Zoom: https://illinois.zoom.us/j/89068564522?pwd=jxbmWEMaCEjkeKr66MPxQEnS5Rhjhw.1
Abstract:
In the US, unhealthy behaviors—such as a sedentary lifestyle, overeating, substance use, and tobacco use—account for approximately 40% of the risk of premature deaths. Modifying these unhealthy behaviors can mitigate the risk of harm, but behavior change is often difficult because of the high self-management burden. Health applications on phones and wearables aim to reduce this burden, but despite their promise, these applications remain ineffective at promoting meaningful behavior change. This inefficacy is partly due to a research focus on predicting health and well-being indicators without attention to modernizing interventions. As a result, interventions have remained generic, non-adaptive, and ineffective at promoting change. My research aims to modernize interventions by delivering the right intervention at the right time to make self-management of behavior easier.
I focus on three types of work to modernize interventions. First, I build personalized recommender systems to replace generic interventions. These systems use machine learning to find patterns in dense mobile data and issue actionable suggestions that are easy to implement and are aligned with a user's health goals. This talk will cover two of these personalized recommender systems. MyBehavior is an app that automatically learns a user’s routine from phone data and recommends small changes to improve physical activity and dietary intake. The sub-goal app is another recommender system that divides a daily goal into personalized sub-goals, suggesting changes when a user is usually active. Second, I work on continuously evaluating and adapting interventions using state-of-the-art sequential experiments and algorithms. This work ensures interventions are delivered when effective and highlights when they are not, prompting new interventions to be tested. This talk will briefly cover several novel interventions I created and evaluated with sequential experimentation. Lastly, I integrate these interventions and signals into traditional healthcare delivery systems both in and between clinic visits. I will provide an overview of a dashboard that balances visualization and AI for reliable and low-burden clinical decision-making. My long-term research plan is to maximize public health impact by creating the right intervention using personalized recommendation, delivering this intervention at the right time using proper evaluation, and delivering it to people in need by integrating with healthcare systems.
Bio:
Mashfiqui Rabbi is a senior research scientist at Optum AI under UnitedHealth Group, the largest payer and one of the largest providers in US healthcare. He received his Ph.D. in Information Science from Cornell University under Professor Tanzeem Choudhury. His Ph.D. thesis led to the creation of the MyBehavior app, the first mobile recommender system to automatically generate personalized physical activity and food suggestions from mobile phone data. Mashfiqui was a postdoctoral fellow at Harvard University, where he worked with Professor Susan Murphy and created the first just-in-time intervention for improving health app engagement. This engagement intervention was later adopted by three NIH-funded grants focusing on youth substance abuse, cancer rehabilitation, and sickle cell disease. Mashfiqui continued his work as a research scientist at Apple Health AI, where he helped establish the AI-based intervention group. His sub-goal app was adopted for the custom plan feature in Apple’s Fitness+ app in iOS17, which is currently used by millions of Apple users. Mashfiqui’s work has been featured in MIT Technology Review, New Scientist, the Economist, Mashable, and the NY Times.
Faculty Host: Jim Rehg
Meeting ID: 890 6856 4522 ; Password: csillinois