Case Western Reserve University, Center for Health Care Research & Policy, MetroHealth Medical Center, Cleveland, Ohio, United States of America.
Case Western Reserve University, Department of Population and Quantitative Health Sciences, Cleveland, Ohio, United States of America.
PLoS One. 2019 Jul 10;14(7):e0219399. doi: 10.1371/journal.pone.0219399. eCollection 2019.
In epidemiology, gerontology, human development and the social sciences, age-period-cohort (APC) models are used to study the variability in trajectories of change over time. A well-known issue exists in simultaneously identifying age, period and birth cohort effects, namely that the three characteristics comprise a perfectly collinear system. That is, since age = period-cohort, only two of these effects are estimable at a time. In this paper, we introduce an alternative framework for considering effects relating to age, period and birth cohort. In particular, instead of directly modeling age in the presence of period and cohort effects, we propose a risk modeling approach to characterize age-related risk (i.e., a hybrid of multiple biological and sociological influences to evaluate phenomena associated with growing older). The properties of this approach, termed risk-period-cohort (RPC), are described in this paper and studied by simulations. We show that, except for pathological circumstances where risk is uniquely determined by age, using such risk indices obviates the problem of collinearity. We also show that the size of the chronological age effect in the risk prediction model associates with the correlation between a risk index and chronological age and that the RPC approach can satisfactorily recover cohort and period effects in most cases. We illustrate the advantages of RPC compared to traditional APC analysis on 27496 individuals from NHANES survey data (2005-2016) to study the longitudinal variability in depression screening over time. Our RPC method has broad implications for examining processes of change over time in longitudinal studies.
在流行病学、老年学、人类发展和社会科学中，年龄-时期-队列（APC）模型用于研究随时间变化的轨迹变化的可变性。同时确定年龄、时期和出生队列效应存在一个众所周知的问题，即这三个特征构成了一个完全共线性系统。也就是说，由于年龄=时期-队列，一次只能估计这三个特征中的两个。在本文中，我们引入了一种替代的框架来考虑与年龄、时期和出生队列相关的效应。特别是，我们不是在存在时期和队列效应的情况下直接对年龄进行建模，而是提出了一种风险建模方法来描述与年龄相关的风险（即，综合多种生物学和社会学影响来评估与变老相关的现象）。本文描述了这种方法的性质，并通过模拟进行了研究。我们表明，除了风险仅由年龄决定的病理情况外，使用这种风险指数可以避免共线性问题。我们还表明，风险预测模型中实际年龄效应的大小与风险指数和实际年龄之间的相关性相关，并且在大多数情况下，RPC 方法可以令人满意地恢复队列和时期效应。我们通过 NHANES 调查数据（2005-2016 年）对 27496 个人进行分析，比较了 RPC 与传统 APC 分析的优势，以研究随时间推移抑郁筛查的纵向变化。我们的 RPC 方法对于检查纵向研究中随时间变化的过程具有广泛的意义。
