Characterizing the mechanistic interactions between exposures and diseases is one of the most critical issues in epidemiologic studies. Previous studies have proposed a stochastic sufficient component cause framework, under which each sufficient cause is treated as a stochastic process instead of a time-invariant random variable. However, different types of mechanistic interactions such as synergism and agonism cannot be further identified. In this study, we proposed a stochastic marginal sufficient component cause model to conceptualize and identify agonism and synergism by exploiting the additional information. We further provided six approaches to identify and estimate agonism and synergism based on an additive hazard model and a complementary log model. Researchers can easily adjust confounding factors by including appropriate covariates into a regression model. Simulations have proven that approaches under three models are all valid tests. The power of an additive hazard model increases as the total follow-up time increases and is higher than that of the other two models. We applied this method to a Taiwanese cohort data set to investigate the mechanistic interaction among hepatitis B and C viruses on the incidence of hepatocellular carcinoma. The hazard of people with agonistic interaction is 1.28×10−5 (95% CI: 6.97×10−6, 1.87×10−5), and the cumulative hazard of those people is 7.41×10−2 (95% CI: 4.09×10−2, 1.07×10−1), which is approximately 3.5 times stronger than that of synergistic interaction. The proposed method makes it possible to quantify different types of mechanistic interactions in longitudinal studies with censored data.
- causal inference
- stochastic process
- sufficient component cause model
- time-to-event data