RMSSD

RMSSD, the root mean square of successive differences, is a heart rate variability statistic that measures how much the gap between one heartbeat and the next changes from beat to beat, reported in milliseconds. Those fast changes come mostly from the vagus nerve, so RMSSD is the standard measure of parasympathetic activity in resting and overnight HRV.

Oura, WHOOP, Garmin and HRV4Training all report HRV as RMSSD. Apple Health stores HRV as SDNN, a different formula on a different scale, which is why an Apple Watch reading and a ring reading from the same night rarely match. Titan uses Apple's SDNN by default and computes RMSSD from your beat-to-beat data when you turn on Use RMSSD for HRV. The HRV entry covers heart rate variability in general, and the SDNN entry covers the other formula.

01The formula, one beat at a time

Start with a run of normal beat-to-beat intervals, RR₁ through RRₙ, in milliseconds.

RMSSD = √( Σ (RRᵢ₊₁ − RRᵢ)² ÷ (n − 1) )

The calculation has four steps. Subtract each interval from the one after it. Square each difference, which makes every value positive and gives big jumps extra weight. Average the squares. Take the square root to get back to milliseconds.

Here are six intervals from a resting heart beating at about 60 beats per minute.

The squares sum to 7,000 ms² across five differences, so the mean square is 1,400 ms² and RMSSD is its square root, 37.4 ms. SDNN, the standard deviation of the same six intervals, is 23.2 ms.

Now take a heart that slows steadily, with intervals of 1,000, 1,010, 1,020, 1,030, 1,040 and 1,050 ms. Every difference is +10 ms, so RMSSD is 10.0 ms. SDNN climbs to 18.7 ms, because the intervals keep spreading apart. RMSSD compares each beat only with its neighbor, so slow drifts in heart rate from posture, temperature or a body settling into sleep barely register. SDNN counts the whole spread. Most of what follows comes from that one property.

Squaring has a cost. Suppose the sensor mistakes one beat and inserts a false 480 ms interval after the 980 ms one. The new difference is 500 ms, its square is 250,000 ms², and RMSSD for the series jumps from 37 ms to 207 ms. Garmin's technical note on its beat-to-beat data says it directly: "RMSSD is very susceptible to outliers due to the squaring operation." Devices screen beats before computing it. Oura's validation paper, for example, used an interval only when 8 consecutive intervals had been labeled normal (Kinnunen and Koskimäki, 2018).

02Why RMSSD tracks the vagus nerve

At rest, the vagus nerve slows the heart, and its signal can change the timing of the very next beat. Vagal output also rises and falls with breathing, which lets the heart speed up slightly as you inhale and slow as you exhale. Successive differences capture those fast swings. HRV4Training's guide describes the vagus nerve modulating pulse "very quickly, on a beat to beat basis. Thus, rMSSD captures parasympathetic activity."

The strongest evidence comes from blocking the nerve. Penttilä and colleagues (2001) gave healthy volunteers glycopyrrolate, a drug that blocks the vagal signal at the heart. RMSSD fell by 97.0 percent and high-frequency spectral power by 99.8 percent. Changing the breathing pattern shifted the spectral measures significantly and left RMSSD unaffected, so the authors judged RMSSD better suited to ambulatory recordings where nobody controls breathing. Shaffer and Ginsberg's 2017 review calls RMSSD "the primary time-domain measure used to estimate the vagally mediated changes reflected in HRV."

The link has a known exception in very fit people. Buchheit's 2014 review notes that in some athletes "a heightened vagal tone may give rise to sustained parasympathetic control of the sinus node, which may eliminate respiratory heart modulation and reduce vagal-related HRV indices." Plews and colleagues' 2013 review in Sports Medicine describes this saturation as a reduction in HRV despite a falling resting heart rate in elite endurance athletes. For most people a lower RMSSD means less vagal activity. For some highly trained athletes with very low resting heart rates, a falling RMSSD can accompany good adaptation.

03Why RMSSD holds up in short recordings

HRV standards were built around 5-minute recordings. Shaffer and Ginsberg write that "while the conventional minimum recording is 5 min, researchers have proposed ultra-short-term periods of 10 s, 30 s, and 60 s." RMSSD survives that cut because it ignores slow waves, and slow waves are what a recording needs time to capture.

Munoz and colleagues (2015) tested this in 3,387 adults. RMSSD from 10 seconds agreed with the 4 to 5 minute reference at r = 0.85 to 0.86, and from 30 seconds at r = 0.93, with small bias. SDNN from 10 seconds reached only r = 0.76, with a much larger bias. In 23 college athletes, Esco and Flatt (2014) found that agreement for ln RMSSD weakened as the window shrank, with intraclass correlation falling from 0.98 at 60 seconds to 0.81 at 10 seconds. They judged 60 seconds acceptable, which is why phone and chest-strap apps can ask for a one-minute reading each morning.

A short window does nothing about day-to-day noise. Buchheit reports that normal variation in training produces a day-to-day coefficient of variation of 10 to 20 percent in ln RMSSD, and he advises endurance athletes to measure at least 3 to 4 times a week and average the results. A 2013 study by Plews and colleagues in the International Journal of Sports Physiology and Performance shows why. In 10 runners over a 9-week training block, the change in ln RMSSD on a single day had a trivial correlation with the change in maximal aerobic speed, r = −0.06. The change in the weekly average correlated at r = 0.72.

04ln RMSSD and why athletes log-transform it

ln RMSSD is the natural logarithm of RMSSD. The 37.4 ms reading from the example becomes 3.62. Sports scientists log-transform RMSSD because the raw values are skewed. Among Dutch men aged 40 to 44 in the Lifelines cohort, mean RMSSD was 35.2 ms and the median 29.0 ms. A long upper tail, reaching 105.5 ms at the 98th percentile, pulled the mean above the median (Tegegne et al., 2020). Marco Altini, who built HRV4Training, puts it plainly: "Since rMSSD is not normally distributed, normally in the scientific literature we report the logarithm of rMSSD (ln rMSSD)."

The log also turns percentage changes into equal steps. A fall from 60 to 54 ms and a fall from 30 to 27 ms are both 10 percent drops, and each is a drop of 0.105 ln units, although one is 6 ms and the other 3 ms. A threshold set in ln units means the same thing for a person with a high baseline and a person with a low one, and one unusually high night pulls a weekly average around less.

Plews and colleagues (2012) followed two elite triathletes for 77 days using the 7-day rolling average of ln RMSSD. One athlete performed poorly in a key race and was diagnosed as non-functionally overreached. That athlete's rolling average declined toward the race, and its coefficient of variation shrank too. The control athlete's values stayed flat. It is a single case comparison, so treat it as an illustration of the method. The overreaching entry covers the condition.

HRV4Training shows a transformed ln RMSSD as its headline score, which Altini says sits "in the 6-10 range, as opposed to the 10-250ms range" of raw RMSSD. Titan's Recovery also runs on the log scale, as described below.

05Devices that report RMSSD

Sources for this table are Oura's HRV help article and Kinnunen and Koskimäki (2018), WHOOP's developer documentation, Garmin's HRV Status page, HRV4Training's quick-start guide, Apple's HealthKit documentation and Titan's HRV not showing article.

Neither Oura nor WHOOP writes HRV to Apple Health, as Titan's Oura and WHOOP help articles explain. Oura's HRV reaches Titan only through the direct Oura connection.

Chest straps are the practical reference. Gilgen-Ammann and colleagues (2019) found that a Polar H10 detected R-R intervals with 99.6 percent signal quality across rest and exercise, against 94.6 percent for a clinical Holter monitor, and held 99.4 percent at high intensity. Plews and colleagues (2017) compared 5-minute resting RMSSD from a phone camera, a Polar H7 strap and ECG in 29 people. Both consumer methods correlated with ECG at about r = 1.00.

06How accurate wearable RMSSD is

Wrist and finger sensors detect the pulse optically, and they miss beats. When Hernando and colleagues (2018) validated Apple Watch intervals against a Polar H7 in 20 people, they found 206 gaps, about 10 percent of all intervals, averaging 5 gaps per recording. With differences across gaps excluded, the time-domain measures, RMSSD among them, were not significantly affected.

Overnight numbers vary more. Miller and colleagues (2022) put 53 young adults in a sleep lab with ECG and six wearables for one night.

Source: Miller et al. (2022), Table 6.

Read this table with its caveats. The Apple Watch values came through a third-party app whose sampling period the authors could not specify, and WHOOP staff supplied the WHOOP data. Every device except WHOOP overestimated low RMSSD and underestimated high RMSSD. Other studies report closer agreement for specific devices. Liang and colleagues (2024) found that Oura's 5-minute RMSSD, in windows that met the authors' signal-quality threshold, correlated with ECG at r = 0.979 in 92 adults under 45 and 0.937 in 22 older adults, and Bellenger and colleagues (2021) put WHOOP's ln RMSSD bias at 1.66 percent.

In practice, a wearable tracks your own trend far better than it matches an ECG in absolute milliseconds. Compare your numbers with your own history on one device, and treat a device switch as a new baseline.

07Population reference values

The largest RMSSD reference comes from 84,772 healthy participants in the Dutch Lifelines cohort, measured with a 10-second resting ECG (Tegegne et al., 2020).

RMSSD from a 10-second resting ECG. Source: Tegegne et al. (2020), Table 1.

Medians fall steadily until about 60 and then level off. The healthy range is enormous at every age, so a single reading says little about your health. The same paper also reports a heart-rate-corrected version, cRMSSD, because RMSSD partly tracks the average length of the beat interval.

Nunan and colleagues' 2010 review of 44 studies of short-term resting recordings, as tabulated by Shaffer and Ginsberg, gives a mean RMSSD of 42 ms with a standard deviation of 15 ms and a range of 19 to 75 ms. Sleep is a different state from a daytime resting ECG. Oura reports that its members' sleeping HRV averages 41 ms overall, with a mean of 39.1 ms and median of 36 ms for women and a mean of 42.8 ms and median of 35 ms for men. Normal HRV by age and sex has the full tables and the SDNN equivalents.

08What moves your RMSSD

Age moves the population median most, as the table shows. Within one person, training load moves it from week to week, and Plews's triathlete case shows a sustained decline preceding non-functional overreaching. Alcohol lowers it the same night. In 4,098 Finnish employees wearing beat-to-beat monitors, Pietilä and colleagues (2018) found that alcohol reduced parasympathetic regulation during the first 3 hours of sleep in a dose-dependent way, and the effect appeared even at low intake of 0.25 g per kg of body weight or less.

Measurement conditions move it too. Garmin's HRV Status page notes that "differences in the timing and duration of the measurement affect the results," which is why Buchheit recommends measuring lying down right after waking and why overnight devices average across the night. The HRV and recovery readiness guide covers measurement protocols and how to turn trends into training decisions.

09How Titan computes RMSSD

Apple Health stores HRV as SDNN. Apple's HealthKit documentation says "HealthKit uses SDNN heart rate variability," and Titan uses that value by default. Turn on Use RMSSD for HRV in Settings under Recovery & Sleep, and Titan reads the Heartbeat Series samples in Apple Health instead. Each sample is a list of beat times, and HealthKit flags any beat "immediately preceded by a gap in the data, indicating that one or more heartbeats may be missing."

For each Heartbeat Series, Titan sorts the beats, measures each interval and squares the change between consecutive intervals. When HealthKit flags a gap, Titan starts a new chain, so no difference is ever taken across missing beats. That is the same rule Hernando's team used for Apple Watch data. A series with fewer than three beats yields no value. Each series produces one RMSSD reading in milliseconds.

Recovery then builds your overnight HRV from those readings.

1. Titan keeps readings that start inside your main sleep session, or between 22:00 and 10:00 when no sleep session was saved.
2. It drops readings below 3 ms or above 300 ms and exact duplicates, and keeps a single source, preferring the device that recorded your sleep.
3. It takes the natural log of each reading and averages them. The Overnight HRV value on the Recovery screen is that average converted back to milliseconds, the geometric mean. Readings of 25, 40, 60 and 35 ms give an arithmetic mean of 40 ms and an Overnight HRV of 38.1 ms, because one high reading counts for less.
4. Your baseline is the median and median absolute deviation of ln Overnight HRV across the measured nights in your Recovery window, 60 nights by default. A baseline needs at least 7 measured nights covering at least 25% of the window, so 15 nights on the default window.
5. Tonight's distance from that median, in scaled median absolute deviations, becomes the HRV term of the score. Sleeping heart rate, the lowest 30-minute average of the night, supplies the second term.

The score is 100 times a logistic function of 0.25 + 1.5 × the HRV term − 0.75 × the heart rate term, with each term capped at 3 deviations. With both inputs at baseline, Recovery scores 56. With HRV one scaled deviation above baseline and heart rate at baseline, it scores 85, and one deviation below scores 22. The baseline entry covers personal baselines in more depth.

Using RMSSD has three practical effects. Scores load more slowly, because Titan reads every beat. Titan needs read access to Heartbeat Series in Apple Health, and it never falls back to SDNN, so HRV stays blank without that permission or when your source writes HRV without beat-to-beat data. Your history also changes scale, and Titan rebuilds Recovery history with the new method when you switch. The same setting applies to the HRV readings behind Stress and Battery. Recovery settings and HRV not showing cover the setup and troubleshooting.

10Reading your own RMSSD

Compare today with your own recent weeks on one device, preferably as a weekly average on the log scale. Single days swing widely for ordinary reasons, and Buchheit puts the day-to-day coefficient of variation of ln RMSSD at 10 to 20 percent. A decline that holds for a week while your resting heart rate rises is worth acting on. A stable reading that sits low on a population chart needs no action.

11References

• Altini M (2023). Heart rate variability (HRV) numbers, what do they mean? https://marcoaltini.substack.com/p/heart-rate-variability-hrv-numbers
• Apple. heartRateVariabilitySDNN and HKHeartbeatSeriesQuery. Apple Developer Documentation. https://developer.apple.com/documentation/healthkit/hkquantitytypeidentifier/heartratevariabilitysdnn
• Bellenger CR et al. (2021). Wrist-based photoplethysmography assessment of heart rate and heart rate variability, validation of WHOOP. Sensors 21(10):3571. https://doi.org/10.3390/s21103571
• Buchheit M (2014). Monitoring training status with HR measures, do all roads lead to Rome? Frontiers in Physiology 5:73. https://doi.org/10.3389/fphys.2014.00073
• Esco MR, Flatt AA (2014). Ultra-short-term heart rate variability indexes at rest and post-exercise in athletes, evaluating the agreement with accepted recommendations. Journal of Sports Science and Medicine 13(3):535-541. https://pmc.ncbi.nlm.nih.gov/articles/PMC4126289/
• Garmin (2023). Garmin Enhanced BBI, an example night. https://www8.garmin.com/garminhealth/news/Garmin-Enhanced-BBI_Final.pdf
• Garmin (2024). HRV Status. Garmin Technology, Health Science. https://www.garmin.com/en-US/garmin-technology/health-science/hrv-status/
• Gilgen-Ammann R et al. (2019). RR interval signal quality of a heart rate monitor and an ECG Holter at rest and during exercise. European Journal of Applied Physiology 119:1525-1532. https://doi.org/10.1007/s00421-019-04142-5
• Hernando D et al. (2018). Validation of the Apple Watch for heart rate variability measurements during relax and mental stress in healthy subjects. Sensors 18(8):2619. https://doi.org/10.3390/s18082619
• HRV4Training. QuickStart guide. https://www.hrv4training.com/quickstart-guide.html
• Hughes L (2025, updated 2026). What is the average HRV? Oura blog. https://ouraring.com/blog/average-hrv/
• Kinnunen H, Koskimäki H (2018). The HRV of the ring, comparison of nocturnal HR and HRV between the Oura ring and ECG. Oura white paper. https://ouraring.com/blog/wp-content/uploads/2018/10/The-HRV-of-the-Ring-Comparison-of-OURA-Ring-to-ECG.pdf
• Liang T et al. (2024). Deriving accurate nocturnal heart rate, rMSSD and frequency HRV from the Oura ring. Sensors 24(23):7475. https://doi.org/10.3390/s24237475
• Miller DJ et al. (2022). A validation of six wearable devices for estimating sleep, heart rate and heart rate variability in healthy adults. Sensors 22(16):6317. https://doi.org/10.3390/s22166317
• Munoz ML et al. (2015). Validity of (ultra-)short recordings for heart rate variability measurements. PLOS ONE 10(9):e0138921. https://doi.org/10.1371/journal.pone.0138921
• Nunan D et al. (2010). A quantitative systematic review of normal values for short-term heart rate variability in healthy adults. Pacing and Clinical Electrophysiology 33(11):1407-1417. https://doi.org/10.1111/j.1540-8159.2010.02841.x
• Oura. Heart rate variability. Oura Help. https://support.ouraring.com/hc/en-us/articles/360025441974-Heart-Rate-Variability
• Penttilä J et al. (2001). Time domain, geometrical and frequency domain analysis of cardiac vagal outflow, effects of various respiratory patterns. Clinical Physiology 21(3):365-376. https://doi.org/10.1046/j.1365-2281.2001.00337.x
• Pietilä J et al. (2018). Acute effect of alcohol intake on cardiovascular autonomic regulation during the first hours of sleep in a large real-world sample of Finnish employees, observational study. JMIR Mental Health 5(1):e23. https://doi.org/10.2196/mental.9519
• Plews DJ et al. (2012). Heart rate variability in elite triathletes, is variation in variability the key to effective training? A case comparison. European Journal of Applied Physiology 112:3729-3741. https://doi.org/10.1007/s00421-012-2354-4
• Plews DJ et al. (2013). Evaluating training adaptation with heart-rate measures, a methodological comparison. International Journal of Sports Physiology and Performance 8(6):688-691. https://doi.org/10.1123/ijspp.8.6.688
• Plews DJ et al. (2013). Training adaptation and heart rate variability in elite endurance athletes, opening the door to effective monitoring. Sports Medicine 43:773-781. https://doi.org/10.1007/s40279-013-0071-8
• Plews DJ et al. (2017). Comparison of heart-rate-variability recording with smartphone photoplethysmography, Polar H7 chest strap, and electrocardiography. International Journal of Sports Physiology and Performance 12(10):1324-1328. https://doi.org/10.1123/ijspp.2016-0668
• Shaffer F, Ginsberg JP (2017). An overview of heart rate variability metrics and norms. Frontiers in Public Health 5:258. https://doi.org/10.3389/fpubh.2017.00258
• Tegegne BS et al. (2020). Reference values of heart rate variability from 10-second resting electrocardiograms, the Lifelines Cohort Study. European Journal of Preventive Cardiology 27(19):2191-2194. https://doi.org/10.1177/2047487319872567
• WHOOP. Recovery. WHOOP for Developers. https://developer.whoop.com/docs/developing/user-data/recovery