What is causal inference and when do you need it?
June 2022 | Written by Michael Grabner, PhD Sarah R. Hoffman, PhD, MPH, MS and Joseph Smith, PhD, MPH
What is causal inference?
Causal inference is used to answer cause and effect research questions and yield estimates of effect (as opposed to associations). This body of methods allows us to answer research questions such as, "Do patients initiating a newly marketed therapy experience more adverse events than patients starting the older therapy and is the newer therapy the cause of these additional events?".
These methods allow researchers to ask and answer explicitly causal questions and rule out the effects of lurking variables (also known as confounders). For example, people taking a newer product might be sicker and thus more likely to experience adverse events in the first place. How do we know that the product itself is the cause of these additional safety events? That is where the causal inference body of methods comes in. Causal inference is the combination of study design, data source selection, and statistical methods that allows scientists to minimize biases when determining the effects of products, devices, policies, and other interventions under consideration in a given study.
Why is causal inference important?
Causal study interpretations have traditionally been restricted to randomized controlled trials; however, causal inference using observational data is growing in importance, driven by the need for generalizable and rapidly delivered real-world evidence to inform regulatory, payer, and patient or provider decision making. The application of causal inference methods leads to stronger and more powerful evidence. Additionally, when these techniques are applied to observational data, the results generated are both from and for the real world.
When is causal inference needed?
Causal inference is needed when you want to answer a research question involving causation (e.g., does X cause Y); otherwise, we are only able to establish a relationship (that is, association). Causal inference is challenging when “real-world” observational data are used, because without randomization, there may be other factors that could drive the differences observed between treatment groups.
Generating a strong body of evidence regarding a product requires both randomized trials and real-world data. Trials are limited in their patient diversity both clinically and demographically as well as in their follow-up time. Real-world, observational studies of product safety are typically required by regulatory agencies upon approval of a new product; moreover, manufacturers – as well as patients, providers, and payers – may want to know if new products are providing the expected health benefits and how cost of care is affected.
What is needed for causal inference?
Causal inference with observational data combines numerous theoretical and technical concepts, necessitating specialized training and competency. The health economics and outcomes research and safety and epidemiology research teams at Carelon Research offers highly specialized scientists, access to a robust source of integrated real-world data, and decades of experience successfully employing these concepts and methods.
Our scientists will collaborate with your team to curate the most appropriate research questions and identify the optimal study design and analytic approach to address your specific needs. Whether it is working to satisfy regulatory requirements or to generate evidence regarding an important public health concern or any other creative scientific endeavor, Carelon Research will help you to find evidence and truth at the core of healthcare.
Carelon Research (formerly HealthCore) presented a workshop on, “Best Practices for Causal Study Designs Using Real-World Data” at ISPOR 2022. A pdf copy of the presentation is available here.