The study, which was pre-registered with the OSF Registries (https://osf.io/w5bf9), was conducted according to the PRISMA guidelines [15], and followed a structured protocol that was agreed among the authors in advance of the literature search. Data eligible for meta-analysis included both original research and existing publications identified by systematic review.

Original research

Unpublished data from two studies undertaken by the authors were included in the meta-analysis. Both studies were cross-sectional and assessed the lifestyle and cardiometabolic risk factors of South Asian and white European men and women, without diabetes, aged 40–70 years, who lived in Scotland (UK). Both studies have been described in detail elsewhere [8, 16], and involved radiological assessment of fat distribution in men and women. The methodology for fat measurement and the demographic characteristics of participants with radiological assessment are shown in electronic supplementary material (ESM) Methods and ESM Table 1). In addition, we included new data from the UK Biobank. UK Biobank is a large prospective study that recruited 502,643 participants (response rate 5.5%) between 2006 and 2010, age range 37–73 years, and consented for their records to be linked with routine data (hospital admissions and death registries). Participants attended one of 22 assessment centres across the UK, where they completed a touch screen questionnaire, had physical measurements taken, and provided biological samples as described in detail elsewhere [17, 18]. The UK Biobank imaging study began in 2014, and intends to collect imaging data of the vital organs, including MRI measures of abdominal body fat, by recalling 100,000 participants. At the time of performing the analyses for this study, abdominal MRI data were available for approximately 30,000 participants. We used abdominal imaging data from South Asians without diabetes who were matched for age, sex and BMI with white Europeans without diabetes in a 1:5 ratio to maximise statistical power. The protocol for abdominal fat measurement in the UK Biobank imaging study has been published elsewhere [19, 20].

Systematic review of published data and selection criteria

To identify existing publications, we searched the Embase and PubMed databases from inception to August 2021, combining the MeSH terms ‘obesity’, ‘adipocyte’, ‘liver’, ‘south asia’, ‘asian continental ancestry group’, ‘caucasian’ and ‘european’, and using the keywords ‘obes*’, ‘fat*’, adipos*’, ‘liver?fat*’, ‘fatty?liver*’, ‘south?asia*’, ‘india*’, ‘bangladesh*’, ‘sri?lanka*’, ‘pakistan*’, ‘caucasian*’, ‘white*’ and ‘european*’ with Boolean rules. A search filter for studies related to humans with a restriction to English language was included. Two researchers (JM and SI) screened all the titles and abstracts, and studies were read in full when they fulfilled the selection criteria. The reference lists of eligible studies were hand-searched to find further relevant studies. Grey literature was also searched via the OpenGrey website (https://opengrey.eu/).

We included studies that met the following criteria: (1) participants were men or women aged over 18 years; (2) participants had measurements of abdominal SAT and VAT, and/or liver fat by computed tomography (CT) or MRI; (3) the study included a South Asian group and a comparison group of white European descent; and (4) any study design apart from case reports. South Asian ethnic background was either reported as such in the studies or participants were of Indian, Pakistani, Bangladeshi or Sri Lankan background. In the meta-analysis, we included studies for which we could extract mean values and standard deviations from published or requested data. We only included data stratified by sex. Two researchers (JM and SI) independently assessed the papers for final selection. Any discrepancies were resolved by discussion. A third reviewer (JMRG) was consulted if any unresolved issues persisted.

Data extraction and quality assessment

We developed a data extraction spreadsheet that included the following information: study characteristics (first author, year of publication, number of people of South Asian descent and number of people of white European descent, study design), study sample characteristics (sex, mean age and BMI, mean fasting glucose and insulin, diagnosis of diabetes [yes or no]), test characteristics (method of measuring abdominal and/or liver fat, mean value for fat quantity and standardised mean difference [SMD] for each group). If the numerical data were not extractable from the published data, the authors were contacted via email. We were unable to obtain data for insulin and glucose concentrations for four studies [19, 21,22,23,24,25]. References [22,23,24] are multiple papers referring to one study dataset.

We used a preliminary version of the ROBINS-E tool (risk of bias in non-randomised studies of exposures) to assess the risk of bias in the individual studies selected across seven domains; the results for the individual studies were then summarised to provide an overall study-level assessment regarding the risk of bias (low, moderate, serious or critical) [26]. We also used the GRADE (Grading of Recommendations, Assessment, Development and Evaluations) approach to assess the overall certainty of evidence of the meta-analysis findings to provide an evidence certainty score (very low, low, moderate or high) [27].

Data analysis

We used Stata software version 14.1 (Stata, USA) for statistical analysis. The weighted SMD (with 95% CI) was calculated by combining the mean differences in fat between the two groups in each study using a random-effects model. One study reported hepatic attenuation to assess liver fat, rather than the liver fat percentage [28]. As lower hepatic attenuation implies higher liver fat, the sign of the standardised mean ethnic difference in hepatic attenuation was reversed to make the findings comparable with other studies. Analyses were stratified by sex. We performed two sensitivity analyses: (1) by separating the studies that included any participants with diabetes from those without diabetes to assess whether the presence of diabetes modified the results, and (2) by only including the studies with matched BMI between the ethnic groups. We also performed an analysis stratified by assessment tool (CT vs MRI). Heterogeneity resulting from the mean difference in each study not being identical with the pooled estimate was quantified using the I2 measure [29].

We assessed the risk of publication bias and potential small-study effect by constructing funnel plots, which plot the mean difference from each study against the SEM as a measure of study size [30].

Ethics

Previously unpublished data from studies by Iliodromiti et al and Ghouri et al were included in these analyses [8, 16]. Both studies were approved by the West of Scotland Research Ethics Committee, and performed according to the Declaration of Helsinki. All participants gave written informed consent to participate. The UK Biobank study was approved by the North West Multi-Centre Research Ethics Committee, and all participants provided written informed consent to participate. Ethical approval was not required for the analysis of data from previously published studies.