A convenient diagnostic tool for discriminating adult-onset glutamic acid decarboxylase antibody-positive autoimmune diabetes from type 2 diabetes: a retrospective study

Background The glutamic acid decarboxylase antibody (GADA) test, commonly used to diagnose autoimmune diabetes, is not cost-effective in areas of low prevalence. The aim of this study was to develop a convenient tool to discriminate adult-onset GADA-positive autoimmune diabetes from type 2 diabetes (T2DM) in patients with newly diagnosed diabetes. Methods This retrospective cross-sectional study, conducted at Changhua Christian Hospital in Taiwan, collected electronic medical record data from January 2009 to December 2018. Patients were divided into a case group (GADA+, n = 152) and a reference group (T2DM, n = 358). Variables that differed significantly between the groups were subjected to receiver operator characteristic analysis to establish cutoff values. Discriminant function analysis was then employed to discriminate the two groups. Results At the onset of diabetes, the GADA+ group was younger, with lower body mass index (BMI), higher hemoglobin A1c (HbA1c), higher high-density lipoprotein cholesterol (HDL-C), and lower total cholesterol and triglycerides (TG). Five major factors were identified to form the linear discriminant functions: BMI, age at onset, TG, HDL-C, and HbA1c. BMI < 23 kg/m2 was the most important factor, followed by TG < 98 mg/dL, HDL-C ≥ 46 mg/dL, age at onset < 30 years, and HbA1c ≥ 8.6%. The overall accuracy of the linear discriminant functions was 87.1%, with 84.2% sensitivity and 88.3% specificity. Conclusions Routine tests in diabetes care were used to establish a convenient, low-cost tool that may assist in the early identification of adult-onset GAD+ autoimmune diabetes in clinical practice.

116 reviewed the medical records from January 1, 2010 to December 31, 2018, for each patient of the 117 reference group to confirm the diagnosis and to reduce the possibility of misclassification to the 118 maximum extent. Thus, 15 were excluded from the 406 patients identified for assessment ( Figure   119 1). 120 121 Data collection 122 The data for both groups were obtained from the hospital's electronic medical record system. 123 Basic data (age at the onset of diabetes, and gender), personal habits (smoking and drinking), 124 height, and body weight were those collected at the initial visit for the diagnosis of diabetes. The 125 laboratory data used in the analysis, including HbA1c, creatinine (Cr), and glutamic pyruvic 126 transaminase (GPT) levels, were those measured closest to the first visit. The lipid profile 127 selected, including total cholesterol, HDL-C and TG levels, was that measured 2-12 weeks after 128 the first visit, documenting the use of any lipid-lowering agents (fibrates or statins) before the 129 lipid profile test.  To establish a convenient diagnostic tool for discriminating the case group (GADA+) from 151 the reference group (T2DM), the major discriminating factors were first identified and then used 152 to form discriminant functions by discriminant function analysis.

153
Step 1: Differences between the two groups for selected variables were evaluated by 154 Student's t-test for continuous variables and chi-square test for categorical variables. The 155 continuous variables with statistical significance were tested for normal distribution using the 156 Kolmogorov-Smirnov test, and data with P-values of <0.05 were regarded as non-normally 157 distributed. The non-normally distributed variables were transformed to ordinary variables for 158 further analysis. A cutoff value for each of these variables was determined from the point on the 159 receiver operating characteristic (ROC) curve with the minimum distance to the upper left 160 corner, calculated as the square root of [(1 -sensitivity)² + (1 -specificity)²].

161
Step 2: Discriminant function analysis using the enter method was then applied to establish Manuscript to be reviewed 163 0.3) in the structure matrix. These major discriminating variables were then analyzed, again with 164 the enter method, to determine the standardized canonical discriminant function. 165 All the tests were two-tailed with a significance level of 0.05. SPSS version 25 software 166 (IBM Corp., Armonk, NY, USA) was used for the analysis.

173
This study included a total of 510 Chinese patients who were ethnically homogeneous. In the 174 GADA+ group, 84 patients (55.3%) had higher GADA titers (≥32 U/mL) than the others. Table   175 1 summarizes the clinical features of the patients at the time of diagnosis of diabetes. The 176 GADA+ group was significantly younger, with lower BMI, higher HbA1c and HDL-C levels, 177 and lower total cholesterol, TG and GPT levels at onset of DM. There were no differences 178 between the two groups in the gender ratio, Cr levels, prevalence of smoking, or the use of lipid-179 lowering agents (fibrates or statins).

184
The discriminant function analysis identified five major discriminating variables from the    The minimum distance between a point on the receiver operating characteristic curve and the upper left corner, calculated as the square root of [(1-sensitivity)² + (1-specificity)²]. The point with the minimum distance was used to define the cutoff value.