Understanding coronary atherosclerosis in relation to obesity: is getting the distribution of body fatness using dual-energy X-ray absorptiometry worth the effort? A novel perspective using Bayesian Networks.

AIM: A relative excess of fat in the upper body region has been proven to be associated with increased coronary artery disease (CAD) risk. Dual-energy X-ray absorptiometry (DXA) is probably the most accurate and precise method available to study fat regional distribution and to directly measure total body fat and lean soft tissue mass. However, while several studies have investigated the abilities of obesity anthropometric measures in predicting CAD, only few studies have evaluated DXA as CAD predictor; particularly, a comparison between a model including information coming from anthropometric measurements and a model in which fat is precisely measured by DXA, is still lacking. In order to verify if CAD severity, as measured by Gensini score, is better predicted when a prognostic model includes DXA measurements rather than anthropometric measures, we compared performance obtained by two Bayesian Networks (BNs) including standard anthropometric measures and DXA, respectively. METHODS: Data come from 58 consecutive patients, 79% of them having suspected and 21% known CAD. Two BNs were implemented: input variables include anamnestic information, biochemical data and obesity measures. In the first model (BN1) obesity was measured by body mass index and waist-to-hip ratio, while in the second one (BN2) it is quantified by DXA-derived parameters. RESULTS: Network graphs and results coming from sensitivity analysis show that in both models lipoproteins and biomarkers of inflammation act as proximal node, while obesity (independently of the chosen measure) appears to be a distal node acting by the intermediation of other variables. Both models show high predictive abilities, the mean percentage classification errors being, respectively, 14.13 and 18.87. CONCLUSIONS: In our study, the BN predictive ability is slightly superior when obesity is measured using anthropometric data instead of DXA measurements. The reason probably relies on the fact that in the BN the obesity role in predicting CAD is mediated by the action of other factors that appear to be more directly influencing the outcome. Thus, the necessity to dispose of a perfect measure becomes less compulsory and the huge effort to precisely estimate body composition with complex methods as DXA could be avoided when using expert system such as BN as predictive tool.