Sharon Jiang, ’22 (Summer ’19)
“We are not limited by the science; we are limited by our ability to make good use of the information and treatments we already have.”From The Death of Cancer by Vincent DeVita, MD
Former Director of the US National Cancer Institute, Physician-in-Chief at Memorial Sloan Kettering Cancer Center, and President of the American Cancer Society
Part I: Hello, World (of drug repurposing)!
This summer, I’ll be interning at Cures Within Reach for Cancer (CWR4C), a nonprofit startup that aims to leverage AI to synthesize evidence from clinical publications, real-world evidence, and pre-clinical data to identify the most promising generic drug candidates to repurpose for cancer treatment.
Whoa. That’s a lot to take in.
That was exactly my first thought when I was introduced to this world of drug repurposing. Let me back up a bit.
My story with CWR4C began at PKG Community Conversations: Healthcare (thanks PKG, I owe you one!), where students could mingle with social organizations that are interested in addressing healthcare challenges. After the event, Laura Kleiman, CWR4C’s Founder and Executive Director, emailed out an internship opportunity to the attendees. I was obviously intrigued, asked to learn more about CWR4C’s mission, and she replied literally 5 seconds later, asking to meet in person. And the rest is history.
Well, not really. We met a couple of days later on a very chilly March afternoon (that’s the Massachusetts weather working its magic). Her spiel went something like this — although I can assure you she said it more eloquently.
Many generic drugs approved by the US Food and Drug Administration (FDA) for non-cancer indications have shown promise for treating cancer in small scale clinical studies. However, these therapies are not yet part of medical practice because pharmaceutical companies lack the financial incentive to fund the large clinical trials testing the inexpensive drugs for the new indication. Unsustainable cancer treatment costs prompt the need for affordable alternatives and the search for more effective treatments by adding these low cost drugs to standard of care.
There have been success stories of repurposing these drugs. Thalidomide, which was originally approved as a sedative in 1957 and later for leprosy, was one of the initial compounds researchers suggested to Cures Within Reach (CWR4C’s fiscal sponsor) for repurposing to treat multiple myeloma. Cures Within Reach, using funding from private foundations, helped support a thalidomide Phase II trial at the Mayo Clinic. Because the drug was already approved for leprosy treatment, the researchers were able to bypass Phase I safety and dosing trials, which can take years to complete. Based on those results, in combination with other successful trials of the drug, the FDA approved thalidomide for multiple myeloma in 2006. It cost only $40–$80 million in total to secure this FDA approval, compared to the average of $1–$2 billion it takes to develop an entirely new drug.
If you’d like to learn more, I highly encourage you to read this article about repurposing!
The primary goal of my project entails developing a scoring model to prioritize thousands of repurposing opportunities based on existing evidence of efficacy; the context of each clinical study varies widely, so details about the patient population, type of intervention, and outcomes impact prioritizations for each drug-cancer pair. The pre-clinical and clinical data need to be considered alongside health economics and outcomes research — which factors in cost-benefit analysis, clinical feasibility and market size — to determine the best overall set of generic drugs for certain cancer subtypes and maximize patient impact.
I’m working with a fantastic team of interns and volunteers from a wide variety of backgrounds in molecular biology, data science, business, technology, and policy who are all dedicated to making this mission successful. All of our projects are interconnected, and there’s so much opportunity to learn and make a real impact in the community. Thanks to the Biden Cancer Initiative Cancer Collaboration Hubs, WeWork has generously offered our team a space to get down to business. Work no longer feels like work when you’re surrounded by an amazing team in an amazing space (with unlimited fruit water!) and collaborating for an amazing cause.
I hope to learn about the effort involved in working towards a social impact startup’s success, as well as gain some insight into the current healthcare system and the drug development process. So many resources are devoted to fueling scientific research into cancer treatment; it is necessary to bridge the gap between academic research and clinical practice, where rigorous testing discovers drugs that can benefit patients’ lives.
I truly believe that the projects this nonprofit is dedicated to working on can have a profound impact on the healthcare landscape. It baffles me that so much of drug development is founded in maximizing profit and not in prioritizing patients’ health. Access to good healthcare is a fundamental human right, regardless of socioeconomic background; providing much cheaper and possibly more effective cancer treatments is a critical step in democratizing medicine.
Alright, that’s enough blogging for now. I’ve got some work to do.
Part II: Let’s get down to business (to defeat the cancer)
First of all, I’d like to apologize for this entry’s really cheesy title that was supposed to be evocative of the classic song from Disney’s Mulan. If you continue reading my blogs, you’ll see some more of where that came from. Now that you’ve been forewarned, I’d like to get down to business, literally and figuratively.
It’s been a couple of weeks since I first started working with CWR4C, and it has been a crazy roller coaster ride — I have been simultaneously exhilarated and stressed out at all times. I have learned not only about the intersection of AI and the healthcare space (which is what originally attracted me to the organization), but I have learned a plethora about startup culture and efficient project management.
When I first started doing background research into my specific project, I already knew it was going to be a difficult task. Once I took a deeper dive into the state-of-the-art literature, I realized that the project was much more complex than I had initially anticipated. The evidence our team had to sort through was broken up into many different subtypes — randomized controlled clinical trials, observational studies, case reports, in vitro and in vivo pre-clinical experiments, and so on. Within each disparate data type, we somehow had to figure out a way to standardize all the details about study designs and outcomes to facilitate direct comparisons for drug intervention efficacy and safety.
To be honest, sometimes I felt very easily overwhelmed by the seemingly never-ending roadblocks in the way, but as I look back on the copious notes and conversations with collaborators, I now understand that each roadblock has taught me something critical that could help me improve the current design of the scoring model. Ultimately, this has been an organic, dynamic learning process, and despite the difficulties along the way, I’m still motivated to work on it because I know it’s for a worthy cause.
This learning journey would not have been possible had it not taken place in a startup environment. This year, CWR4C has been selected as a MassChallenge Finalist, one of the biggest startup accelerators in the world; one of the many MassChallenge resources we have access to is the generous support from our mentors, one of whom is actually a cancer survivor. The energy that our mentors, advisers, and team members have brought to this effort has snowballed, and everyone’s passion for this undertaking has pooled together to create a one-of-a-kind collaboration.
This passion among all those involved in the startup ecosystem was most apparent at MassChallenge’s Startup Showcase, which featured all of the startup finalists presenting their work to interested investors, potential collaborators, and members of the public. I not only was able to act as a representative of CWR4C, but I was also able to interact with individuals from other startups. When I had these conversations, I noticed a common thread: at a startup, especially an early-stage one, time is valuable, and everything that you do has a higher chance of direct impact. At a startup, everything is built from scratch, and oftentimes, you have to find the answers you’re looking for, instead of being given an established set of rules to work with.
Here’s an über interesting (and better written) blog about the utter chaos and joy that one startup founder has experienced. You know, for light nighttime reading.
With all that I’ve learned from working in a startup, probably the most important takeaway is proper project management. Our team has planning and review meetings, where we lay out all the tasks and deliverables we’re aiming to achieve for every two weeks. We have daily standups, which are quick debriefs about yesterday’s progress, today’s plans, and obstacles (we don’t actually stand up for them, but we call them standups because they force you to be brief. Our standups aren’t that brief, but that’s because we’re special). These practices weren’t something I was used to as a student, but incorporating them into my regular routine has greatly helped my time management and key focus for each day so that I can finish what needs to be done.
Long story short, if you’re reading this, it’s not too late (Drake reference, anyone?) to pursue something you’re really passionate about and perhaps get involved in a startup if that’s in your potential interest. As a person with a mostly academic background, it’s been really refreshing and rewarding to try something outside of my comfort zone.
Part III: How to Change Cancer Care in 18 Months
Let’s be honest — there’s no guarantee that you can revolutionize the way medicine has been practiced in such a short period of time, but our team is aiming to be that ambitious, and here is a brief overview of how we are going to do it.
Here’s just a brief rundown of the problem we’re trying to tackle. There are about 2,000 FDA-approved generic drugs, and 300 of them have already shown anti-cancer evidence. My job is to decide which 10 drugs are the most worthwhile to test in clinical trials and how to make that decision in a smart, well-informed way.
Manual systematic reviews of scientific literature take a lot of time and present a static view of evidence. As hundreds or thousands of new relevant studies are added to the corpus over time, it becomes infeasible to integrate evidence from new findings into evidence summaries for each drug.
Using novel machine learning methods, we can now identify and synthesize this evidence at scale. We then run our scoring pipeline on the synthesized evidence to generate a prioritized list of the most promising repurposing opportunities. As new studies are added to the corpus, evidence summaries continue to be automatically updated and reprioritized. The most promising therapies will then be clinically validated to gather additional evidence of their efficacy, and these generic drugs will be incorporated into standard of care.
As you can imagine, there are many layers of complexity involved in this process. From data sources of differing strength (clinical trials, observational studies, pre-clinical studies, conference reports, etc.), I’m responsible for extracting the necessary information to aggregate a score that represents the overall efficacy of a drug for a particular cancer subtype. Our overall scoring methodology involves three steps:
- Evidence-based prioritization: Generating a score for each drug-cancer pair and a preliminary ranked list of promising drug repurposing candidates
- Health economics and outcomes research: Incorporating pragmatic factors such as market size, patient impact, cost, and feasibility and re-prioritizing the list
- Expert review: Presenting the list and evidence to experts from relevant fields to determine which candidates to pursue further
This summer, we have been designing a two-step scoring methodology and analyzing clinical data for the evidence-based prioritization step. For our simple model, we consider only three inputs: number of studies testing drug X for cancer Y, study type, and therapeutic effect. For the intermediate model, we add more granularity to the scores by analyzing the total number of patients, study design, outcome measures, level of significance, study quality, and drug class.
Once we have this prioritized list of drugs, we can represent a detailed breakdown of these scores in an interactive visual format; this way, doctors and cancer patients can be more informed about the positive and negative evidence for repurposed drugs. This will have an enormous impact, as some patients simply don’t have enough time to wait for clinical trials to be completed. The wide availability and low cost of these drugs mean that patients may have more treatment options now.
We’ve already seen many cancer patients who have taken the initiative to seek out answers for treatments when they have exhausted all other options. On social media, patients are sharing their data on generic drug use and subsequent outcomes. Patients obviously want more information about possible treatment options, and our work helps them interpret all this evidence in an easily understandable way.
Ok, so I might not have given detailed steps on how to change cancer care, but I think this overview will suffice. There’s not a manual for everything — and that’s the fun part. I’ve really enjoyed finding answers to help cancer patients who are trying to find their own answers.
Part IV: Reflection (not to be confused with the classic Mulan song)
It’s the end of an era.
We’ve just wrapped up giving presentations about our projects to one of our key advisors and it’s pretty emotional here at work.
As my summer internship with CWR4C comes to a close, I can’t help but look back on all of the progress our team has made and the learning journeys that we’ve embarked upon. All of our projects started as merely ideas, but now they have taken the form of meaningful, tangible progress.
Here’s an update on the work we’ve done since Blog #3. Our Northeastern University collaborators generously gave us a filtered and expertly annotated list of randomized controlled clinical trials that mentioned a drug-cancer relationship. They computationally extracted information about efficacy association, drug intervention, comparator, and relevant outcomes.
From there, we assigned weights to the inputs for the simple model (refer to Blog #3 for that information) and generated a preliminary list of drugs with the strongest anti-cancer evidence. We then took a sample of the highest scoring drugs and did a deep dive to extract more specific data points about clinical endpoints (progression free survival, time to progression, and overall response rate) and their corresponding significance levels.
To directly compare these clinical outcomes between different studies, we utilized standardization techniques for hazard ratios, which are used for time to event outcomes, and Peto odds ratios, which are used for dichotomous outcomes. We then aggregated all of the weighted inputs to reprioritize the list. We have still got work to do in terms of validating the robustness of these results, and this iterative process will continue to be refined as we acquire more data.
We are still laying the groundwork for data visualization; once we have revised our scoring models to analyze the enormous corpus of relevant publications that our IBM Research collaboration has generated, we will then incorporate our findings in an interactive tool that can represent rankings of drug-cancer pairs based on multiple attributes. This platform provides an easy-to-use visualization of the factors that inform a score and subsequent ranking changes after altering weights for these factors.
A couple of months ago, I wouldn’t have imagined that I could do this research and present it in an articulate way to our team, our advisers, or the general public. Through this experience, I’ve learned that so much of the science has already been done, and that more effort needs to be placed on how to interpret it and get it to the users for direct impact.
On a personal note, I’ve gained lifelong skills in conducting independent research, taking part in meaningful teams and collaborations, and effective communication and leadership. Don’t get me wrong, I love and value learning in a classroom setting, but I couldn’t gain these skills in the same way on campus. This opportunity has afforded me an invaluable hands-on learning experience, where I always have to look for my own answers. Another thing that I’ve learned that will probably stay with me forever is that the people around you are almost always the best resource to turn to when you’re confused or lost.
So this might be the end of the beginning, but there’s still a lot of work to be done, and I’m still as excited to work on this project and with CWR4C as I was on my first day.
P.S. Here’s my chance to shamelessly plug CWR4C. If you’re even remotely interested in what we’re doing, we’re always looking for enthusiastic members to join our mission! Here’s our website, and feel free to reach out!
Want to support students like Sharon? Donate to the PKG Center and help us keep all PKG programs cost-free for students.