3-1- Results of the search
We initially identified 3,325 citations. After discarding duplicates and publications before 2000, 2,936 studies remained for title and abstract screening, which resulted in 66 articles for second round of screening. In addition, 7 studies were manually added. After reading 73 full texts, 35 papers discarded and 38 articles remained entered into the meta-analysis. The reasons for excluding that 35 articles were: studies out of EMRO (n = 4), reporting incidence of nephropathy (n = 2), review or meta-analysis study (n = 5), non-English papers (n = 5), full text unavailable (n = 5), and unclear results or irrelevant papers (n = 14). The flowchart of data extraction is shown in figure 1 and basic characteristics of the included studies in meta-analysis is shown in table 1. Our dataset consists of 38 studies that were published from 2000 to 2019 and contained data on the population-based prevalence of nephropathy in type two diabetes in EMRO. The sample sizes of the included studies ranged from 99 to 54,670 patients, with a cumulative size of 112,235 patients.
3-2- Results of the meta-analysis
3-2-1- Heterogeneity
The output of heterogeneity analysis showed that tau2 was 0.04 (95% CI= 0.02, 0.06), I2 was 99.69% (95% CI= 99.52, 99.81), and the Q-statistic was 9,418.874 (p <0.0001), which all suggested a high heterogeneity in the effect sizes. Also, the high value of I2 indicated that almost all heterogeneity could be due to the between-study variance (fig. 2).
3-2-2- Prevalence of nephropathy in type two diabetes
We found that the summary proportion was 0.2634 (95% CI= 0.2104, 0.3200), which represented 26.34% prevalence of nephropathy in type two diabetes (fig. 2). Also, by performing separate meta-analyses in male and female datasets, the prevalence of diabetic nephropathy was 30.42% [0.3042 (95% CI= 0.2338, 0.3794)] in males (fig. 3) and 22.10% [0.2210 (95% CI= 0.1605, 0.2879)] in females (fig. 4).
3-2-3- Prevalence of nephropathy in diabetes based on sub-group analysis by HDI
In sub-group analysis by HDI, the summary effect proportions were 0.2117 (95% CI= 0.0110, 0.562), 0.2618 (95% CI= 0.1777, 0.3557), 0.2039 (95% CI= 0.1134, 0.3127), and 0.3093 (95% CI= 0.2216, 0.4044) for the four subgroups of low, medium, high, and very high, respectively. Low HDI countries and very high HDI countries had the lowest (21.17%) and the highest (30.93%) nephropathy prevalence in type two diabetes.
As a nature of our analysis (separate random effects models in each subgroup), within-group estimates of tau2 were 0.105 [Q(df = 2)= 426.085, p <0.001], 0.032 [Q(df = 11)= 2723.9, p <0.001], 0.031 [Q(df = 7)= 350.631, p <0.001], and 0.038 [Q(df = 14)= 2912.522, p <0.001], for sub-groups low, medium, high, and very high, respectively. We did not find a significant difference between the four subgroup summary estimates (Q(df = 3)= 2.325, p =0.508), representing that HDI did not share on the true heterogeneity of the summary proportion.
3-3- Results of the Meta-regression
The meta-regression analysis was performed for five different variables, include HDI, publication year, mean duration of diagnosed diabetes, diagnostic test, and mean age.
The slope of the estimated regression line for all of the five different variables were near to horizontal, which did not suggested significant moderating effects on the prevalence of nephropathy in type two diabetes ([HDI] Q(df = 1)= 0.9398, coefficient = 0.0297, Z = 0.9695, p = 0.332; [publication year] Q(df = 1)= 0.124, coefficient = -0.002, Z = -0.3522, p = 0.725; [mean duration of diagnosed diabetes] Q(df = 1)= 1.166, coefficient = 0.0095, Z = 1.0798, p = 0.280; [mean age] Q(df = 1)= 1.5048, coefficient: 0.0088, Z = 1.2267, p = 0.220; [diagnostic test] Q(df = 1)= 0.5268, coefficient: 0.0625, Z = 0.7258, p = 0.468) (fig. 5).
Importantly, in all meta-regression plots, most of the studies were outside the 95% CI boundaries, which implied on the presence of unknown or missed parameters affecting the prevalence of nephropathy in type two diabetes. A zero value of R2, which represents the amount of between-study heterogeneity explained by a modulator, supports it.
3-4- Publication bias
Visual inspection of the funnel plot showed asymmetry in our data (fig. 6). Also, the Egger test shows that the funnel plot was significantly asymmetrical (Z = 2.0983, p = 0.0359). Furthermore, the funnel plot of proportions against sample sizes showed that small-study effect was present in our meta-analysis (appendix 10). Importantly, it should be noticed that the just mentioned asymmetry might not necessarily indicate publication bias; that is, other parameters that interfere with the inclusion of small studies may contribute to this asymmetry in observational studies [7]. First, we previously showed a high between-study heterogeneity. Again, a substantial number of studies fall out of the two limit lines of in forest plot, which confirmed the high heterogeneity. In other words, this high between-study heterogeneity might be due to the particular reasons. Second, we excluded foreign languages’ small studies, the so-called English language bias. Finally, irrespective of the sensitive search strategy, gray literature search and manual search in references for relevant studies, the citation bias might be happened.