Energy Availability and RED-S: A Data-Literate Guide to Recognizing Underfueling

You’ve been training more. Sleep’s been rougher. Your ring says your HRV trend is down. Your period’s late. Your last three lifts felt heavier than the numbers suggest they should.

Are you underfueled? Or are you stressed, slightly sick, under-slept, premenstrual, and in a heavy training block all at once?

This is the hard problem of consumer-grade energy availability surveillance: the biomarker signatures that suggest underfueling are almost indistinguishable from the signatures of a dozen other things. The goal of this guide is to help you read your own data with honest skepticism — and to know when the pattern is worth a conversation with a clinician rather than another week of watching a dashboard.

This is the pillar post for Omnio’s energy availability cluster. For the companion deep-dives on the cluster scope, see adaptive training intelligence, nutrition intelligence, and composite scores with confidence. Every claim in this guide is framed around what consumer data can surface — not diagnose.

---

What Energy Availability Is, and Why It Matters Outside Elite Sport

Energy availability (EA) is the kilocalories left over for basic physiological function after exercise energy expenditure is subtracted from dietary intake, expressed per kilogram of fat-free mass per day. The original research framework, set out in papers by Anne Loucks, Jaci VanHeest, and colleagues in the late 1990s and early 2000s, proposed a threshold around 30 kcal/kg FFM/day below which hormonal disruption begins to measurably accumulate in controlled laboratory conditions. The number is widely quoted, frequently misunderstood, and deserves context.

EA is not a body composition target. It is not a daily TDEE subtraction. It is a balance equation between intake and exercise-specific expenditure, normalized to the metabolically active tissue. A 60 kg woman at 18 percent body fat has roughly 49 kg of fat-free mass. Eating 1,700 kcal and burning 800 kcal during training would leave her at (1,700 − 800) / 49 ≈ 18 kcal/kg FFM/day — well inside the low-EA range that the literature associates with hormonal consequences. Eating 2,400 kcal with the same training load would leave her around 33, which sits above the threshold but below what most researchers consider comfortably repleted.

Two facts make EA difficult to measure outside a lab. First, dietary intake is consistently under-reported in self-report studies by 10 to 30 percent, often more among athletes and dieters. Second, exercise energy expenditure estimates from consumer wearables are approximations — training-log TRIMP, GPS watches, and strap HR-based calorie estimates differ by 10 to 25 percent from indirect calorimetry in validation studies. Stack those errors and a computed EA number has a plausible range of roughly ±5 kcal/kg FFM/day on a single day. That’s wide enough that a single reading of 28 and a single reading of 33 may reflect the same physiology.

That uncertainty is why the serious literature uses EA as a multi-day average and a research tool, and why responsible consumer surveillance treats the computed number as directional rather than categorical.

RED-S (Relative Energy Deficiency in Sport) is the syndromic consequence of sustained low EA — it is what happens when the body, denied enough fuel to support training, basal metabolism, and all of its regulatory systems simultaneously, begins shutting things down. The 2023 International Olympic Committee consensus statement (Mountjoy et al.) expanded the scope explicitly beyond the original Female Athlete Triad (Nattiv et al.) framing to include male athletes, recreational athletes, and adolescents. That matters for two reasons: first, it means RED-S is not a niche diagnosis confined to elite endurance athletes; second, it means more people are now looking at their wearable data wondering if they qualify.

They often don’t. Sustained low EA is clinically serious and relatively rare at population scale. What’s common is a few weeks of elevated training and somewhat under-fueled days — a pattern that lowers EA temporarily without triggering the cascade the RED-S framework describes. The distinction matters because the treatment is different. Short-term under-fueling responds to a few deliberate higher-intake days. Sustained RED-S requires clinical nutrition support and, in many cases, a medical workup for bone health, endocrine function, and mental health.

The job of a surveillance tool is to flag trajectories, not to assign diagnoses. The next section explains why.

RED-S: Scope, Symptoms, Clinical Criteria, and the Limits of Consumer Data

RED-S is a clinical syndrome, defined by consensus rather than a single lab test. The 2023 IOC consensus lists impacts across ten-plus body systems: menstrual function, bone health, metabolic rate, immune function, protein synthesis, cardiovascular, gastrointestinal, psychological, endocrine, growth and development. Symptoms include secondary amenorrhea (absence of menstrual period in someone who previously menstruated regularly), stress fractures and low bone mineral density, persistent fatigue unexplained by training, decreased training response, gastrointestinal complaints, mood disturbance, low libido and erectile dysfunction in male athletes, sleep disruption, and cold intolerance.

No single symptom is pathognomonic. A runner can have low EA and regular menses. A lifter can have RED-S and still set PRs for six more months before the training response collapses. An athlete can be amenorrheic for reasons unrelated to EA. A stress fracture can happen to a well-fueled athlete with poor form or excessive running volume on hard surfaces.

This is why the clinical literature emphasizes risk stratification rather than binary diagnosis. The IOC RED-S CAT 2 risk assessment tool scores an athlete across multiple domains — menstrual history, bone history, metabolic markers, disordered eating, body mass trajectory, training load — and places them into low-, moderate-, or high-risk categories, each with different return-to-sport and clinical follow-up recommendations. The tool is explicitly designed for use by qualified practitioners in sport medicine and dietetics.

Consumer wearable data does not feed into this tool. It cannot, because the signals the tool relies on — DEXA-measured bone mineral density, resting metabolic rate from indirect calorimetry, detailed menstrual and fracture history, psychometric assessment — are not available from a ring or a watch. What wearables do see are downstream physiological consequences: altered autonomic balance, changes in resting heart rate, disrupted sleep, changes in recovery markers.

These signals correlate with, but do not equal, RED-S. A study by Stellingwerff and colleagues on elite endurance athletes showed detectable changes in heart-rate variability and resting heart rate during periods of low EA; multiple studies since have replicated the directionality. But the same studies, and others, show large inter-individual variance, non-trivial day-to-day noise, and substantial overlap with physiology driven by training load, sleep deprivation, and illness. No study has demonstrated that wearable-derived signals alone can classify RED-S risk at clinically useful sensitivity and specificity. Most have not tried, because the confounders are too strong.

So what can consumer data actually do? Two things honestly.

First, it can show trends. An athlete whose HRV has drifted down by 20 percent over six weeks, whose resting heart rate is up 8 bpm, whose sleep has become more fragmented, and whose body mass has dropped unintentionally is in a trajectory that’s worth investigating. That pattern doesn’t diagnose anything, but it argues for a conversation.

Second, it can rule things in and out by pattern. If the autonomic signals have drifted and the athlete has also been traveling across time zones, drinking on weekends, and in a heavy training block, the signals might be entirely explained by those confounders. If the signals have drifted without those confounders — especially if body mass has dropped, menses are absent, and fatigue is persistent — the pattern is more specific, and the case for a clinical conversation is stronger.

The rest of this guide is about how surveillance tools — Omnio included — try to make that second judgment defensibly.

Biomarker Signatures of Underfueling, and the Confounders That Mimic Them

The physiological cascade from low EA to visible biomarker changes is well-described in the research literature. When energy is chronically insufficient, the body responds by down-regulating metabolic and reproductive function. Thyroid hormone levels drop. Leptin falls. Sympathetic drive falls, parasympathetic drive becomes more variable, and autonomic balance shifts in ways that show up in HRV. Menstrual function is suppressed through hypothalamic suppression of GnRH. Bone turnover rises relative to formation. Resting heart rate rises modestly. Sleep becomes more fragmented, particularly deep sleep. Cold intolerance, constipation, and dry skin follow from reduced metabolic rate.

The biomarkers a wearable can actually see are a small subset of that list: heart-rate variability (typically RMSSD from overnight PPG), resting heart rate, sleep duration and staging, and activity. A connected scale adds body mass trajectory. Women tracking menstrual cycles add cycle length and period occurrence. Lifting logs add strength trajectories. That’s a useful subset — but every one of those signals has at least three non-RED-S explanations.

Here is the honest mapping, signal by signal, based on published physiology. The full deep-dive on each signal is in biomarker signatures of underfueling.

Resting heart rate elevation. Low EA is associated with elevated RHR in some studies and lowered RHR in others, reflecting the stage and severity of the cascade. Luteal phase of the menstrual cycle elevates RHR by 2–5 bpm in most women even when fueling is adequate. Viral illness elevates RHR for days before other symptoms appear. Alcohol elevates RHR for 24–48 hours after drinking. Sleep debt elevates RHR. Heat exposure elevates RHR. Overreaching weeks elevate RHR. A single morning with RHR 7 bpm above baseline tells you almost nothing about EA on its own.

Heart rate variability decline. The autonomic shift from sustained low EA shows up as lower time-domain HRV (RMSSD, SDNN) and altered frequency-domain balance. But: luteal phase lowers HRV. Illness lowers HRV, often sharply and early. Alcohol lowers HRV for up to two days. Acute training load lowers HRV transiently. Sleep debt lowers HRV. Psychological stress lowers HRV. Air travel lowers HRV. The directionality alone is not diagnostic — what matters is the duration, the slope, and whether confounders are in play.

Sleep disruption. Low EA disrupts sleep in the published literature — reduced total sleep time, reduced deep sleep, more fragmentation. But sleep is also disrupted by alcohol, late meals, illness, stress, travel, temperature, medication changes, and every other normal-life factor. Sleep is probably the signal most vulnerable to confounding.

Body mass trajectory. Unintentional weight loss during training is one of the higher-specificity signals — it’s hard to lose weight while eating enough. But: weight loss during cutting phases is intentional and expected. Weight fluctuation on a single scale reading is dominated by hydration, glycogen, and gut contents, which can swing 1–2 kg in a day. Multi-week downward trends matter; single-week dips don’t.

Menstrual irregularity. Secondary amenorrhea (three or more consecutive missed periods in someone who was regularly menstruating) is one of the stronger specific signals in the Female Athlete Triad / RED-S framework. But oligomenorrhea (irregular periods) has many causes, including PCOS, thyroid disease, hormonal contraceptive use, and stress. Any persistent menstrual change warrants clinical evaluation, not a wearable flag.

Strength plateau or regression. The RED-S framework predicts that strength training adaptation blunts under low EA. But strength progression plateaus for many reasons: technique ceilings, sleep, programming errors, regression to the mean after a novice gains phase, specific muscle recovery deficits. Stalled bench pressing is not a smoking gun.

Blunted mood, low libido, cold intolerance. Each of these is reported by a subset of RED-S patients. Each also has dozens of other causes — depression, sleep deprivation, relationship stress, thyroid disease, weather. Self-reported mood data is a valuable confounder-resolution signal but not a primary flagging signal.

The takeaway is not that these signals are useless. It is that any single signal, at a single point in time, is nearly worthless. What matters is compound trajectories — multiple signals moving in a consistent direction over weeks, while known confounders (cycle phase, illness, alcohol, recent training spikes, travel, sleep debt) can be reasonably ruled out.

Compound trajectories are what makes surveillance possible. They are also what makes it hard. The next section describes what honest surveillance looks like in practice.

Surveillance vs. Diagnosis: What Omnio’s Watch and Refeed Flags Mean

Omnio’s energy availability surveillance does not claim to detect RED-S. It monitors for patterns consistent with underfueling, gates those patterns against confounders, and surfaces them for user review — explicitly not as diagnoses, and explicitly with a recommendation to discuss with a professional when the pattern is persistent.

Here is how the surveillance logic is structured. The specifics are a work in progress and are deliberately conservative; the feature is feature-flagged internally until the empirical false-positive rate drops below a documented threshold.

Baseline establishment. For each biomarker (resting HR, HRV, sleep duration, body mass), the system computes a personalized baseline from at least 30 days of clean data. Baselines are phase-aware for menstruating users: when at least three prior cycles of data are available, the baseline is computed by menstrual phase so that luteal-phase elevations in RHR don’t get mistaken for underfueling. This is a direct implementation of the confounder-gating principle: the system knows that luteal physiology mimics low-EA physiology and adjusts for it. For a deeper look at personalized baselines across all signals, see composite scores with confidence.

Signal detection. The system looks for sustained multi-signal drift. A single morning of high RHR is ignored. A seven-day rolling mean RHR more than 1 standard deviation above phase-matched baseline, co-occurring with a seven-day rolling HRV more than 1 SD below baseline, over two consecutive weeks, is a pattern worth flagging — but only after the confounder gates are checked.

Confounder gating. Before any flag is surfaced, the system checks:

• Has the user logged illness in the past 5 days? If yes, the autonomic rule is suppressed.
• Has the user logged alcohol consumption above a daily threshold in the past 48 hours? If yes, suppress.
• Is the user in a week where TRIMP (training load) z-score is ≥ 2.0 — i.e. a clear training spike? If yes, suppress the autonomic rule; overreaching explains the signal.
• Has the user recently traveled across time zones? Surveillance pauses.

This gating is the difference between a surveillance tool you can trust and one that cries wolf. The detailed version is in confounders that mimic RED-S.

Signal-specific weighting. Weight stall has low confounder load and receives more weight. Autonomic signals (RHR, HRV) have high confounder load and receive less weight when unaccompanied by weight signal. Subjective reports (fatigue, mood) are evaluated as corroborating signals when the user has logged them consistently.

Shadow mode. Until the empirical false-positive rate of the flagging logic is demonstrated to be below a documented acceptable level (assessed from user feedback on flags via a confounder-affordance dismiss flow), user-visible flags are disabled. The logic runs in shadow; developers and a small internal cohort see the flags; users don’t. This is a deliberate choice — releasing a high-FPR flag would cause more harm than the feature’s potential benefit.

User-facing surface, when enabled. When the feature comes out of shadow, a flag is surfaced as: “Over the past 2–3 weeks, your data shows patterns consistent with underfueling. Known confounders we could rule out: [list]. Please review the details, consider whether any of the following recent events apply: [illness, travel, alcohol, cycle phase, heavy training], and if the pattern is persistent and unexplained, speak with a sport dietitian.” It is never phrased as “You have RED-S” or “You are underfueling.”

This is what we mean by surveillance over diagnosis. The job of the tool is to tell you when a pattern is worth thinking about — and to be silent when it isn’t. The job of the clinician is to diagnose.

Similar principles apply to other early-warning systems on the platform. The general philosophy is described in predicting health dips before they happen: trend-to-threshold forecasts, transparent confounders, and a clear handoff to the human when the data is ambiguous.

When to Escalate: Sport Dietitians, Endocrinologists, and the Limits of Self-Tracking

A surveillance dashboard is a starting point. Clinical care is the finish. Knowing when to make the jump is the practical skill this guide wants to leave you with.

See a clinician now if any of the following apply, regardless of what your wearable says:

• Missed three or more consecutive menstrual periods, or any previous diagnosis of amenorrhea. Hypothalamic amenorrhea from low EA is reversible when treated; prolonged untreated amenorrhea compromises bone health in ways that are not fully reversible.
• Stress fracture or bone stress injury, particularly in a low-impact runner or non-contact athlete, particularly if you are already managing training volume.
• Unintentional weight loss over more than 4 weeks while trying to maintain or gain.
• Persistent fatigue that doesn’t respond to a full off-day, a full off-week, or extra sleep.
• Any suspicion that you are restricting food in a way that causes distress, or any history of disordered eating.
• Cold hands and feet that are new and persistent, dry skin, constipation, or low resting body temperature that weren’t there a year ago.

The right professionals, in rough order for most cases:

• Sport medicine physician: frontline provider for RED-S workup. They can order DEXA, bone turnover markers, hormonal panels, metabolic assessment, and coordinate referrals.
• Sport dietitian with CSSD credentials (in the US) or equivalent: for nutritional rehabilitation, intake planning, performance-aware refeeding.
• Endocrinologist: when hormonal workup suggests deeper reproductive or metabolic endocrine dysfunction.
• Mental health professional with sport or disordered-eating expertise: disordered eating is frequently comorbid with low EA and requires coordinated care.
• Primary care provider: for the initial workup if a sport medicine physician isn’t immediately accessible; they can rule out other causes of fatigue and menstrual disruption.

What consumer data can and can’t contribute. It can show trends. It can provide a history of training load, sleep, and cardiovascular markers that a clinician would otherwise have to ask you to reconstruct from memory. Many sport dietitians will explicitly ask for this. It can support a conversation with evidence.

It cannot provide a diagnosis. It cannot rule out or confirm RED-S. It cannot measure bone mineral density, resting metabolic rate, hormone levels, or inflammatory markers. It cannot replace a trained clinician’s pattern recognition across dozens of patients.

The right framing for a clinic visit is: “Here’s a six-month picture of my training, sleep, heart-rate data, and body mass. These are the patterns I’m seeing. Can you help me figure out whether this warrants further workup?” That kind of conversation is where consumer data earns its keep.

Note: Omnio is a hosted health data aggregation platform. It unifies your data, runs surveillance on it, and shows you the patterns. It doesn’t diagnose or treat. That’s intentional.

Refeed Protocols: What Research Supports, What’s Experimental

Refeeding — a deliberate temporary increase in energy or carbohydrate intake — shows up in the strength and endurance literature in two distinct contexts. It’s useful to separate them.

Context 1: Glycogen restoration for performance. High-carbohydrate refeed days before endurance competition are well-established. Sherman and colleagues’ work in the 1980s on carbohydrate loading, updated by more recent reviews, supports a 24–72 hour high-carbohydrate phase before long-duration events to maximize muscle glycogen. This is not treatment for RED-S; it’s a performance protocol, and it’s not controversial.

Context 2: Metabolic and autonomic recovery after low-EA periods. This is where the evidence gets thinner. The underlying hypothesis is that a few days of higher energy intake can restore leptin, thyroid, and autonomic function after a period of under-fueling. Several small studies have shown acute leptin and thyroid responses to short-term refeeding. Whether these acute responses translate into sustained recovery — or whether intermittent refeeding is a reasonable treatment strategy for sustained low EA — has not been established at the level of consensus.

Here’s what the honest reading of the evidence suggests:

• Short-term refeeding days are a reasonable tool for breaking training stagnation when the pattern suggests a few weeks of mild under-fueling. A day or two at 15–25 percent above normal intake, biased toward carbohydrate, is a low-risk intervention.
• As a treatment for chronic RED-S, periodic refeeding is not an adequate substitute for sustained adequate intake and clinical care. The 2023 IOC consensus does not endorse periodic refeeding as a primary intervention. Nutritional rehabilitation is the primary intervention. Refeeding may play a supporting role during transition phases, under professional supervision.
• As a data-driven surveillance-plus-refeed loop — “my data says I’m underfueled, I’ll eat more for two days, watch the data, and see if it recovers” — this is entirely experimental. No controlled study has validated this loop as a clinical tool. The reason this kind of loop is tempting is that it fits the quantified-self aesthetic; the reason it’s risky is that it conflates pattern-recognition with treatment. A couple of things are worth noting: (a) if the signals do recover, it doesn’t prove EA was the cause — it might have been coincidental confounders resolving; (b) if the signals don’t recover, you have not ruled out RED-S, and the next step is clinical, not another attempted refeed. The depth on this is in refeed protocols: what the research supports.

Practically, if surveillance flags a pattern and confounders are ruled out, the right action for a moderately-competitive recreational athlete is usually: add a higher-intake day, re-examine your weekly training and eating structure with fresh eyes, watch the data for another week or two, and if the pattern persists, see a clinician. That’s a reasonable self-managed response to a moderate-risk flag. The high-risk flag response — amenorrhea, stress fracture, persistent fatigue — is to see a clinician now.

Putting It Together

The honest version of the energy availability story, from a data-literate consumer perspective, looks like this:

1. Low energy availability is a real physiological state with real consequences. RED-S is the clinical syndrome that follows from sustained low EA. Both concepts are research-supported and clinically recognized (Mountjoy et al. 2023 IOC consensus; Nattiv et al. 2007 Female Athlete Triad; Logue et al. 2019 for male athletes; Loucks and Thuma 2003 for the EA threshold research).

2. The biomarker signatures of underfueling — altered HRV, elevated RHR, disrupted sleep, weight stall, menstrual irregularity, blunted training response — are real but nonspecific. Every one of them has at least three other common causes, most notably cycle phase, illness, alcohol, travel, sleep debt, and acute training spikes.

3. Consumer wearable data can reveal trajectories; it cannot diagnose RED-S. Surveillance tools that claim diagnostic power are overclaiming.

4. Defensible surveillance looks like compound-signal trajectory detection with explicit confounder gating, personalized baselines, and a feature-flagged shadow mode until empirical false-positive rate is known. Omnio’s surveillance is built this way by design; its user-facing layer is intentionally cautious.

5. Escalation to a sport medicine physician, sport dietitian, or endocrinologist is the right response when the pattern is persistent, unexplained by confounders, or accompanied by hard signals like missed periods or stress fractures. Self-tracking complements clinical care; it does not replace it.

6. Refeed protocols have evidence for glycogen restoration and acute metabolic response. They do not have evidence as a sustained treatment for chronic low EA. Self-directed refeed loops are experimental and should not substitute for clinical evaluation when the pattern is concerning.

The goal of any tool in this space — Omnio’s surveillance included — is to give you better-calibrated information, not more confident conclusions. If a surveillance flag makes you more informed when you talk to a clinician, it’s doing its job. If it makes you stop calling the clinician because the app already told you what’s wrong, it’s failing.

Be careful with your data. Be skeptical of single-signal claims. Know the confounders. Know the professionals to escalate to. And remember that the most important output of any pattern-recognition system for health is better questions — not automated answers.

---

References and Reading

The foundational literature for this article includes:

• Mountjoy M. et al. (2023). 2023 International Olympic Committee’s (IOC) consensus statement on Relative Energy Deficiency in Sport (REDs). British Journal of Sports Medicine. The current clinical consensus document; expands scope to male athletes and recreational athletes.
• Nattiv A. et al. (2007). The Female Athlete Triad Position Stand. American College of Sports Medicine. The original triad framework of energy availability, menstrual function, and bone health.
• Loucks A.B., Thuma J.R. (2003). Luteinizing hormone pulsatility is disrupted at a threshold of energy availability in regularly menstruating women. Journal of Clinical Endocrinology and Metabolism. The foundational study establishing the 30 kcal/kg FFM/day threshold.
• Logue D. et al. (2019). Low energy availability in athletes 2020: An updated narrative review of prevalence, risk, within-day energy balance, knowledge, and impact on sports performance. Sports Medicine. Among the best syntheses of the evidence for male-athlete RED-S.
• Stellingwerff T. et al. (2021). Overtraining Syndrome (OTS) and Relative Energy Deficiency in Sport (RED-S): Shared Pathways, Symptoms and Complexities. Sports Medicine. Useful for understanding the overlap between RED-S signatures and overreaching.
• De Souza M.J. et al. (2014). 2014 Female Athlete Triad Coalition Consensus Statement on Treatment and Return to Play. British Journal of Sports Medicine. The clinical framework for return-to-sport decisions after triad diagnosis.
• Keay N., Francis G. (2019). Infographic. Energy availability: concept, control and consequences in relative energy deficiency in sport (RED-S). British Journal of Sports Medicine. Accessible clinical reference.

Professional resources to investigate for clinical support:

• Sport medicine physician — initial workup, coordinating specialist referrals, RED-S CAT 2 risk assessment.
• Sport dietitian — look for the CSSD credential (Certified Specialist in Sports Dietetics) in the US or the Registered Sport Dietitian (SENr) in the UK.
• Endocrinologist — when hormonal workup indicates deeper evaluation is needed.
• Mental health professional — particularly one with experience in eating disorders and athlete populations.

Omnio’s surveillance logic is described in general terms in predicting health dips before they happen and at the scoring level in composite scores with confidence. The biomarker foundation is covered in what HRV is and how wearables measure it and how wearables measure stress and strain. For training-load context, see adaptive training intelligence and what ACWR is. For nutrition integration, see nutrition intelligence and the calorie target guide. Omnio’s feature overview is at features — most relevant to this cluster are analytics, adaptive training, and composite health scores.

---

More in this Series

This is the pillar post for a cluster of deep-dives on energy availability and RED-S. The companion posts are scheduled over the following weeks:

• What Is RED-S? A Literacy Guide for Athletes and Active Adults — coming soon
• How Energy Availability Is Calculated (and Why It’s Harder Than It Looks) — coming soon
• Biomarker Signatures of Underfueling: Sleep, HRV, RHR, and Their Limits — coming soon
• Confounders That Look Like RED-S: Cycle Phase, Illness, Alcohol, Training Spikes — coming soon
• Refeed Protocols: What the Research Supports (and What It Doesn’t) — coming soon

Each spoke goes deeper on one dimension of the surveillance problem. The cluster philosophy is the same as the pillar: surveillance, not diagnosis; skepticism over confidence; clinical escalation when the pattern matters.