Background. Because nearly one third of diabetic individuals go undetected well past actual disease onset and much pathophysiologic damage can occur when diabetes is uncontrolled, protocols are needed to target populations at higher risk. The use of anthropometry to develop these protocols is appealing because obesity is the primary modifiable risk factor for diabetes. In addition, several instruments have been developed using variables routinely available at clinic visits, but they have been developed mostly in Caucasian populations.

Aims. In this dissertation, we investigated the use of anthropometry and other established risk factors for diabetes screening purposes using an escalating level of complexity in a diverse array of populations. These pre-screening protocols were designed to identify subjects who should undergo glucose testing according to their level of obesity, determined by simple binary rules based on body mass index (BMI), waist circumference (WC) or waist-hip ratio (WHR) alone, by combining anthropometric indicators, and by including other variables available at routine clinic visits.

Methods. The analysis is based on data collected by the Collaborative Study of Obesity and Diabetes in Adults (CODA). CODA is a consortium of cross-sectional and prospective studies from different parts of the world. To establish the collaboration, we performed a literature search to identify studies reporting the association of general and central obesity indicators with diabetes incidence. Twenty eight cross-sectional studies from CODA were used to develop the pre-screening protocols. We stratified the study sample into seven population groups. We developed models stratified by gender and population group using generalized mixed models to account for the hierarchical structure of the data. We analyzed discriminatory properties and ability to predict screen-detected diabetes of a set of models with different levels of complexity. Meta-analytic tools were used to describe results.

Results. Data collected from the literature search perform a meta-analysis. We compared the strength of association of three obesity indicators (BMI, WC and WHR) with incident diabetes. We found that all three anthropometric variables had similar strength of association with diabetes and there was much heterogeneity of the association unexplained by study characteristics. Using individual study data (CODA data), diabetes pre-screening protocols using anthropometric variables alone are presented in a second paper. We found that the ability to detect diabetes in screening based only on anthropometrics is limited but better with population-specific anthropometric cut-points than with the “one size fits all” established cut-points. The third manuscript (CODA data) evaluates strategies with different levels of complexity, using anthropometric variables alone, linear combinations of two anthropometrics, and linear combinations of variables available at clinic visits. Findings suggest that the performance of any anthropometric variable is similar, little is gained by combining anthropometric variables, and the use of additional information available at clinic visits could (full model) provide further discriminatory ability for screen-detected diabetes. Using the full model, we found that the specificity at the 80% sensitivity varied from 43% to 71% depending on the gender and population group. A common finding in all three papers is the significant differences across population groups in the ability to predict and discriminate risk. Because many people are missed when any pre-screening protocol for diabetes screening is implemented, other factors such as cost, resources and availability of information may be considered.