CMStatistics 2020: Start Registration
View Submission - CMStatistics
Title: Using negative controls to estimate the effect of treatment on survival when everyone is treated Authors:  Ruth Keogh - London School of Hygiene and Tropical Medicine (United Kingdom) [presenting]
Abstract: Treatments are sometimes introduced for all patients in a particular cohort. When an entire cohort of patients receives a treatment, it is difficult to estimate its effect because there are no directly comparable untreated patients. The application that motivates this relates to a disease-modifying treatment in cystic fibrosis, ivacaftor, which has been available for everyone in the UK with a specific genotype since 2012. It is of interest to understand the causal effect of treatment on survival, which has not been assessed in randomized controlled trials, and also to project the potential long-term impact on life expectancy. To investigate this, we use observational longitudinal data on treatment use, genotype, survival and measures of patient health status from the UK Cystic Fibrosis Registry. Negative control outcomes observed in patients who do not receive the treatment are used to enable estimation of the causal effect of ivacaftor on survival, including patients eligible for the new treatment but before its availability (historical controls) and patients with an ineligible genotype (genotype controls). Causal diagrams and the potential outcomes framework are used to define the causal estimand of interest and to discuss the assumptions under which it can be estimated using the available control groups. We will discuss the use of different analysis models, including Cox regression and the additive hazards model.