A comparison of principal component analysis, partial least-squares, and reduced-rank regressions in the identification of dietary patterns associated with hypertension: YaHS-TAMYZ and Shahedieh cohort studies

Abstract

Limited data exist on the advantage of data reduction hybrid methodologies for evaluating the relationship between dietary patterns (DPs) and chronic diseases and they have led to inconsistent results. This study aimed to investigate the association between DPs extracted using principal component analysis (PCA), partial least-squares (PLS), and reduced-rank regressions (RRRs) in identifying DPs associated with hypertension (HTN) risk. The current study was conducted in the context of two cohort studies accomplished in Iran. DPs were generated using PCA, PLS, and RRR methods. Log-binomial logistic regression test was used to assess the association between DPs and the risk of HTN. From a total of 12,403 included participants aged 20–70 years, 507 incident cases of confirmed HTN were identified. The PCA-DP2 was associated with HTN in the fully adjusted model (T3 vs. T1: RR: 0.737, 95% CI: 0.57–0.93, Ptrend = 0.013). The PLS-DP2 and HTN risk were inversely associated in the multivariate model (T3 vs. T1: RR: 0.704, 95% CI: 0.54–0.91, Ptrend = 0.013). The RRR-DP2 was associated with an increased risk of HTN (T3 vs. T1: RR: 1.412, 95% CI: 1.11–1.80, Ptrend = 0.007). Our findings suggest that the RRR method reveals stronger results in association with HTN risk. However, further investigations are required to confirm the association between DPs derived by PLS and RRR methods by incorporating biomarkers related to HTN as the response variables.