sion, in a real life clinical set of patients with type 2 diabetes mellitus, the .... MODY 2 (GCK) and 3 (HNF1A) in Brazilian diabetic families and describe clini-.
EPIDEMIOLOGY—CLINICAL—DIAGNOSIS AND SCREENING
&
up=7.4 and 5.5 years, respectively), 161 (19.9%) and 220 (23.7%) patients died, with an age and sex adjusted mortality annual incidence rate of 2.1% and 2.8%, respectively. In both study samples the target risk score tended to be linearly associated with all-cause mortality (HR for SD increment 1.08, 95% CI: 1.03-1.14, p=0.001, and HR 1.02, 95% CI: 0.98-1.07, p=0.243, respectively). When the two cohorts were pooled and analyzed together, a clear association between target risk score and all-cause mortality was observed (HR for SD increment 1.05, 95% CI:1.02-1.08, p=0.004). This counterintuitive association was no longer observable in a model including age, sex, body mass index, smoking habit, estimated-glomerular filtration rate, albuminuria and anti-diabetic, anti-hypertensive and anti-dyslipidemic treatment as covariates (HR for SD increment 1.00, 95% CI: 0.96-1.04, p=0.852). In conclusion, in a real life clinical set of patients with type 2 diabetes mellitus, the combination of recommended target values of established cardiovascular risk factors is not associated with all-cause mortality.
RAMON CASANOVA, SANTIAGO SALDANA, SEAN SIMPSON, MARY B. LACY, ANGELA SUBAUSTE, CHAD BLACKSHEAR, LYNNE E. WAGENKNECHT, ALAIN BERTONI, Winston-Salem, NC, Providence, RI, Jackson, MS
Statistical models for prediction of incident diabetes are often based on a few variables selected by experts. Here we pursued two main goals: 1) Investigate relative performance of a machine learning method such as Random Forests (RF) and a logistic regression (LR) model for detecting incident diabetes, and 2) Uncover potential predictors of diabetes. The Jackson Heart Study collected data at baseline and in two follow-up visits from 5,301 African Americans. We excluded those with baseline diabetes and no follow-up, leaving 3,633 for analyses. Over a mean 8-year follow-up, 584 participants developed diabetes. Diabetes was defined as current use of insulin or oral antidiabetic agent, self-report or physician’s diagnosis, fasting glucose > 126 mg/dl, or HbA1C > 6.5%. We used a LR model published by ARIC (Schmidt et al., Diabetes Care 2005) which included BMI, waist circumference, systolic BP, age, sex, glucose, HDL cholesterol, triglycerides, and parental diabetes history. The RF model evaluated 95 variables (including those provided to the LR model) from demographic, anthropometric, blood biomarkers, medical history, and echocardiogram data. The dataset was partitioned into training and testing sets 100 times. Each training dataset included 400 participants who developed diabetes during follow-up and 400 who did not. Both RF and LR were estimated for the training sets; the testing sets were used to evaluate performance based on accuracy, sensitivity, specificity and area under the curve (AUC). We also used RF measures to rank the importance of the variables to the model. RF produced Accuracy = 77%, Sens. = 70%, Spec. = 76% and AUC = 0.81 (mean values) versus, for LR, Accuracy = 71%, Sens. = 67%, Spec. = 71% and AUC = 0.76. Among the top-ranked variables HA1C , glucose, renin and waist were detected. This work shows the potential of RF for incident diabetes prediction and detecting subtle patterns in data. Our findings confirm well-known predictors of diabetes and suggest a role for renin in prediction of diabetes.
EPIDEMIOLOGY—CLINICAL—DIAGNOSIS AND SCREENING Guided Audio Tour: Diabetes Advances in Clinical Diagnosis and Screening (Posters: 1522-P to 1529-P), see page 13.
&
1522-P
Irisin and Vaspin as Novel Markers to Predict Metabolic Syndrome in Patients with Type 2 Diabetes
POSTERS
SHIWEI LIU, MINGMING WANG, XIAOHUA LOU, JUN WANG, LI ZHANG, SHUJUN ZHAO, KUI FU, Taiyuan, China, Houston, TX
Epidemiology/ Genetics
1523-P
Prediction of Incident Diabetes in the Jackson Heart Study Using Random Forests
Metabolic syndrome (MS) is a popular public health issue currently, for which insulin resistance and obesity are considered as the major causative factors. Energy metabolism dysfunction with enhanced reactive oxygen species (ROS) production may contribute to the metabolic abnormality of adipose tissue in obesity and diabetes. Irisin, a newly identified myokine and adipokine, can increase energy expenditure and improve insulin sensitivity and glucose tolerance. Vaspin as an adipocytokine was identified with insulin-sensitizing effects. Our aims were to examine the serum irisin and vaspin levels in type 2 diabetes (T2D) patients, investigate the correlation of irisin and vaspin with clinical parameters pertaining MS, and evaluate the performance of irisin and vaspin as markers to predict MS in T2D patients. A total of 260 T2D patients were enrolled. Age, gender, anthropometry, biochemistry parameters, HOMA-IR, and levels of irisin, vaspin and ROS in fasting serum were assessed. Compared to T2D patients without MS, T2D patients with MS had lower serum level of irisin, higher level of vaspin and ROS (P
