Correlations among Diabetic Microvascular Complications: A Systematic Review and Meta-analysis

Early detection of diabetic microvascular complications is of great significance for disease prognosis. This systematic review and meta-analysis aimed to investigate the correlation among diabetic microvascular complications which may indicate the importance of screening for other complications in the presence of one disorder. PubMed, Embase, and the Cochrane Library were searched and a total of 26 cross-sectional studies met our inclusion criteria. Diabetic retinopathy (DR) had a proven risk association with diabetic kidney disease (DKD) [odds ratio (OR): 4.64, 95% confidence interval (CI): 2.47–8.75, p < 0.01], while DKD also related to DR (OR: 2.37, 95% CI: 1.79–3.15, p < 0.01). In addition, DR was associated with diabetic neuropathy (DN) (OR: 2.22, 95% CI: 1.70–2.90, p < 0.01), and DN was related to DR (OR: 1.73, 95% CI: 1.19–2.51, p < 0.01). However, the risk correlation between DKD and DN was not definite. Therefore, regular screening for the other two microvascular complications in the case of one complication makes sense, especially for patients with DR. The secondary results presented some physical conditions and comorbidities which were correlated with these three complications and thus should be paid more attention.


Result
Description of studies. A total of 1,086 articles were identified, and their records were included in Endnote X8 (Clarivate Analytics, Philadelphia, PA, US). After removing 207 duplicates, the remaining 879 articles were screened based on the titles and abstracts (separately) by two reviewers according to our inclusion criteria. Any disagreements about the inclusion of an article for full review were resolved by the third researcher. A full-text assessment was conducted on the rest of the 62 articles. Finally, 26 articles  were included in this meta-analysis. The article search and selection process are summarized in Fig. 1.
The characteristics and JBI scores of these 26 studies are summarized in Table 1. A total of 60,136 participants were involved, and all of the articles included were of high quality according to their JBI scores. Besides, the methods of diagnosis DR, DKD and DN in the included studies were summarized in the Supplementary Table S1.
The association between DKD and DN was also studied. After the pooled analysis, DKD did not have a determined relation with DN (pooled OR: 1.19, 95% CI: 0.95-1.49, p = 0.13) (Fig. 4). Stratification analysis was based on the severity of DKD which included any DKD and overt DKD, yet the subgroup difference was not statistically significant. Furthermore, there was only one article involving the influence of DN on DKD (OR: 1.44, 95% CI: 0.97-2.13, p = 0.07), yet the 95% CI of the OR overlapped 1, and the p value was over 0.05. www.nature.com/scientificreports www.nature.com/scientificreports/ Secondary clinical outcome. The secondary outcome of this study was an analysis of the correlation factors of the diabetic microvascular complications, which was produced by a meta-analysis ( www.nature.com/scientificreports www.nature.com/scientificreports/ only certain correlation factor for DKD was hypertension. For DN, the determined correlation factors were the diabetes duration, HbA1C%, microalbuminuria, age, cardiovascular disease, peripheral vascular disease, and dyslipidemia. Sensitivity analysis. The heterogeneity was large in the outcome of the association between DR and DKD and between DR and DN (I 2 ≥ 50) thus sensitivity analysis was conducted (Fig. 5). When omitting each study, we found no obvious changes to the results thus draw a conclusion that our results on vision efficacy were stable and reliable. Since the outcome of the impact of DKD on DN did not suffer apparent heterogeneity, we did not conduct sensitivity analysis on that.
Publication bias. Begg's test and Egger's test were appropriate for assessing publication bias when the included papers were larger than 10, thus they were only conducted on the outcome of the impact of DR on DN (n = 13). The asymmetry in the Begg's funnel plot (Fig. 6A) and the p value of the Egger's test (p < 0.05) indicated publication bias. Because of this, a sensitivity analysis using the trim and fill method was conducted (Fig. 6B) [%]. The pooled analysis incorporating the hypothetical studies continued to show a statistically significant influence of DN on DR by both fixed effect (OR: 1.25, 95% CI: 1.19-1.31) and random effect (OR: 1.40, 95% CI: 1.08-1.81), thus indicated the stability and robustness of our results.

Discussion
DR and DKD were found to be correlation factors for each other. When studying DR's impact on DKD, a total of 4 original data sets were involved, one of which possessed an abnormal OR value (1.187 × 10 9 , 95% CI: 0-∞). After consideration according to our inclusion criteria, it was involved in the meta-analysis. This resulted in 0% weight, and thus, it played no role in the final pooled value. In the analysis of the influence of DKD on DR, after dividing DR into the any DR and PDR subgroups, we found that along with DR progression, the correlation degree significantly increased (p < 0.01).
DR and DN are each other's correlation factors. When studying the impact of DR on DN, we stratified DN into cardiac autonomic neuropathy, peripheral neuropathy and polyneuropathy and found statistically significant subgroup differences. Accordingly, cardiac autonomic neuropathy and peripheral neuropathy, being the two most common types of diabetic polyneuropathy, had diverse association with DR.
DKD is not a determined correlation factor for DN, and DN did not exert an impact on DKD. The 95% CIs and p values were close to the marginal values, and only one article involved the impact of DN on DKD. Therefore, more studies are needed, and their association still needs to be proven.
We also analyzed the correlation factors for these three complications. Among the DR's correlation factors, hypertension and diabetes duration were in accordance with the results from another meta-analysis 39 , while microalbuminuria, maleness, and age were recognized for the first time in a meta-analysis. Moreover, HbA1C% is commonly considered to be related to DR 3,39,40 ; yet, in this meta-analysis, their correlation was unclear (OR: www.nature.com/scientificreports www.nature.com/scientificreports/ 1.12, 95% CI: 0.81-1.55, p = 0.51). Since only two articles considered this data, more articles are needed, and the results might be different. Moreover, this is the first meta-analysis to identify the correlation factors of DKD and DN, with the exception of high-density lipoprotein cholesterol (HDL-C), which has been reported to decrease the risk of DN in type 1 diabetes 41 . In our study, HDL-C could be regarded to have protective influence on DN in view of its OR (0.99) and 95% CI (0.97-1.00), but the p value was 0.11, which was over 0.05. This is a controversial issue that requires further study, because several articles have reported its protective effect [42][43][44][45] as well as its risk influence 46 on DN.  www.nature.com/scientificreports www.nature.com/scientificreports/ Our study did have some limitations. First of all, potential publication bias was indicated by the asymmetry of Begg's funnel plot on the outcome of DR's influence on DN. The trim and fill sensitivity analysis did not change the general result that DR had a risk correlation with DN (although the strength was slightly attenuated), suggesting that those unpublished negative studies did not influence our results. Besides, we focused on cross-sectional studies which may only indicate the risk correlation but not the inferences of cause and effect. In addition, the disease diagnostic criteria varied across the studies, although they were all standardized and did not influence the stability of the outcomes. Fourthly, the adjusted confounding factors, which were the controlled covariates for multivariable logistic regression, varied in the included studies and thus might cause some heterogeneity. At last, the statistical method of our included studies was confined to multivariate logistic regression which might lead to some bias, however, we could obtain high quality data for meta-analysis.
By contrast, this was the first meta-analysis to evaluate the risk correlations among the diabetic microvascular complications. With data from 26 articles, a significant correlation was found between DR and DKD, as well as between DR and DN, which demonstrated that screening for the other two microvascular complications in the presence of DR is essential. In the future, more studies are needed to further analyze the association between DKD and DN. In addition, certain physical condition and comorbidities may relate to these three complications, and more attention should be paid to them.

Methods
Search strategy. Two independent reviewers (J. Li and Y. Cao) performed a systematic search of PubMed, Embase, and the Cochrane Library on January 9, 2018 for articles evaluating the associations among the diabetic microvascular complications. Using the MeSH or Emtree terms as well as free words, the search strategy included the following: [("diabetic retinopathy" AND "diabetic nephropathy") OR ("diabetic retinopathy" AND "diabetic neuropathy") OR ("diabetic nephropathy" AND "diabetic neuropathy")] AND "cross-sectional studies. " Selection criteria. The eligibility criteria were as follows: (1) cross-sectional studies, (2) multivariable logistic regressions were used to analyze the interactions among the diabetic microvascular complications, and (3) the ORs and 95% confidential intervals (CIs) could be obtained. Those studies with no full text, irrelevant topics, and different statistical methods were excluded. Any disagreements about the inclusion of an article for full review were resolved by a third researcher (P. Lu). The rigorous inclusion criteria were established and strictly followed by two independent reviewers (J. Li and Y. Cao) in order to control the selection bias.
Quality assessment. In order to examine the validity of the included data for the meta-analysis, these articles were assessed using the Joanna Briggs Institute (JBI) Prevalence Critical Appraisal Tool, which contains 9  www.nature.com/scientificreports www.nature.com/scientificreports/ items 47 . The evaluation scores ranged from 0-9, with <3 defined as "low quality, " 3-6 as "moderate quality, " and >6 as "high quality. " Data extraction. The characteristics extracted from the eligible articles included the first author's name, publication year, country where the study was conducted, year of data collection, study setting, number of samples, gender, mean age of the participants, diabetes subtypes, diabetes duration, glycated hemoglobin and the adjusted confounding factors.
The main outcome of this meta-analysis was the interactions among the diabetic microvascular complications. The secondary outcome was the correlation factors of these three diabetic complications. Therefore, the ORs and 95% CIs were extracted for further analysis. Data synthesis. The results from our included studies were combined using Review Manager (RevMan) version 5.3 (The Nordic Cochrane Centre, Copenhagen, Denmark). The effect value in this meta-analysis was the OR, which was obtained through a multivariate logistic regression. Before the analysis, the study heterogeneity