The science behind Vital

Why this page exists

Health apps make a lot of claims. Most don't show their work. Vital scores your day across Recovery, Exertion, Stress, Health and Readiness, and every weight in those formulas traces back to peer-reviewed research. This page lists what each score is built on, links every paper, and the full reference library sits at the bottom. There's nothing proprietary buried here: this page is the scoring document.

Where a weight is a judgement call, an engineering decision rather than something a paper validates, we say so.

The three principles

Personal baselines beat population norms. A healthy adult's HRV can sit anywhere from 20 ms to over 100 ms. Comparing user A against user B isn't a comparison of recovery; it's a comparison of age, fitness history and genetics. Vital scores almost everything against your own rolling distribution, not against a one-size table.

State signals and trait signals get different weights. Resting heart rate and HRV change day-to-day (state). VO₂ max changes month-to-month (trait). The Recovery score weights state signals more heavily because the question it answers (how recovered am I today?) is a state question.

The lack of a paper isn't a free pass. Several weights in Vital are engineering decisions, not citations. Where that's true, this page calls it out, and the honest caveats section collects them in one place.

Recovery

Recovery formula35% HRV + 25% sleep + 15% RHR + 15% (inverse) accumulated fatigue + 10% fitness composite

HRV (35%)

HRV is parasympathetic tone, and parasympathetic tone is the physiological substrate of "recovered." The 1996 European Society of Cardiology / NASPE Task Force standards (Circulation) established HRV as an independent mortality predictor and the canonical autonomic-balance marker. Shaffer & Ginsberg 2017 (Frontiers in Public Health) documents the wide healthy range (RMSSD 20–100+ ms) that mandates personal-baseline scoring: there's no useful absolute cut-off.

For daily scoring, Aubert 2003 (Sports Medicine) and Plews 2013 (Sports Medicine) are load-bearing: in trained populations, rolling averages outperform single-day reads, and trend deviations matter more than absolute values. This is why Vital uses a personal percentile of the recent window.

One honest caveat on the metric itself. The research above is built largely on RMSSD, but Apple Health exposes HRV as SDNN from roughly one-minute samples. SDNN and RMSSD are correlated but not interchangeable, and a one-minute window sits below the Task Force's short-term recording standard. Vital works in SDNN against your own SDNN baseline, where the trend is what matters, but the cross-metric mismatch is real and we'd rather name it than paper over it.

Sleep quality (25%)

Duration, efficiency, and deep+REM each carry a sub-citation. Hirshkowitz 2015 (NSF, Sleep Health) and Watson 2015 (AASM/SRS consensus, Sleep) set the 7–9 h adult target. Ohayon 2004 (Sleep, meta-analysis of 65 studies) is the canonical lifespan reference for deep% and REM%. Xie 2013 (Science) showed glymphatic clearance roughly doubles during slow-wave sleep, and Van Cauter 2000 (JAMA) showed roughly 70% of nocturnal growth-hormone pulses coincide with SWS episodes. The case that "deep sleep" specifically is the recovery-relevant stage isn't ornamental.

Caveat: consumer sleep-stage classification from a wrist or phone is approximate against laboratory polysomnography, so the deep% and REM% figures Vital reads carry more noise than the studies they lean on. We lean on duration and your own night-to-night trend more than any single night's stage split, which softens the error without pretending it's gone.

Van Dongen 2003 and Belenky 2003 establish that sleep debt is asymmetric: three recovery nights don't undo a week of restriction. That's why Recovery applies a chronic-bad-night penalty on top of last night's score, and why Readiness includes a sleep-trend factor.

Resting heart rate (15%)

Zhang 2016 (CMAJ), a meta-analysis of 46 studies (n > 1.2 million), established that each 10 bpm increase in RHR raises all-cause mortality risk by 9%. We don't use that as an absolute threshold; we use it inversely against your own baseline, because absolute RHR is dominated by fitness and genetics. Karvonen 1957's heart-rate-reserve framework is the basis for every personalised HR system since.

Accumulated fatigue (15%, inverse)

A single-day read misses multi-day stacking. Vital's fatigue accumulator runs an exponentially-weighted moving average over recent stress and sleep signals; Recovery subtracts the inverse. Borg 1982 (MSSE) validated the psychophysical foundations of subjective exertion scales; Krupp 1989 (Archives of Neurology) validated short Likert fatigue instruments. Vital's single 0–10 slider is a deliberate simplification of these (daily friction matters for adherence), but we acknowledge it carries more measurement noise than a 9-item scale.

Fitness composite (10%)

VO₂ max and 1-minute heart-rate recovery split this weight. Kodama 2009 (JAMA, n = 102,980) and Mandsager 2018 (JAMA Network Open, n = 122,007) establish that low cardiorespiratory fitness is among the largest single mortality predictors, larger than smoking or diabetes, with no plateau at the elite end. Cole 1999 (NEJM) and Jouven 2005 (NEJM) establish 1-minute HRR as an autonomic-function marker, not just a fitness proxy. Lambe 2025 (PLOS ONE) is an honest caveat: Apple Watch VO₂ max underestimates lab values with a mean absolute percentage error of ~13%, which is why we weight this composite at 10% rather than higher.

Exertion

Exertion formulaEdwards TRIMP across continuous HR zones + sport-impact bonus + ring-based floor, optionally blended with logged RPE

Banister 1975 introduced the fitness-fatigue impulse-response model that underlies all modern training-load monitoring. Edwards 1993 proposed the 5-zone simplification Vital ships. Borresen & Lambert 2009 (114-article review in Sports Medicine) showed Banister and Edwards correlate r = 0.80–0.99, so the simplification trades little physiological precision for huge data-input feasibility on consumer wearables.

Karvonen 1957 is the basis for the heart-rate-reserve concept. Norton 2010 (J Sci Med Sport, ESSA position) standardises moderate/vigorous intensity descriptors. Murray 2017 (British Journal of Sports Medicine) provides peer-reviewed validation that exponentially-weighted moving averages of acute-to-chronic workload predict injury risk more sensitively than rolling averages. This is why Vital's CTL/ATL/TSB pipeline uses EWMAs.

Stress

Stress formulaHRV depression up to 40 + RHR elevation up to 20 + sleep deficit up to 15 + accumulated fatigue up to 15 + SpO₂ up to 5 + respiratory rate up to 5 + user-reported up to 20

Kim 2018 (Psychiatry Investigation, meta-analysis of 37 studies) confirms that acute stress reliably reduces HF-HRV and RMSSD across cognitive, social-evaluative, and pain stressors. Thayer 2012 (Neuroscience & Biobehavioral Reviews) is the neurovisceral-integration model: HRV indexes vmPFC–amygdala–brainstem coupling. Low HRV is paired with elevated cortisol and disinhibited HPA-axis reactivity. This is the strongest single piece of evidence in the stress algorithm.

For user-reported stress, Cohen 1983 (J Health Soc Behav) is the canonical PSS reference. Vital's 5-level daily check-in is a simplified version validated against the full PSS-10 by Elo 2003 (Scand J Work Environ Health, r ≈ 0.6–0.7).

The menstrual-cycle modifier

Schmalenberger 2019 (J Clin Med), a meta-analysis of 37 studies in 1,004 women, established the magnitude: vagal HRV decreases from follicular to luteal with d = −0.39, roughly a 15–20% reduction in HF-HRV power. Progesterone is the within-person driver (Schmalenberger 2020). Tenan 2014 (Psychophysiology) confirmed parallel RHR elevation; Baker & Driver 2020 (Temperature) documented the luteal +0.3–0.7 °C core-temperature rise plus altered sleep architecture.

What this means in practice: Vital scales the stress score down by 15% in luteal phase because the underlying physiology is hormonal, not psychogenic. We're working on phase-aware baselines as a more principled long-term answer.

Health

Health formula25% recovery + 25% activity + 10% sleep + 18% nutrition + 12% hydration + 10% (inverse) stress, with adaptive weight-shift toward weak domains

The Health score is a long-horizon trait-level composite. Each input is covered above (recovery, sleep, stress) or here:

• Activity: Ekelund 2019 (BMJ, harmonised meta-analysis of 8 cohorts) shows the activity–mortality dose-response is non-linear with steepest gains in the first ~25 min/day of moderate-vigorous activity. The WHO 2020 guidelines (Bull et al., BJSM) set the 150–300 min/week aerobic target.
• Nutrition: Drewnowski 2009 (J Am Coll Nutr) defines the Nutrient Rich Foods Index: 9 nutrients-to-encourage minus 3 to limit, validated against the Healthy Eating Index. It's the canonical nutrient-density framework.
• Hydration: EFSA 2010 sets 2.5 L/day (males) and 2.0 L/day (females) total water Adequate Intakes; IOM 2004 sets parallel figures. Cheuvront & Sawka 2005 (and the ACSM Position Stand on fluid replacement, Sawka 2007) provides the clinically validated dehydration markers.

The 18%/12% nutrition/hydration ratio is an editorial weight, not a paper-validated one. We say so honestly.

Readiness

Readiness formula35% recovery + 15% (inverse) stress + 20% Training Stress Balance + 15% (inverse) fatigue + 8% sleep trend + 7% HRV trend

The TSB term comes from the Banister 1991 / Coggan generalisation of the 1975 model. Murray 2017 (BJSM) validates exponentially-weighted moving averages for acute-to-chronic workload as injury-risk predictors. The mapping of TSB to a 0–100 contribution via hyperbolic tangent is a Vital engineering choice, since the literature treats TSB as banded (peaked >+15, productive overload −10 to −30) rather than continuously mapped, and we document it that way.

The sleep-trend and HRV-trend factors are deliberately small (8% and 7%) because they may have fewer than three snapshots in a new user's history, in which case they renormalise out. Plews 2014 establishes that ≥3 valid HRV readings per week is the minimum-validity threshold for meaningful trend interpretation.

TDEE

TDEE formulaMifflin–St Jeor BMR × activity multiplier (1.2–1.9)

Mifflin 1990 (Am J Clin Nutr) is the BMR equation derived from n = 498 healthy adults. Frankenfield 2005 (J Am Diet Assoc) is the systematic cross-equation comparison the American Dietetic Association cites: Mifflin produces RMR within ~10% of measured values in normal-weight and obese populations, with the narrowest error band versus Harris-Benedict and Owen. The activity multipliers (PAL) come from the FAO/WHO/UNU 2004 Human Energy Requirements consultation, which defines three lifestyle bands (sedentary 1.40–1.69, moderate 1.70–1.99, vigorous 2.00–2.40) from doubly-labelled-water studies.

For users with reliable body-composition data, the Katch–McArdle / Cunningham 1991 lean-mass equation (BMR = 370 + 21.6 × lean body mass in kg) is more accurate than Mifflin, particularly at the lean and obese tails of the distribution. We branch where the data supports it.

What the literature doesn't validate: the honest caveats

Not every choice in Vital is paper-derived. We name them rather than hide them:

• The exact 35/25/15/15/10 Recovery weight ratio. Each component is supported individually; the specific ratio is an engineering judgement built around the state-vs-trait framing.
• The HRV metric mismatch: the recovery research is RMSSD-based, but Apple Health exposes SDNN from ~1-minute samples, below the Task Force 1996 short-term standard. Vital scores SDNN against your own SDNN baseline; trend absorbs some of the gap but not all of it.
• Consumer sleep-stage accuracy. Wrist and phone deep%/REM% estimates are approximate versus polysomnography, so the stage inputs are weighted behind duration and personal trend.
• The 65/35 blend between logged RPE and HealthKit-derived exertion. Foster 1998 (MSSE) supports the combination principle but typically at closer to 50/50. The 65 favouring logged is a user-trust call.
• The hyperbolic-tangent mapping of Training Stress Balance to a 0–100 Readiness contribution. Defensible (symmetric saturation matches the empirical observation that very-high and very-low TSB both saturate) but it's not a paper citation.
• The 18%/12% nutrition-to-hydration ratio in the Health score.
• The adaptive Health-score weight-shift toward weak domains. The principle has support in clinical wellness frameworks (PROMIS, SF-36); the exact magnitudes (5 pts per weak domain, capped at 50% of strong-side pool) are engineering.
• Apple Watch as a measurement instrument. The HRV recordings (~1-minute samples) sit below the Task Force 1996 short-term threshold; VO₂ max underestimates lab values (Lambe 2025). Vital absorbs some of this slack through personal baselines and trend analysis; it does not eliminate it.

The full reference library

Every source named on this page, in one place, with a link to each. Around 40 papers across HRV, sleep, training load, autonomic stress, menstrual-cycle physiology, BMR equations, hydration and nutrient density. The list is maintained alongside the code and updated whenever an algorithm weight changes.

• Aubert 2003, Sports Medicine. Rolling HRV averages outperform single-day reads in trained populations.
• Baker & Driver 2020, Temperature. Luteal +0.3–0.7°C core-temperature rise plus altered sleep architecture.
• Banister 1975. Fitness-fatigue impulse-response model underlying modern training-load monitoring. (foundational text; no stable DOI)
• Banister / Coggan 1991. TSB term; generalisation of the 1975 model for Training Stress Balance. (foundational text; no stable DOI)
• Belenky 2003. Sleep debt asymmetric; recovery nights don't undo a week of restriction.
• Borg 1982, MSSE. Validated psychophysical foundations of subjective exertion (RPE) scales.
• Borresen & Lambert 2009, Sports Medicine. Banister and Edwards TRIMP correlate r=0.80–0.99 (114-article review).
• Cheuvront & Sawka 2005. Clinically validated dehydration markers.
• Cohen 1983, J Health Soc Behav. Canonical Perceived Stress Scale (PSS) reference for user-reported stress.
• Cole 1999, NEJM. 1-minute heart-rate recovery as an autonomic-function marker.
• Drewnowski 2009, J Am Coll Nutr. Nutrient Rich Foods Index; canonical nutrient-density framework.
• Edwards 1993. 5-zone TRIMP simplification Vital ships for Exertion.
• EFSA 2010. Total-water Adequate Intakes: 2.5 L/day males, 2.0 L/day females.
• Ekelund 2019, BMJ. Non-linear activity–mortality dose-response, steepest first ~25 min/day (8-cohort meta-analysis).
• Elo 2003, Scand J Work Environ Health. Single-item daily stress check validated against PSS-10 (r≈0.6–0.7).
• European Society of Cardiology / NASPE Task Force 1996, Circulation. HRV as independent mortality predictor and canonical autonomic-balance marker; short-term sampling threshold.
• FAO/WHO/UNU 2004, Human Energy Requirements consultation. PAL activity multipliers / three lifestyle bands for TDEE.
• Foster 1998, MSSE. Supports combining logged RPE with HealthKit exertion (typically ~50/50).
• Frankenfield 2005, J Am Diet Assoc. Cross-equation comparison; Mifflin within ~10% of measured RMR (ADA-cited).
• Hirshkowitz (NSF) 2015, Sleep Health. 7–9 h adult sleep-duration target.
• IOM 2004. Parallel total-water intake figures for hydration.
• Jouven 2005, NEJM. 1-minute HRR as autonomic-function marker, not just a fitness proxy.
• Karvonen 1957. Heart-rate-reserve framework behind personalised HR systems and zones.
• Katch & McArdle. Lean-mass BMR equation (370 + 21.6 × LBM); more accurate at lean/obese tails.
• Kim 2018, Psychiatry Investigation. Acute stress reliably reduces HF-HRV and RMSSD (37-study meta-analysis).
• Kodama 2009, JAMA. Low cardiorespiratory fitness among largest mortality predictors (n=102,980).
• Krupp 1989, Archives of Neurology. Validated short Likert fatigue instruments behind the fatigue slider.
• Lambe 2025, PLOS ONE. Apple Watch VO₂ max underestimates lab values (~13% MAPE); caps composite at 10%.
• Mandsager 2018, JAMA Network Open. Low fitness a leading mortality predictor with no elite plateau (n=122,007).
• Mifflin 1990, Am J Clin Nutr. Mifflin–St Jeor BMR equation (n=498 healthy adults) for TDEE.
• Murray 2017, British Journal of Sports Medicine. EWMA acute-to-chronic workload predicts injury risk; CTL/ATL/TSB and Readiness.
• Norton (ESSA position) 2010, J Sci Med Sport. Standardises moderate/vigorous intensity descriptors.
• Ohayon 2004, Sleep. Canonical lifespan reference for deep% and REM% (meta-analysis of 65 studies).
• Plews 2013, Sports Medicine. Trend deviations matter more than absolute HRV; personal percentile windowing.
• Plews 2014. ≥3 valid HRV readings/week as minimum-validity threshold for trend interpretation.
• PROMIS (Cella 2007), Medical Care. Support for adaptive Health-score weight-shift toward weak domains.
• Sawka (ACSM Position Stand on fluid replacement) 2007. ACSM fluid-replacement position stand; validated dehydration markers.
• Schmalenberger 2019, J Clin Med. Vagal HRV drops follicular→luteal d=−0.39 (meta-analysis, 1,004 women).
• Schmalenberger 2020. Progesterone is the within-person driver of luteal HRV reduction.
• SF-36 (Ware 1992), Medical Care. Support for adaptive Health-score weight-shift toward weak domains.
• Shaffer & Ginsberg 2017, Frontiers in Public Health. Wide healthy HRV range (RMSSD 20–100+ ms) mandating personal-baseline scoring.
• Tenan 2014, Psychophysiology. Confirmed parallel luteal RHR elevation.
• Thayer 2012, Neuroscience & Biobehavioral Reviews. Neurovisceral-integration model; HRV indexes vmPFC–amygdala–brainstem coupling.
• Van Cauter 2000, JAMA. ~70% of nocturnal growth-hormone pulses coincide with SWS episodes.
• Van Dongen 2003. Sleep debt is asymmetric; supports chronic-bad-night penalty.
• Watson (AASM/SRS consensus) 2015, Sleep. 7–9 h adult sleep-duration target.
• WHO (Bull et al.) 2020, BJSM. 150–300 min/week aerobic activity target.
• Xie 2013, Science. Glymphatic clearance roughly doubles during slow-wave sleep.
• Zhang 2016, CMAJ. Each 10 bpm RHR rise raises all-cause mortality 9% (46-study meta-analysis).

Spotted a citation we're missing, or a finding that contradicts a Vital weight? Write to business@lmslabsltd.com and we'll read it.

This page is informational. Vital is a wellness app, not a medical device. Nothing here is a substitute for advice from a qualified clinician.