Few diabetes risk scores are derived from and designed for use in the electronic medical record (EMR). We developed an EMR-derived random blood glucose (RBG) diabetes risk score to detect dysglycemia (diabetes+prediabetes) and improve case identification. A RBG-risk score to detect prevalent dysglycemia (A1C≥5.7%) was derived from EMR data in an integrated safety-net health system. Primary care patients (age 18-64; resulted A1C in 2011-2014; ≥1 RBG 24 months prior to the A1C) were eligible. We excluded patients with diagnosed dysglycemia using ICD codes and lab results. Most recent RBG and EMR-extracted demographics, comorbidities, and diabetes risk factors on the date of the A1C were used to build logistic regression models predicting dysglycemia using backwards selection. Predictors retained in the final regression model were converted to a points-based risk score (Age 0-6; race 0-3, BMI 0-5; hypertension 0-1; RBG 0-13). We report dysglycemia risk, sensitivity, specificity, and PPV for clinically relevant cutpoints. The cohort included 11,387 patients (mean age 48, 65% female, 42% Hispanic, 36% black, mean BMI 32, 29% with hypertension), of whom 42% had dysglycemia. Screening criteria were met by 86% (ADA), 65% (USPSTF), and 40-60% of patients at clinically relevant Risk Scores (9-11). The Risk Score c-statistic was significantly better than both ADA (0.75 vs. 0.56; p<0.001) and USPSTF (0.75 vs. 0.60; p<0.001) guidelines. An example patient with a Risk Score of 9 would be: age 50, BMI 29, white, has hypertension and RBG 112. A score of 9 identified 61% of the sample as high risk (35%) for undiagnosed dysglycemia and yielded 80% sensitivity, 53% specificity, PPV 56%, and NPV 78%. Using a cutpoint of 9, if the Risk Score were run on 1000 patients, 610 would undergo screening, 335 new cases of dysglycemia would be detected, and 85 cases would be missed. An EMR-derived RBG-risk score capable of automation in the EMR outperforms screening guidelines to detect dysglycemia. Further validation studies are needed to determine clinical usefulness.


M.E. Bowen: None. J. Sanders: None. S. Zhang: None. N.O. Santini: None. I. Lingvay: Advisory Panel; Self; Novo Nordisk A/S, Eli Lilly and Company, AstraZeneca, Intarcia Therapeutics, Inc., Sanofi-Aventis. Research Support; Self; Novo Nordisk A/S, Merck Sharp & Dohme Corp., Pfizer Inc., GI Dynamics Inc., Novartis AG. Other Relationship; Self; Sanofi, AstraZeneca, Boehringer Ingelheim GmbH, Novo Nordisk A/S. E. Halm: None.

Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at http://www.diabetesjournals.org/content/license.