Shaun R Barber1, Melanie J Davies2, Kamlesh Khunti2, Laura J Gray3. 1. Department of Health Sciences, University of Leicester, United Kingdom. Electronic address: srb43@le.ac.uk. 2. Diabetes Research Centre, University of Leicester, United Kingdom. 3. Department of Health Sciences, University of Leicester, United Kingdom.
Abstract
AIM: To describe and evaluate risk assessment tools which detect those with pre-diabetes defined as either impaired glucose tolerance or impaired fasting glucose using an OGTT or as a raised HbA1c. METHODS: Tools were identified through a systematic search of PubMed and EMBASE for articles which developed a risk tool to detect those with pre-diabetes. Data were extracted using a standardised data extraction form. RESULTS: Eighteen tools met the inclusion criteria. Eleven tools were derived using logistic regression, six using decision trees and one using support vector machine methodology. Age, body mass index, family history of diabetes and hypertension were the most frequently included variables. The size of the datasets used and the number of events per variable considered were acceptable in all the tools. Missing data were not discussed for 8 (44%) of the tools, 10 (91%) of the logistic tools categorised continuous variables, external validation was carried out for only 7 (39%) of the tools and only 3 tools reported calibration levels. CONCLUSIONS: Several risk scores are available to identify those with pre-diabetes. Before these are used in practice, the level of calibration and validity of the tools in the population of interest should be assessed.
AIM: To describe and evaluate risk assessment tools which detect those with pre-diabetes defined as either impaired glucose tolerance or impaired fasting glucose using an OGTT or as a raised HbA1c. METHODS: Tools were identified through a systematic search of PubMed and EMBASE for articles which developed a risk tool to detect those with pre-diabetes. Data were extracted using a standardised data extraction form. RESULTS: Eighteen tools met the inclusion criteria. Eleven tools were derived using logistic regression, six using decision trees and one using support vector machine methodology. Age, body mass index, family history of diabetes and hypertension were the most frequently included variables. The size of the datasets used and the number of events per variable considered were acceptable in all the tools. Missing data were not discussed for 8 (44%) of the tools, 10 (91%) of the logistic tools categorised continuous variables, external validation was carried out for only 7 (39%) of the tools and only 3 tools reported calibration levels. CONCLUSIONS: Several risk scores are available to identify those with pre-diabetes. Before these are used in practice, the level of calibration and validity of the tools in the population of interest should be assessed.