Development of a screening tool using electronic health records for undiagnosed Type 2 diabetes mellitus and impaired fasting glucose detection in the Slovenian population

• Published in: Diabetic medicine 2018-05, Vol.35 (5), p.640-649
• Main Authors: Štiglic, G., Kocbek, P., Cilar, L., Fijačko, N., Stožer, A., Zaletel, J., Sheikh, A., Povalej Bržan, P.
• Format: Article
• Language: English
• Subjects: Adult; Age Factors; Aged; Blood Glucose - metabolism; Body Mass Index; Cross-Sectional Studies; Diabetes; Diabetes mellitus; Diabetes mellitus (non-insulin dependent); Diabetes Mellitus, Type 2 - diagnosis; Diabetes Mellitus, Type 2 - metabolism; Electronic Health Records; Electronic medical records; Fasting; Fasting - metabolism; Female; Glucose; Health risk assessment; Humans; Laboratory testing; Male; Mass Screening - methods; Middle Aged; Models, Statistical; Physical activity; Population; Prediabetic State - diagnosis; Prediabetic State - metabolism; Regression analysis; ROC Curve; Sex Factors; Slovenia; Sugar; Waist Circumference
• ISSN: 0742-3071; 1464-5491
• DOI: 10.1111/dme.13605

Aim To develop and validate a simplified screening test for undiagnosed Type 2 diabetes mellitus and impaired fasting glucose for the Slovenian population (SloRisk) to be used in the general population. Methods Data on 11 391 people were collected from the electronic health records of comprehensive medical examinations in five Slovenian healthcare centres. Fasting plasma glucose as well as information related to the Finnish Diabetes Risk Score questionnaire, FINDRISC, were collected for 2073 people to build predictive models. Bootstrapping‐based evaluation was used to estimate the area under the receiver‐operating characteristic curve performance metric of two proposed logistic regression models as well as the Finnish Diabetes Risk Score model both at recommended and at alternative cut‐off values. Results The final model contained five questions for undiagnosed Type 2 diabetes prediction and achieved an area under the receiver‐operating characteristic curve of 0.851 (95% CI 0.850–0.853). The impaired fasting glucose prediction model included six questions and achieved an area under the receiver‐operating characteristic curve of 0.840 (95% CI 0.839–0.840). There were four questions that were included in both models (age, sex, waist circumference and blood sugar history), with physical activity selected only for undiagnosed Type 2 diabetes and questions on family history and hypertension drug use selected only for the impaired fasting glucose prediction model. Conclusions This study proposes two simplified models based on FINDRISC questions for screening of undiagnosed Type 2 diabetes and impaired fasting glucose in the Slovenian population. A significant improvement in performance was achieved compared with the original FINDRISC questionnaire. Both models include waist circumference instead of BMI. What's new? The increasing prevalence of Type 2 diabetes mellitus calls for more effective screening tests. This study introduces the first predictive models for undiagnosed Type 2 diabetes mellitus and impaired fasting glucose for the Slovenian population. Waist measurement is preferred to BMI in both screening tools. Two simplified models based on data from electronic health records were obtained by removing two questions from the Finnish Diabetes Risk Score (FINDRISC) questionnaire. The proposed models achieved a significantly higher predictive performance in comparison with the routinely used FINDRISC questionnaire.