A comparison of principal component analysis, reduced-rank regression, and partial least-squares in the identification of dietary patterns associated with cardiometabolic risk factors in Iranian overweight and obese women

Background According to epidemiological studies, unhealthy dietary patterns and lifestyle lead to rising obesity and cardiometabolic diseases in Iran. Hybrid techniques were used to identify a dietary pattern characterized by fiber, folic acid, and carotenoid intake due to their association with cardiometabolic risk factors such as anthropometric measurements, blood pressure, lipid profile, C-Reactive Protein (CRP), Plasminogen Activator Inhibitor (PAI), Homeostatic Model Assessment Index (HOMA Index), cardiometabolic index (CMI), and monocyte chemoattractant protein (MCP-1). So, the objective of the recent study is to compare the reduced-rank regression (RRR) and partial least-squares (PLS) approaches to principal component analysis (PCA) for estimating diet-cardiometabolic risk factor correlations in Iranian obese women. Methods Data on dietary intake was gathered from 376 healthy overweight and obese females aged 18 to 65 years using a 147-item food frequency questionnaire (FFQ). In this cross-sectional study, participants were referred to health centers of Tehran. Dietary patterns were developed using PCA, PLS, and RRR, and their outputs were assessed to identify reasonable patterns connected to cardiometabolic risk factors. The response variables for PLS and RRR were fiber, folic acid, and carotenoid intake. Results In this study, 3 dietary patterns were identified by the PCA method, 2 dietary patterns by the PLS method, and one dietary pattern by the RRR method. High adherence to the plant-based dietary pattern identified by all methods were associated with higher fat free mass index (FFMI) (P < 0.05). Women in the highest tertile of the plant-based dietary pattern identified by PLS had 0.06 mmol/L (95% CI: 0.007,0.66, P = 0.02), 0.36 mmHg (95% CI: 0.14,0.88, P = 0.02), and 0.46 mg/l (95% CI: 0.25,0.82, P < 0.001), lower FBS, DBP, and CRP respectively than women in the first tertile. Also, PLS and RRR-derived patterns explained greater variance in the outcome (PCA: 1.05%; PLS: 11.62%; RRR: 25.28%), while the PCA dietary patterns explained greater variance in the food groups (PCA: 22.81%; PLS: 14.54%; RRR: 1.59%). Conclusion PLS was found to be more appropriate in determining dietary patterns associated with cardiometabolic-related risk factors. Nevertheless, the advantage of PLS over PCA and RRR must be confirmed in future longitudinal studies with extended follow-up in different settings, population groups, and response variables.