基于神经网络方法的区间删失数据回归分析：应用于24 h身体活动模式的死亡风险建模

Regression analysis of interval-censored data based on a neural network approach: application to modelling mortality risk in 24-hour physical activity patterns

摘要:

目的 基于身体活动数据预测每个观测对象的生存概率。方法 从美国国家健康与营养调查（NHANES）数据库中提取NHANES 2003—2006年不同对象的身体活动监测数据。通过综合Brier评分、受试者工作特征（ROC）曲线及曲线下面积（AUC）等多种评价指标来评价模型的预测性能，比较了条件生存森林、自适应LASSO、深度学习生存模型等三种生存分析模型的预测性能。结果 本研究所建立的深度学习生存模型的预测性能优于其他两个模型。同时，本研究还分析了各项身体活动变量在深度学习生存模型中的重要性，其中年龄和总活动计数对人的身体影响最大，在内部和外部验证中均观察到类似的结果。结论 深度神经网络生存模型可以作为预测身体活动对身体影响的有效工具。

关键词:神经网络;区间删失;生存分析;美国国家健康与营养调查（NHANES）;身体活动预测;身体表现

Abstract:

Objective To predict the survival probability of each observed object based on physical activity data.Methods Physical activity monitoring data from National Health and Nutrition Examination Survey (NHANES) 2003–2006 were extracted from the NHANES database for different study subjects. The predictive performance of the models was evaluated by combining various evaluation metrics such as Brier score, receiver operating characteristic (ROC) curve and area under the curve (AUC), and the predictive performance of three survival analysis models such as conditional survival forest, adaptive least absolute shrinkage and selection operator (LASSO), and deep learning survival model were compared.Results The prediction performance of our deep learning survival model is better than the other two models. We also analyzed the importance of each physical activity variable in the deep learning survival model, with age and total activity count having the greatest impact on a person's body. Similar results were observed in both internal and external validation.Conclusion Deep neural network survival models can be used as an effective tool for predicting the effects of physical activity on the body.

Key words:neural networks;interval censoring;survival analysis;National Health and Nutrition Examination Survey (NHANES);physical activity prediction;physical performance