1248-P: Estimating Diabetes Duration from Electronic Medical Records (EMRs) with an Intrinsic Diabetes Risk Prediction Model: From Clinical Trial to Real-World Clinical Practice

• DOI: 10.2337/db19-1248-p
• Publication Date: 2019-06-05T00:46:36Z
• Authors: HUI SHAO; SHUANG YANG; CHARLES STOECKER; VIVIAN FONSECA; XIANG CHENG; LIZHENG SHI
• Keywords: Lasso; Medical record; Ordinary least squares; Predictive modelling

We have previously described the utility and high predictive value of Building, Relating, Assessing, Validating Outcomes (BRAVO) model, which was developed from Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial data. However, it uses diabetes duration as a key factor for modelling progression, is not routinely collected EMRs. This study use machine learning estimate with EMR information. Among individuals (N=8,026) complete records ACCORD trial, estimated an ordinary least squares (OLS) absolute shrinkage selection operator (LASSO) pruned decision tree random forest model using 37 baseline predictors such biomarkers, demographics, disease history, medications. Model performance compared root-mean-square Error (RMSE), mean-absolute error (MAS), R-squared. All three indicators were calculated through 10-fold cross-validation 1000 iterations order address over-fitting. Age, current HbA1c value, retinopathy, number oral antidiabetic drugs, Insulin types, sulfonylureas identified patient-reported duration. The achieved best on all (RMSE=6.16, MAE= 4.71, R2=0.30), significantly higher than (RMSE=6.57, 5.03, R2=0.21) LASSO (RMSE=6.60, 5.08, R2=0.20). can be predicted fields included Integration results this prediction equation into models (such BRAVO) will allow patient-specific estimates based solely increase precision when inferring risk level aid clinical making among patients type 2 diabetes. Disclosure H. Shao: None. S. Yang: C. Stoecker: V. Fonseca: Board Member; Self; American Association Clinical Endocrinologists. Consultant; Abbott, Asahi Kasei Corporation, Novo Nordisk Inc., Sanofi US, Takeda Pharmaceutical Company Limited. Research Support; Bayer US. Stock/Shareholder; Amgen BRAVO4HEALTH, Mellitus Health. X. Cheng: L. Shi: