Establishment of Clinical Diagnosis Model of Type 1 Diabetes Mellitus

Shuang Chen,Qi Fu,Yong Gu,Jin Liu,Xin Li,Lei Xiao,Jing Zhu,Yun Shi,Heng Chen,Min Sun,Wei Gu,Mei Zhang,Xiangting Xu,Jian Wang,Feng Chen,Tao Yang

DOI: https://doi.org/10.3760/cma.j.issn.1674-5809.2016.10.006

2016-01-01

Abstract:Objective To develop a satisfactory clinical diagnosis model of type 1 diabetes mellitus (T1DM) for diagnosis and etiological classification of uncertain cases, achieve quantitative criteria for initial judgment without the standard autoantibody test results and identify specific patients needing accurate islet autoantibody assay. Methods 532 newly-diagnosed diabetic patients (diagnosed less than 1 year) were retrospectively studied, including 203 patients were diagnosed as T1DM, 329 patients were diagnosed as type 2 diabetes mellitus (T2DM), who were hospitalized in Department of Endocrinology and Metabolism, the First Affiliated Hospital of Nanjing Medical University between Jun 2012 to Apr 2016. Eligible patients underwent randomization in a 2∶1 ratio to establish(n=355) or validate (n=177) diagnosis models. Logistic regression with common variable to identify T1DM and T2DM was applied to establish clinical diagnosis model of type 1 diabetes(CODE). The receiver operator characteristic (ROC) curves were analyzed to test the model performance. Results Three models(CODE-C0, CODE-C120, CODE-Cauc) containing age at diagnosis time, body mass index (BMI), C-peptide (fasting C-peptide, C0; 120 min postprandial C-peptide, C120;area of the curve of C-peptide, Cauc, respectively) were created. In the establish cohort, the area of the ROC curve(AUC) of these three models were 0.832(95%CI 0.786-0.878), 0.840 (0.793-0.887) and 0.846(0.799-0.892), respectively. In the validation cohort, AUC of these three models&nbsp;were 0.825(0.753-0.897), 0.848(0.782-0.914) and 0.851(0.785-0.917), respectively. With two cut off points, 10.80 and 7.30, the specificity of CODE-C0 diagnosing T1DM and T2DM was 95%.In the CODE-C120 and CODE-Cauc models, both CODE-C120>10.91 and CODE-Cauc>10.85 were indicative of T1DM (95%specificity). A low CODE-C120 (<7.65) and low CODE-Cauc (<7.87) were indicative of T2DM (95%specificity). Patients with scores between high and low cut off points were undetermined type needing further examination of autoantibody and long-term follow-up. Conclusion Through three clinical diagnosis models based on age at diagnosis time, BMI and C-peptide under the circumstance without high-sensitive islet autoantibody results, the specificity of diagnosing T1DM from T2DM can be improved.