Suggestions
Sandy Aronson
Executive Director of IT and AI Solutions at Mass General Brigham Personalized Medicine
Sandy Aronson is a prominent figure in healthcare technology and artificial intelligence, currently serving as the Executive Director of IT and AI Solutions for Mass General Brigham Personalized Medicine and the Senior Director of IT for their Accelerator for Clinical Transformation.12
Professional Background
Aronson's career spans various roles in technology and healthcare:
- He leads teams that develop and deploy technology to improve clinical processes in areas such as clinical genetics, remote health, and decentralized clinical trials.1
- His work focuses on building infrastructure that supports the evolution and practice of genetic-based personalized medicine in both patient-facing and laboratory settings.12
- Aronson's team has developed the GeneInsight Suite of applications, which were registered as a medical device and distributed.1
Current Projects and Research
Aronson is involved in several cutting-edge initiatives:
- Developing and validating Generative AI functionality to support variant assessment in whole genome sequence management and interpretation.1
- Enhancing remote patient management capabilities using Generative AI to improve trial quality and cost-effectiveness in decentralized clinical trials.1
- Investigating the use of Generative AI in clinical trial screening and scientific literature review for genetic variant assessment.34
Education
Aronson holds multiple advanced degrees:
- Masters in Organizational Behavior from Stanford University
- Bachelors in Computer Science from Stanford University
- Masters focused in Biology from Harvard Extension School12
Recent Publications
Aronson has contributed to significant research in the field of AI in healthcare:
- Co-senior author of a study published in NEJM AI examining the use of Generative AI for clinical trial screening.3
- Corresponding author of a paper in NEJM AI investigating the use of Generative AI in scientific literature review for genetic variant assessment.4
Sandy Aronson's LinkedIn username is sandy-aronson-13995, reflecting his professional presence on the platform.
Highlights
Generative AI inconsistency can present a risk to clinical processes that should be considered in hazard analyzes. This may be an obvious statement, but it appears to me that the risk of inconsistency increases with task complexity. This figure from our new @MassGeneralBrigham study shows metrics generated through repeated runs of a dataset focused on testing GPT-4 series models’ ability to assess evidence of variant pathogenicity contained in genetic literature. In my view the models perform surprising well, but they are also surprisingly inconsistent. It was great fun to work with Matt Lebo, PhD, FACMG, Kalotina Machini, Jiyeon Shin, Pranav Sriraman, Sean Hamill, Emma H., Charlotte Mailly, Angie N., Sami Amr and Michael Oates on this analysis. Thank you NEJM AI for helping us spread the word on this issue! Note to API users: The risk of inconsistency is present even when steps are taken to reduce inconsistency including setting the temperature parameter to 0, setting seed to a constant value, and monitoring the fingerprint.
I appreciate Tatyana Kanzaveli including me in this panel proposal. The potential for AI in this area is significant, and there are a lot of lessons learned in this space that may be applicable to other fields. If the panel sounds interesting to you, it would be great if you could vote for it!
