Back
Jeff Hammerbacher
Cofounder, Cloudera

Centerstone Research Institute - Mapping Hypothesis Testing in a Health Care Environment

🎥 Jun 14, 2017 📺 Centerstone Health ⏱ 1m 👁 39 views
Jeff Hammerbacher, co-founder of Cloudera, asks "How does the ability to do empirically driven hypothesis testing in a web ...
Watch on YouTube

About Jeff Hammerbacher

Jeff Hammerbacher, cofounder of Cloudera and an assistant professor at the Icahn School of Medicine at Mount Sinai, has focused his recent work on applying data science to biomedical research, particularly cancer immunotherapy. In a 2020 talk, he described a relation extraction project on biomedical literature aimed at understanding T cell function and differentiation, noting that immune checkpoint blockade is forecast to generate over $100 billion in sales by 2024 and that over 2,000 clinical trials for such therapies are active. He has emphasized the importance of open science, stating that his lab’s pipeline, epiD, is open source under an Apache 2.0 license and that all development occurs publicly on GitHub to allow others to reperform analyses. Hammerbacher has also spoken about the challenges of translating high-throughput web experimentation to healthcare, arguing that the field needs to conceive of healthcare delivery as a high-frequency, low-cost touchpoint with patients to enable rapid learning. He has criticized the tendency of institutions to outsource data infrastructure to large tech companies, calling such use cases “just marketing” from a Silicon Valley perspective. In earlier talks, he discussed the ethical implications of data collection, stating that “the best minds of my generation are thinking about how to make people click ads” and that decisions about what to measure involve “implicit political, moral, and ethical choices.”

Source: AI-verified profile updated from Jeff Hammerbacher's recent appearances. Browse all interviews →

Transcript (1 segments)
J
Jeff Hammerbacher0:09
Health care is nothing more than a time series of interventions on an animal system, right? So if we wanted to intelligently make interventions into that time series, one way to do it would be to never sample. You wait for that time series to wander into your office and then make a really aggressive intervention into it. But probably a much better way to build a control system would be to frequently sample, figure out a way to at low cost frequently sample that system, and take much more lightweight interventions. So until we start to conceive of healthcare provision as a high-frequency, low-cost touch point with the patient, then you're right, we will never be able to do high-throughput experiments. I think there's some work that has to go into how we conceive of healthcare delivery before we can start to bring in the tools of high-throughput experimentation.