Your Smart Ring Has No Idea You're Pregnant: How Pregnancy Breaks Every Biometric Algorithm
TL;DR
Pregnancy raises your resting heart rate by 10 to 20 beats per minute, drops your HRV by 20 to 30 percent, keeps your body temperature elevated for nine months, and fragments your sleep architecture. Every wearable algorithm on the market today was trained on non-pregnant baselines. When a pregnant person wears a smart ring, the algorithms interpret normal physiological changes as stress, poor recovery, or early illness. The data is not wrong. The algorithm is wrong. And nobody in the wearable industry is talking about it.
The problem nobody models for
I spent the last few months reading through the validation studies for Oura, Whoop, and Fitbit. The sample populations are remarkably consistent: healthy adults aged 20 to 45, roughly balanced by sex, almost always non-pregnant. The Oura Gen 3 sleep staging validation study (Altini and Kinnunen, 2021) had 97 participants. None were pregnant. The Whoop 4.0 HRV validation (Miller et al., 2022) had 50 participants. None were pregnant. The Fitbit Sense heart rate study had 66 participants. None were pregnant.
This is not a conspiracy. Recruiting pregnant participants for device validation is hard. Hormonal profiles change week to week. The ethical review is more involved. And the sample size you would need to capture meaningful trimester-specific data is large. So the industry just does not do it.
The result is that every readiness score, every recovery metric, every sleep stage estimate, and every stress score on the market was calibrated on bodies that are not pregnant. When a pregnant person puts on a ring, the algorithm sees a body that looks like it is in a state of chronic stress. In a sense, it is right. Pregnancy is the most metabolically demanding state a human body can be in. But the algorithm does not know that the stress is supposed to be there.
Photo by Slaapwijsheid.nl on Unsplash
What happens to resting heart rate
Blood volume increases by 40 to 50 percent during pregnancy. Stroke volume goes up. Cardiac output increases by 30 to 50 percent by the third trimester. The heart works harder because it is supporting two circulatory systems.
The numbers are well established in obstetrics literature. Resting heart rate rises by about 10 bpm in the first trimester and another 5 to 10 bpm by the third. A person whose pre-pregnancy resting heart rate was 65 bpm will typically sit at 75 to 85 bpm by the time they are full term. Some studies report increases up to 20 bpm in the third trimester.
Now consider what a wearable algorithm does with that. Most readiness scores use resting heart rate as a key input. A rising resting heart rate signals incomplete recovery, increased stress load, or early illness. The algorithm flags the trend as negative. The user sees a lower readiness score, a warning about elevated cardiovascular strain, and a recommendation to rest more.
The recommendation to rest more is actually fine. Pregnant people should rest. But the framing is wrong. The algorithm is telling the user something is wrong with their recovery when the reality is that their body is doing exactly what it should be doing. Over nine months, that is a lot of incorrect feedback.
Photo by Juan Encalada on Unsplash
What happens to HRV
Heart rate variability drops during pregnancy. This is well documented. A 2020 systematic review in PLOS One found that HRV decreases progressively across all three trimesters, with the largest drop in the third. The mechanism is straightforward: the autonomic nervous system shifts toward sympathetic dominance to support the increased cardiovascular demand. The parasympathetic withdrawal shows up as lower HRV.
The numbers vary by study, but the pattern is consistent. RMSSD (the most common time-domain HRV metric used by wearables) drops by roughly 20 to 30 percent from pre-pregnancy baselines by the third trimester. Some studies report drops of 40 percent or more in individual participants.
Every wearable on the market treats a sustained HRV drop as a red flag. Low HRV means poor recovery. Low HRV means accumulated stress. Low HRV means you should take it easy. The Oura readiness score weights HRV heavily. Whoop's recovery score is built around HRV. When a pregnant user's HRV drops into what the algorithm considers the red zone, the device tells them to rest. For nine months.
The problem is not that the advice is dangerous. The problem is that the algorithm is pathologizing a normal physiological state. The user gets constant negative feedback about their body for something that is not a problem. That has real psychological effects. We wrote about health tracking anxiety before, and this is a specific instance of the same pattern.
What happens to body temperature
Basal body temperature rises after ovulation and stays elevated if conception occurs. Progesterone, which remains high throughout pregnancy, has a thermogenic effect. The result is that body temperature runs about 0.3 to 0.5 degrees Celsius higher than pre-pregnancy baselines for the entire duration of the pregnancy.
Smart rings track skin temperature with enough precision to detect this shift. Oura's temperature trend feature is actually quite good at showing the sustained elevation. The problem is how the data is interpreted. Most wearables flag sustained temperature elevation as a potential sign of illness or inflammation. Some devices use temperature trends as an early illness detection signal.
A pregnant user sees their temperature trend line sitting above baseline for weeks and months. The device may flag it. The user may wonder if they are getting sick. They are not. They are pregnant. The temperature elevation is normal and expected.
I think this is actually one of the more fixable problems. If a wearable knows the user is pregnant, it can adjust the temperature baseline. But most wearables do not ask. And even if they did, the algorithm would need trimester-specific baselines, not a single pregnancy adjustment.
Photo by Janko Ferlic on Unsplash
What happens to sleep
Sleep architecture changes dramatically during pregnancy. The first trimester brings increased sleep need, more daytime sleepiness, and more frequent nighttime awakenings. The second trimester is often the best sleep of the pregnancy, but still worse than pre-pregnancy baselines. The third trimester is where things fall apart: frequent urination, physical discomfort, fetal movement, leg cramps, and acid reflux fragment sleep into short cycles.
Polysomnography studies show that pregnant people spend less time in deep sleep (N3) and REM sleep, with more time in light sleep (N1 and N2) and more frequent awakenings. Total sleep time may stay the same or even increase, but sleep efficiency drops.
A wearable algorithm trained on non-pregnant sleep patterns sees this and flags it as poor sleep quality. The sleep score drops. The algorithm recommends improving sleep hygiene, going to bed earlier, reducing caffeine. These are reasonable suggestions for a non-pregnant person with poor sleep. For a pregnant person in the third trimester, they are mostly useless. You cannot sleep through the need to urinate every two hours. You cannot deep sleep through fetal movement.
The algorithm does not know the difference between bad sleep because of bad habits and bad sleep because you are growing a human. It just sees the pattern and scores it negatively.
What happens to breathing rate
Respiratory rate increases during pregnancy. Progesterone stimulates the respiratory center, and the growing uterus physically limits diaphragmatic excursion. Minute ventilation increases by 30 to 50 percent by the third trimester. Tidal volume goes up. The subjective experience is feeling short of breath even when nothing is wrong.
Most wearables estimate respiratory rate from the PPG signal during sleep. The normal non-pregnant range is 12 to 16 breaths per minute. During pregnancy, 16 to 20 breaths per minute is common. Some studies report rates up to 24 in the third trimester.
A wearable that tracks respiratory rate as a health signal may flag this elevation. Elevated respiratory rate is associated with fever, infection, anxiety, and respiratory conditions. The algorithm does not know the user is pregnant. It just sees the number and scores it accordingly.
What the industry should do
I do not think the answer is wearables should not be used during pregnancy. The answer is that wearables should be validated on pregnant populations and should adjust their algorithms when the user reports pregnancy.
This is not hard. It requires three things.
First, the device needs to ask. A simple checkbox in the app: Are you pregnant? with a date or due date. No cloud upload required. The data stays on the phone.
Second, the algorithm needs trimester-specific baselines. First trimester baselines are different from third trimester baselines. A single pregnancy mode that shifts everything by a fixed offset is not enough. The algorithm needs to know where the user is in the pregnancy and adjust the expected ranges accordingly.
Third, the algorithm needs to stop flagging normal pregnancy changes as problems. The readiness score should not drop because HRV is lower. The temperature trend should not trigger illness alerts. The sleep score should account for the expected fragmentation.
I am not sure how many wearable companies will actually do this. It is a small market within a market. The ROI on validating devices for pregnancy is not obvious to most hardware companies. But the people who are pregnant and wearing these devices are getting actively misleading feedback about their health. That seems worth fixing.
What Pulsyn is doing
We are building pregnancy-aware baselines into the Pulsyn app. The data stays on the device. The user can optionally tell the app they are pregnant, and the algorithm adjusts the expected ranges for heart rate, HRV, temperature, sleep, and respiratory rate. No data leaves the phone. No cloud processing required.
The adjustment is not a single offset. It is a trimester-aware model that shifts the expected ranges progressively. First trimester: smaller adjustments to heart rate and HRV, larger adjustment to temperature. Second trimester: moderate adjustments across all metrics. Third trimester: full adjustments, especially for sleep and respiratory rate.
We are also building a pregnancy-specific readiness score that separates normal pregnancy changes from actual warning signs. The goal is to flag things that are genuinely concerning (sustained blood pressure spikes, extreme HRV drops beyond the expected range, temperature spikes above the pregnancy-adjusted baseline) while not flagging the normal physiological changes.
I would be lying if I said we have this fully validated. We have the model. We have the trimester-specific ranges from the literature. But we have not run a clinical validation study on pregnant users yet. That is on the roadmap for after the Kickstarter ships. If you are pregnant and willing to test, I would love to hear from you.
The bigger pattern
This is one instance of a much larger problem in wearable health. The algorithms are trained on narrow populations and then applied to everyone. Pregnant people. Older adults. People with chronic conditions. People on medications that affect heart rate. People with darker skin tones. The list goes on.
The industry treats the standard healthy adult as the default and everything else as an edge case. But edge cases are most of the population. Most people are not a 28-year-old non-pregnant healthy adult with no medications and no chronic conditions. Most people have something that shifts their baselines.
The fix is not complicated. It is just work. It means validating on more populations. It means building adjustable baselines. It means asking users about their context instead of assuming the default. And it means being honest about the limits of the algorithm instead of presenting every score as objective truth.
About the author
James Hoffmann is the founder of Pulsyn. He has been reverse-engineering BLE health devices for two years and is building a smart ring that does not assume you are a 28-year-old non-pregnant healthy adult.
References
- Altini, M., and Kinnunen, H. (2021). The utility of consumer-grade wearables for sleep assessment. Sleep Medicine Reviews, 59, 101455.
- Miller, D. J., et al. (2022). Validation of a wearable device for HRV measurement. Sensors, 22(3), 1012.
- Moyer, C., et al. (2020). Heart rate variability in pregnancy: A systematic review. PLOS One, 15(6), e0234532.
- Sanghavi, M., and Rutherford, J. D. (2014). Cardiovascular physiology of pregnancy. Circulation, 130(12), 1003-1008.
- Wilson, D. L., et al. (2019). Sleep in pregnancy: A systematic review of subjective and objective studies. Sleep Medicine Reviews, 45, 1-10.
- LoMauro, A., and Aliverti, A. (2015). Respiratory physiology of pregnancy. Breathe, 11(4), 297-301.
- Charkoudian, N., et al. (2017). Thermoregulation during pregnancy. Temperature, 4(3), 238-249.
