Professor Susan Murphy Combines AI and Statistical Methods to Transform Healthcare Interventions
As your dentist slowly pulls out the probe, you know what’s coming next: the pronouncement that there’s a new cavity in your mouth. While you’ve tried to ward off cavities and to faithfully brush your teeth every night, often you let it go when juggling many responsibilities. The problem of sustaining dental health, intimately linked to general health and well-being, is a common one. According to the NIH Website, the annual total costs of dental disease to work productivity around the world was estimated to be about $545.4 billion in 2015 (Righolt et al. 2018).
To address the widespread challenge of improving dental health and to provide healthcare support for many other conditions (e.g., hypertension, bone marrow transplants, and addiction), Professor Susan Murphy and her research lab have developed new digital app interventions for clinical trials. These mobile apps are powered by sequential decision-making, reinforcement learning (RL) algorithms. During the trials, the algorithms update their decision-making process for sending patient notifications based on positive patient behavior data (e.g. brushing their teeth). Murphy’s research combines AI tools, such as RL, with statistical methods, such as confidence intervals and hypothesis tests, to generate forward-thinking healthcare interventions that can make a real difference in people’s lives. The most recent recognition of Murphy’s leadership across statistics, computer science, and clinical trials was her selection for the Mosteller Statistician of the Year Award by the Boston Chapter of the American Statistical Association (BCASA). Murphy’s award ceremony/talk and her recent interview with us provide additional insights into her research and how its cross-disciplinary nature is promoting healthcare innovation.
Embodying a “Full Statistician,” “Full Scientist”
Named in honor of Fred Mosteller, the founding chair of the Department of Statistics at Harvard, the Mosteller Statistician of the Year Award is given to “individuals from academia, industry, and government who have made exceptional contributions to the field of statistics and who have shown outstanding service to the statistical community” (source: BCASA's Website). In her introduction to Murphy’s talk, Professor Xihong Lin cited Murphy’s honors, such as her election to the National Academy of Sciences, MacArthur Award, and contributions to the community, such as her tenure as president of the Institute of Mathematical Statistics and the Bernoulli Society. Lin praised Murphy for operating as a “full statistician” and “full scientist” by developing and employing statistical and algorithmic methods to have real-world positive impacts.
Pursuing Sequential Decision-Making Problems
In an interview, Murphy explained the driving force behind her research spanning computer science, statistics, and medical fields: “I want very much to contribute to social good and improve society; healthcare is a natural area for achieving this goal.” Her interest in problems related to sequential decision-making in the medical domain grew out of her interest in working on causal inference problems more broadly. In the early 2000s, Murphy’s work focused on designing sequential decision-making trials to help clinicians make improved, repeated decisions for patients with chronic diseases (e.g., cardiovascular diseases, mental illness and substance abuse). Then, as AI tools developed further, Murphy adapted her approach by incorporating more automated methods, such as RL algorithms that learn from user data over time to deliver appropriate healthcare interventions.
Murphy’s receipt of a MacArthur Award in 2013 was a turning point for implementing and generating concrete results in clinical trials with digital healthcare apps (watch Murphy’s MacArthur Award Video). Given annually to recipients across the arts, sciences, and humanities, the MacArthur Award is a “no-strings-attached grant for individuals who have shown exceptional creativity in their work and the promise to do more” (source: MacArthur Award website). Reflecting on the experience of winning the award, Murphy said, “it was an enormous confirmation of my efforts in sequential decision-making; I found it very meaningful to see numerous investigators incorporating these experimental designs into clinical trials.”
Combining AI and Statistical Tools to Transform Healthcare Interventions
Murphy’s Mosteller Award talk, “Online Reinforcement Learning in Digital Health Interventions,” elaborated on her work by discussing two examples of clinical trials with digital healthcare apps: HeartSteps and Oralytics. While HeartSteps encourages patients to walk to reduce the need to take medication for hypertension, Oralytics prompts patients to improve toothbrushing. Both trials involve “wrap-around” care (complementary support for patients and their healthcare providers) and include collaboration with clinicians throughout the trial planning and implementation process. At the start of a trial, the RL algorithms learn from the patient data and then issue notifications to encourage positive behavior (e.g., walking or toothbrushing). After each implementation, Murphy and her collaborators perform statistical analyses on the algorithm results to improve the algorithm for the next implementation. “The goal in these implementations,” explained Murphy in her talk, “is to use autonomous algorithms so that these apps can adapt on the fly to user behavior and provide appropriate interventions at the right time; however, the challenge is how do you help an artificial agent understand how a human agent makes decisions?” Through Murphy’s creative, combined use of AI and statistical methods and collaboration with medical scientists, she is helping to answer this question and refine AI decision-making that will transform the delivery of healthcare interventions.
Advancing Future Digital Healthcare App Interventions:
For the next stage of her lab’s work, Murphy sees the focus shifting to how AI tools and sequential decision-making can leverage patients’ social networks to improve patient outcomes. In her interview with us, Murphy elaborated on her interest in this area of research: “For a long time, I’ve pushed for research on sequential decision-making on social networks surrounding someone who is struggling with a chronic disease, particularly with addiction, because social networks are critical for addressing addiction. I’m excited that we have funding for three studies that look at small social networks that surround a target person.” The three studies involve patients’ caretakers in sequential decision-making trials to improve the patients’ health outcomes. One trial is for young adults with bone marrow transplants who need help with sticking to medication with severe side effects. In another trial, adults with early-stage dementia and dental sores will use electronic toothbrushes with sensors to collect data on and improve their toothbrushing habits. The third trial is for adolescents who have been using illicit drugs; the app will help parents to time more effective interventions.
Prof. Murphy’s most recent focus on leveraging social networks to improve patients’ health demonstrates the strong potential of AI and statistical tools to support more effective healthcare interventions. While dental disease, cancer, and addiction have a devastating impact on people’s lives, current innovations in wraparound care (as well as in other areas of healthcare) will continue to improve people’s lives and health outcomes.
