Dietary patterns (DPs) have long been recognized as a useful tool for describing overall diet, as they synthesize multiple related dietary components in one or more combined variables. A major drawback of DPs is their limited replicability in different populations (i.e. external reproducibility). This is especially true for the a posteriori DPs, which are derived using standard multivariate analysis techniques and reflect existing dietary behavior in a population. A few papers examined external reproducibility of a posteriori DPs derived with exploratory factor analysis. Although they supported the idea that some DPs are reproducible across populations, there is still no consensus on the statistical approach to measure external reproducibility. In addition, study-specific DPs could be relevant for disease risk. In 2017, de Vito et al. proposed multi-study factor analysis as a generalization of exploratory factor analysis able to handle multiple studies simultaneously. This approach learns the so-called shared DPs, which are common to all studies, as well as extra study-specific DPs for some of the studies, in an integrated model based on the maximum likelihood. External reproducibility is therefore assessed from a different perspective: the reproducible DPs are those that each study population shares with all the others. In the current application, we used individual-level pooled data from 7 case-control studies (3,844 cases; 6,824 controls) participating in the International Head and Neck Cancer Epidemiology (INHANCE) consortium. We derived shared and study-specific DPs from the application of multi-study factor analysis on the study-specific correlation matrices of a common set of 23 nutrients. We then examined DP association with cancers of the oral cavity and pharynx (OCP) and larynx, using mixed-effects logistic regression models, applied to quantiles of factor scores. We identified 3 shared DPs which were common to all the studies; all of them were associated with cancers of the OCP and/or larynx. Each of the 4 US studies expressed an additional study-specific DP; two of these DPs were associated with OCP cancer. Multi-study factor analysis is a promising statistical approach that provides insight into DP reproducibility and supports previous evidence on DPs across populations.
Shared and study-specific dietary patterns / R. De Vito, C. La Vecchia, G. Parmigiani, V. Edefonti. ((Intervento presentato al 29. convegno International Biometric Conference tenutosi a Barcelona nel 2018.
|Titolo:||Shared and study-specific dietary patterns|
LA VECCHIA, CARLO VITANTONIO BATTISTA (Secondo)
EDEFONTI, VALERIA CARLA (Ultimo) (Corresponding)
|Data di pubblicazione:||2018|
|Settore Scientifico Disciplinare:||Settore MED/01 - Statistica Medica|
|Enti collegati al convegno:||International Biometric Society|
|Citazione:||Shared and study-specific dietary patterns / R. De Vito, C. La Vecchia, G. Parmigiani, V. Edefonti. ((Intervento presentato al 29. convegno International Biometric Conference tenutosi a Barcelona nel 2018.|
|Appare nelle tipologie:||14 - Intervento a convegno non pubblicato|