Objective: To identify real-world physical activity patterns in older adults using objective measurements from wearable devices, and to analyze the associations between these patterns and gut microbiota composition.
Methods: Based on data collected from a real-world health management project, a total of 743 participants from Eastern, Central, and Northern China were enrolled between January 2018 and June 2025. A 180-day objective physical activity dataset prior to fecal sampling was collected via smart wearable devices to extract features including mean daily steps, coefficient of variation of steps, and the proportion of active days. Fecal samples underwent 16S ribosomal RNA (rRNA) gene (V3-V4 region) amplicon sequencing to obtain genus-level relative abundance matrices. Covariates, including demographics, lifestyle, and chronic disease history, were collected via questionnaires and physical examinations. The discriminative dimensionality reduction via learning a tree (DDRTree) algorithm combined with K-means clustering was applied to identify physical activity phenotypes. Alpha diversity was evaluated using the Shannon index (Kruskal-Wallis test), and beta diversity was assessed using covariate-adjusted permutational multivariate analysis of variance (PERMANOVA) based on Bray-Curtis distance. Multivariable linear regression with false discovery rate (FDR) correction was used to screen differential taxa. A microbial risk score (MRS) was constructed based on taxa with a raw P < 0.05, defined as the difference between the standardized abundance of beneficial and harmful taxa. Co-occurrence networks were constructed to evaluate micro-ecological topological structures.
Results: The cohort comprised 381 (51.3%) individuals aged 60-74 years and 362 (48. 7%) aged ≥75 years. Compared with the 60-74 group, the ≥75 group had higher prevalences of hypertension (45.9% vs. 36.7%, P=0.045) and heart disease (34.0% vs. 25.2%, P=0.032), higher systolic blood pressure (median 130 mmHg vs. 120 mmHg, P < 0.001), and fewer mean daily steps (median 6 200 steps vs. 7 000 steps, P < 0.001). Clustering identified three activity patterns: active group (n=143, 19.2%; high steps, low variation, high adherence), moderate group (n=429, 57.7%), and irregular group (n=171, 23.0%; low steps, high variation, low adherence). The active group exhibited the lowest prevalences of hypertension (35.0%) and heart disease (21.7%), and the lowest systolic blood pressure (mean 124.4 mmHg), whereas the irregular group showed the highest values (51.5%, 40.4%, and 127.6 mmHg, respectively). Alpha diversity showed no significant differences among the groups. After adjusting for covariates, physical activity patterns showed no statistically significant effect on beta diversity (R2=0.003 7, P=0.115). Compared with the irregular group, two genera in the active group showed significant differences (P < 0.05). Specifically, the relative abundance of Roseburia in the active group was significantly lower than that in the irregular group (P < 0.05), and the relative abundance of Butyricimonas was also significantly lower than that in the moderate group (P < 0.01). However, these differences did not remain statistically significant after FDR correction. The MRS exhibited a significant gradient distribution across the groups, with the active group scoring the highest (P < 0.001). Co-occurrence network analysis revealed that the active group had the highest network density and proportion of positive correlations (84.5%), whereas the irregular group had the lowest (60.3%).
Conclusion: Physical activity patterns identified from wearable device data are associated with gut microbiota composition and ecological network characteristics in older adults. Active and regular physical activity patterns indicate a higher MRS and more stable microbial co-occurrence networks, suggesting potential associations between activity regularity and gut microbial ecology, though causal inference requires longitudinal confirmation.
目的: 利用可穿戴设备的测量数据，识别老年人群的身体活动模式，并分析不同身体活动模式与肠道菌群的关联。
方法: 基于一项真实世界健康管理项目，于2018年1月至2025年6月在中国东部、中部、北部等地区共收集743名研究对象的数据进行分析。通过智能可穿戴设备收集粪便采样前180 d的身体活动数据，提取日均步数、步数变异系数及活跃天数比例等特征。采集粪便样本进行16S核糖体RNA (ribosomal RNA，rRNA)基因(V3~V4区)扩增子测序，获得属水平的相对丰度矩阵。协变量通过问卷与体格检查收集，包括人口学特征、生活习惯及慢性病史。采用判别降维树(discriminative dimensionality reduction via learning a tree, DDRTree)降维结合K均值聚类分析构建身体活动模式。采用Shannon指数衡量菌群α多样性，组间比较采用Kruskal-Wallis检验；采用基于Bray-Curtis距离的多因素置换多元方差分析检验β多样性；采用多元线性回归[结合错误发现率(false discover rate, FDR)校正]筛选差异菌属。基于原始P < 0.05的差异菌属构建微生物风险评分(microbial risk score, MRS)，定义为有益菌与有害菌标准化丰度之和的差值，构建菌群共现网络评估微生态拓扑结构。
结果: 743名研究对象中，60~74岁者381人(51.3%)，≥75岁者362人(48.7%)。≥75岁组与60~74岁组相比，高血压患病率(45.9% vs. 36.7%，P=0.045)、心脏病患病率(34.0% vs. 25.2%，P=0.032)、收缩压水平(中位数130 mmHg vs. 120 mmHg，P < 0.001)均较高，日均步数较少(中位数6 200步vs. 7 000步，P < 0.001)。基于身体活动特征聚类识别出三种身体活动模式：活跃型(143人，19.2%；特征为高步数、低变异、高依从性)、中等型(429人，57.7%)和不规律型(171人，23.0%；特征为低步数、高变异、低依从性)。活跃型模式组的高血压患病率(35.0%)、心脏病患病率(21.7%)和收缩压水平(平均值124.4 mmHg)均为三组中最低，不规律型模式组最高(分别为51.5%、40.4%和127.6 mmHg)。三组间菌群α多样性差异无统计学意义，调整协变量后，身体活动模式分组对β多样性的影响无统计学意义(R2=0.003 7，P=0.115)。活跃型模式组肠道内罗斯氏菌属(Roseburia)的相对丰度显著低于不规律型模式组(P < 0.05)，丁酸单胞菌属(Butyricimonas)的相对丰度也显著低于中等型模式组(P < 0.01)，但FDR校正后差异无统计学意义。MRS在三组间差异有统计学意义，活跃型模式组评分最高(P < 0.001)。菌群共现网络分析显示，活跃型模式组的网络密度及正相关边比例(84.5%)最高，不规律型模式组最低(60.3%)。
结论: 基于可穿戴数据识别的身体活动模式与老年人肠道菌群组成和生态网络特征存在关联，活跃且规律的身体活动模式显示较高的MRS和更稳定的菌群共现网络，提示身体活动规律性与肠道微生态特征存在关联，其因果方向需纵向研究以进一步验证。
Keywords: Gastrointestinal microbiome; Microbial risk score; Older adults; Physical activity patterns; Wearable electronic devices.