Vorland Colby J, O'Connor Lauren E, Henschel Beate, Huo Cuiqiong, Shikany James M, Serrano Carlos A, Henschel Robert, Dickinson Stephanie L, Ejima Keisuke, Bidulescu Aurelian, Allison David B, Brown Andrew W
Department of Epidemiology and Biostatistics, Indiana University School of Public Health-Bloomington, Bloomington, IN, USA.
Agricultural Research Service, United States Department of Agriculture, Beltsville Human Nutrition Research Center.
Crit Rev Food Sci Nutr. 2025 Jul 11:1-16. doi: 10.1080/10408398.2025.2525459.
Nutrition epidemiological models involve many analytic decisions, such as defining exposures, selecting which covariates to include, or configuring variables in different ways. We explored the impact of analytical decisions on conclusions in nutrition epidemiology using self-reported beef intake and incident coronary heart disease as a case study. We used REasons for Geographic and Racial Differences in Stroke (REGARDS) data, and selected covariates and their configurations from published literature to recapitulate common models used to assess associations between meat intake and health outcomes. Three model sets were designed: sets one and two used continuous and quintile-defined beef intakes, respectively, each with ∼500,000 randomly sampled specifications. Set three models directly emulated published covariate combinations. Few models (<1%) were statistically significant at < 0.05. More hazard ratio (HR) point estimates were >1 when beef was polychotomized via quintiles (95% of models) vs. continuous intake (79% of models). Including covariates for race or multivitamin use shifted HRs toward the null with similar confidence interval widths. Models emulating existing published associations were all above HR of 1. For our case study, exposure configuration and exposure inclusion resulted in substantially different HR distributions, illustrating how analytical decisions can affect nutrition-related exposure/outcome associations. The finding of few statistically significant models does not prove, but may suggest, minimal association between beef and CHD. Singular assessments of nutritional epidemiology questions should therefore be interpreted with caution. Modeling many analytical approaches may better establish and investigate the uncertainty of nutritional epidemiology questions and provisional answers.
营养流行病学模型涉及许多分析决策，例如定义暴露因素、选择要纳入的协变量，或以不同方式配置变量。我们以自我报告的牛肉摄入量和冠心病发病情况为案例研究，探讨了分析决策对营养流行病学结论的影响。我们使用了中风地理和种族差异原因（REGARDS）数据，并从已发表的文献中选择协变量及其配置，以重现用于评估肉类摄入量与健康结果之间关联的常见模型。设计了三组模型：第一组和第二组分别使用连续的和五分位数定义的牛肉摄入量，每组有大约50万个随机抽样的规格。第三组模型直接模拟已发表的协变量组合。很少有模型（<1%）在<0.05水平上具有统计学意义。当牛肉摄入量通过五分位数进行多分类时（95%的模型），与连续摄入量（79%的模型）相比，更多的风险比（HR）点估计值>1。纳入种族或多种维生素使用的协变量会使HR向无效值偏移，同时置信区间宽度相似。模拟现有已发表关联的模型的HR均高于1。对于我们的案例研究，暴露配置和暴露纳入导致了HR分布的显著差异，说明了分析决策如何影响营养相关的暴露/结果关联。很少有具有统计学意义的模型这一发现并不能证明，但可能表明牛肉与冠心病之间的关联最小。因此，对营养流行病学问题的单一评估应谨慎解释。对多种分析方法进行建模可能会更好地确定和研究营养流行病学问题及临时答案的不确定性。
