Department of Biostatistics

People

Prospective Students

Academics

Research

News & Events

Calendar

Consulting

Employment Opportunities

Resource Quicklinks

Computing Environment

Contact

Many of the factors affecting all-cause mortality are both confounders for each other, and intermediate variables on the causal pathway. For example, smokers may have higher blood pressure, which may in turn increase the chance of death. However, those with high blood pressure may have increased pressure to quit smoking, which may in turn reduce the chances of death. Thus high blood pressure may be a confounder for the effect of smoking, and also an intermediate variable on the causal pathway of smoking on mortality. Longitudinal data are needed to assess such relationships. However, conventional methods (such as survival models with time-varying covariates) may be biased where this form of time-varying confounding exists. A suggested approach is the g-estimation of causal effects (1,2). Here I describe this approach, using longitudinal data on blood pressure, cholesterol and smoking (from the ARIC study). The resulting estimates of the effect of smoking cessation on mortality are compared to those obtained ignoring the time-varying confounders.

1. Witteman JC, D'Agnostino RB, Stijnen T, Kannel WB, Cobb JC, de Ridder MAJ, Hofman A, Robins JM. G-estimation of causal effects: isolated systolic hypertension and cardiovascular death in the Framingham Heart Study. American Journal of Epidemiology 1998; 148(4): 390-401.
2. Mark SD, Robins JM. Estimating the causal effect of smoking cessation in the presence of confounding factors using a rank preserving structural failure time model. Statistics in Medicine 1993; 12: 1605-1628.