Supplementary information files for "Multilevel analysis of individual heterogeneity and discriminatory accuracy (MAIHDA) to understand how obesity risk varies according to multiple lifestyle behavior recommendations"

Supplementary files for article "Multilevel analysis of individual heterogeneity and discriminatory accuracy (MAIHDA) to understand how obesity risk varies according to multiple lifestyle behavior recommendations"<br><br><b>Background</b>The combined and interactive effects of multiple lifestyle behaviours on obesity risk are not well understood. We used Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy (MAIHDA) to examine how adherence to public health recommendations for five lifestyle behaviours affects BMI and obesity risk.<b>Methods</b>The sample included 139,540 men and 125,455 women from the UK Biobank. We categorized fruit and vegetable intake, physical activity, sleep duration and alcohol intake as binary variables (meeting vs. not meeting guidelines), and smoking status into three categories (previous, current, never). These categories were combined to form 48 unique strata, representing all possible combinations of the five behaviours. Linear and binary logistic MAIHDA models were used, with individuals nested within strata, and BMI and obesity status (obesity vs. normal weight) as outcomes. Three models were employed: Model 1 (null), Model 2 (with fixed effects for lifestyle behaviours), and Model 3 (with confounders and fixed effects). Variance Partition Coefficient (VPC), Proportional Change in Variance (PCV), and predicted BMI and obesity risk were estimated.<b>Results</b>For both sexes, strata with the lowest obesity risk were associated with meeting most recommendations, while strata with the highest risk were linked to meeting few. Logistic Model 1 VPCs revealed 7% of variance in obesity risk among males and 5% among females was explained by between-strata differences. In Model 3, VPCs attenuated to 0.5% among males and 0.1% among females, suggesting differences in obesity risk were largely additive effects. PCVs from Model 3 also indicated primarily additive rather than interactive effects. Results were similar for BMI in the linear models.<b>Conclusions</b>Using a novel statistical approach, this study shows that additive effects of multiple lifestyle behaviours predominantly explain differences in BMI and obesity risk. Meeting more public health lifestyle recommendations is important in mitigating obesity risk.<br><br>© The Author(s), CC BY 4.0

论文《探究肥胖风险随多种生活方式建议变化规律的个体异质性与判别精度多层级分析》补充材料

**背景**
目前学界对多种生活方式行为对肥胖风险的联合作用与交互效应尚未形成充分认知。本研究采用个体异质性与判别精度多层级分析（Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy，MAIHDA）方法，探究遵循5种生活方式行为的公共卫生建议对身体质量指数（Body Mass Index，BMI）及肥胖风险的影响。

**方法**
本研究样本来自英国生物银行（UK Biobank），包含139540名男性与125455名女性。研究将蔬果摄入、体力活动、睡眠时长与饮酒量归类为二分类变量（遵循指南vs未遵循指南），并将吸烟状态划分为3个类别（既往吸烟、当前吸烟、从未吸烟）。上述类别组合后共形成48个独特层级，覆盖5种行为的所有可能组合。本研究采用线性与二分类logistic MAIHDA模型，以个体嵌套于层级中的结构进行建模，以BMI与肥胖状态（肥胖vs正常体重）作为结局指标。共设置3类模型：模型1（空模型）、模型2（纳入生活方式行为固定效应）与模型3（纳入混杂变量与固定效应）。研究估算了方差分解系数（Variance Partition Coefficient，VPC）、方差变化比例（Proportional Change in Variance，PCV），以及预测BMI与肥胖风险。

**结果**
对于两性而言，肥胖风险最低的层级均与遵循多数指南建议相关，而风险最高的层级则与遵循极少指南建议相关。二分类logistic模型1的VPC结果显示，男性肥胖风险的7%、女性肥胖风险的5%可由层级间差异解释。在模型3中，VPC分别衰减至男性0.5%、女性0.1%，表明肥胖风险差异主要由累加效应导致。模型3的PCV结果同样表明，生活方式行为的作用以累加效应为主，而非交互效应。线性模型中BMI的分析结果与此一致。

**结论**
本研究采用新颖的统计方法，证实多种生活方式行为的累加效应可主导解释BMI与肥胖风险的差异。遵循更多公共卫生领域的生活方式建议，对降低肥胖风险具有重要意义。

© 作者本人，CC BY 4.0