At Mount Sinai, researchers have turned the quiet data of daily movement into a kind of foreknowledge — a system that can sense, an hour before it happens, when a woman living with chronic pelvic pain is about to settle into the stillness that worsens her condition. For the one in seven women whose days are shaped by endometriosis, fibroids, and the particular fatigue that makes rest feel necessary, this is not merely a technical achievement but a reframing of what care can look like: not advice given after the fact, but a gentle signal arriving at the right moment. The work asks an old questi
Mount Sinai AI Predicts Prolonged Sitting in Women With Chronic Pelvic Pain
Chronic pelvic pain affects approximately 1 in 7 women, causing pain, fatigue, and reduced quality of life through conditions like endometriosis and adenomyosis.
More complex AI is not always better.
Mark
So the AI is predicting sitting before it happens. How far ahead are we talking?
Mimi
About an hour. The model looks at the patterns in your movement and heart rate and sleep, and it says: in the next sixty minutes, you're probably going to sit down for a while.
Mark
And then what? It tells you to get up?
Mimi
Not exactly. It prompts you to move—a small break, what they call an exercise snack. The idea is to interrupt the pattern before it locks in.
Luke
But we don't know yet if the prompts actually work. The study shows the prediction is accurate. That's different from showing it changes behavior.
Mimi
Right. That's what the clinical trials are for. They're testing whether the prompts reduce sitting time and improve symptoms.
Mark
Why does sitting matter so much for women with pelvic pain?
Mimi
Prolonged sitting can worsen pain and fatigue. But the standard advice—just move more—doesn't fit the reality of living with chronic pain. You can't always move.
Luke
One thing worth noting: they tested this on 134 women with pelvic pain and 61 healthy women. The comparison group is smaller. That's fine for a feasibility study, but it's worth keeping in mind.
Mark
And the AI runs on the device itself, not in the cloud?
Mimi
Yes. The simpler models work just as well as the complex ones, and they can live on your phone or watch. That means your data stays with you.
Luke
That's a real privacy win. Though I'd want to know: what happens to the data over time? Is it stored? Deleted? That matters for actual deployment.
Mark
When do we know if this actually helps people?
Mimi
The clinical trials are next. That's where they'll see if the prompts change behavior and improve quality of life.
Der Puls
- Chronic pelvic pain traps roughly one in seven women in a cycle where sitting brings relief but prolonged stillness deepens their suffering — and standard medical guidance has largely failed to account for this cruel paradox.
- Mount Sinai researchers trained personalized AI models on just ten days of Fitbit data, achieving the ability to forecast sedentary periods one hour in advance with accuracy that matched far more computationally demanding systems.
- The discovery that simple models outperformed complex deep-learning approaches is itself consequential — lightweight algorithms can run locally on a phone or watch, keeping sensitive health data private and eliminating dependence on cloud infrastructure.
- The system proved resilient against the imperfections of real life, maintaining predictive accuracy even when devices were removed, forgotten, or left unsynced for a day.
- The team is now designing clinical trials to answer the harder question: whether AI-generated movement prompts delivered at precisely the right moment can actually reduce sitting time, ease symptoms, and restore quality of life.
At Mount Sinai, researchers have turned the quiet data of daily movement into a kind of foreknowledge — a system that can sense, an hour before it happens, when a woman living with chronic pelvic pain is about to settle into the stillness that worsens her condition. For the one in seven women whose days are shaped by endometriosis, fibroids, and the particular fatigue that makes rest feel necessary, this is not merely a technical achievement but a reframing of what care can look like: not advice given after the fact, but a gentle signal arriving at the right moment. The work asks an old question in a new way — can understanding a person's patterns, deeply enough, help them interrupt the ones that harm them?
A research team at the Icahn School of Medicine at Mount Sinai has developed an artificial intelligence system capable of predicting, roughly an hour in advance, when a woman with chronic pelvic pain is about to enter a prolonged period of sitting. Published in npj Women's Health, the work addresses a quiet but significant gap: for women living with endometriosis, adenomyosis, or uterine fibroids, extended sedentary periods worsen symptoms, yet the standard instruction to simply move more fails to reckon with the exhaustion and pain that makes stillness feel like the only option.
The study enrolled 134 women with chronic pelvic pain disorders alongside 61 healthy participants, each wearing a Fitbit for up to ninety days. Using only the first ten days of recorded movement, heart rate, and sleep data, the team built personalized predictive models for each individual. Their target was specific: identify the approach of a fifteen-minute or longer sitting period during waking hours — the window in which a brief movement break might interrupt the pattern before it solidifies.
The most striking finding was not the prediction itself but what achieved it. The simplest, least computationally intensive models performed just as well as sophisticated deep-learning systems. This matters practically: a lightweight model can live on a person's phone or watch, processing data locally without transmitting sensitive health information to external servers. The system also held its accuracy when data was incomplete — when a device was removed or a day was missed — a crucial quality for any tool meant to function in the texture of real life.
Led by assistant professor Ipek Ensari, the team is now moving toward clinical trials designed to test whether prediction translates into meaningful change. They are building a just-in-time adaptive intervention — personalized prompts delivered at the precise moment the algorithm anticipates a long sit — to determine whether such nudges can reduce sedentary time, ease pain, and improve daily life. The researchers also envision the approach extending to other chronic conditions where prolonged sitting compounds harm.
A team at the Icahn School of Medicine at Mount Sinai has built an artificial intelligence system that watches the patterns in a woman's daily movement and predicts, roughly an hour in advance, when she is about to sit down for an extended stretch. The work, published September 30 in npj Women's Health, emerges from a straightforward observation: women living with chronic pelvic pain often find themselves stuck in long sedentary periods, and the standard advice to "move more" does not account for the particular weight of their condition.
Chronic pelvic pain touches one in seven women. It arrives alongside endometriosis, adenomyosis, uterine fibroids—conditions that bring pain, fatigue, and a kind of exhaustion that makes sitting feel safer than standing. The researchers wanted to know whether the data streaming from a simple wearable device could do more than count steps after the fact. Could it anticipate the moment before stillness takes hold?
They enrolled 134 women with chronic pelvic pain disorders and 61 healthy women as a comparison group. Everyone wore a Fitbit for up to ninety days. The devices recorded, minute by minute, physical activity, heart rate, and sleep. Using the first ten days of data from each person, the team trained personalized models—mathematical blueprints specific to each individual—to forecast activity levels one hour ahead. The goal was narrow and practical: could the system spot an incoming fifteen-minute stretch of sitting during waking hours, the kind of moment when a brief movement break—what the researchers called an "exercise snack"—might interrupt the pattern?
What emerged surprised the team. The simplest models, the ones that required the least computational muscle, predicted prolonged sitting just as accurately as the far more complex deep-learning systems they tested alongside them. That matters. A lightweight model can run directly on a person's phone or watch. It does not need to send sensitive health data to a distant server. It does not depend on cloud infrastructure. It stays with the person who owns it.
The models held up even when the data was incomplete—when someone removed their device for a shower, or forgot to sync it, or simply went a day without wearing it. Real life is messy. The system worked anyway.
Ipek Ensari, an assistant professor of artificial intelligence and human health at Mount Sinai, framed the next question plainly: the system can predict. But does prediction change anything? The team is now building what they call a just-in-time adaptive intervention—a framework that will deliver personalized movement prompts at the moment the algorithm senses a long sit coming on. Clinical trials will test whether those prompts actually work: whether they reduce sedentary time, whether they ease symptoms, whether they improve the quality of life for women whose days are shaped by chronic pain. The researchers also see the approach extending beyond pelvic pain, to any chronic condition where prolonged sitting makes things worse.
Bemerkenswerte Zitate
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.— Ipek Ensari, Ph.D., Assistant Professor of Artificial Intelligence and Human Health, Mount Sinai
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.— Mount Sinai research team
Möchten Sie die ganze Geschichte?
Das Original lesen bei Pulse 2.0 ↗