Treatment Effects and the Representativeness of Medical Trials
Faculty Supervisor(s):

Jason Abaluck

Project Description:

The population of patients in medical trials often differs systematically from the population of patients that will be treated with a drug following approval. Groups may be under or overrepresented given demographics (age, sex, race) or given prior medical conditions (e.g. whether a patient previously had a stroke or was recently hospitalized). We will develop systematic ways to measure whether a lack of representativeness leads trial results to mislead in a way that leads to incorrect treatment decisions. Our ultimate goal is to better inform the interpretation of existing trials as well as to develop tools to assess whether new trials are appropriately representative.

A key element of our analysis is that representativeness depends not only on whether groups are proportionately represented in trials, but also on whether treatment effects vary with the factor in question. For example, if the effect of treatment does not vary with age, then representativeness with respect to age need not be a concern. Of course, one does not necessarily know whether treatment effects vary with age prior to conducting a trial. We will analyze a database of existing medical trials after the fact to determine on which dimensions: a) treatment effects vary substantially and b) the factor is underrepresented in the trial. When groups are underrepresented but not absent entirely, there is often enough information to infer treatment effects for that group -- or at least, across multiple trials -- to infer whether treatment effects systematically differ for that group. This allows us to characterize groups that are both underrepresented and for which the average treatment effect estimated in trials is likely to mislead.

Our results will inform the design of new trials in two ways: first, we will identify groups based on demographics and medical history that are systematically underrepresented and have varying treatment effects across many conditions. Second, we will discuss how to use existing trials and observational evidence for a given condition to assess whether treatment effects differ for underrepresented patients with that specific condition.


Requisite Skills and Qualifications:

Applicants should be completing or have completed a Bachelors or Master’s degree and have excellent programming skills, especially in Stata. Knowledge of R is also helpful. Preference will be given to detail-oriented applicants with previous research and programming experience, particularly in working with large datasets using Stata and/or R. We prefer candidates who can work for two years.


Special Application Instructions:

This principal investigators on this research are Jason Abaluck (Yale), David Chan (Stanford), Leila Agha (Dartmouth), and Sachin Shah (UCSF).

To apply, please email Emily Crawford (emily.crawford@yale.edu) a pdf document titled “lastname_firstname” and containing the following material:

1. A cover letter describing your interest, dates available, your familiarity with programming languages, your prior research experience, and the names, email addresses, and phone numbers for 2-3 references;

2. A Current CV;

3. A transcript;

4. A writing sample;

5. A sample of your coding (preferably Stata).

LINK TO APPLICATION PAGE:

https://tobin.yale.edu/fellowships/pre-doctoral-fellows-program/apply