Comparative analysis of a-priori and a-posteriori dietary patterns using state-of-the-art classification algorithms: A case/case-control study

AbstractObjective pare the accuracy of a-priori and a-posteriori dietary patterns in the prediction of acute coronary syndrome (ACS) and ischemic stroke. This is actually the first study to employ state-of-the-art classification methods for this purpose. s and materials 2009–2010, 1000 participants were enrolled; 250 consecutive patients with a first ACS and 250 controls (60 ± 12 years, 83% males), as well as 250 consecutive patients with a first stroke and 250 controls (75 ± 9 years, 56% males). The controls were population-based and age-sex matched to the patients. The a-priori dietary patterns were derived from the validated MedDietScore, whereas the a-posteriori ones were extracted from principal components analysis. Both approaches were modeled using six classification algorithms: multiple logistic regression (MLR), naïve Bayes, decision trees, repeated incremental pruning to produce error reduction (RIPPER), artificial neural networks and support vector machines. The classification accuracy of the resulting models was evaluated using the C-statistic. s e ACS prediction, the C-statistic varied from 0.587 (RIPPER) to 0.807 (MLR) for the a-priori analysis, while for the a-posteriori one, it fluctuated between 0.583 (RIPPER) and 0.827 (MLR). For the stroke prediction, the C-statistic varied from 0.637 (RIPPER) to 0.767 (MLR) for the a-priori analysis, and from 0.617 (decision tree) to 0.780 (MLR) for the a-posteriori. sion ietary pattern approaches achieved equivalent classification accuracy over most classification algorithms. The choice, therefore, depends on the application at hand.