SDNN

SDNN is the standard deviation of the intervals between normal heartbeats, measured in milliseconds. It is the total amount of beat-to-beat timing variation in a recording, and it is the heart rate variability figure Apple Health stores for every Apple Watch reading.

If you have looked at HRV in the Health app, you have looked at SDNN. The number is easy to misread for two reasons. Its size depends on how long the recording lasted, and it mixes fast breathing-linked changes with slower swings from posture, temperature and the time of day. A 40 ms reading from your watch and a 40 ms result from a 24-hour hospital recording describe different things, and neither converts cleanly into the RMSSD figure that WHOOP and Oura report.

01The formula

Take every interval between normal beats in the recording, called NN intervals (normal-to-normal, which excludes ectopic beats and artifacts). Find their mean. Measure how far each interval sits from that mean, square those distances, average them and take the square root.

SDNN = √( Σ (NNᵢ − NN̄)² / (N − 1) )

NNᵢ  = each interval between normal beats, in ms
NN̄   = mean of all intervals in the recording
N    = number of intervals

The squared quantity, SDNN², is the variance of the interval series. Kleiger, Stein and Bigger point out in their 2005 review that total spectral power is equivalent to that same variance. SDNN is therefore a single number that sums every rhythm present in the recording, fast and slow, without telling you which rhythm contributed what. The HRV glossary entry covers the full family of time-domain and frequency-domain measures.

02What SDNN captures

Your heart rate is never still. Several control loops push it around at once, and each has its own tempo. Shaffer and Ginsberg's 2017 review in Frontiers in Public Health describes four frequency bands.

Frequency ranges and sources from Shaffer and Ginsberg (2017).

SDNN picks up all four. Shaffer and Ginsberg write that "both SNS and PNS activity contribute to SDNN and it is highly correlated with ULF, VLF and LF band power, and total power." SNS and PNS are the sympathetic and parasympathetic nervous systems. RMSSD, by contrast, compares each beat only with the one before it, so slow drifts cancel out and the fast vagal component dominates. The same review calls RMSSD "more influenced by the PNS than SDNN."

What SDNN contains therefore depends on the recording. In a short, still, seated reading, most of the slow bands have no time to appear, and Shaffer and Ginsberg note that "the primary source of the variation is parasympathetically-mediated RSA, especially with slow, paced breathing." Over a full day, the day-night difference in heart rate becomes the largest single term. Kleiger and colleagues estimate that roughly 30 to 40 percent of 24-hour SDNN comes from the gap between daytime and nighttime intervals alone.

03Why SDNN grows with recording length

A standard deviation measures spread around a mean. A longer recording gives slow rhythms time to move that mean, so the spread grows even if nothing about your autonomic state changed.

Here is an illustration with round numbers. Suppose breathing moves your intervals with a standard deviation of 40 ms whenever you are at rest. A 5-minute recording at your desk returns an SDNN near 40 ms. Now record for 24 hours. Your mean interval sits around 850 ms during the day and 1,050 ms at night, so half the beats cluster 100 ms below the grand mean and half 100 ms above it. The same 40 ms breathing swing now sits on top of that 100 ms day-night split, and the 24-hour SDNN comes out near √(40² + 100²), about 108 ms. RMSSD barely moves, because a slow shift in the mean hardly changes the difference between one beat and the next.

Shaffer and Ginsberg put it plainly. "It is inappropriate to compare metrics like SDNN when they are calculated from epochs of different length." The 2015 European position statement led by Sassi adds that "recording conditions (duration, body position, free or controlled breathing, etc.) can substantially affect short-term metrics."

Going shorter makes things worse. Munoz and colleagues compared ultra-short ECG segments with 4 to 5 minute recordings in 3,387 Dutch adults (PLOS ONE, 2015). A single 10-second SDNN correlated with the longer reference at r = 0.76, against r = 0.85 for RMSSD, and SDNN carried a larger bias. At 120 seconds SDNN reached r = 0.956. "For all recording lengths and agreement measures, RMSSD outperformed SDNN," the authors wrote, and they recommended at least 30 seconds, or several 10-second strips, for SDNN.

0424-hour clinical bands and short recordings

SDNN's clinical reputation traces largely to one study. Kleiger and colleagues analyzed 24-hour Holter recordings from 808 people who had survived a heart attack, taken 11 ± 3 days after the event, and followed them for a mean of 31 months (American Journal of Cardiology, 1987). Patients with SDNN below 50 ms had a 5.3 times higher risk of death than those above 100 ms, and SDNN predicted mortality after adjustment for ejection fraction and other clinical factors.

Those cut points became the bands you still see quoted online. Shaffer and Ginsberg summarize them as SDNN below 50 ms unhealthy, 50 to 100 ms compromised health and above 100 ms healthy, and they call 24-hour SDNN the "gold standard" for medical stratification of cardiac risk. Kleiger's later review notes that most laboratories require at least 18 hours of usable data before they will calculate a 24-hour SDNN.

The bands apply to 24-hour recordings only. A watch reading covers a short window with almost none of the day-night swing that makes up a large share of a 24-hour value, so a 35 ms wrist reading sits below the 50 ms line by construction and says nothing about your heart. Even among 24-hour recordings, Bigger and colleagues found that values linked to death in coronary patients turn up in only about 1 percent of healthy middle-aged people (Circulation, 1995). If a low reading worries you, take it to a clinician, who can order the right test.

05Typical short-term values

For short resting recordings, the best pooled reference is Nunan, Sandercock and Brodie's 2010 systematic review of 44 studies and 21,438 healthy adults. As tabulated by Shaffer and Ginsberg, short-term SDNN averaged 50 ms with a standard deviation of 16 ms and a range of 32 to 93 ms across studies. Nunan's group found values lower than the 1996 Task Force norms and large differences between laboratories, driven partly by how each lab cleaned its beat data.

Age moves SDNN steadily. In Voss and colleagues' 2015 analysis of 5-minute supine recordings from the German KORA S4 cohort, mean SDNN fell from 48.7 ms in women aged 25 to 34 to 27.8 ms in women aged 65 to 74, and from 50.0 ms to 29.6 ms in men across the same decades. The HRV by age article has the full tables and the RMSSD equivalents. Over 24 hours the decline is slower. Umetani and colleagues found 24-hour SDNN in 260 healthy people still at 60 percent of its second-decade value in the tenth decade, while RMSSD had already fallen to 47 percent by the sixth.

Treat these as a sense of scale. Apple Watch readings come from shorter optical windows taken at moments the watch chooses, and no published study gives population norms for them.

06Why Apple Health stores SDNN

Apple's HealthKit documentation says it directly. "While there are multiple ways of computing HRV, HealthKit uses SDNN heart rate variability, which uses the standard deviation of the inter-beat (RR) intervals between normal heartbeats." The same page notes that "the system automatically records samples on Apple Watch." Apple's support page adds that HRV "is sampled while you're still."

Apple has not published its reasoning for picking SDNN. What the record shows is that SDNN is the time-domain measure with the longest clinical history, from Kleiger's 1987 cohort onward. Apple Watch can also save the beat-to-beat timing behind each reading as a Heartbeat Series, which is how Hernando and colleagues extracted interval data for their 2018 validation. That series lets an app compute RMSSD or any other statistic from the same beats.

07How accurate Apple Watch SDNN is

The strongest evidence covers seated, still recordings. Hernando and colleagues compared Apple Watch interval data with a Polar H7 chest strap in 20 healthy adults during five minutes of relaxation and five minutes of a mental stress task (Sensors, 2018). Across 12,109 paired intervals the bias was 0.06 ms, with a concordance correlation of 0.989 at rest and 0.977 under stress. The watch missed about 10 percent of intervals, in roughly five gaps per recording averaging 6.5 seconds, but those gaps produced no significant difference in SDNN or the other time-domain measures. Both devices showed SDNN falling from rest to stress.

That result describes a controlled five-minute session. No published study has validated the background SDNN values the watch writes to Health on its own schedule against a simultaneous ECG. The Apple Watch HRV explainer covers overnight validation data, fit, tattoos and cold skin in detail.

08What moves your SDNN

Anything that changes either fast vagal activity or slow drift in your heart rate shows up in SDNN. Hard training, short sleep, alcohol, illness and psychological stress lower vagal activity and usually lower the number. Two more factors can change SDNN with no change in how recovered you are.

Heart rate itself. Because the relationship between heart rate and interval length is inverse, Sacha's 2013 analysis in Frontiers in Physiology shows that "the same changes of HR cause much higher fluctuations of R-R intervals for the slow average HR than for the fast one." A night with a lower resting heart rate tends to produce a higher SDNN even with the same autonomic input.

What happened during the window. Posture changes, a short walk to the kitchen or slow paced breathing all add low-frequency swing. A reading taken while you follow a breathing animation measures how large an oscillation you can produce on command. Daytime background readings also land at different points of your circadian rhythm, so two readings hours apart on the same day may differ for reasons unrelated to recovery.

09Why SDNN and RMSSD can't be converted

People often ask for a multiplier that turns Apple's SDNN into a WHOOP-style RMSSD. No such factor exists, because the ratio between the two moves with age, sex, heart rate, breathing and recording length.

Means from Voss et al. (2015), Tables 5 and 7. Ratios calculated from those means.

Inside one standardized protocol, the ratio still ranges from 1.14 to 1.65 between groups, and it varies again from person to person within each group. A free-living watch reading adds slow drift that inflates SDNN and leaves RMSSD mostly alone, so the gap changes with each reading's circumstances. Plews and colleagues built their approach to monitoring elite endurance athletes on vagal-derived HRV indices averaged over several days (Sports Medicine, 2013), and tracking the vagal component is the job RMSSD does best.

The practical consequence is simple. Compare SDNN with SDNN from the same device and the same kind of recording, and keep one metric for the life of any baseline.

10How Titan uses SDNN

Titan reads the SDNN values Apple Watch writes to Apple Health, and SDNN is the default metric for every HRV figure in the app. For Recovery, Titan builds one Overnight HRV value per night. It takes the HRV readings from one source recorded inside the main sleep session that ended that morning, or from 22:00 to 10:00 when no sleep session was saved. It then averages them on a log scale, which gives the geometric mean. Readings outside 3 to 300 ms are discarded as artifacts. Using only overnight readings removes most of the posture, activity and time-of-day noise described above, since every night is recorded in the same state.

Titan compares that value with your own baseline. The baseline is the median of your measured nights before tonight within your Recovery window, 60 nights by default. Distance from the median is measured in units of your own typical night-to-night spread, the median absolute deviation, on the same log scale. On the log scale a drop from 30 to 25 ms counts for more than a drop from 80 to 75 ms, so a change is judged relative to your usual level. Your sleeping heart rate, the lowest 30-minute average during the night, is scored the same way, and HRV carries twice its weight in the score. A baseline needs at least 7 measured nights covering at least 25% of the window, so 15 nights on the default window. If HRV is missing entirely, HRV not showing lists the usual causes.

The Use RMSSD for HRV setting

Settings > Recovery & Sleep has one HRV control, Use RMSSD for HRV, and it is off by default. Turn it on and Titan stops reading Apple's SDNN. It computes RMSSD itself from the Heartbeat Series in Apple Health, one value per series. Titan breaks the calculation at every gap in the series so that two beats separated by missing data are never treated as neighbors, and it skips any series with fewer than three beats. The in-app description warns that it loads more slowly, because Titan has to read the raw beat data.

The setting applies to Recovery, Stress and Battery on your iPhone. When you change it, Titan recalculates your Recovery history with the new method. It stores each night's inputs separately for SDNN and RMSSD, so a baseline never mixes the two scales. Titan needs read access to Heartbeat Series, and without it you get no HRV at all. The Titan Apple Watch app shows the Recovery score your iPhone calculated.

Pick RMSSD if you want numbers that line up with published training research or with a WHOOP or Oura history. Keep SDNN if you want Titan to match what the Health app shows. Recovery works on your own baseline either way, so the choice changes the scale of the number, and the scoring logic stays the same.

11Reading your own number

Your SDNN means most when you compare it with your own recent nights, measured the same way. A single low night can come from a late meal, a drink or a loose band. Several nights below your usual range, with sleeping heart rate rising alongside, deserve more attention than any one reading. Population tables and clinical bands tell you the scale of the number and little else. The HRV and training readiness guide covers how to act on a sustained drop.

12References

• Apple (n.d.). heartRateVariabilitySDNN. HealthKit developer documentation. https://developer.apple.com/documentation/healthkit/hkquantitytypeidentifier/heartratevariabilitysdnn
• Apple (n.d.). Monitor your heart rate with Apple Watch. Apple Support. https://support.apple.com/en-us/HT204666
• Bigger JT et al. (1995). RR variability in healthy, middle-aged persons compared with patients with chronic coronary heart disease or recent acute myocardial infarction. Circulation 91(7):1936-1943. https://doi.org/10.1161/01.CIR.91.7.1936
• 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
• Kleiger RE et al. (1987). Decreased heart rate variability and its association with increased mortality after acute myocardial infarction. American Journal of Cardiology 59(4):256-262. https://doi.org/10.1016/0002-9149(87)90795-8
• Kleiger RE et al. (2005). Heart rate variability, measurement and clinical utility. Annals of Noninvasive Electrocardiology 10(1):88-101. https://doi.org/10.1111/j.1542-474X.2005.10101.x
• 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
• Plews DJ et al. (2013). Training adaptation and heart rate variability in elite endurance athletes, opening the door to effective monitoring. Sports Medicine 43(9):773-781. https://doi.org/10.1007/s40279-013-0071-8
• Sacha J (2013). Why should one normalize heart rate variability with respect to average heart rate. Frontiers in Physiology 4:306. https://doi.org/10.3389/fphys.2013.00306
• Sassi R et al. (2015). Advances in heart rate variability signal analysis, joint position statement by the e-Cardiology ESC Working Group and the European Heart Rhythm Association co-endorsed by the Asia Pacific Heart Rhythm Society. EP Europace 17(9):1341-1353. https://doi.org/10.1093/europace/euv015
• Shaffer F et al. (2017). An overview of heart rate variability metrics and norms. Frontiers in Public Health 5:258. https://doi.org/10.3389/fpubh.2017.00258
• Umetani K et al. (1998). Twenty-four hour time domain heart rate variability and heart rate, relations to age and gender over nine decades. Journal of the American College of Cardiology 31(3):593-601. https://doi.org/10.1016/S0735-1097(97)00554-8
• Voss A et al. (2015). Short-term heart rate variability, influence of gender and age in healthy subjects. PLOS ONE 10(3):e0118308. https://doi.org/10.1371/journal.pone.0118308