L10: Simple time-varying exposures

Content for the week of Monday, April 12, 2021–Friday, April 16, 2021

Readings

Notes

We are often interested in estimating the effects of treatments or interventions that are maintained over time, but exposures to these treatments and interventions often vary over time for participants in our datasets. Reasons may include

  • treatment changes or discontinuation
  • nonadherence
  • loss to follow-up
  • mobility and migration

among others.

This week, we will focus on a simple scenario in which we want to compare risk of death under continuous treatment and under no treatment, but some people in the treatment arm discontinue treatment. In the exercise, we will compare risk under 2 treatments, where people in both arms discontinue treatment.

Videos

Recordings of lectures are available on the EPID 722 Sakai site.

Exercise

Exercise 10 may be found on the course Teams site.

Optional readings

In the reading and during lecture, we discussed using IPW to estimate “per protocol” effects in the presence of a simple time-varying exposure. The reading below illustrates use of g computation to achieve the same goals.

Naimi AI, Perkins NJ, Sjaarda LA, Mumford SL, Platt RW, Silver RM, Schisterman EF. The Effect of Preconception-Initiated Low-Dose Aspirin on Human Chorionic Gonadotropin–Detected Pregnancy, Pregnancy Loss, and Live Birth: Per Protocol Analysis of a Randomized Trial. Annals of Internal Medicine. 2021 Jan 26.

The reading also introduced the idea of emulating a randomized trial using observational data. For more on this idea, see

Hernán MA, Robins JM. Using big data to emulate a target trial when a randomized trial is not available. American journal of epidemiology. 2016 Apr 15;183(8):758-64.

Questions

Please use the form below to submit your questions about this week’s reading.