**2. Methods**

This review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta Analyses (PRISMA) statement [26].

#### *2.1. Search Methods*

Two researchers (Z.C. and W.Y.L.) independently searched the following online electronic databases: PubMed, Cochrane Library, Embase (Ovid), CINAHL, Web of Science Core Collection, CNKI, and Wanfang (accessed on 15 April 2020); the latter two being Chinese. The search period was restricted from the earliest records to 15 April, 2020. Key search terms were "sarcopenia", "epidemiology", "prevalence", "aged", "elderly", "older", "China", and "Chinese". Logical operators (i.e., AND, OR) were used to shape the search strategy. Detailed search strategies for PubMed and Wanfang are presented in the Supplementary Materials as examples (S1: Search strategy). Relevant studies were identified through manual searches in the reference list of eligible studies.

Inclusion criteria were as followed: (1) the prevalence of sarcopenia was reported or could be calculated; (2) participants were of Chinese ethnicity; (3) the age was 60 years and over; (4) participants were recruited from the community, clinical settings, or nursing homes; (5) EWGOSP, AWGS, IWGS or FINH definitions of sarcopenia were adopted; (6) muscle mass was measured with DXA or BIA; (7) primary research irrespective of design. Exclusion criteria were (1) studies published as reviews, letters to editors, conference abstracts, expert opinions, case reports; (2) studies published in languages other than English and Chinese.

First, duplicate records were identified and removed. Then, titles and abstracts were screened to remove studies irrelevant to the research questions. Next, full-text articles were retrieved and reviewed for eligibility according to the selection criteria. When multiple papers came from the same dataset, we chose the one with the largest sample size which might be closer to the original cohort in order to avoid those using a particular subset of the whole dataset. The above process was performed independently by the two researchers (Z.C. and W.-Y.L.). Any discrepancies were adjudicated by a third researcher (P.-H.C.).

#### *2.2. Data Extraction*

The following details were extracted from each of the eligible studies: the year of publication, country or area in which the data were collected, research design, sample size, settings where participants were recruited, diagnostic criteria for sarcopenia, measurement of muscle mass, muscle strength, and physical performance and prevalence of sarcopenia. For cohort or intervention studies, only baseline data were extracted. For studies using more than one diagnostic criterion to define sarcopenia, all estimates of prevalence were extracted.

#### *2.3. Critical Appraisal*

Two researchers (Z.C. and W.Y.L.) independently assessed the quality of included studies using a validated tool developed by Hoy et al., which was tailored to assess the risk of bias for prevalence studies with different types of design [27]. The tool evaluates the risk of bias through 10 items. The first four items mainly focus on external validity, which involve the representativeness of the target population and the sampling frame, selection of the sample, and non-response bias. The remaining items address the issue of internal validity, which involves the use of proxy respondent or not, acceptable case definition, validated measurement, consistent mode of data collection, the length of the shortest prevalence period, and correct calculation of prevalence. Each item was rated as "low risk" or "high risk". When information in the article was not sufficient for judgement, that item was rated as "high risk". Following previous literature, the study was considered to have a low risk of bias when 9 or 10 items were rated "low risk"; a moderate risk of bias when 6 to 8 items were rated "low risk"; and a high risk of bias when 5 or less items were rated "low risk" [28,29].

#### *2.4. Statistical Analysis*

The prevalence of sarcopenia varied in different genders and populations recruited from different settings [18]. Accordingly, in this study, the pooled prevalence was obtained separately for each gender and for the population from each setting. For studies which recruited participants from mixed settings, only those which reported the prevalence for each setting were included in the meta-analysis. While more than one estimate was extracted from studies which used more than one diagnostic criterion to define sarcopenia, only the one that was more frequently used in other eligible studies was used for calculation of the pooled estimate.

Random-effect models take into account the possibility that the parameters for the population may vary among studies [30]. Therefore, because of the variations in populations, diagnosis criteria, muscle mass assessments, etc., among the eligible studies, we chose a random-effect model to calculate the pooled prevalence. The *metaprop* command in Stata 15.0 (Stata Corp, College Station, TX, USA) was used to obtain pooled estimates of prevalence [31,32].

We assessed the heterogeneity in the estimates across studies using Cochran's Q test, with a *p*-value < 0.10 indicating heterogeneity [33]. We quantified the heterogeneity using Isquare, with I2 statistics of 25%, 50%, and 75% set as the cut-offs for low, moderate, and high heterogeneity, respectively [33]. If heterogeneity existed, random-effect meta-regression was conducted to explore the potential source of variability in prevalence estimates across studies. Meta-regression can explore the effects of multiple study characteristics on the variance of pooled estimates simultaneously [30]. For this study, we performed multivariable meta-regression (using the *metareg* command) to model the adjusted association between multiple explanatory variables and prevalence estimates. Based on previous evidence, we considered the following study characteristics to be potential sources of heterogeneity: populations from different settings, diagnostic criteria for sarcopenia, assessment of muscle mass, muscle strength and physical performance, and the area of study. In order to have a sufficient power, meta-regression was only performed for covariates reported in at least ten studies [33].

Publication bias refers to the phenomenon that studies with significant results are more likely to ge<sup>t</sup> published than those with negative results, which can result in systematic differences between published and unpublished studies [34]. In the case of observational studies that report prevalence, however, there are no positive or negative results, and no well-established method is recommended to test for this bias in meta-analysis of prevalence studies. Therefore, in this study, we did not check for publication bias.
