Population screening for glucose intolerant subjects using decision tree analyses.

Abstract

The purpose of this study was to develop a method of screening for impaired glucose tolerance and previously undiagnosed NIDDM that could be used preliminary to the administration of an oral glucose tolerance test (OGTT) for final classification of glucose tolerance status. The purpose of a preliminary screening of this type would be to reduce the number of OGTT's needed to identify cases of IGT and NIDDM in the population. We used NIDDM risk indicators and decision tree analysis methods (CART software) to identify subgroups of the population at increased risk. We examined a population of Hispanic (n = 583) and non-Hispanic white (n = 768) subjects without a prior history of diabetes. Subjects were classified as normal, IGT or NIDDM (WHO criteria) based on results from a 75 g oral glucose tolerance test (OGTT). Sensitivity (SEN) and specificity (SPE) of the CART models were calculated using the OGTT as the 'gold standard.' Two approaches to screening were simulated. In the simultaneous approach all risk variables were entered into CART models at once. In the serial approach, risk variables were grouped according to degree of effort required for data collection, and were entered into CART models in stages. Fasting glucose, age and body mass index (BMI) were selected as risk variables by CART when simulating the simultaneous approach (SEN = 91%, SPE = 55%). In the serial approach, CART used age and BMI to eliminate 35% of the population from further screening, and then used fasting glucose, glycohemoglobin, age and BMI to classify the remaining higher risk subjects (SEN = 85%, SPE = 64%). These models suggest that screening for IGT and previously undiagnosed NIDDM can be based on measurement of relatively simple indicators, and yet maintain a level of both sensitivity and specificity acceptable for this type of preliminary screening.