Health research has been increasingly focused on discovering the mechanisms through which an exposure or intervention affects health outcomes. Social science has become a part of this picture as the importance of social behavior and other social factors in health has been increasingly recognized. Causal mediation analysis methods have been developed to explore such mechanisms. However, these approaches are rooted in discrete time models, and fail to account for the continuous time nature of many, if not most, social and behavioral, as well as biological, processes. The proposed research will develop a new continuous time causal mediation analysis model (CMM). In addition to allowing a description of the dynamic interplay among variables in time, the new model will allow the incorporation of continuous time features such as time lags and extinction effects. The model will accommodate arbitrary data patterns, multiple causal links, continuous and categorical variables, and interaction/modification effects. We start with a differential equation model based on a potential outcome (causal) framework, and show the connection to a nonlinear model form that may be fit to data. An integration approach may be used to estimate mediation/path effects and predict the effect of new interventions. Secondly, we will extend CMMs to handle multi-level data, for example, family, neighborhood or other social variables, which may themselves unfold over time. In addition, we will model the impact of individuals on one another, providing a explanation for emergent social behavior. We will apply our new methods to a randomized study of family interventions to improve dental care use in children and a longitudinal observational study of family, neighborhood, behavioral and biologic factors in the progression of dental caries in disadvantaged and high risk (including very low birth weight) children. The new methods will be evaluated and refined through simulation studies. The new continuous time mediation models will be further used to study design issues, including the number and timing of measurements, and to investigator new designs, for example that randomize different subjects to different measurement times. Finally, we will develop user-friendly computer programs to allow behavioral and other health researchers to apply the new methods. The new methods will allow a more valid and thorough exploration of causal mechanisms involving social behavior, with broad applicability across heath areas. They will also facilitate the development of effective behavioral and other interventions to improve health outcomes and decrease health disparities.
This project will develop new statistical methods that will help researchers better understand the progression of certain diseases and why particular interventions work or fail to work. The new methods, intended to reveal complex causal relationships among risk factors and outcomes as they unfold over time, will be applied to data from a randomized study of family interventions to improve use of dental care and a longitudinal study of behavioral and biological factors leading to dental caries in disadvantaged children. The new methods will facilitate the design of effective new behavioral and other interventions with the potential to improve health outcomes and reduce health disparities.
|Wang, Wei; Albert, Jeffrey M (2017) Causal Mediation Analysis for the Cox Proportional Hazards Model with a Smooth Baseline Hazard Estimator. J R Stat Soc Ser C Appl Stat 66:741-757|