Causal null hypotheses of sustained treatment strategies: What can be tested with an instrumental variable?

Abstract

Sometimes instrumental variable methods are used to test whether a causal effect is null rather than to estimate the magnitude of a causal effect. However, when instrumental variable methods are applied to time-varying exposures, as in many Mendelian randomization studies, it is unclear what causal null hypothesis is tested. Here, we consider different versions of causal null hypotheses for time-varying exposures, show that the instrumental variable conditions alone are insufficient to test some of them, and describe additional assumptions that can be made to test a wider range of causal null hypotheses, including both sharp and average causal null hypotheses. Implications for interpretation and reporting of instrumental variable results are discussed.Electronic supplementary materialThe online version of this article (10.1007/s10654-018-0396-6) contains supplementary material, which is available to authorized users.

Guidelines for performing Mendelian randomization investigations

“…The straightforward statement of the causal hypothesis is that interventions in the exposure variable will affect the outcome. If the genetic associations with the exposure vary with time, then there are some nuances in terms of what causal hypotheses can be tested 8 ; we discuss the impact of time-varying relationships between variables in Section 9.…”

“…A particular concern arises if the two samples represent different ethnic groups, as patterns of linkage disequilibrium can differ between populations, meaning that a genetic variant may not be as strongly (or even not at all) associated with the exposure in the outcome dataset. Alternatively, the two samples could differ substantially according to population characteristics such as age, sex, socio-economic background, and so on 8 , 30 . Such differences can affect not only the interpretation of causal estimates, but also the validity of causal inferences 31 .…”

“…Its consideration of testing, however, does not make clear what types of null hypotheses can really be assessed (see Swanson et al . 2018) 1 , and the description on effect estimation only suggests qualitatively weighing whether the genetic variants act like the intervention of interest. More guidance could be provided on question framing, especially by giving more specifics on the types of causal questions MR is able to answer.…”

“…(Note that this type of concern also applies to causal null hypothesis testing, as described in Swanson et al . 2018) 1 . At very least, some qualitative guidance on how to interpret results in light of the time-varying nature of exposures and outcomes over the life-course is needed.…”

Do SNAP Recipients Get the Best Prices?

“…Using our IV approach, we found no significant effects of SNAP participation on the expensiveness index across all model specifications. It should be mentioned that the test of whether a causal treatment effect is null only requires the validity of the instrument (Swanson, Labrecque, and Hernán, 2018;VanderWeele et al, 2014). Thus, the validity of the test for the null effect of SNAP participation on prices is robust to additional assumptions such as homogeneity of treatment effects.…”