Can a wearable predict prolonged sitting for people with chronic pelvic pain?
A peer-reviewed study used Fitbit data from 134 participants to forecast sedentary periods one hour ahead. It demonstrates a modelling pipeline—not that prompts improve pain, activity or health outcomes.
Editorial responsibility: The Impact of AI Editorial Desk · Report a factual concern
Research topic
Whether participant-specific wearable models can forecast sedentary periods accurately enough to support future just-in-time prompts
At a glance
- 1The study analysed Fitbit data from 134 participants with chronic pelvic pain disorders: 19 in a tuning cohort and 115 in the main evaluation cohort, plus a separate 61-person healthy control cohort.
- 2Simple adaptive and statistical models performed similarly to a long short-term memory network, suggesting recent activity and daily patterns carried much of the useful signal.
- 3No movement prompt was delivered and no change in pain, sitting time, activity or health was tested, so clinical benefit remains unproven.
Living evidence record
Impact record IAI-146UBL6
Evidence stage
Studied
Confidence
Supported
Reporting basis
Source analysis
Independent support
Present
Record status
Monitoring
Last checked
30 September 2026
Source trail
2 direct sources across 2 source types.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Related-source reporting disclosure
This record analyses 2 linked source records around the same underlying development. The extra records add method, date or context, but they do not by themselves constitute independent replication of every performance claim or predicted outcome.
The study forecast behaviour; it did not test a treatment
Researchers at Mount Sinai and Columbia used minute-level Fitbit data to forecast periods of sedentary behaviour among participants with chronic pelvic pain disorders. The peer-reviewed paper was published on 30 September in npj Women's Health. Its practical aim is a future system that could recognise when someone is likely to remain inactive and send a timely prompt, rather than issue generic reminders throughout the day. The present study stops before that intervention: it evaluates prediction from recorded behaviour and does not deliver prompts or measure whether anyone moves more as a result.
The parent observational study collected daily and weekly information for 90 days using Fitbit Inspire 2 trackers. Eligible participants were assigned female at birth, menstruating, aged 18 to 64, reported a surgical or clinician diagnosis of a chronic pelvic pain disorder and had experienced pain for at least six months. The modelling analysis used 134 participants: 19 with at least 90% data completeness formed the tuning cohort, while 115 with sufficient valid data formed the evaluation cohort. A separate 61-person healthy control cohort was used for an additional consistency check.[1][2]
Recent activity and daily routine carried most of the signal
The team translated Fitbit activity levels into a physical activity score and compared an adaptive recursive least-squares model with two offline approaches: SARIMA, a statistical time-series model, and an LSTM neural network. All three beat simple mean and persistence baselines for one-hour-ahead forecasting. In the 115-person evaluation cohort, median root-mean-square error was 0.090 for SARIMA, 0.091 for recursive least squares and 0.095 for the LSTM. Those small differences led the researchers to favour the adaptive model because it can update on-device without repeated central retraining.
For 15-minute sedentary bouts, the authors report an operating point of roughly one true alert a day alongside 0.6 false alerts. That is a model-selected trade-off, not a patient-tested tolerance. A reminder system that interrupts someone during pain, work, travel or sleep may create burden even when the prediction is technically correct. Conversely, an alert can be missed precisely when pain makes movement difficult. The study therefore establishes that timing information can be extracted from these wearable traces; it does not establish which alert threshold is acceptable or helpful in daily life.[1]
On-device learning could protect privacy, but the device still defines the data
Participant-specific learning can reduce the need to pool raw activity traces in a central service. The chosen adaptive model is deliberately lightweight and could update locally as routines change. That design is relevant because continuous movement, heart-rate and sleep data can reveal sensitive patterns about work, home life and health. Genuine privacy, however, depends on the full product architecture: where raw data are stored, whether model updates leave the device, who can retrieve logs, how long records persist and whether a user can pause or delete the system.
The physical activity score also inherits limitations from the Fitbit measurements used to create it. The researchers treated minutes without heart-rate data as missing, aggregated activity into time bins and excluded evaluation participants without at least 480 valid hours. That improves stability but means the reported performance applies to people who generated enough usable data. People with poorer device fit, irregular wear, different mobility patterns or limited access to consumer wearables may experience different accuracy. The sample came from one research programme and does not demonstrate performance across countries, devices or clinical services.[1]
What patients should expect—and what would change our assessment
This work should not change treatment today. It does not diagnose chronic pelvic pain, recommend a safe activity level or show that reducing sitting improves symptoms for an individual. A future tool would need co-design with patients, clear advice for pain flares and clinical review of who should not receive movement prompts. It should also distinguish a prediction from a medical instruction and let users control timing, intensity and data sharing. Accessibility matters: a system that assumes every user can respond by walking may exclude people with disability, fatigue or severe pain.
Our assessment would strengthen after a prospective trial randomises people to model-timed prompts, simpler scheduled reminders or no prompts and measures sedentary time, pain, function, adherence, unwanted interruptions and dropout. External validation on other wearables and more diverse populations would show whether the signal transfers. It would weaken if real-time deployment produces substantially more false alerts, if benefits disappear outside the observed sample or if prompts worsen pain or anxiety. For now, this is a credible forecasting study and a design candidate—not evidence of a clinically effective digital intervention.[1][2]
What this means for people
- People with chronic pelvic pain could eventually receive fewer, better-timed movement prompts, but the system has not yet shown that it improves symptoms or activity.
- Continuous wearable data can expose intimate routines, so local processing and meaningful control over collection and deletion are important.
Global context
The study was conducted within a US research programme and reports no country-level comparison. Wearable access, clinical guidance, privacy law and the feasibility of responding to movement prompts vary widely, limiting direct transfer to other health systems.
What the evidence does not yet show
- This is a secondary analysis of observational Fitbit data rather than a randomised trial of prompts or a treatment study.
- Only participants with sufficient valid wearable data entered the main evaluation, which may favour people able to wear and sync the device consistently.
- The work does not establish generalisability across countries, consumer devices, genders, disabilities or clinical settings.
What to watch next
- A prospective trial comparing model-timed prompts with simpler reminders and measuring pain, function, adherence and unwanted interruptions.
- External validation using different wearable devices and more diverse patient populations.
- A complete privacy design covering local processing, cloud transfer, retention, deletion and user control.
Evidence trail
Sources used for this report
Links checked 30 September 2026
This report is labelled source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Continue the story
Related reporting
Health & Life Sciences
Can AI link fragmented health records without a patient ID?
A peer-reviewed Brazilian study matched death, hospital and notification records with very high accuracy in one state. Its shared blocking-and-labelling pipeline means nationwide performance is still unproven.
7 min · 2 sources
Health & Life Sciences
Medical-AI privacy attacks do not affect every patient equally
A Nature study using seven real-world clinical datasets found that membership-inference attack success can vary between patients, challenging the idea that one average privacy score protects everyone equally.
4 min · 1 source
Health & Life Sciences
Did machine learning beat logistic regression on surgical sepsis?
A 328,292-case US study found no significant predictive advantage for LASSO, XGBoost or random forest. Sepsis was rare, the best risk group still had low absolute incidence, and no model has external validation.
8 min · 2 sources
Reader discussion
Add evidence, experience or a question
No account is required. Reader notes are published after a brief civility, relevance and safety check; disagreement is welcome.
Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.