest plots (DerSimonian and Laird, 1986). Study heterogeneity was evaluated through χ ² tests and the I² statistic (Borenstein et al., 2009). In cases of substantial heterogeneity (I² ≥ 75%), prevalence estimates were summarized using the median and interquartile range (IQR). To identify potential sources of heterogeneity, subgroup analyses and univariate random-effects meta-regression were conducted. Publication bias was assessed using a funnel plot and Egger’s asymmetry test (Mavridis and Salanti, 2014). All statistical analyses were performed using STATA 17.0 (Stata Corp LLC, College Station, TX, United States), with a two-sided significance threshold of P < 0.05.

A total of 4,007 articles were identified, with 4,003 retrieved from database searches and 4 from manual searches. After duplicate removal, 3,848 articles remained for eligibility screening. Following title and abstract review, 3,813 articles were excluded. The remaining 35 articles underwent full-text assessment, leading to the exclusion of nine studies for the following reasons: focusing on heavy metal contamination (six articles), not being chemical-laboratory studies (two articles), and could not retrieve data on weight-loss products (one article). The final selection process is summarized in Figure 1.

A total of 26 studies were included in the systematic review (Table 1) published in English between 2012 and 2021. It includes nine studies conducted in Asia (35%) (Girish et al., 2020; Hoa et al., 2020; Jin et al., 2017; Karamahito et al., 2021; Phattanawasin et al., 2012; Saichanapan et al., 2020; Wong et al., 2021; Yun et al., 2018; Zhong et al., 2017), seven in Europe (27%) (Ancuceanu et al., 2013; Croitoru et al., 2017; Hachem et al., 2016; Mathon et al., 2014; Monakhova et al., 2012; Reeuwijk et al., 2014; Wu et al., 2020), six in Middle East (23%) (Dastjerdi et al., 2018; Firozian et al., 2021; Jairoun et al., 2021; Kamil, 2016; Khazan et al., 2014; Salahshour et al., 2020), three in the America (12%) (Carvalho et al., 2012; Moreira et al., 2013; Wilson and Masse, 2016) and one in Africa (Ahmed et al., 2019).

Regarding sampling locations, three studies did not specify the location of the sample (Jairoun et al., 2021; Kamil, 2016; Zhong et al., 2017). Out of 23 studies, six (26.1%) obtained samples randomly from community pharmacies (Ahmed et al., 2019; Ancuceanu et al., 2013; Carvalho et al., 2012; Hoa et al., 2020; Moreira et al., 2013; Wong et al., 2021). Six studies (21.6%) collected samples from the internet (Girish et al., 2020; Jin et al., 2017; Karamahito et al., 2021; Mathon et al., 2014; Monakhova et al., 2012; Wu et al., 2020). Three studies (13%) obtained samples from local markets (Khazan et al., 2014; Reeuwijk et al., 2014; Saichanapan et al., 2020) or herb/grocery stores (13%) (Croitoru et al., 2017; Firozian et al., 2021; Salahshour et al., 2020). The remaining five studies (21.7%) gathered samples from mixed sources such as pharmacies and internet, herb/grocery stores and internet or local markets and internet (Dastjerdi et al., 2018; Hachem et al., 2016; Phattanawasin et al., 2012; Wilson and Masse, 2016; Yun et al., 2018).

Various analytical techniques have been applied to detect adulteration. Among these, chromatographic methods are the most commonly used, accounting for 77% of studies (20/26 studies), including GS-MS, LC-MS, (high performance) thin-layer chromatography, HPTLC-UV, HPLC-DAD-MS/MS, HPLC, distance-based paper chromatography, LC-MS/MS, UPLC–MS/MS, LC-ESI–MS/MS, HPLC-DAD, HPLC-PDA, and RP-HPLC-MS/MS. Spectroscopic techniques, such as H-NMR and LF-NMR, were employed in 12% of studies (3/26 studies), while electrophoretic methods, including capillary electrophoresis and porous graphene ink-modified electrodes (PGr-ink/GCE), were also used in 12% of studies (3/26 studies).

Regarding the identified adulterants, a diverse range of synthetic and natural compounds has been detected, including stimulants (e.g., amphetamine, β-Methylphenethylamine [BMPEA], ephedrine, β-Phenylethylamine [β-PEA], yohimbine, caffeine), laxatives (e.g., sennosides, phenolphthalein), diuretics (e.g., furosemide, hydrochlorothiazide, amiloride, chlorthalidone), anorexiants (e.g., sibutramine, orlistat, fenfluramine), antidepressants (e.g., fluoxetine, paroxetine, sertraline, bupropion), and anxiolytics (e.g., diazepam, alprazolam, flurazepam, chlordiazepoxide, lorazepam, bromazepam, clonazepam). Among these substances, sibutramine was the most frequently identified (23/26 studies, 88.5%).

Using the ToxRTool for quality assessment, 17 out of the 26 included studies (59%) met most of the criteria and were categorized as high quality (Ancuceanu et al., 2013; Girish et al., 2020; Hachem et al., 2016; Hoa et al., 2020; Jairoun et al., 2021; Jin et al., 2017; Khazan et al., 2014; Mathon et al., 2014; Monakhova et al., 2012; Moreira et al., 2013; Phattanawasin et al., 2012; Reeuwijk et al., 2014; Saichanapan et al., 2020; Wilson and Masse, 2016; Wu et al., 2020; Yun et al., 2018; Zhong et al., 2017). Eight studies (28%) (Ahmed et al., 2019; Carvalho et al., 2012; Croitoru et al., 2017; Dastjerdi et al., 2018; Firozian et al., 2021; Karamahito et al., 2021; Salahshour et al., 2020; Wong et al., 2021) were rated as being of moderate quality. One study (3%) (Kamil, 2016) was classified as low quality.

Among the 26 studies reviewed, 15 studies (58%) quantified the levels of adulterants detected, with sibutramine being the most frequently identified (12/15 studies, 80%) (Ahmed et al., 2019; Ancuceanu et al., 2013; Carvalho et al., 2012; Firozian et al., 2021; Hachem et al., 2016; Hoa et al., 2020; Karamahito et al., 2021; Khazan et al., 2014; Mathon et al., 2014; Monakhova et al., 2012; Reeuwijk et al., 2014; Yun et al., 2018), followed by phenolphthalein (5/15 studies, 33%) (Ahmed et al., 2019; Ancuceanu et al., 2013; Hachem et al., 2016; Khazan et al., 2014; Yun et al., 2018). The measurement units differed among studies. In five studies, the median concentration of sibutramine was 72.7 mg/g (IQR 16.1–97.3). Meanwhile, six other studies reported a median amount of 8.6 mg/capsule (IQR 2.7–17.8). One study identified adulteration levels ranging from 4–36 mg/day. Phenolphthalein was detected at a median level of 105.6 mg/g (IQR 22.0–189.8) and 51.58 mg/capsule (IQR 34.9–825.0).

A total of 22 studies with sufficient data for estimating the prevalence of adulteration were included in the meta-analysis. A meta-analysis of 22 studies revealed a pooled prevalence rate of synthetic adulteration in weight-loss natural products with a median of 37.5% (IQR 25.9%–49.6%) with a significant heterogeneity (χ² = 295.5, df = 21, P = 0.00, I ² = 92.9%). The median prevalence rates showed substantial variability, ranging from 0% in studies conducted in India (0%, IQR 0.0%–10.4%) (Girish et al., 2020) and Australia (0%, 0.0%–25.9%) (Wong et al., 2021) to 100% in studies in Romania (100%, IQR 34.2%–100.0%) (Ancuceanu et al., 2013) and Thailand (100%, IQR 60.9%–100.0%) (Saichanapan et al., 2020) (Figure 2A).

In terms of sibutramine, the estimated prevalence of sibutramine adulteration in natural weight-loss products was found to have a median rate of 21.8% (IQR 11.9%–33.5%), with substantial heterogeneity (χ² = 115.9, df = 13, P = 0.00, I ² = 88.8%) (Figure 2B).

From a geographical perspective, the highest prevalence rate of adulteration was observed in Europe, with a median of 54.7% (IQR 48.0%–61.4%), followed by the Middle East at 41.3% (IQR 19.4%–64.9%), America at 31.1% (IQR 0.3%–78.4%) and Asia at 24.7% (IQR 9.2%–43.8%).

In terms of publication period, more recent studies (publications during year 2020–2021) reported a lower median prevalence rate of adulteration at 23.3% (IQR 11.3%–37.7%) compared to studies published during 2016–2018 and 2012–2014, which documented median prevalence rates of 54.4% (IQR 28.2%–74.3%) and 45.9% (IQR 17.3%–75.9%), respectively.

The lowest prevalence of adulteration was observed in samples collected from pharmacies, with a median prevalence of 14.8% (IQR 0.44%–38.5%). In contrast, the highest prevalence was found in samples obtained from local markets, where the median prevalence was 79.4% (IQR 38.2%–100.00%). Samples purchased from the internet and general stores (such as herb or grocery stores) exhibited comparable prevalence rates, with medians of 35.8% (IQR 3.7%–76.4%) and 27.7% (IQR 17.8%–38.8%), respectively (Table 2).

Meta-regression analyses were performed to explore potential contributors to heterogeneity in the pooled prevalence of synthetic adulteration. The analysis examined covariates such as geographic region, publication period, and sampling source. However, none of these factors showed a statistically significant relationship with the overall prevalence of adulteration (P > 0.05) (Table 3).

The funnel plot analysis (Figure 3) showed no indication of publication bias, as evidenced by its symmetrical distribution. Furthermore, Egger’s test did not detect any significant small-study effects (P = 0.153).

This study has several strengths, such as a comprehensive literature search without language restrictions, a dual data extraction process, and a rigorous evaluation of study quality. Nevertheless, there are also limitations. First, due to a lack of sufficient information, we were unable to gather detailed data on weight-loss products such as brands, origins, and herbal components. Such information provides essential insights for healthcare providers and regulatory agencies in overseeing the importation and supply of these products, particularly given that preparations originating from Asia have been reported as the most adulterated (Posadzki et al., 2013). Future studies should aim to provide as much detail as possible to offer useful information to the public. Second, we do not address unintentional adulteration, and this will certainly add to the risks to patients and consumers. Third, the included studies lacked detailed information on the sampling technique, making it possible that the sampling was not random in some cases which may have contributed to the high heterogeneity in the prevalence rate. Therefore, the results should be interpreted with caution. Future systematic review and meta-analysis on this topic could utilize more comprehensive databases - such as Scopus, Web of Science, Embase and/or OVID - to address the limitations identified in this study.