A comparison of principal component analysis, partial least-squares and reduced-rank regressions in the identification of dietary patterns associated with bone mass in ageing Australians

The relative advantages of dietary analysis methods, particularly in identifying dietary patterns associated with bone mass, have not been investigated. We evaluated principal component analysis (PCA), partial least-squares (PLS) and reduced-rank regressions (RRR) in determining dietary patterns associated with bone mass. Data from 1182 study participants (45.9% males; aged 50 years and above) from the North West Adelaide Health Study (NWAHS) were used. Dietary data were collected using a food frequency questionnaire (FFQ). Dietary patterns were constructed using PCA, PLS and RRR and compared based on the performance to identify plausible patterns associated with bone mineral density (BMD) and content (BMC). PCA, PLS and RRR identified two, four and four dietary patterns, respectively. All methods identified similar patterns for the first two factors (factor 1, prudent and factor 2, western patterns). Three, one and none of the patterns derived by RRR, PLS and PCA were significantly associated with bone mass, respectively. The prudent and dairy (factor 3) patterns determined by RRR were positively and significantly associated with BMD and BMC. Vegetables and fruit pattern (factor 4) of PLS and RRR was negatively and significantly associated with BMD and BMC, respectively. RRR was found to be more appropriate in identifying more (plausible) dietary patterns that are associated with bone mass than PCA and PLS. Nevertheless, the advantage of RRR over the other two methods (PCA and PLS) should be confirmed in future studies.