There has been increasing attention to 1,5-AG as a new kind of glycemic index. Several studies have explored the possibility of using 1,5-AG to diagnose diabetes for its relatively convenient and low-cost measuring procedure. Yamanouchi et al.  found the best cut-off point was 14.0 µg/mL and Goto’s group  recommended 14.2µg/mL. Our previous study explored the reference intervals of 1,5-AG and found that the inferior threshold of the healthy population was around 14–16 ug/mL . One study from Shanghai  pointed out that the best point of 1,5-AG for screening diabetes was 15.9 ug/mL, and the sensitivity and specificity were 69.2% and 72.3%, respectively.
As we know, this research explored the cut-off points of 1,5-AG for diagnosing diabetes in China based on diabetic-specific retinopathy for the first time. We found 13.05 µg/mL was the most effective point for diagnosis. This was in accordance with former conclusions [4, 5, 13, 21]. The sensitivities were partly sacrificed to ensure higher specificities to avoid overdiagnosis of diabetes, causing an extra burden on the healthcare system. This was why our recommended point was a little lower than some previous results. We also analyzed other points with a relatively high Youden index and found that elevation of the cut-off point may raise the sensitivity a little, but reduce the specificity and PPV greatly. Thus, 13.05µg/mL was a reasonable cut-off point.
We further examined this point after excluding those participants using hypoglycemic drugs and found 1,5-AG had similar ability with FPG in discriminating diabetic-specific retinopathy with a specificity of 93.5%. Moreover, 1,5-AG had even higher sensitivity and PPV than FPG in this group. Epidemiology  showed that 50% of Chinese patients with diabetes only had elevated postprandial blood glucose, which meant using FPG only would miss large quantities of diabetes. A previous study found that 1,5-AG had a good correlation with blood glucose, especially postprandial glucose, and this relationship existed even in impaired glucose tolerance and pre-diabetes groups [23–25]. In our study, we did not practice OGTT among patients who had already been diagnosed with diabetes or had an FPG ≥ 7.0 mmol/L for the sake of safety. However, we still found a significant negative correlation between 1,5-AG and 2hPG in the diabetic-specific retinopathy group, with a correlation coefficient of -0.753. Therefore, we supposed that 1,5-AG provided an advantage in diagnosing diabetes with elevated postprandial glucose.
According to the guideline , two abnormal test results from the same sample or in two separate samples are required for diagnosis in those who do not have typical clinical symptoms. Many early-stage patients often have an absence of classic symptoms in the clinic, which means a repeated FPG test on another day or an OGTT test is required. And this may lead to a waste of money and time. Since 1,5-AG was effective with postprandial glucose, the combination of 1,5-AG and FPG could substitute for OGTT in some way. We found that 1,5-AG in combination with FPG or HbA1c could both improve the specificities over 95%, and also increase the PPVs, thus preventing overdiagnosis and overtreatment of diabetes. Although a previous study  found that glucose load could slightly elevate extracellular 1,5-AG levels, it had a negligible impact on serum 1,5-AG levels due to the existence of a large 1,5-AG pool in the human body . Thus, we considered the level of 1,5-AG to be relatively stable in one day and independent of fasting. Random blood samples were qualified enough for diagnosis. We suggested choosing two to three indicators based on the individual conditions of patients to simplify the diagnostic process.
Previous studies tended to choose moderate DR as the gold standard when exploring the threshold of one glycemic indictor. This is because in addition to hyperglycemia, hypertension, and hyperlipidemia are also risk factors for DR [28, 29]. In this research, among 196 patients with any kind of DR (ETDRS ≥ 21), only 88 had diabetes (including newly diagnosed diabetes and self-reported diabetes). Among the remaining 108 participants, 80 had hypertension or hyperlipidemia. The area under the ROC curve based on ETDRS ≥ 21 was only 0.66, which meant 1,5-AG had a poor ability to discriminate between non-specific DR. Therefore, we chose ETDRS ≥ 31, identified as diabetic-specific retinopathy, as the gold standard, referring to some previous studies [10, 30, 31]. According to Wisconsin research , the definition of level 31 was “Microaneurysms and one or more of the following: venous loops 31 µ or greater; questionable soft exudate, intraretinal microvascular abnormalities or venous beading, and retinal hemorrhages.” However, there were still 11 participants without diabetes in the group of ETDRS ≥ 31 (Supplementary Fig. 3). Interestingly, 7 of these 11 individuals had 1,5-AG levels under 13.05 µg/mL. We could not explain whether these people were non-diabetic individuals with shifting 1,5-AG levels or true diabetic patients missed by the OGTT test unless further follow-up investigations were conducted. If raising the level of ETDRS level up to 43 (moderate NPDR), only 20 individuals remained in this group, and only 1 of them was without diabetes. This is in accordance with the comparatively low prevalence of DR in China . The area under the ROC curve based on level 43 could be up to 0.952 (Supplementary Fig. 2). The most effective point to diagnose was still 13.05µg/mL, with sensitivity and specificity up to 95% and 90.7%, respectively. DR became more specific with the rising ETDRS level, becoming more representative of diabetes. However, excessive strict definition for diabetic-specific retinopathy would cause a large amount of missing data on diabetes. Taken together, we still chose retinopathy with ETDRS ≥ 31 to indicate the existence of diabetes.
Our research had several limitations. First, we considered whether to exclude those participants using hypoglycemic drugs, as many previous studies did. In this study, most diabetic-specific retinopathy individuals were on hypoglycemic treatment. Excluding these participants would lead to huge numbers of missing data and cause bias. However, inclusion of these people could influence the level of 1,5-AG since it was sensitive to glucose fluctuations, resulting in a predicted threshold higher than the true level. This contradiction has been discussed at length in many other studies [10, 11, 32, 33]. Our decision was to establish the threshold in the total population and then examined it in the non-treated group. We found that the optimal diagnostic point remained effective after excluding individuals under treatment. Second, this was a cross-sectional study, and we could not tell the relationship between 1,5-AG and the incidence of DR. Third, we found that 1,5-AG was influenced by gender, uric acid, and renal function in our previous study , but we did not adjust these factors in this study. We expect to make improvements in future research.