Metrics, Formulas &
Normal Values

The percentage of hemoglobin molecules carrying oxygen, measured by reflectance pulse oximetry at 1 Hz from the finger. The O2Ring uses two wavelengths (red ~660 nm and infrared ~940 nm) and Beer–Lambert absorption to estimate the oxygenated-to-total hemoglobin ratio. Device accuracy: ±2% (manufacturer spec), reliable 70–100% range.

Events per hour where SpO₂ drops ≥ N% from a 5-minute rolling baseline and recovers. A rolling baseline (rather than a fixed threshold) compensates for individual differences in baseline SpO₂ and slow overnight drifts. The suffix (1, 2, 3, or 4) is the minimum drop depth in percentage points.

The percentage of total recording time where SpO₂ was below a given absolute threshold. Reported at 10 levels: T95, T94, T93, T92, T91, T90, T89, T88, T85, T80. Precise cumulative seconds are also shown as CT<90, CT<89, CT<88, CT<85.

The integral of SpO₂ deficit below 90% over the recording, normalized per hour. Captures both depth and duration — a brief dip to 85% contributes 5× more than a dip to 89%.

Severity-weighted sum of time below SpO₂ thresholds, integrating T95–T80 with quadratic depth penalties. Higher than AUC-90 for predominantly mild hypoxemia because it captures burden across the full desaturation range.

SD / mean × 100 for the whole-night SpO₂ signal. Captures intra-night fluctuation amplitude independent of absolute level. Elevated CoV with near-normal mean may indicate periodic breathing not yet severe enough to flag via ODI.

Pearson correlation between consecutive SpO₂ values (SpO₂[t] vs SpO₂[t+1]). Values near 1.0 indicate a slow-moving, strongly correlated signal (normal). Values near 0.5 suggest rapid random fluctuation. Used internally as a signal quality screen and as a periodicity proxy for FFT validation.

Kurtosis measures the tail weight of the SpO₂ distribution (high = frequent extreme dips). Skewness measures asymmetry (negative skew = tail toward low values). Together they characterise whether desaturation is dominated by rare severe events or frequent mild dips.

Two metrics computed only on samples below 94%: (1) Cond. Mean <94 — the average SpO₂ during hypoxemic samples, indicating typical dip depth during exposure; (2) Cond. % <94 — fraction of total recording time spent below 94%, identical to T94 but emphasised as a CHA-94 companion.

Desaturation nadir events are bucketed into five depth tiers: above 91%, 90–91%, 88–89%, 85–87%, and below 85%. Provides a richer picture than simple ODI counts — an ODI-4 of 20/hr is clinically very different if all events land in the above-91 bin vs. the below-85 bin.

p75 − p25 of the whole-night SpO₂ distribution. Robust to extreme outliers unlike SD. A narrow IQR (1–2%) with a low median indicates consistently poor saturation; a wide IQR suggests cyclic oscillation.

Mean and standard deviation of the time between consecutive desaturation nadir timestamps (seconds). Regular IEI (low SD) with a period matching the clinical CS range (40–90 s) supports a periodic breathing pattern. Highly variable IEI favours obstructive fragmentation.

SD / mean × 100 of the inter-nadir interval series. Identical source data as IEI but expressed as a percentage so short and long recordings are comparable. Low CV = metronomic respiratory cycling (strong CS/PB signal). High CV = irregular event spacing.

Ratio of absolute mean dip slope to absolute mean recovery slope. Values >1.5 indicate the SpO₂ falls faster than it recovers — characteristic of abrupt obstructive apneas. Values <0.7 indicate slow descent with rapid recovery — more consistent with central/hypoventilation events.

Timestamps (hours from recording start) of the first and last detected desaturation nadir. Events concentrated in the first third of the recording may reflect sleep-onset hypoxemia or positional effects. Events concentrated in the final third often reflect REM-related events or CPAP pressure inadequacy.

Count of samples where SpO₂ is fixed at exactly 100% for ≥30 consecutive seconds. Consumer-grade devices clip the signal at 100; long runs of exactly 100% indicate sensor contact issues, firmware clamping, or device removal rather than true perfect saturation. Flagged as a signal quality warning.

Ratio of SpO₂ recovery magnitude to mean HR during recovery windows. Higher values mean the cardiovascular system achieves more SpO₂ resaturation per unit cardiac work. Reduced efficiency may reflect impaired cardiac output or poor pulmonary reserve.

Mean HR delta (bpm) measured between the nadir moment and 60 seconds later, averaged across all detected desaturation events. A positive response (>3 bpm) indicates preserved arousal coupling. A flat or negative response suggests blunted autonomic arousal — associated with worse cardiovascular outcomes.

Count and per-hour rate of rapid SpO₂ drops where the signal falls ≥5 percentage points within any 30-second window, with a 30-second refractory cooldown to prevent re-counting sustained drops. Captures acute, fast-onset desaturations that may not meet the 10-second ODI minimum-duration criterion.

Epidemiological AHI approximation derived from ODI-4 using the regression coefficient from Azarbarzin et al. 2021 (AHI ≈ ODI-4 × 1.1). This is a population-level proxy, not an individual measurement. A confirmed AHI requires attended or validated home polysomnography.

The lowest mean SpO₂ found in any 10-minute sliding window across the recording. More clinically relevant than the absolute minimum, which may reflect a single-second artefact. A low worst-10-min value with a near-normal overall mean indicates severe but brief nocturnal dipping.

The highest T95 (% time below 95%) found in any 30-minute rolling window. Identifies the worst sustained hypoxemic episode within the night. Useful when OSA is positional or REM-concentrated and the whole-night T95 is diluted by asymptomatic periods.

Number of consecutive 5-minute windows where SpO₂ SD <1% and mean >94%. Represents the fraction of the night with truly stable oxygenation. Complements LCSP — LCSP measures the longest single run while stable windows counts the total number of high-quality periods.

An internally-calibrated approximation of the published Sleep Breathing Impairment Index (Hui et al., Respirology 2024), which was validated in 4,485 SHHS participants — there its top quintile carried roughly double the CVD-mortality risk of the lowest. This card is an internal depth²×duration estimate, not the validated algorithm — directional, not diagnostic.

A custom adult oximetry severity grade based on ODI-4 rate and CT<90. Not the published McGill Oximetry Score, which is a pediatric tool (Brouillette/Nixon) scored by visual clusters of desaturations plus nadir depth (≥3 clusters with ≥3 drops <90 / <85 / <80%). This proxy is inspired by it but uses different inputs and is not validated. Grade ≥ 2 = abnormal nocturnal oximetry.

SD of SpO₂ within non-overlapping 30-second motion-free windows, averaged across the night. Because respiration modulates both SpO₂ and HR, this captures the respiratory coupling amplitude as a low-resolution RSA surrogate. Higher values suggest stronger respiratory–cardiac coupling.

p75 − p25 of all motion-free HR values. A robust spread metric less sensitive to artifact spikes than SDNN. Wide HR IQR reflects strong HR variability across the night; narrow IQR with elevated mean may indicate sustained sympathetic activation.

Difference between mean HR inside detected periodic breathing oscillation windows and mean HR outside those windows. A positive contrast (HR higher during PB) reflects the sympathetic activation associated with cyclic apneas. A near-zero contrast suggests the autonomic arousal pathway is blunted.

Dominant frequency of the motion-free HR signal in the 0.15–0.40 Hz band, converted to breaths per minute. At 1 Hz, this is a coarse estimate sensitive to noise; treat as directional only. Elevated proxy rates may reflect tachypnea or signal artifact.

Difference between mean HR acceleration rate (rise/s) and mean HR deceleration rate (fall/s) over the whole night, computed on 10-sample rolling windows. Positive asymmetry = HR rises faster than it falls (sympathetic dominance). Negative = HR falls faster (parasympathetic dominant, normal sleep).

Difference between mean HR in the last quartile (Q4) and first quartile (Q1) of the recording. Negative value (Q4 < Q1) = HR declining across the night, which is normal. Positive or flat trend may indicate progressive autonomic stress or REM-rich later sleep.

SD / mean × 100 of motion-free HR. Normalises SDNN proxy by mean HR, making it more comparable across individuals with different resting rates. A high CV with high HR floor may indicate sustained nocturnal sympathetic tone.

Least-squares cosine fit (y = A·cos(2πt/T) + C) to the nightly HR vector. Amplitude = A (bpm), magnitude of the circadian HR oscillation. Nadir Hour = time of minimum HR as fraction of recording (hours from start). Larger amplitude indicates stronger cardiac circadian drive; nadir before 4 AM is normal.

Hour from recording start at which the lowest 5-minute smoothed HR is observed. Normally occurs in the first half of sleep during deep NREM. Late nadir (after 60% of recording) may indicate REM-heavy sleep architecture or delayed parasympathetic activation.

Percentage by which HR descends from the night mean to its floor: (mean − floor) / mean × 100. Note: this is an intra-night dip, not the standard clinical nocturnal dip (which requires daytime HR). It measures autonomic flexibility within the sleep period only.

Weighted composite of three parasympathetic markers: pNN3 (short-term vagal drive), HR floor (low floor = strong vagal tone), and longest clean run (data quality / sustained calm period). Formula intentionally non-linear to reflect the multiplicative nature of autonomic recovery.

Weighted sum of three arousal markers, each normalised to a 0–1 scale: HR spike rate (weight 0.4), post-dip HR response (weight 0.4), and AAI load (weight 0.2). Captures the overall sympathetic burden per night. High SSI with low ODI-4 points to UARS or non-apneic arousal; high SSI with high ODI-4 indicates OSA-driven arousal.

Number of runs where HR decreases by at least 3 bpm from its local peak and sustains that decrease for ≥30 consecutive seconds. Reflects prolonged parasympathetic activation episodes. Elevated count may suggest strong vagal rebound after arousal clusters (post-arousal bradycardia).

Percentage of 30-second windows in which SpO₂ and HR change in the same direction (both rise or both fall). Normal physiology expects an inverse relationship (SpO₂ falls → HR rises). High decoupling % suggests impaired autonomic reflex arc or central event pattern.

Transient HR accelerations against a 3-minute rolling baseline. Each spike is a candidate autonomic arousal. Motion-contaminated spikes (motion>0 in a 12-sample lookahead window) are automatically excluded and counted separately as positional arousals.

Combined rate of HR spikes and ODI-4 events per hour. High AAI with low ODI-4 → non-apneic arousal causes (UARS, PLMS, environmental). High AAI tracking closely with ODI-4 → arousal driven by obstructive events.

WASO expressed as a percentage of total recording time after the detected sleep-onset window. Complements absolute WASO minutes — 40 min WASO is very different in a 5-hour vs. a 9-hour recording. Derived from the motSleep sub-object.

Number of 5-minute windows after sleep onset that contain any motion sample (motion > 0). Provides a discrete count of interrupted sleep periods rather than a continuous time measure. Each window represents a distinct awake episode.

Number of abrupt motion bursts (≥3 seconds of motion signal after a ≥5-minute quiet period) detected across the night. Proxy for major position changes. Very high counts may indicate restless legs syndrome or periodic limb movements. Very low counts (<2) in a long recording suggest complete immobility.

Weighted composite of three sleep-continuity disruptors: WASO duration (weight 0.4), motion burst count (weight 0.15), and sleep-onset latency (weight 0.25). Higher SPI = greater residual sleep pressure accumulation, suggesting the sleep period did not provide adequate restoration.

Coefficient of variation of the inter-nadir interval series (see IEI above), expressed as a percentage. Identical source data as Recovery CV — presented here in the sleep section as a sleep-breathing regularity summary. Low biCV with IEI in the CS range is the strongest pattern indicator for Cheyne–Stokes.

Percentage of all samples where the motion channel is non-zero. Used as a global sleep restlessness indicator. High motion% reduces the reliability of all HR and SpO₂ metrics because artifact rejection removes those intervals from calculations.

Number of separate runs of consecutive motion > 0 samples, each separated by ≥30 seconds of quiet. Distinct from motion% — a single 10-minute movement period counts as 1 burst, whereas 30 brief startle movements count as 30 bursts. Higher burst counts at stable motion% suggest fragmented microarousals.

Longest uninterrupted run of samples where motion = 0, SpO₂ is in the 70–100% range, and no HR artifact flags are set. Indicates the best-quality data window in the recording. Short clean runs (<10 min) indicate poor recording quality throughout and reduce confidence in all derived metrics.

Ratio of mean absolute recovery slope to mean absolute dip slope. Values near 1.0 indicate symmetric events. Values >1.5 mean the SpO₂ recovers faster than it drops (fast recovery — typically obstructive events with intact arousal). Values <0.8 indicate slow recovery (possible impaired cardiac output or central pattern).

Count and percentage of detected periodic breathing (PB) oscillation windows that occur without a corresponding HR spike within 60 seconds. High diverge % means PB is present but arousals are absent — the BLUNTED AROUSAL pattern associated with central sleep apnea and heart failure.

Binary flag set when both conditions are met: (1) PB Diverge % ≥ 75%, and (2) oscillation episode count ≥ 6. The dual threshold was tuned on a small internal set (n = 18 nights) to eliminate false positives on CPAP nights where arousal suppression is expected by design.

Aerobic capacity estimate (ml/kg/min) from the HRmax/HRrest ratio. HR floor = HRrest proxy. Age-predicted HRmax via Tanaka formula. Overridable with measured HRmax from the profile panel.

Five training intensity zones from HRR (= HRmax − HRrest), where HRrest = nocturnal HR floor. Each zone is a percentage range of HRR added to HRrest. HRR-based zones are more individualized than age-predicted fixed-percent zones.