Like having a personal healthcare mentor in your pocket

Individuals battling cancer who undergo stem cell transplantations must navigate a protracted recovery, necessitating medications that come with severe side effects and constant support. The process is taxing, with research indicating that over 70 percent of patients fail to follow medication schedules. 

Statistician Susan Murphy dedicates her workdays to assisting those grappling with such daunting conditions. The Mallinckrodt Professor of Statistics and Computer Science and associate faculty member at the Kempner Institute and her team address healthcare challenges not through traditional medicine, but via mobile applications. 

Murphy’s laboratory focuses on developing advanced computational strategies known as reinforcement learning algorithms, which form the foundational technology behind next-gen applications designed to aid individuals in adhering to medication protocols, maintaining consistent dental hygiene, or decreasing cannabis consumption.

And if this resembles one of those prevalent applications that monitor steps or calorie intake, reconsider. 

“If you’ve ever installed a health application, those tend to be rather ineffective,” Murphy remarked. “For instance, you might download a physical activity app, you sprain your ankle, yet it continues to suggest that you take a walk.” 

By harnessing innovations in artificial intelligence and sensing technologies to move past uniform interventions, the laboratory’s apps can provide real-time personalizations, offering psychological incentives, and in certain instances, utilizing social networks to assist users in reaching their objectives.

This methodology is referred to as “just-in-time adaptive intervention” because it seeks to deliver assistance precisely when needed by recognizing evolving needs and circumstances. 

At present, the Murphy lab is collaborating with software developers, cancer care specialists, and behavioral scientists to craft an application for patients undergoing stem-cell transplants and their primary caregivers, often parents. 

Effective health management, particularly for the most seriously ill, generally requires the involvement of others. For instance, as many as 73 percent of family-caregivers bear the primary responsibility for overseeing cancer-related medications. 

The researchers are currently in the initial phases of creating the algorithm, which is to be tested in a first clinical trial this year in partnership with collaborators at the University of Michigan and Northwestern University. This trial, known as ADAPTS HCT, will concentrate on teens and young adults who have received stem-cell transplants within 14 weeks post-operatively.

The algorithm will guide sequential decisions, including when and whether to issue motivational reminders to the patient, as well as whether to send communications and alerts to both the patient and caregiver. The application features a word-guessing game that promotes social support and teamwork between the patient and their caregiver. 

“We surmise that by enhancing the bond between patients and caregivers, patients will be able to function more effectively and manage their medications appropriately,” stated Harvard postdoctoral fellow Ziping Xu, who is spearheading the ADAPTS HCT algorithm development.

The app will utilize reinforcement machine learning, whereby the software will “learn” from past interactions. For example, instead of merely dispatching pre-set reminders regarding medications, the algorithm will adjust timing and content based on instances when they have been most effective for patients. This minimizes the likelihood of notifications being considered irrelevant or poorly timed and thereby ignored over time.

“We employ the algorithm to discover the most effective method to engage with each patient,” Xu remarked. 

The Murphy lab is extending its algorithmic expertise across additional fields. Alongside their University of Michigan partners, they have recently conducted a pilot initiative named MiWaves, which targets young individuals grappling with cannabis misuse. 

Similar to the ADAPTS HCT application, MiWaves continuously evolves and adjusts through interactions with each patient to refine its decision-making rules, aspiring to assist them in decreasing their daily consumption.

The laboratory is also several years into a project titled Oralytics, which recently concluded a 10-week randomized trial aimed at optimizing the delivery of push notifications to support patients in adhering to a dental hygiene routine: two sessions lasting two minutes each day, covering all four quadrants of the mouth. 

The initial Oralytics clinical trial included approximately 70 participants who received the mobile application along with a wireless-enabled toothbrush that transmitted data to the team’s collaborators at Proctor and Gamble. 

Graduate student Anna Li Trella, who led the Oralytics project through its first trial, expressed that the newly gathered data will aid the team in devising strategies to better tackle complicated issues like missing data and software glitches. 

“There are numerous challenges to executing an algorithm in real-world scenarios,” Trella noted. “Now that we have completed the first trial, we can enhance our approach to improve data collection and learning efficacy for the algorithm.”

Murphy envisions her lab as creating practical pocket mentors  who can assist individuals in achieving their objectives. 

“Very, very few individuals can afford a personal coach. Moreover, some may prefer not to engage in such intensive human interaction,” Murphy stated. “That’s where the notion of these digital supports comes into play.”