Each week, we publish lay abstracts of new articles from our prestigious portfolio of journals in statistics. The aim is to highlight the latest research to a broader audience in an accessible format.

The article featured today is from the Canadian Journal of Statistics with the full article now available to read here.

Jiang, C., Wallace, M.P. and Thompson, M.E. (2023), Dynamic treatment regimes with interference. Can J Statistics. https://doi.org/10.1002/cjs.11702

In statistics, causal inference identifies the relationship between a cause (such as a medical treatment) and an effect (the result of that treatment). Many statistical methods used for causal inference assume that the data being analyzed contains no ‘interference’. This term describes situations where the treatment received by one individual might affect another individual. For example, suppose a parent and child share a household. If the parent receives a vaccination against an infectious disease, this may reduce the risk of the child contracting the disease, even if the child is not vaccinated. As another example, if an adolescent smokes cigarettes, their schoolmates may be affected by second-hand smoking (interference) even if they themselves do not smoke. Interference can make it difficult to estimate the effect of a treatment (such as a vaccine) or a behaviour (such as smoking). Problems arise because we have to consider both the effect on the person being treated, and the effect the treatment has on others. Building on a particular method called dynamic weighted ordinary least squares (dWOLS), which boasts easy implementation plus robustness in terms of model misspecification, this work develops new statistical methods that can be used when interference may be present. It identifies and estimates the effects a treatment may have on an individual, and any unintentional effects that treatment may have on others associated with the individual. These estimations provide the data-driven evidence needed to improve personal health care by making treatment decisions specific for individuals. The method’s properties are demonstrated via different simulation studies with different social network structures. In addition, data from the Population Assessment of Tobacco and Health (PATH) study are analyzed to investigate the effects of e-cigarettes on cigarette dependence within two-person household networks, where interference may result in one resident’s treatment affecting the other.