Drug repositioning (or repurposing) is not new.
What is somewhat new is the following:
The 21st Century Cures Act (Cures Act), signed into law on December 13, 2016, is designed to accelerate medical product development and bring new innovations and advances faster and more efficiently to the patients who need them. Among other provisions, the Cures Act added section 505F to the Federal Food, Drug, and Cosmetic Act (FD&C Act). Pursuant to this section, the Food and Drug Administration (FDA) has created a framework for evaluating the potential use of real-world evidence (RWE) to help support the approval of a new indication for a drug already approved under section 505(c) of the FD&C Act or to help support or satisfy drug postapproval study requirements.
This is a boon to drug companies. Clinical drug trials are very expensive to run, and avoiding a full-blown trial saves money and time—while improving and/or saving more lives.
I wondered, how this Act came to be.
It seemed to me like a very good idea, and as I did not expect this type of goodness to come from our government, I decided to look into it.🤷🏽♂️
Introduced by Representative Suszanne Bonamici on January 6, 2015.
After some amendments, the House voted 392-26 in favor on November 30, 2016.
After further amendements, the Senate voted 94-5 in favor on December 7, 2016.
President Barack Obama signed it into law on December 13, 2016.
Companies are looking for the highest probability of success with the least amount of additional work.
One way to do this is to extract the most information they can from the enormous amount of data they already have or can readily obtain.
This is called observational data.
It is an untapped gold mine. Much of it, however, is:
This is where our research on Causal Inference comes in.
Causal Inference is a very old field of research. What is it that we are doing that is new?
Until now, causal inference research did not focus on scenarios where large amounts of confounded data could later be selectively deconfounded.
We can provably reduce the number of samples needed to accurately estimate the effectiveness of the drug in treating a disease.
As we write in the introduction of our paper:
The fundamental problem in causal inference is to estimate causal effects using observational data. This task is particularly motivated by scenarios when experiments are infeasible. While the literature typically addresses rigid settings in which confounders are either always or never observed, in many applications we might observe confounders for a subset of samples.
Due to the high cost of genetic tests, we might only be able to afford to reveal the value of the genetic confounder for a subset of patients. Note that for a variable such as a genetic mutation, we might observe retrospectively, even after the treatment and outcome have been observed. We call this process of revealing the value of an (initially unobserved) confounder deconfounding, and the samples where treatment, outcome, and confounders are all observed deconfounded data.
So motivated, this paper addresses the middle ground along the confounded-deconfounded spectrum. Naively, one could estimate the ATE with standard methods using only the deconfounded data. First, we ask: how much can we improve our ATE estimates by incorporating confounded data over approaches that rely on deconfounded data alone? Second, motivated by the setting in which our confounders might be retrospectively observed for cases with known treatments and outcomes, we introduce the problem of selective deconfounding—allocating a fixed budget for revealing the confounder based upon observed treatments and outcomes. This prompts our second question: what is the optimal policy for selecting data to deconfound? To our knowledge, this is the first paper that focuses on the case where ample (cheaply-acquired) confounded data is available and we can select only few confounded samples to deconfound (expensive).
Although our work was motivated by the 21st Century Cures Act, and predates COVID-19, it was exciting to note that our research may be helpful to the current crisis (see a report from March 2020):
More than 80 clinical trials have been launched to test coronavirus treatment, including 24 drug repurposing or repositioning studies (clinicaltrials.gov database) involving more than 20 medicines, such as human immunoglobulin, interferons, chloroquine, hydroxychloroquine, arbidol, remdesivir, favipiravir, lopinavir, ritonavir, oseltamivir, methylprednisolone, bevacizumab, and traditional Chinese medicines (TCM).
Perhaps Gabriel Garcia Marquez would have called our (“mathematical realism” 😉) research:
Math in the time of coronavirus. 😏