Andrew Lo
Video: Can changing how we fund drug research change our outcomes?
Andrew Lo sees a rather important relationship between finance and healthcare – one that might also allow for more prolific and better drug research.
Talking about how data science and financial engineering can improve the healthcare industry, he talked to the audience about what these disciplines have to do with using clinical development successfully.
“We’re living through a very unusual period,” Lo said. Biomedicine, he added, is at an inflection point, a “convergence” in which life science and engineering can complement one another.
In explaining his approach to healthcare, Lo mentioned what some call the “omics” revolution, with fields like genomics, epigenomics, transcriptomics and metabolomics informing research. Still, he said, there’s one of these ideas being left out – economics.
“You have to pay for stuff,” Lo said, referring to the money that goes into advancing in healthcare in general.
Business models, he said, are slow to change, and new approaches could improve results.
Lo presented an equation showing healthcare finance metrics like profits and costs in context.
“The expected value of a drug program is given by just three terms:” he said, “the present value of profits if your drug gets approved, multiplied by the probability that it does get approved, minus the cost of developing that drug. That’s it. Now, economists can tell you a lot about the present value of profits, we can tell you a lot about the costs of developing drugs. But we can tell you nothing about the probability of success, because that’s a scientific and engineering problem. And so the question is, how difficult is it to develop a drug?”
A new model, he said, might help unlock potential for investors to get more excited about clinical development and clinical trial processes.
First, he showed some of the metrics around a single drug development process – that the average project, he estimated, would cost $200 million over 10 years, with approximately a 5% probability of success.
“This is not a very exciting endeavor,” he said. “From an investment point of view, it’s pretty low odds of success.”
However, Lo pointed out, the payoff, if the project is successful, would be around $12.3 billion. A caveat, he noted, is that the payoff happens over 10 years.
“What’s the risk? … we measure risk in terms of standard deviation, and the volatility or standard deviation is 423.5%,” he said. “Way too risky? Most investors will not willingly invest in a project like this over such a long time span.”
Then, Lo presented a different kind of math – where a collective investment pot of $30 billion could be used to generate a series of development processes, some of which, presumably, would pay off.
“The likelihood of getting at least three successes out of 150 tries is a whopping 98%,” he said. “I’ve got a 98% chance of producing $36.9 billion over a 10-year period. And so, this is incredibly attractive.”
Crowdsourcing capital this way and combining project risks can dramatically increase the payoff potential chances, but Lo suggested that there’s also another way to do better in modeling, that has to do with machine learning.
He presented how his company, QLS Advisors, uses the same kinds of tools used in credit scoring or other analyses to focus on clinical trial outcomes.
Using tried-and-true statistical principles like random forest and nearest neighbor models, Lo showed how companies can target particular results, and gave viewers a QR code to an article published in the Harvard Data Science Review.
The code for some of these projects, he said, is public and accessible in github.
He talked about a partnership with Novartis, in which partners including MIT professionals put this theory into practical use, in explaining some of the ways that crowdsourcing can enhance the modeling power of new technologies.
“Over 300 participants in 50 different teams submitted 3000 different models to try to improve on ours,” he said. “Now they had a head start, they had our models… the data vendor licensed the data to them. And over that period of time, they developed a number of interesting alternatives. … the results were useful and interesting.”
In closing, Lo went back to the idea of what finance, as a rule, is supposed to help people to achieve.
“Finance does not have to be a zero sum game, if we don’t let it,” he said. “Those of us who are in the financial industry, we may think that finance is the ‘be all and end all.’ But for most … people, it is a means to an end, not an end unto itself. With the right business models, the right amount of financing, we can achieve that mythical state of actually being able to do well.”
Lo is a Charles E. and Susan T. Harris Professor at Massachusetts Institute of Technology, and co-founder of QLS Advisors, with academic bonafides from Harvard and Yale.
Read the full article here
