Is the Presence of Retinopathy of Practical Value in Defining Cases of Diabetic Nephropathy in Genetic Association Studies?: The Experience With the ACE Insertion/Deletion Polymorphism in 53 Studies Comprising 17,791 Subjects

We used a preexisting dataset based on 47 studies published from January 1994 through March 2004 that examined the association between ACE I/D and diabetic nephropathy (3). This dataset was subsequently expanded in 2006 by the addition of six later studies to a total of 53 studies comprising 17,791 subjects (5–10). Briefly, studies were considered if they provided sufficient information for a comparison of the ACE I/D genotype distribution between case and control subjects. Case subjects were type 1 or type 2 diabetic subjects fulfilling the minimal criterion of microalbuminuria, whereas control subjects were defined predominantly on the basis of normoalbuminuria. Of the 53 studies, 11 specifically required the concomitant presence of diabetic retinopathy when defining cases of diabetic nephropathy (Table 1).

Funnel plots of the effect estimate based on log-odds ratio were plotted against its SE to evaluate the possibility of publication bias (11). The magnitude of the genetic association between ACE I/D and diabetic nephropathy was obtained by calculating the odds ratio (OR) and its associated 95% CI. A random effects model was used based on the assumption that the studies represented a random sample from the larger population of such studies, with each having its own underlying effect size. Under this model, it is assumed that the study-specific OR varies in response to a mean population effect size. Because the random effects model takes into account the interstudy heterogeneity, such as differences in study design and case definitions for diabetic nephropathy, it provides a more conservative evaluation of the significance of the association than one based on fixed effects (12).

A total of 53 studies (n = 9,556 case and 8,235 control subjects) fulfilled the criteria for inclusion in this review (Table 1). Twenty-one studies involved type 1 diabetic subjects (n = 4,154), while the remaining 32 studies were conducted on patients with type 2 diabetes (n = 13,637). The potential presence of publication bias was assessed using funnel plots of the estimate of log-odds ratio for the genotype II versus DD/ID against its SE (Fig. 1A). Considerable scatter was observed around the pooled log-odds ratio estimate when the reciprocal of the SE was small and approached convergence to form a symmetrical funnel, as this reciprocal increased when all 53 studies were assessed. Similarly, there was no evidence of such bias when the 53 studies were analyzed separately, depending on whether they required the concomitant diabetic retinopathy in the case definitions (Fig. 1B and 1C).

The overall pooled OR for all 53 studies was 0.78 (95% CI 0.70–0.87; P < 0.001), which indicated a significant protection against diabetic nephropathy for genotype II compared with carriage of the D-allele (Fig. 2). The pooled OR for the 11 studies (n = 3,413) requiring diabetic retinopathy in the case definition was 0.68 (0.53–0.86; P = 0.002), and this was not significantly different from the pooled OR of 0.81 (0.71–0.92; P = 0.001) obtained from the 42 remaining studies (n = 14,378) that eschewed the corroborative presence of diabetic retinopathy (P = 0.198) (Fig. 2).

In subgroup analyses on 21 studies conducted on 4,154 type 1 diabetic patients, the overall pooled OR was 0.84 (95% CI 0.68–1.05; P = 0.119). The pooled OR for five studies (n = 1,759) requiring diabetic retinopathy status in case subjects was 0.78 (0.58–1.06; P = 0.110), and this was similar to the pooled OR of 0.85 (0.63–1.13; P = 0.255) for the remaining 16 type 1 diabetes studies (n = 2,395) (P = 0.704). In corresponding subgroup analyses, the overall pooled OR was 0.75 (0.66–0.86; P < 0.001) for the 32 studies comprising 13,637 type 2 diabetic patients. The pooled OR for six studies (n = 1,654) requiring diabetic retinopathy in the case definitions was 0.54 (0.36–0.82; P = 0.004), which was not significantly smaller than the pooled OR of 0.79 (0.69–0.91; P = 0.001) for the 26 remaining type 2 diabetes studies (n = 11,983) (P = 0.087).

We considered the possibility that this lack of effect of diabetic retinopathy may be due to the fact that case subjects in some studies may have had a high prevalence of this complication even though it was not explicitly required in the case definition. Because these studies would have been placed under the category of studies not requiring diabetic retinopathy in the preceding analyses, the anticipated outcome would have been to drive any apparent effect of diabetic retinopathy toward the null. To clarify this issue, we scrutinized the published reports and found that of the 42 studies that did not specifically stipulate diabetic retinopathy in the case definition, 25 did provide sufficient clinical information for us to determine the prevalence of diabetic retinopathy among the case subjects (Table 1). Seven studies, in which diabetic retinopathy was present in at least a majority (80%) of cases, were selected from among these (Table 1). These studies were combined with the 11 studies that specified diabetic retinopathy in their case definitions for comparison with the other remaining studies. The overall pooled ORs for these 18 studies (n = 4,414) was 0.71 (95% CI 0.58–0.87; P = 0.001) compared with 0.82 (0.72–0.94; P = 0.003) for the 35 remaining studies (n = 13,377) (P = 0.249) (online appendix supplementary Fig. 1 available at http://dx.doi.org/10.2337/db08-0581]). Confining our analyses to just the 36 studies that provided information about diabetic retinopathy also yielded similar findings (data not shown). No significant differences associated with the requirement for diabetic retinopathy were observed in either patients with type 1 (P = 0.448) or type 2 diabetes (P = 0.236).

The promise of new insights into the pathogenesis of diabetic nephropathy is fuelling intense efforts to identify genes conferring risk of the complication (13–15). While much of the attention has been placed on attaining large sample sizes to provide power for detecting small effects, another key consideration is the case definition of diabetic nephropathy. In this study, we reviewed the literature on the association of ACE I/D and diabetic nephropathy and found evidence suggesting that stipulating the concomitant presence of diabetic retinopathy in order to corroborate a diagnosis of diabetic nephropathy is unlikely to yield significant benefits when searching for genetic associations.

The inclusion of diabetic retinopathy in the case definition is commonplace in the published literature on ACE I/D. Of the 53 studies, 21% imposed this requirement, and this was comparable in studies focusing on either type 1 or type 2 diabetic patients (24 and 19% respectively). This practice is likely based on several studies that found that only a subset of patients with proteinuria and/or azotemia have kidney biopsies that substantiated a diagnosis of diabetic glomerulopathy, which has subsequently been taken to mean that proteinuria per se is insufficient as conclusive evidence of diabetic nephropathy (16–20). However, in a systematic review of nine published reports and their data (21), Oslen and Mogensen deliberated on this issue and proposed that a very likely reason for the high prevalence of nondiabetic kidney disease was the fact that most of the reports were based on biased groups of patients who were inadvertently selected for such nondiabetic kidney conditions (21). Another potential explanatory factor was the application of a differing criterion for diagnosing glomerulonephritis, a major contributor to nondiabetic kidney disease (21).

In our literature review, several points emerged that should be highlighted. Of the 11 studies that required diabetic retinopathy in the case definition, 9 studies did not require that their control subjects have diabetic retinopathy as well. It was also striking that one study specifically required that its control subjects be free of diabetic retinopathy when all its case subjects had this eye complication (Table 1). Understandably, one would be hard pressed to determine whether any observed association between ACE I/D and diabetic nephropathy, diabetic retinopathy, or even a combination of both complications truly exists.

In practical terms, the requirement for diabetic retinopathy in control subjects will inadvertently diminish the overall size of the study population, which is already limited by the requirement that case subjects have diabetic retinopathy. Unfortunately, on the basis of our present results, this drop in sample size and consequent drop in power come without any tangible reciprocal benefit that would be expected if disease misclassification among cases had been rampant in the absence of diabetic retinopathy as previously suggested (20). Moreover, because recent studies suggested that the majority (70–74%) of albuminuric type 2 diabetic patients do indeed have diabetic glomerulopathy even in the absence of diabetic retinopathy (22,23), it becomes questionable whether genetic associations found in studies using diabetic retinopathy can be readily extrapolated to these diabetic nephropathy patients. Nevertheless, it is noteworthy that the overall pooled OR was slightly but consistently higher in studies where diabetic retinopathy was prevalent, although even with the large dataset under review, this difference failed to reach statistical significance. One may thus consider the possibility that including diabetic retinopathy helps in the identification of potential genetic factors for common underlying traits that may manifest as a joint retinal-renal phenotype.

Several strengths and limitations of our study should be discussed. On a positive note, the meta-analysis was conducted on a substantial dataset comprising 17,791 patients from 53 studies. Moreover, there was no overt sign of publication bias that would argue against the validity of our results, with funnel plot analyses indicating that small studies with negative findings were as likely to be published as large studies with positive findings. In addition, we performed subgroup analyses according to whether the patients had type 1 or type 2 diabetes. This distinction was relevant because of the debate as to whether nondiabetic kidney disease is more common in albuminuric patients with type 1 or type 2 diabetes (16,24).

A main limitation is that our study was restricted to ACE I/D. This decision was borne of necessity because ACE I/D is the most extensively studied polymorphism to date with regards to diabetic nephropathy, and there is a severe lack of extensive studies into other genetic markers. Despite this situation, our study manages to render a first critical insight into the issue. Finally, reports of late have provided evidence that diabetic nephropathy may be associated with specific risk haplotypes at the ACE locus. However, a meta-analysis on ACE haplotypes is precluded due to a paucity of such reports (10,25).

In conclusion, our study using real-life association data suggests that the presence of diabetic retinopathy may be of limited practical value for defining cases of diabetic nephropathy when seeking genetic associations. In addition, the reduced sample sizes arising from such a stipulation may make it harder to detect these associations. Interpretation of the results from such studies could also be hampered by the possible confounding presence of diabetic retinopathy if left uncontrolled.

This study was supported by the National Medical Research Council, Singapore (NMRC/0850/2003, NMRC/1018/2005).

We thank Siti Nurbaya (National University of Singapore) for assisting with the literature review.