The prevalence of diabetes in Afghanistan: a systematic review and meta-analysis

The results were defined using the Preferred Reporting Items for Systematic and Meta-analyses (PRISMA) guidelines [11]. The PRISMA 2009 checklist is attached in supplementary file S1. The protocol of this study was registered with the International Prospective Register of Systematic Reviews (PROSPERO), with protocol registration number CRD42020172624.

A comprehensive literature search was conducted to identify potential articles published on the prevalence of diabetes in Afghanistan. The search was carried out systematically using the following electronic websites: PubMed, Medline, EMBASE, Google Scholar and the Cochrane Library. We considered studies published from inception to April 2020. Using MeSH headings, the terms “Type-2 diabetes”, “Type-II diabetes”, “T2D”, “prevalence”, “Impaired glucose tolerance (IGT)”, “risk factors”, “risk factor”, “glucose intolerance”, “glucose abnormalities”, “non-communicable diseases”, “Afghanistan”, and “Afghan” as well as variations thereof were searched for. Additionally, articles were also searched from reference lists in previously included studies.

The criteria for studies to be included were as follows: (i) articles were published in peer-reviewed journals and reported the prevalence of diabetes; (ii) articles reported population-based or community-based surveys; (iii) articles were published in either English or Pashto.

The criteria used to exclude studies were: (i) irrelevant to diabetes; (ii) not providing clear data; (iii) review articles, case series, or case reports; (vi) relating to the Afghan community living outside of Afghanistan; (v) Articles reported duplicated data (for data published in more than one article) the more up-to-date version was considered and the other articles excluded.

After selecting the relevant articles, two investigators independently screened the titles and abstracts to consider articles for full text review. The investigators then extracted all the necessary data using a standardized data extraction format in Microsoft Office Excel 2013. Any disagreement between the two investigators was resolved by the third investigator whose decision was regarded as final. The extracted information was: first author surname, year of publication, sex ratio (male/female), age, sample size, the prevalence rate of diabetes, the prevalence of smoking, sampling method, study design and the province in which the study was performed. An extract of these collected data is presented in Table 1, using the Preferred Reporting Items for Systematic and Meta-analyses (PRISMA) guidelines [11].

Two authors independently assessed the methodological quality of each of included studies by using Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [18]. We categorized the quality of each included study as Good (G) if it scored at least 70% of the points, Fair (F) if it scored 50–69% of the points, and Poor (P) if its points score was less than 50%.

Meta-analysis was performed using the statistical software R version. 3.6.1 [19]. for Microsoft Windows, using two packages (“meta” and “metafor”). Random effects meta-analysis models were used to investigate the pooled prevalence of diabetes using DerSimonian and Laird’s approach with 95% confidence intervals (CIs) [20]. The inverse of the Freeman-Tukey double arcsine transformation was used to stabilize the variance of each study [21]. A forest plot was used to assess visually the prevalence estimates and corresponding 95% confidence intervals (CIs) across included studies. For the evaluation of statistical heterogeneity across studies, the I²-statistic was used [22]. Heterogeneity was considered as high, moderate, low with I² values of 75, 50 and 25% respectively. To identify potential sources of heterogeneity, meta-regression and subgroup analyses were conducted by age, year of publication, hypertension, sex, province, literacy rate, obesity and the methodological quality of the included studies. Publication bias was initially assessed by visual inspection of the funnel plot and then tested by the Egger regression test [23, 24].

Initially, a total of 64 potential articles was retrieved. Out of these, 36 duplicate articles were removed. After careful review of the titles and abstracts, 12 articles were found irrelevant and then excluded from the process. As the result, only 16 studies were considered for full text reading. Later, 10 studies were excluded after full text reading because they provided no quantitative measure of the prevalence diabetes; overlapped data set; or they were not based in Afghanistan. In the end, only 06 studies were met the inclusion criteria and data were extracted accordingly for the analysis. The flow chart of study selection process is presented in Fig. 1, using the PRISMA flow diagram [11].

The main characteristics of selected studies are summarized in Table 1 [12‐17]. All studies used cross-sectional designs and a random sampling procedure. The included studies were published between 2012 and 2017 while the period to which the data relate was from December 2011 to November 2015. The number of subjects per study ranged from 1129 to 1231, for a total of 7071 subjects across studies. Six studies reported the prevalence of diabetes as an outcome of interest. The existence of diabetes was tested with biological measures in all studies. Considering cut-point definitions of diabetes, all studies used a fasting glucose level of 126 mg/dL. Data were collected from five provinces of Afghanistan. Two studies were conducted in Kabul province [12, 14], one study each in Nangarhar province [13], Kandahar [15], Herat [16], and Balkh [17] provinces. The female proportion varied from 51.2% [14] to 66.50% [12]. The average age of participants varied from 38.3 to 49.5 years. The percentage with hypertension ranged from 30.5% [13] to 33.14% [12], based on five studies. The percentage of obese people varied from 15% [16] to 30.1% [12] (six studies). The percentage of illiterate people ranged from 54 to 73.2% (6 studies). The percentage of smokers ranged from 5.1 to 13.7%. The percentage of overweight people ranged from 32.2 to 48.5% (five studies).

Statistical analysis of the prevalence of diabetes is described in Table 2. The pooled prevalence of diabetes was 12.13% (95% CI: 8.86–16.24%, I² = 99.3%, based on 06 studies) in a total sample of 7071 individuals. The forest plot of the prevalence estimates and their respective 95% confidence intervals (CIs) is presented in Fig. 2. The funnel plot (Fig. 3) showed almost no publication bias which is confirmed by the Egger regression test (p = 0.732). Furthermore, the no publication bias was confirmed by ‘Trim and Fill’ sensitivity analysis—as we did not find any hypothetical missing study. The prevalence of undiagnosed diabetes was 9.70% (95% CI: 4.99–15.74, I² = 97.5%, based on four studies) with a total sample size of 4697 participants.

The pooled prevalence among females was 12.56% (95% CI: 8.48–17.29), while for males, it was 12.0% (95% CI: 8.98–15.55). When stratified by age, the pooled prevalence in age groups 25–34 years, 35–44 years, 45–54 years and 55 years and over was 8.45% (95% CI: 3.65–14.97), 13.02% (95% CI: 7.54–19.70), 17.12% (95% CI: 9.42–26.51), and 19.23% (95% CI: 16.20–22.44), respectively. The pooled prevalence in the 55 years and over age group was the highest of the four age groups, which shows that the prevalence of diabetes increases with age (Table 2).

When stratified by province, the prevalence of diabetes was highest (22.40%) in Kandahar, compared with 11.08% (95% CI: 7.42–15.36) in Kabul, 11.80% in Nangahar, 9.92% in Herat, and 9.18% in Balkh. The difference in the pooled prevalence of diabetes between males and females was insignificant. There was no significant publication bias in all subgroup analyses.

The result of the univariate meta-regression analysis showed that the prevalence of diabetes increased with a mean age (β = 0.0107, 95% CI: 0.0002–0.0208, p = 0.0343, R² = 42.34%), hypertension (β = 0.0377, 95% CI: 0.0034–0.0720, p = 0.0311, R² = 48.94%) and obesity (β = 0.0064, 95% CI: − 0.0007–0. 0135, p = 0.070, R² = 35.64%). The univariate meta-regression analysis also revealed that there was no significant difference by sex (male vs female), sample size, year of publication, smoking, education (illiterate versus literate) or the methodological quality of included studies.