Statistical method:
|
For the blood- and urine based biomarkers, the primary analysis will be a two-sample t-test for comparing the change in each biomarker (including total cholesterol, LDL, HDL, triglyceride, glucose, insulin, and CRP levels) from pre-intervention to Week 16 after intervention, between the two study Arms. The nonparametric Wilcoxon rank sum test will be used if assumptions for the t-test cannot be met. Multivariable analysis will be performed using a linear regression model, where the dependent/outcome variables are the biomarker levels at Week 16 and the primary independent variable is the intervention arm, adjusted for baseline biomarker level and other potential confounding factors such as dietary intake, medication use, smoking and physical activity. Changes in biomarkers from baseline to Weeks 16 post-intervention within each arm will also be evaluated using the same univariate analysis methods mentioned above. Multivariate analysis will apply a linear mixed model to account for the correlated data collected from repeated measurements (baseline, and Week 16).
For the microbiome data, the impact of peanut intervention with changes in overall microbial diversity, including α-diversity, β-diversity, and overall stability of human gut microbiota was evaluated. The species-level absolute abundance was rarefied using the R function VEGAN::RAREFY. The α-diversity and β-diversity were estimated based on the rarefied species-level absolute abundance data and relative abundance data of gut microbial metabolic pathways using the R functions VEGAN::DIVERSITY and VEGAN::VEGDIST, respectively. α-diversity was measured by the Chao1 richness index, Shannon - Wiener diversity index and Pielou evenness index. Differences across groups and time were compared using non-parametric tests. β-diversity was measured by the Bray-Curtis dissimilarity matrix and Jaccard dissimilarity matrix. The Permutational Multivariate Analysis of Variance (PERMANOVA) test was applied to assess whether there was a difference in β-diversity by study group and time with 999 permutations using the R functions vegan. The associations of peanut intervention with changes in the Chao1, the Shannon, Pielou, Bray-Curtis, and Jaccard indexes were examined using general linear regression models via the R package MicrobiomeStat. Covariates included in the models are age, gender, body mass index, and other legume intake with exclusion of peanut consumption. The overall stability of human gut microbiota from baseline to post-16 weeks was determined based on β-diversity distance within-sample and subtracted them from 1. Species stability and the stability of gut microbial metabolic pathway was determined using formula 1 – Bray-Curtis Index within-sample (baseline vs. 16-week follow-up). Beta regression analysis was performed using the R package “betareg.” β coefficients, standard error (SE), and p values were calculated in models with adjustment for the aforementioned covariates. The statistical analyses were performed at two-sided tests, and associations with p value<0.05 were considered statistically significant. The association of peanut intervention with individual gut microbial taxa from phylum to species for those microbial taxa that were present in >10% of samples and had a median relative abundance >0.001% were evaluated. Linear mixed-effects models (LMM) within a linear regression for differential abundance analysis (LinDA) were applied with adjustment for age, gender, BMI, and another legume intake via the R package “MicrobiomeStat” v.1.1.3. LinDA-LMM used the centered log-ratio (clr) transformation to normalize the absolute abundance of taxa at each taxonomic level from phylum to species, with zeros replaced by the next minimal read count value of the whole dataset. β coefficients, standard error (SE), and p values for individual microbial taxa were produced. False discovery rate (FDR) was calculated at each taxonomic level for multiple testing sets. The same approach was applied to evaluate peanut intervention with individual microbial metabolic pathways for those microbial metabolic pathways observed in > 10% of samples and median relative abundance >0.001%. All associations with FDR-corrected p-values of <0.1 were considered statistically significant.
The impact of peanut intervention on gut microbiome sub-community structure was also examined. Latent Dirichlet Allocation (LDA), a three-level Bayesian probabilistic generative model that reveals latent structures in unlabeled data, was utilized to infer sub-communities in gut microbiota. The abundance data of species-level subgroups and evaluated the associations of gut microbial taxa with peanut intervention for species-level subgroups was estimated using Linear mixed-effects models.
The Bray-Curtis metric was used to calculate the stability of those individual species and metabolic pathways that were present in >20% of samples and median of relative abundance >0.01%, via comparing the absolute abundance/ relative abundance of each species/metabolic pathway across all participants at baseline and post-16 weeks. Beta regression analysis was applied to estimate the β coefficients, standard error (SE), p values, and FDR of each single feature in association with peanut intervention. All statistical analyses were performed using R version 4.3.2.
|
Calculated Results ater
the Study Completed:
|
This study will provide direct evidence on how peanut consumption affects cardiovascular risks. It will also generate new knowledge on how peanut consumption influences the gut microbiome, the newly recognized human organ that plays a critical role in human health. The research findings will be essential for promoting peanut consumption to improve cardiovascular health, particularly among low-income populations who bear a high risk of CVD but are less able to afford tree nuts on a regular basis
|