Diabetic retinopathy in predicting diabetic nephropathy in patients with type 2 diabetes and renal disease: a meta-analysis
- First Online:
- Cite this article as:
- He, F., Xia, X., Wu, X.F. et al. Diabetologia (2013) 56: 457. doi:10.1007/s00125-012-2796-6
- 3.5k Downloads
The aim of this meta-analysis is to determine the predictive value of diabetic retinopathy in differentiating diabetic nephropathy from non-diabetic renal diseases in patients with type 2 diabetes and renal disease.
Medline and Embase databases were searched from inception to February 2012. Renal biopsy studies of participants with type 2 diabetes were included if they contained data with measurements of diabetic retinopathy. Pooled sensitivity, specificity, positive predictive value, negative predictive value and other diagnostic indices were evaluated using a random-effects model.
The meta-analysis investigated 26 papers with 2012 patients. The pooled sensitivity and specificity of diabetic retinopathy to predict diabetic nephropathy were 0.65 (95% CI 0.62, 0.68) and 0.75 (95% CI 0.73, 0.78), respectively. The pooled positive and negative predictive value of diabetic retinopathy to predict diabetic nephropathy were 0.72 (95% CI 0.68, 0.75) and 0.69 (95% CI 0.67, 0.72), respectively. The area under the summary receiver operating characteristic curve was 0.75, and the diagnostic odds ratio was 5.67 (95% CI 3.45, 9.34). For proliferative diabetic retinopathy, the pooled sensitivity was 0.25 (95% CI 0.16, 0.35), while the specificity was 0.98 (95% CI 0.92, 1.00). There was heterogeneity among studies (p < 0.001), and no publishing bias was identified.
Diabetic retinopathy is useful in diagnosing or screening for diabetic nephropathy in patients with type 2 diabetes and renal disease. Proliferative diabetic retinopathy may be a highly specific indicator for diabetic nephropathy.
KeywordsDiabetic nephropathy Diabetic retinopathy Meta-analysis Non-diabetic renal disease Renal biopsy Type 2 diabetes
Diagnostic odds ratio
End-stage renal disease
Non-diabetic renal disease
Proliferative diabetic retinopathy
Quality assessment of studies of diagnostic accuracy included in systematic reviews
Summary receiver operating characteristic
The incidence and prevalence of diabetes mellitus have been increasing. New figures indicate that, if no urgent action is taken, the number of people living with diabetes will rise from 366 million in 2011 to 552 million by 2030 . In adults, type 2 diabetes mellitus accounts for 90–95% of all diagnosed cases of diabetes in the USA . Approximately 40% of people with diabetes develop diabetic nephropathy (DN), which has become the leading cause of end-stage renal disease (ESRD) in developed countries . From United States Renal Data System reports, the adjusted rate of prevalent ESRD due to diabetes rose 2.2% to 647 per million people in 2009, and the total medical care expenditure for ESRD rose 3.1%, reaching US$29 billion dollars .
Kidney biopsy can discriminate DN from non-diabetic renal disease (NDRD), but it is invasive and not suitable for every patient. NDRD is rare in type 1 diabetes mellitus, particularly in patients with a history of diabetes of >10 years ; however, reports of the prevalence of NDRD in type 2 diabetes mellitus have varied widely from 10% to 85% [6, 7, 8, 9, 10]. One joint analysis of available data on the prevalence of NDRD among patients with type 2 diabetes mellitus revealed that NDRD was evident on kidney biopsy in ∼22% of European and 26.7% of Asian patients . Furthermore, the treatment and prognosis of DN and NDRD are different. Research has shown that diabetic patients with NDRD have significantly better renal outcomes than patients with biopsy-proven DN, since many NDRDs are treatable, and even remittable .
Assessment of diabetic retinopathy (DR) is inexpensive and could be routinely performed during outpatient screening for chronic complications of diabetes. Indeed, previous literature has shown that DR may be helpful in distinguishing the type of kidney pathology in patients with type 2 diabetes mellitus and renal disease [13, 14, 15]. However, the results of these studies are diverse and have been found to have variable predictive value in the different series. In addition, most of the available data are from retrospective studies with small samples that lack a quantified standard. Therefore, it is worth comprehensively reviewing the data on the predictive role of DR in biopsy studies. This meta-analysis focused on both prospective and retrospective biopsy studies, and aims to estimate the overall capacity of DR for predicting DN in type 2 patients with diabetes mellitus and renal disease.
The databases searched included Medline and Embase, from the time of their inception to February 2012. The medical subject headings (MeSH) were ‘Biopsy’ or ‘Pathology’ and ‘Diabetic nephropathy/diagnosis/aetiology/pathology’. The references from retrieved articles and reviews identified in the search were manually inspected to verify further articles. One reviewer (F.H.) performed the search, while a second (X.X.) confirmed the process.
The search yielded 3,361 articles, which were assessed using titles, abstracts and/or full articles. Only papers published in English were included. The inclusion criteria were: (1) patients with type 2 diabetes mellitus and renal disease; (2) identification of renal diseases based on kidney biopsy findings; (3) presence of DR and numbers of patients classified in each renal disease group. We included the latest publication when more than one paper was published on a study. While screening the citations, two reviewers (F.H. and X.X.) independently reviewed the search results to determine article inclusion. In cases of discord, a consensus was reached through discussion with the senior author (F.X.H.).
Data extraction and quality assessment
The same investigators (F.H. and X.X.) each retrieved data using standardised forms, obtaining information on study design, author, publication year, percentage of men, duration of diabetes, presence of baseline proteinuria, methods of evaluating DR, and the inclusion criteria to select patients. Data were collected at baseline in the case of longitudinal studies. The numbers of true-positive, false-positive, true-negative and false-negative results were calculated for each study. Study quality was assessed with the quality assessment of studies of diagnostic accuracy included in systematic reviews (QUADAS) checklist (maximum score 14) . The checklist is structured as a list of 14 questions that should be answered ‘yes,’ ‘no’ or ‘unclear.’
Sensitivity, specificity, positive predictive value, negative predictive value and diagnostic odds ratios (DORs) were calculated for each study after construction of a 2 × 2 table. Cells with a value of ‘0’ in the 2 × 2 tables were replaced with ‘0.5’ for pooling purposes. The pooled estimates with 95% CIs were calculated using a random-effects model . A summary receiver operating characteristic (sROC) curve was performed to assess the interaction between sensitivity and specificity. A weighted AUC was obtained to estimate the diagnostic performance. The I2 test was used to quantify the degree of heterogeneity among studies, with I2 values of 25%, 50% and 75% being tentatively considered low, moderate and high heterogeneity, respectively . The potential presence of publication bias was tested for using the Egger test .
Analyses were performed using Stata statistical software v.11.0 for Windows (Stata Corp, College Station, TX, USA) and Meta-DiSc software (Madrid, Spain) . Statistical tests were two-sided and used a significance level of p < 0.05.
Literature search results and study characteristics
The literature search initially identified 3361 articles, which were reduced to 48 after titles and abstracts had been read. After full-text evaluation, 26 papers remained for analysis [8, 9, 10, 12, 14, 15, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40], including nine prospective studies and 17 retrospective ones.
Characteristics of the 26 studies included in the meta-analysis
First author, year
Patients included (n)
Duration of diabetes (years)
Retinopathy DN/NDRD (n)
No retinopathy DN/NDRD (n)
Quality score (QUADAS)
Consecutive series of patients with type 2 diabetes mellitus and proteinuria
Brocco, 1997 
Ophthalmoscopy after mydriasis
Mak, 1997 
Ophthalmoscopy after mydriasis
Christensen, 2000 
Suzuki, 2001 
Direct ophthalmoscopy after mydriasis
Zhou, 2008 
> 300 mg/day
Patients with type 2 diabetes mellitus using criteria of biopsy for T1DM
John, 1994 
Christensen, 2001 
Fundus photography after mydriasis
Wong, 2002 
Serra, 2002 
Amoah, 1988 
Direct ophthalmoscopy after mydriasis
Kleiknechet, 1992 
Richards, 1992 
Olsen, 1996 
Nzerue, 2000 
Tone, 2005 
Soni, 2006 
Huang, 2007 
Pham, 2007 
Akimito, 2008 
Direct ophthalmoscopy after mydriasis
Lin, 2009 
Ophthalmoscopy after mydriasis
Ghani, 2009 
Chawarnkul, 2009 
Mou, 2010 
Chang, 2011 
>0.3 g/day (50%)
Bi, 2011 
Chong, 2012 
DR predicting DN
The I2 test detected moderate to high heterogeneity among studies. Therefore, random-effects models were used for the meta-analysis. The Egger test showed no significant publication bias (p = 0.77).
Proliferative DR (PDR) predicting DN
This meta-analysis shows that the pooled sensitivity and specificity for the presence of DR differentiating DN from NDRD among patients with type 2 diabetes mellitus and renal diseases were 0.65 and 0.75, respectively. Meanwhile, the pooled positive and negative predictive values were 0.72 and 0.69. In addition, DOR was 5.67 and AUC was 0.75. With regard to the subgroup analysis, DR had a higher predictive value in the screening studies than in those with selected patients based on biopsy criteria for type 1 diabetes mellitus. Moreover, predictive results of DR between prospective and retrospective studies were not very different. Our data also show that PDR (an advanced stage of DR) had a high pooled specificity (0.98) and high pooled positive predictive value (0.96), although the pooled sensitivity and negative predictive value were 0.25 and 0.48, respectively. Furthermore, the AUC of 0.99 represented good discrimination.
The strength of this analysis is that it is the first meta-analysis combining data from previous prospective and retrospective biopsy studies on the predictive accuracy of DR for the clinical differentiation of DN. Guidelines for diabetes and chronic kidney disease (CKD) summarised the predictive value of DR for diagnosis of diabetic kidney disease in biopsy studies until 2005 . Estimates of renal disease in diabetic patients are defined by functional abnormalities, such as microalbuminuria. However, it is important to exclude NDRD because some cases require targeted therapy. Kidney biopsy is the gold standard method for identifying DN, but it cannot be performed on all patients because of contraindications. Furthermore, at least 24 h of observation are recommended after a percutaneous kidney biopsy to assess potential complications . In contrast, assessment of DR is very convenient and is routinely performed as part of a physical examination in outpatient departments. DR and DN are both microvascular complications of diabetes, and some authors have identified correlations of anatomical measures between them. Retinopathy severity was found to be associated with renal anatomical measures when other risk factors were controlled for in patients with type 1 diabetes mellitus . Although DR was found to be an important predictor, it is not known whether its presence can completely differentiate DN from NDRD. Along these lines, our previous study showed that the absence of DR together with a short duration of diabetes may be a useful indication for renal biopsy in patients with type 2 diabetes mellitus and overt proteinuria . One study suggested that diabetes mellitus duration of >10 years together with retinopathy did not exclude NDRD in patients with type 2 diabetes mellitus . A differential diagnostic model composed of five clinical indices (diabetes duration, systolic blood pressure, HbA1c, haematuria and DR) had an advantage in the clinical prediction of DN, with a sensitivity of 90.0% and a specificity of 92.0% . The present studies indicated that DR alone had an imperfect predictive value, and that perhaps more clinical features should be confirmed to construct more precise diagnostic criteria for making a distinction between DN and NDRD.
The 26 studies identified for our meta-analysis varied in certain characteristics. For instance, studies included patients with microalbuminuria (20–200 mg/day) , macroalbuminuria with ranges defined as from >300 mg/day to >3 g/day [9, 14, 15, 22, 23, 24, 25, 26, 27, 28, 29, 31, 34, 36, 37, 38, 39], or not defined clearly. In addition, there was significant diversity in the methods used to assess DR, and only seven studies applied gold standards for DR screening [21, 22, 24, 27, 30, 36, 37, 43]. A third issue was the different categories of renal pathology between studies. Most studies divided patients into two groups (DN group and NDRD group). Otherwise, a near-normal renal structure was taken as NDRD in one study for analysis , and one study omitted a small proportion of overlapping cases (1.8%) and an ambiguous case , whereas in the other studies these overlapping cases were classified in the NDRD groups.
We also found significant heterogeneity, which may be explained by the following limitations. First, only articles published in English were included, although the Egger test did not indicate publication bias. Second, baseline risk factors were not standardised between studies. Of these, ethnic origin affected the susceptibility to DR development even after adjustment for other risk factors [44, 45], although DR showed no significant difference in diagnostic accuracy among the Asian, European and African-American populations (data not shown) in our analysis. The duration of diabetes, a strong risk factor for the development of DR, varied widely from 5 to 12 years in all the studies . Moreover, hyperglycaemia, hypertension and dyslipidaemia have all been confirmed to have an effect on DR [47, 48, 49]. However, individual patient data on these risk factors were not available to allow us to explore the heterogeneity in more detail. Third, our results showed that PDR was a high specific indicator for the diagnosis of DN. However, the findings should be interpreted cautiously because they were based on a small sample (169 patients).
In conclusion, current evidence suggests a potential role for DR in predicting DN in type 2 diabetes mellitus with renal disease. Although the overall test performance was not as high as expected, measuring DR may be considered useful for predicting DN in the light of its simplicity and non-invasiveness.
This work was supported by the grants from the National Key Basic Research Program of China (grant number 2011CB504005), the National Key Technology Research and Development Program of the Ministry of Science and Technology of China (grant number 2011BAI10B05), the National Natural Science Foundation of China (grant number 81170765), and the Guangdong Natural Science Foundation (grant number S2011020002359).
Duality of interest
The authors declare that there is no duality of interest associated with this manuscript.
FH originated the study, acquired data, analysed statistics, interpreted data, drafted the manuscript, and critically revised the manuscript. XX originated the study, acquired data, analysed statistics, interpreted data, drafted the manuscript, and critically revised the manuscript. FXH designed the study, interpreted data and critically revised the manuscript. XFW analysed and interpreted data, and critically revised the manuscript. XQY interpreted data and critically revised the manuscript for important intellectual content. All authors have approved the final version.