Diagnosing type 2 diabetes using Hemoglobin A1c: a systematic review and meta-analysis of the diagnostic cutpoint based on microvascular complications

Aims Diabetic microvascular complications of retinopathy, nephropathy and neuropathy may occur at hemoglobin A1c levels (HbA1c) below the 6.5% (48 mmol/mol) diagnostic threshold. Our objective was to assess the validity of the HbA1c diagnostic cutpoint of 6.5% based upon published evidence of the prevalence of retinopathy, nephropathy and neuropathy as markers of diabetes. Methods Data Sources PubMed, Embase, Cochrane, Scopus and CINAHL from 1990-March 2019, grey literature sources. Study Selection All studies reported after 1990 (to ensure standardized HbA1c values) where HbA1c levels were presented in relation to prevalence of retinopathy, nephropathy or neuropathy in subjects not known to have diabetes. Data Extraction Studies were screened independently, data abstracted, and risk of bias appraised. Data Synthesis Data were synthesized using HbA1c categories of < 6.0% (< 42 mmol/mol), 6.0–6.4% (42–47 mmol/mol) and ≥ 6.5% (≥ 48 mmol/mol). Random-effects meta-analyses were conducted for retinopathy, nephropathy and neuropathy prevalence stratified by HbA1c categories. Random-effects multivariable meta-regression was conducted to identify predictors of retinopathy prevalence and sources of between-study heterogeneity. Results Pooled mean prevalence was: 4.0%(95% CI: 3.2–5.0%) for retinopathy, 10.5% (95% CI: 4.0–19.5%) for nephropathy, 2.5% (95% CI: 1.1–4.3%) for neuropathy. Mean prevalence when stratified for HbA1c < 6.0%, 6.0–6.4% and ≥ 6.5% was: retinopathy: 3.4% (95% CI: 1.8–5.4%), 2.3% (95% CI: 1.6–3.2%) and 7.8%(95% CI: 5.7–10.3%); nephropathy: 7.1% (95% CI: 1.7–15.9%), 9.6% (95% CI: 0.8–26.4%) and 17.1% (95% CI: 1.0–46.9%); neuropathy: 2.1% (95% CI: 0.0–6.8%), 3.4% (95% CI: 0.0–11.6%) and 2.8% (95% CI: 0.0–12.8%). Multivariable meta-regression showed HbA1c ≥ 6.5% (OR: 4.05; 95% CI: 1.92–8.57%), age > 55 (OR: 3.23; 95% CI 1.81–5.77), and African-American race (OR: 10.73; 95% CI: 4.34–26.55), to be associated with higher retinopathy prevalence. Marked heterogeneity in prevalence estimates was found across all meta-analyses (Cochran’s Q-statistic p < 0.0001). Conclusions The prevalence of nephropathy and moderate retinopathy was increased in subjects with HbA1c values ≥ 6.5% confirming the high specificity of this value for diagnosing T2DM; however, at HbA1c < 6.5% retinopathy increased at age > 55 years and, most strikingly, in African-Americans, suggesting there may be excess microvascular complication prevalence (particularly nephropathy) in individuals below the diabetes diagnostic threshold. Electronic supplementary material The online version of this article (10.1007/s00592-020-01606-5) contains supplementary material, which is available to authorized users.


Introduction
The prevalence of diabetes has reached epidemic proportions globally, with 424.9 million affected adults (20-79 y), representing 8.8% of the global adult population. Current projections indicate that this figure will rise to 628.6 million by the year 2045, affecting almost 10% of the worldwide adult population [1]. Type 2 diabetes (T2DM) accounts for Managed by Antonio Secchi.
Stephen L Atkin and W. Garry John are Joint senior authors.
Extended author information available on the last page of the article the vast majority (90-95%) of diabetes cases and is commonly characterized by the inability of pancreatic beta cells to meet the demand for insulin secretion due to a relative deficit of functional beta cells in a setting of peripheral insulin resistance. There has been much debate over the years as to how T2DM should be diagnosed; what should be measured, and the diagnostic targets have changed, being refined as our understanding of the disease has improved along with improvement in analytical methods. The diagnostic criteria for T2DM are established [2], but it is clear that a continuum in blood glucose level exists from normoglycemia to frank diabetes. As such, in 1997 and 2003, the Expert Committee on Diagnosis and Classification of diabetes mellitus recognized a cohort of subjects whose glucose levels did not meet the criteria for diabetes but were too high to be considered as normal [3,4]; this "prediabetic" group exhibited impaired fasting glucose (IFG) [fasting plasma glucose (FPG) levels of 100-125 mg/dL (5.6-6.9 mmol/L) and/or impaired glucose tolerance (IGT)] defined as a 2-h plasma glucose following a 75 g oral glucose tolerance test (OGTT) of 140-199 mg/dL (7.8-11.0 mmol/L)] and represents individuals at high risk for development of T2DM.
Hemoglobin A1c (HbA1c) is considered key for assessing glycemic control in patients known to have diabetes, and several prospective studies using HbA1c have shown a strong, continuous association between HbA1c and the development of diabetes and complications [5][6][7]. An International Expert Committee (IEC) recommended an HbA1c level of 6.5% (48 mmol/mol) as the diagnostic threshold for T2DM diagnosis [8], purporting that individuals with HbA1c levels above this cutoff have a much higher probability of having retinopathy than those below. Both the American Diabetes Association (ADA) and, subsequently, the World Health Organization (WHO) endorsed this opinion, which was entirely based upon the risk of observing diabetic retinopathy, without consideration for other diabetic microvascular complications; however, the subject has to have had diabetes for a period of time for microvascular complications to develop that may occur at a lower HbA1c than 6.5%.
Key studies contributing to the IEC/ADA/WHO diagnostic threshold decision were the results of the cross-sectional Evaluation of Screening and Early Detection Strategies for Type 2 Diabetes and Impaired Glucose Tolerance (DETECT-2) study [9] and three epidemiological studies undertaken in the 1990s on Pima Indians, Egyptians and US subjects enrolled in the National Health and Nutrition Examination Survey (NHANES) study [3,10,11].
The current HbA1c threshold was based on data available at the time of these groups' reports and, even at the time, there was debate as to whether a level of 6.5% (48 mmol/ mol) may be too high since, in many studies, it identified fewer patients as having diabetes than the traditional blood glucose criteria [12]. Since then, a number of studies have been performed with the aim of better characterizing the HbA1c threshold for prevalent or incident retinopathy. The studies prior to 2013 were summarized in a publication by Kowall and Rathmann who looked at retinopathy, nephropathy and neuropathy [13]; since this time, there have been a number of relevant publications which can now be considered.
The key aims of this study were to perform a contemporary systematic review and meta-analysis to verify the HbA1c cutpoint of 6.5% (48 mmol/mol) for the diagnosis of T2DM using currently available retinopathy data and, secondly, to extend the analysis to establish the prevalence of nephropathy and neuropathy at differing levels of HbA1c.

Data sources and search strategy
This systematic review was guided by the Cochrane Collaboration Handbook [14], and followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines [15]. The PRISMA checklist is shown in Supplementary Table 1. A comprehensive systematic search for literature was conducted in the academic databases PubMed, Embase, Cochrane, Scopus, CINAHL and in sources for grey literature in October 2019 (Supplementary Table 2).
Pre-searches to identify relevant search terms, search strategies and information sources were performed in May-October 2019. PubMed's MeSH was used to systematically identify search terms that encompass the variations in terminology for glycated hemoglobin, including thresholds or cutpoints, as well as terms that describe diabetic microvascular complications (Tables 1, 2, 3). The search strategy developed in PubMed was replicated in all databases with eventual search terms or technical variations documented (Supplementary Table 2). The PRESS peer review of electronic search strategies: 2015 guideline statement was used to peer review the search string [16].
All selected search terms were searched in a combination of "Abstract" and "Article Title" (alternatively "Topic" or "Title, Abstract and Keyword") and in MeSH/Subject Headings/Thesaurus when available. A publication year filter to include studies from 1990 up to the search date was applied to account for the predominant lack of standardized/ harmonized HbA1c measurement before this time. In order to ensure literature saturation and inclusion of pre-indexed materials, no additional filters or limitations were included.
In addition to the search in academic databases, Open Grey, Clinical Trials.gov, The New York Academy of Medicine-Grey Literature Report and ProQuest Dissertation and Theses were searched for grey literature. Hand screening of  A review protocol for this study was registered in the Prospero international prospective register of systematic reviews: https ://www.crd.york.ac.uk/prosp ero/displ ay_recor d.php?Recor dID=99410 PROSP ERO20 18CRD 42018 09941 0.

Study selection
Search results were imported into the Covidence systematic review tool where duplicate publications were identified and duplicates excluded. Study eligibility was independently determined by two investigators (AEB and SLA). Where there was a discrepancy, both investigators revisited the publication in question, discussed the results with the group and re-entered their decision into the database.
The criterion for inclusion was for a publication to report the prevalence of microvascular complications by levels of HbA1c. More specifically, studies were only considered if they provided a prevalence measure for a microvascular complication stratified by HbA1c of < 6% (< 42 mmol/mol), 6-6.4% (42-47 mmol/mol) and ≥ 6.5% (48 mmol/mol) in the same study. The reason for that is to ensure rigor in the association with microvascular complications. Exclusion criteria were studies that did not report microvascular complications by different HbA1c level and those published before 1990.
For terminology, a "publication" is a document containing a relevant outcome measure, while a "study" indicates all details pertaining to a specific outcome measure-one publication may contribute multiple studies. A study, such as "prevalence of retinopathy," could include multiple stratified "measures," such as prevalence by age group.

Data extraction and quality assessment
Following screening of records for eligibility, eligible studies had full-text screening (Supplementary Figure 1). Extracted variables included: author(s), publication title, year(s) of data collection, publication year, country of origin, country of survey, study design, study sampling procedure, study population and its characteristics (e.g., sex, age and ethnicity), sample size, HbA1c measurement method, HbA1c stratification outcome measures, complications, retinal photography method, number of eyes photographed, determination of albumin creatinine ratio, measure of peripheral neuropathy. Two investigators (AEB and SLA) independently assessed the full-text articles and determined the eligibility of studies for inclusion in the systematic review.

Quality assessment
Risk of bias (ROB) and precision assessments were performed for all studies included in the review. Guided by the Cochrane approach [17], studies were classified as having "low" vs. "high" ROB on two quality domains assessing (1) consistency in HbA1c diagnostic measurement across all study participants (consistent vs. not consistent) and (2) rigor of sampling methodology (probability-based vs. nonprobability-based). Studies with unavailable information for any given domain were classified as having "unclear" ROB for that domain. Studies including at least 100 participants were considered as having higher precision. For a prevalence of microvascular complications of 1% (see prevalence by HbA1c in Table 1) and a sample size of 100, the 95% confidence interval (95% CI) is 0-5%, an acceptable CI for the

Table 4
Results of meta-analyses of studies reporting complications of diabetes in patients with different HbA1c (%) A minimum of two studies was necessary to conduct a meta-analysis *Imputation of sample size at the median of studies in that category was performed for two studies to avoid loss of information a Q: the Cochran's Q statistic is a measure assessing the existence of heterogeneity in effect size (here, prevalence) across studies b I 2 : a measure assessing the magnitude of between-study variation that is due to differences in effect size (here, prevalence) across studies rather than chance c prevalence of microvascular complications (a lower sample size is needed to detect a higher prevalence). Results of the quality assessment are shown in Supplementary Tables 3  and 4.

Meta-analysis methodology
Meta-analyses were conducted to estimate the pooled mean of diabetes complications in relation to the stratified HbA1c values (extracted overall outcome measures for a specific HbA1c category, such as 6.0-6.4, were substituted with stratified measures such as by sub-levels of HbA1c, that is 6.0, 6.1, 6.2…, or population characteristics). Forest plots were constructed to visualize prevalence measures and 95% confidence intervals (CIs) associated with each of the three outcomes of interest (retinopathy, nephropathy and neuropathy). Estimates for the pooled mean prevalence and 95% CIs were then calculated using random-effects metaanalyses. Here, variances of prevalence measures were first stabilized using a Freeman-Tukey-type arcsine square-root transformation [18,19]. Inverse variance weighting [19,20] was subsequently applied prior to pooling estimates using a DerSimonian-Laird random-effects model [21]. This model assumes a normal distribution for true effect sizes (prevalence) across studies and thus factors in sampling variation and true heterogeneity in effect size [22]. A heterogeneity assessment was further conducted using Cochran's Q statistic to confirm existence of heterogeneity across studies and I 2 to quantify magnitude of between-study variation that is due to true differences in effect size rather than chance [22,23].

Meta-regression methodology
Associations with prevalence and sources of between-study heterogeneity were identified using random-effects univariable and multivariable meta-regression analyses. Predictors considered a priori included: HbA1c levels, age and race. Factors associated with prevalence at p value ≤ 0.10 in univariable analysis were eligible for inclusion in the multivariable model. In the latter, a p value ≤ 0.10 but > 0.05 was considered as showing "good evidence" for an association with prevalence, while a p value ≤ 0.05 indicated strong evidence for an association with prevalence. The magnitude of the association between these predictors and prevalence was determined by calculating, in the univariable analysis, odds ratios (ORs) and their associated 95% CIs, and in the multivariable analysis, by calculating adjusted odds ratios (AORs) and their associated 95% CIs.

Quality assessment
Supplementary Tables 3 and 4 show the results of the summarized and study-specific quality assessments. In sum, 72.7% (n = 16 out of 22) of retinopathy studies, 50.0% (n = 2 out of 4) of nephropathy studies and 66.7% (n = 2 out of 3) of neuropathy studies included at least 100 participants and therefore were considered as having higher precision.
Low risk of bias, assessed as consistency in measuring HbA1c across study participants, was found in 81.8% of studies assessing retinopathy, all studies assessing nephropathy, and 66.7% of studies assessing neuropathy. The majority of studies assessing retinopathy (95.5%) and all studies assessing nephropathy and neuropathy used probabilitybased sampling and hence were also classified as having low ROB on that quality domain.
Overall, studies reporting the prevalence of microvascular complications of T2DM were of acceptable quality: 81.8% of retinopathy studies, all nephropathy studies and 66.7% of neuropathy studies had low ROB on both quality domains. High ROB on both domains was found in only 4.5% of retinopathy studies and none of nephropathy or neuropathy studies.
There was evidence for heterogeneity in prevalence estimates across all meta-analyses (Table 4). P value for Cochran's Q statistic was almost always < 0.0001. I 2 was > 90% indicating that most variability is due to true differences in prevalence across studies rather than chance.

Subgroup Analysis
As a higher prevalence of complications may be anticipated with increasing age, the meta-analysis data of studies reporting retinopathy were stratified by age using both two strata (age ≤ 55 years and > 55 years) and three strata (age ≤ 55 years, age 55-59 years and > 60 years) with levels of HbA1c (Table 5). As anticipated, the pooled means did tend to increase with age, though this trend did not differ substantially from the original analysis.
Certain ethnic populations are also known to develop diabetes complications more frequently, notably African-Americans [46]. A sensitivity analysis was therefore performed on the results of the meta-analysis for studies reporting retinopathy separately for African-Americans and non-Black populations ( Table 6). The results showed the same upward trend for retinopathy with increasing levels of HbA1c in both the Black and non-Black populations, though the African-Americans populations had a notably higher prevalence of retinopathy at every level of HbA1c; the prevalence of retinopathy in non-African-Americans in those studies did not differ to the overall non-African American prevalence.
There were insufficient studies for both nephropathy and neuropathy to perform subgroup analyses.

Meta-regression results
Only retinopathy prevalence had a sufficient number of studies to warrant conduct of meta-regression analysis ( Table 7).
The univariable analysis showed an association with prevalence for HbA1c levels, age and race; these were therefore included in the multivariable model. Here, HbA1c stratification showed a marked and significant increase of retinopathy prevalence at ≥ 6.5% with an AOR of 4.0 (95% CI: 1.9-8.5; p < 0.001) after controlling for the confounding effect of age and race. The model also showed that individuals > 55 years of age had threefold higher odds for retinopathy than younger individuals (AOR: 3.3; 95% CI: 1.8-6.0; p < 0.001). African-Americans also had significantly higher odds for retinopathy prevalence than those of other race (AOR: 11.0; 95% CI: 4.4-27.5; p < 0.001). This model explained 44.8% of the variation in retinopathy prevalence. Fig. 4 Pooled mean prevalence and adjusted odds ratio of retinopathy according to HbA1c. a Prevalence of retinopathy (%) stratified by HbA1c levels of < 6.0%, 6.1-6.4% and 6.5% or greater, showing marked increase in retinopathy prevalence in the 6.5% or greater group. b Adjusted odds ratio for retinopathy prevalence (for age, sex and ethnicity) stratified by HbA1c levels of < 6.0%, 6.1-6.4% and 6.5% or greater, showing marked increase in the odds ratio in the 6.5% or greater group Fig. 5 Pooled mean prevalence of retinopathy stratified in 0.1% increments of HbA1c. The prevalence of retinopathy stratified in 0.1% increments showed no trend of increasing retinopathy for HbA1c ranging from 6.0 to 6.4%, but a marked increase in prevalence at HbA1c of 6.5% and greater

Discussion
The meta-analysis relating stratified HbA1c to the prevalence of retinopathy, nephropathy and neuropathy showed clearly that for moderate retinopathy, the recent data are in accord with a HbA1c diagnostic cutoff of 6.5% (48 mmol/ mol and above) for T2DM, with the inflection point for the increase in diabetic retinopathy prevalence being at 6.5% (48 mmol/mol), as also shown by the stratified analysis using the J-shaped curve. Pooled mean retinopathy prevalence by 0.1% HbA1c changes showed no obvious signal of a rising retinopathy prevalence for HbA1c 6.0 to 6.4%, with the inflection point being at 6.5% or greater. This being said, there were too few studies assessing retinopathy prevalence at HbA1c increments of 6.3% and 6.4% for evidence to be conclusive. The latter analysis also showed considerable retinopathy prevalence at below 6.0% (42 mmol/mol) HbA1c. It should be also emphasized that the stratification of HbA1c and retinopathy was based on the detection of moderate retinopathy and it is unknown whether the diagnostic cutpoint for T2DM may be altered by prevalence studies on minimal diabetic retinopathy. These results differ from that reported in a previous meta-analysis that concluded that HbA1c thresholds could not be identified from the studies Table 5 Results of meta-analyses of studies reporting retinopathy in patients with different HbA1C stratified by age a Q: the Cochran's Q statistic is a measure assessing the existence of heterogeneity in effect size (here, prevalence) across studies b I 2 : a measure assessing the magnitude of between-study variation that is due to differences in effect size (here, prevalence) across studies rather than chance on microvascular complications, though the threshold of HbA1c of 6.5% was strong for severe retinopathy [13]. The heterogeneity in the retinopathy data was controlled for in the meta-regression and showed, in both the univariate and multivariate analyses, the increased odds ratio of 3.2 and 4.05, respectively, with an HbA1c of 6.5% or greater; however, as noted the stratified HbA1c data revealed a J-shaped curve showing accountable retinopathy prevalence at below 6.0%. The meta-regression also revealed the increase in retinopathy prevalence with those aged over 55 years, with an OR of 3.23 and a striking increase in retinopathy prevalence for African-Americans with an OR of 10.73 on multivariate analysis. While this is in accord with the previous data [3,[9][10][11], retinopathy clearly occurs earlier than the cutoff of 6.5%. The magnitude of the association of retinopathy with race found here may not be representative as only 8 studies were available among African-Americans, though it raises the question whether a different diagnostic cutoff for T2DM for African-Americans aged greater than 55 years may be appropriate, as they would appear to be the group at greatest risk [54]. It should also be emphasized that the stratification of HbA1c and retinopathy was based on the detection of moderate retinopathy and it is unknown whether the diagnostic cutpoint for T2DM may be altered by prevalence studies on minimal diabetic retinopathy. Thus, the T2DM diagnostic cutpoint for HbA1c may not be a universal "one size fits all," but may need to be stratified according to age and race.
The main limitations of the 22 studies of retinopathy included in the meta-analysis were that retinal photography was not standardized and that the degree of retinopathy was often reported as moderate or poorly specified [27,38]. As different studies have measured the complications in different ways, this may have affected the outcome and we may therefore have underestimated the prevalence of the diabetic complications. There were few studies focusing specifically on the onset of minimal diabetes retinopathy. In addition, minimal retinopathy may be difficult to differentiate between diabetes and non-diabetes-related retinopathy and therefore only moderate diabetes changes were identified in all studies analyzed. This could suggest that with a standardized procedure specifically identifying minimal diabetic retinopathy, that the inflection point for the onset of retinopathy would be a HbA1c lower than 6.5% (48 mmol/mol), but this is clearly speculative as there is insufficient evidence to support this at present.
Four studies were evaluated which considered the onset of nephropathy, the data showing that the HbA1c cutoff of 6.5% (48 mmol/mol) was appropriate with a significant increase in the urine albumin/creatinine ratio. Again, it is evident that there were significant numbers of subjects with albuminuria in the range of "prediabetes" between 6.0-6.4% (42-47 mmol/mol); however, the main limitations were the small number of studies, heterogeneity of the population and lack of division of the HbA1c level into 0.1% ranges that prevented a more stringent determination. In addition, all of the studies were cross sectional with a need for longitudinal studies to be performed.
There were only three studies with sufficient data that could be evaluated for the onset of neuropathy, and these were inconclusive with no inflection point seen for the HbA1c cutoff, with a prevalence of neuropathy at an HbA1c of 6.0-6.4 (42-46 mmol/mol) being the same as that of 6.5% (48 mmol/mol) and above. The main limitations here include the limited amount of data in too few studies, heterogeneity of the population and poor division of the HbA1c range into 0.1% ranges; however, given the limited data, it is unlikely that division of HbA1c into 0.1% ranges would have allowed a more stringent determination.

Conclusions
In conclusion, based on the strongest data of moderate retinopathy, this systematic review and meta-analysis is in accord that the HbA1c diagnostic cutpoint of 6.5% (48 mmol/ mol) is highly specific for diagnosing T2DM, though the increased prevalence of retinopathy in those aged 55 years or greater and in African-Americans may suggest a lower threshold is appropriate for those groups. However, this analysis highlights that diabetic microvascular complications commonly occur at lower HbA1c values, and the sensitivity of diagnosing T2DM may improve with a lower diagnostic cutpoint. There is a need for studies that standardize the definition of minimal diabetic retinopathy with more sensitive measures of nephropathy and neuropathy to determine if a lower threshold is appropriate. by the Biostatistics, Epidemiology, and Biomathematics Research Core at the Weill Cornell Medicine-Qatar. SLA is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Author contributions AEB and SLA performed the systematic analysis, interpreted the data and wrote the manuscript. WGJ, EE, EK and KGMMA conceived of the study and contributed to the writing of the manuscript. HC and LJAR analyzed and interpreted the data and contributed to the writing of the manuscript. LO performed the search and assisted with writing the manuscript. All authors reviewed and approved the final version of the manuscript.
Funding Open Access funding provided by the Qatar National Library.

Availability of data and materials
All the data for this study will be made available upon reasonable request to the corresponding author.

Compliance with ethical standards
Conflict of interest No authors have any conflict of interest or competing interests to declare.

Consent to participate Not applicable.
Consent for publication All authors gave their consent for publication. Table 7 Results of univariable meta-regression analyses for the prevalence of retinopathy by key factors a The multivariable model explained 44.3% of the variation in retinopathy prevalence b Strength of evidence for an association with prevalence was deemed "good" at p value ≤ 0.1 and "strong" at p Informed consent Since the studies included in this systematic review and meta-analysis are already published work and therefore in the public domain with no individual identifying information, no informed consent was required to undertake this study.

Human and Animal Rights
No animals were used in the performance of this systematic review and meta-analysis.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.