Wearable AI System Predicts Sedentary Periods for Women with Chronic Pelvic Pain
Researchers at Mount Sinai have developed an innovative artificial intelligence system designed to predict when women with chronic pelvic pain are likely to remain sedentary. This personalized approach to digital healthcare utilizes data from consumer wearables, such as Fitbit devices, to forecast inactivity before it occurs. The findings of this research were published in the peer-reviewed journal npj Women’s Health.
The AI system aims to provide timely prompts for movement, encouraging users to stand, stretch, or take short walks before they enter prolonged periods of sitting. While the study successfully evaluated the accuracy of the forecasting method, it did not assess whether acting on these predictions could alleviate pain or yield other clinical benefits.
Chronic pelvic pain can stem from various conditions, including endometriosis, adenomyosis, and uterine fibroids. The World Health Organization estimates that endometriosis affects around 10 percent of women of reproductive age globally, equating to approximately 190 million individuals. Symptoms of this condition often include severe menstrual pain, chronic pelvic pain, heavy bleeding, and fatigue, which can significantly hinder daily activities and overall well-being.
The Mount Sinai project seeks to address the inadequacy of generic advice to "sit less and move more" for individuals managing persistent pain. Instead, it explores the potential of wearable devices to learn individual activity patterns and suggest appropriate times for movement that align with users' routines and comfort levels.
Senior author Ipek Ensari, an Assistant Professor of Artificial Intelligence and Human Health at the Icahn School of Medicine, stated, "Our goal was to determine whether everyday wearable devices could serve as an early-warning system for prolonged sitting in women living with chronic pelvic pain. Rather than offering generic advice after the fact, we wanted to anticipate these moments and support people with simple, well-timed prompts that fit naturally into their daily lives."
The research involved 134 women diagnosed with chronic pelvic pain disorders, primarily endometriosis, alongside 61 healthy participants. Over a period of up to 90 days, participants wore Fitbit devices, which collected minute-by-minute data on their physical activity, heart rate, and sleep patterns. This data was used to train a personalized model capable of forecasting activity levels one hour in advance, specifically identifying 15-minute sedentary periods during waking hours.
Importantly, the model does not diagnose medical events or determine the reasons behind inactivity. Instead, it identifies patterns in wearable data that suggest a likely period of limited movement. Future interventions could leverage these forecasts to recommend brief activities, termed "exercise snacks," tailored to the user’s symptoms and preferences.
Interestingly, the study found that simpler, more interpretable models were nearly as effective in forecasting sedentary periods as more complex deep-learning models. Lead author Jannes Jegminat remarked, "We were surprised by how well the simplest models performed. More complex AI is not always better. Lightweight, interpretable models can accurately forecast sedentary behavior while being practical enough to run directly on a person’s own device, which also helps protect privacy."
This finding has significant implications for the design of wearable healthcare technology. A lightweight model could operate on personal devices, minimizing the need to transmit sensitive health information to remote servers. While on-device processing does not eliminate all privacy concerns, it can reduce exposure and reliance on cloud computing, as well as lower the computational resources required for predictions.
The study also acknowledges the challenges of wearable datasets, which are often incomplete due to device removal, battery issues, or synchronization failures. The forecasting models demonstrated robustness even with missing data, suggesting that useful predictions can still be made under typical usage conditions.
However, the researchers caution that the model does not guarantee accurate forecasts and is not yet ready for routine clinical use. The findings provide a foundation for further development, with prospective clinical trials needed to establish whether these predictive prompts can effectively reduce sedentary time, alleviate symptoms, or enhance quality of life.
If successful, the forecasting framework may be adapted for other chronic conditions where pain, fatigue, or limited mobility contribute to prolonged inactivity. The immediate significance of this study lies in its potential to transform ordinary wearable data into proactive alerts, rather than merely recording past inactivity. It also highlights that effective healthcare AI does not always require complex models; sometimes, a system that understands individual routines and prioritizes data privacy can be more beneficial.




