Dietary Fat and Fiber Energy Density and Cancer in NHANES 2021–2022: Parallel Survey-Weighted and Bayesian Estimates

ABSTRACT
Background Diet–cancer links are often modest and design-sensitive. We evaluated energy-density metrics for fiber and total fat under two parallel frameworks (survey-weighted and Bayesian), with mutual adjustment, and quantified method agreement.
Methods NHANES 2021–2022 adults; outcomes were self-reported any cancer and site-specific (breast, prostate, colorectal). Exposures were 2-day mean energy density (g/1000 kcal; z-scaled). Survey models preserved PSU/strata/weights; Bayesian GLMMs used a PSU random intercept with dietary weights rescaled to mean=1. We reported ORs with 95% CI/CrI, posterior probability of direction (P_dir), and ROPE probability; method agreement usedΔlog(OR) and Bland–Altman. Sensitivity analyses spanned density+total energy, residual models, isoenergetic %E substitution, and plausible-reporters subset.
Results (all effects per 1-SD increase in energy density, g/1000 kcal): Any cancer was near-null and concordant across methods: fiber OR≈1.03 (Survey 95% CI 0.91,1.17; Bayes 95% CrI 0.93,1.14); total fat OR≈0.95 (0.82,1.09; 0.86,1.06). Site-specific: Breast (women): fiber 1.00 (0.86,1.15); Bayes 0.99 (0.75,1.26); total fat 1.04 (0.77,1.42); Bayes 1.10 (0.81,1.51)—no clear associations. Prostate (men): fiber 0.99 (0.66,1.47); Bayes 0.93 (0.62,1.37); total fat 0.94 (0.75,1.16); Bayes 0.97 (0.68,1.37)—no clear associations.
Colorectal: total-fat density showed a robust inverse association—Survey 0.46 (0.29–0.74), Bayes 0.55 (0.35–0.84); fiber was null (Survey 0.98 [0.64,1.50]; Bayes 0.95 [0.60,1.48]). Agreement was high: mostΔlog(OR)s clustered near 0, with a more negative difference for colorectal×total fat (Δlog(OR)<0), indicating a stronger protective estimate from survey models.
Conclusions: In a nationally representative sample, associations between “any cancer” and energy-density exposures were near-null. After excluding implausible reporters and across multiple specifications, the colorectal×total-fat density inverse association was directionally consistent and statistically persuasive in both frameworks. The parallel design (Pdir/ROPE complementing P-values) underscores that statistical significance and practical significance diverge here, with absolute risk changes estimated to be near zero. Findings should be interpreted within energy-adjusted, isoenergetic-substitution contexts rather than as a blanket endorsement of higher total fat. Cross-sectional design precludes causal inference.