The smartwatch will make a beep sound to alert the user to take a short walk. The bedtime app recommends calming breathing exercises before bed. Such just-in-time adaptive interventions (JITAI) provide individualized support at the moment when people are most likely to benefit. Advances in machine learning (ML) have made these tools even more powerful, allowing them to predict user needs and adjust recommendations in real time.
“Even the most sophisticated patient health tools are limited by people’s willingness to use them,” said Maia Jacobs, assistant professor of computer science and preventive medicine. “While research and industry are focused on making these systems more accurate and personalized, little is known about what people expect from predictive health tools and what they need to use them with confidence.”
To address this gap, researchers at Northwestern University’s Health Aware Bits (HABits) Lab and Personalized and Adaptive Technology for Health (NU-PATH) Lab investigated how pregnant people envision interacting with future predictive stress management tools. They are co-designing a JITAI system called Lowering Unwanted Cortisol Activity (LUCA) and are working directly with pregnant women to tailor its support features to end-user expectations, preferences for when and how to use technology in daily life, and concerns about uncertainty in features and algorithms.
In the study, “I don’t just like being told what to do, I need to know why,” the researchers found that participants wanted more than personalized recommendations: They wanted to understand how the technology worked and be able to decide when to trust and rely on its guidance.
“Our philosophy is to co-shape the future of ML-driven health tools based on how pregnant people perceive and understand these systems by centering lived experiences,” said Mara Ulloa, a computer science doctoral student at Northwestern Engineering and lead author of the study.
Ulloa explained that the study revealed two important requirements. They are the desire for a consistent mental image of how the system works and the need for personal autonomy.
“Just as patients engage more in their medical care with active cooperation from their doctors, JITAI users may be able to engage more meaningfully with technology if they understand its logic,” she said.
Ulloa presented the paper at the inaugural ACM Interactive Health Conference in Porto, Portugal, July 5-8, where he also presented it at a series of doctoral colloquium workshops. Co-author of the study, published April 27 ACM Transactions on Computing for Health Special issue on human-centered computing in healthcare. Mr. Jacobs also participated. Nabil Al-Shurafa, associate professor of preventive medicine and (courtesy) computer science and electrical and computer engineering. Negar Kamali, a doctoral student in computer science; Glenn Fernandez (PhD ’26, MS ’22); Elizabeth Soyemi, research assistant at Northwestern University Feinberg School of Medicine. Miranda L. Belzer, former postdoctoral fellow at the Feinberg Center for Behavioral Intervention Technology;
make a meaningful impact
Building on more than a decade of basic research on the effects of prenatal stress reduction on maternal well-being and early childhood neurodevelopment, the collaboration between HABits Lab and NU-PATH began in 2023 with support from Northwestern University’s Center for Safety Improvement in Machine Intelligence.
Alshhrafa and the HABits Lab team first developed an ML algorithm that incorporated wearable devices and app-based surveys to predict next-day stress in pregnant people and provide preventive cognitive behavioral therapy (CBT)-based interventions.
In this new study, led by Ulloa as part of her dissertation research, the team at NU-PATH and HABits Lab aimed to turn that potential into meaningful impact by recruiting 20 pregnant women to explore their interest and expectations in ML-driven JITAI and incorporate their perspectives and attitudes into future system designs.
“As predictive algorithms that can predict stress advance, our understanding of what people actually want and need from those algorithms should advance at the same pace,” Ulloa said. “The real contribution of this paper is that rather than technology dictating the human experience, human-centered needs dictate the direction of prenatal preventive health technology.”
Based on participatory design principles, the researchers developed a storyboard illustrating the types of functionalities and interactions that are technically possible. The co-design session guided participants through the relatable story of a pregnant individual, “Aria,” and “LUCA,” a text-based agent that embodies an ML model. LUCA can predict future instances of stress and provide timely CBT and mindfulness recommendations.
Participants viewed interacting with these tools as a two-way process. Just as they wanted technology to help them understand recommendations, they also wanted the opportunity to help AI understand its decisions by distinguishing intentional choices from situations in which it simply cannot act or explicitly disagrees.
“People realize that AI learns from people’s actions,” Jacobs said. “Participants wanted a way to tell the system the difference between ‘I can’t do this’ and ‘I decided not to do this,’ so the technology could better understand participants’ preferences instead of making incorrect assumptions.”
Practical use
Moving from co-design to implementation, Ulloa and the research team are currently building a functional mobile application prototype that directly incorporates the study participants’ design requirements. The goal of this phase is to assess whether supporting the mental models and controls that users explicitly request will lead to sustained transparency and measurable increases in engagement. Ulloa also wants to identify new or unresolved user issues.
In the next stage, the team plans to introduce the prototype directly into the daily lives of pregnant people. Participants will download the application and use it continuously in their natural environment, providing researchers with high-fidelity data about how their systems endure the unpredictable rhythms, stressors, and daily changes of pregnancy.
This initial deployment is designed to lay the foundation for a multi-year research exploration in the NU-PATH lab, with the resulting data, user feedback, and other insights ultimately translated into a large-scale randomized controlled trial that will clinically validate human-centered, ML-driven prenatal care.
“Integrating human-computer interaction, behavioral science, and machine learning with real-world users is extremely rewarding,” Ulloa says. “What makes this research unique is that it allows pregnant women to define how the personal technologies that are truly valuable to them work. Ultimately, the most impactful medical technologies are the result of true collaboration across research disciplines and, most importantly, with the people they ultimately hope to help.”
