Confounders That Look Like RED-S: Cycle Phase, Illness, Alcohol, Training Spikes

Your HRV is down 22 percent. Your resting heart rate is up 7 bpm. Your sleep score has been below baseline for five nights out of seven.

Before you conclude you’re underfueling, check: Is this week 3 of your cycle? Did you have a cold last weekend? Did you drink on Saturday? Was Tuesday’s interval session a personal-record effort? Did you fly across three time zones recently?

Any of those answers being “yes” is enough to explain the signal. This is the core literacy point of consumer EA surveillance: the signal is real, and so are the alternatives.

This post is a spoke in the energy availability and RED-S pillar. It catalogs the confounders that produce biomarker signatures indistinguishable from low EA and describes how to think about them.

Why Confounders Matter So Much Here

Most medical screening has the luxury of reasonable specificity. A positive pregnancy test almost never occurs without pregnancy. An elevated troponin is almost always cardiac. A high HbA1c is almost always diabetes or prediabetes.

The signals consumer wearables capture are not like that. HRV, RHR, and sleep are general-purpose markers of autonomic state. They respond to every common stressor the body encounters. They are useful for pattern recognition over time. They are not useful as single-signal diagnostics for specific conditions, including RED-S.

The underlying problem is that the autonomic shifts caused by low EA, luteal-phase progesterone, viral illness, alcohol, overreaching, sleep deprivation, and travel all push the autonomic system in similar directions. The physiology is not identical, but the downstream wearable readouts are.

The implication for surveillance is unambiguous: if you don’t explicitly account for these confounders, you will flag them as RED-S signatures. In internal evaluation of naive rule logic, flags fire on the majority of women during luteal phase, on anyone with a recent viral illness, and on anyone in a heavy training week. Without gating, the false-positive rate is high enough that the tool loses its value.

Omnio’s EA surveillance is designed around explicit confounder gating. The feature is flagged off until empirical false-positive rate (measured via user feedback on flags with confounder-affordance dismiss flow) drops below an acceptable threshold. This is deliberate and conservative. The rest of this post explains what each confounder looks like and how gating works in practice.

Confounder 1: Menstrual Cycle Phase (Luteal)

What happens. In the luteal phase of the menstrual cycle — roughly days 15 to 28 in a 28-day cycle, or the second half after ovulation — progesterone rises sharply, peaks around day 21, and then falls before menstruation. Progesterone has several autonomic effects.

• Resting heart rate rises 2–5 bpm above follicular-phase baseline.
• HRV drops proportionally — typically 10–20 percent reduction in RMSSD from follicular.
• Core body temperature rises by approximately 0.3–0.5 degrees C.
• Sleep quality often decreases, with more fragmentation and, in some women, reduced deep sleep.
• Resting metabolic rate increases by roughly 5 percent, which can alter subjective hunger and energy balance perception.

Published support. The autonomic effects of cycle phase are well-documented in physiology literature. Many wearable validation studies exclude female subjects or analyze them separately to avoid this confound. Cycle phase is the single strongest day-to-day modifier of autonomic signals in menstruating women.

How it mimics low EA. The directional changes — higher RHR, lower HRV, fragmented sleep — match the low-EA signature almost exactly. The magnitudes also overlap. Without phase awareness, any naive low-EA detector will fire on luteal days for most women.

How to handle it.

• Personally: Track your cycle and check phase before reading a data pattern. An elevated RHR on day 22 is expected; an elevated RHR on day 5 that persists into day 12 is different.
• Surveillance tools: Build phase-matched baselines. When at least 3 prior cycles of data are available, the personalized baseline for each biomarker is computed by cycle phase. Luteal-phase days are compared to luteal-phase baseline, not to the annual mean. When fewer than 3 cycles of data are available, default to suppressing autonomic-only flags during luteal phase.

The surveillance implementation for this is one of the harder pieces of the pipeline to get right, because cycles vary in length, ovulation day shifts across cycles, and hormonal contraception disrupts the biology entirely (pharmacologically controlled cycles remove the signal). The robust approach uses cycle data when reliable, falls back to not firing autonomic flags when cycle data is insufficient, and frames cycle-related patterns in the UI explicitly.

Men and non-menstruating populations do not have this confounder. Women on hormonal contraception have a different picture — there is no meaningful cyclic phase signal because the hormonal state is pharmacologically controlled — but this also means cycle phase cannot be used as a confounder-exclusion signal.

Confounder 2: Acute Illness

What happens. Viral illness (upper respiratory infection, gastroenteritis, any febrile illness) produces immediate and often pronounced autonomic shifts.

• HRV drops sharply, often within 24 hours of viral exposure and before overt symptoms.
• Resting heart rate rises, typically 5–15 bpm, and stays elevated through recovery.
• Sleep is disrupted — more awakenings, more fragmentation.
• Subjective fatigue and malaise accompany.
• Body temperature may rise.

Published support. Multiple studies have documented HRV and RHR changes as early warnings of viral illness. The same changes are observed in COVID infection, influenza, and common upper respiratory infections.

How it mimics low EA. The autonomic signature of illness is qualitatively similar to low EA — just acutely. The distinction is usually temporal: illness produces a sharp single-event deviation, while sustained low EA produces a gradual drift. But an illness onset followed by incomplete recovery can produce a multi-week trend that mimics low EA onset.

How to handle it.

• Personally: When you feel unwell or test positive for a virus, note it. Expect 1–2 weeks of autonomic suppression depending on severity. Ignore data patterns that coincide with or follow recent illness.
• Surveillance tools: Provide an easy illness self-report log. Suppress autonomic-based EA flags for 5–14 days following logged illness. The exact suppression window depends on severity; a 5-day cold and a 2-week flu should suppress the rule for different durations.

Omnio’s trackers support illness logging explicitly. A flag that fires during an illness-suppression window is by design impossible — the logic checks the tracker first.

Confounder 3: Alcohol Consumption

What happens. Alcohol has well-documented acute autonomic and sleep effects.

• HRV is suppressed for 24–48 hours after moderate drinking, with dose-dependence.
• Resting heart rate is elevated for a similar window.
• REM sleep is suppressed the night of drinking, often followed by rebound and poor overall sleep quality.
• Sleep architecture is disrupted in ways that consumer sleep scores reliably detect as “worse sleep.”

Published support. Pietilä et al. and subsequent work using consumer wearables have characterized alcohol effects on HRV at the population level. The dose-response is well-established.

How it mimics low EA. A week with one or two drinking occasions produces autonomic signal pulses that, averaged across the week, mimic a low-grade sustained autonomic depression. A weekly drinking pattern produces a chronic autonomic depression that can be easily confused with low-EA autonomic drift.

How to handle it.

• Personally: Log drinking days. Expect 24–48 hours of data impact per drinking event. Don’t read a Monday morning HRV in isolation without knowing what happened Saturday and Sunday.
• Surveillance tools: Provide alcohol consumption self-report. Suppress autonomic EA flags if drinking has occurred within the past 48 hours. For users with regular weekend drinking patterns, weight the weekday signal more heavily and discount weekend noise.

This is the kind of confounder that’s trivially easy to track and meaningfully moves the false-positive rate. The dismiss-with-confounder-affordance UX in Omnio’s flag surface asks explicitly about recent drinking before allowing a flag to persist.

Confounder 4: Acute Training Load Spikes

What happens. Hard training — PR attempts, long runs, tournament weekends, very high TRIMP weeks — produces transient autonomic suppression. The “fitness-fatigue” model from the training literature describes this explicitly: acute fatigue suppresses the readiness signal proportionally to training stress, and it resolves over 24–72 hours as fatigue dissipates.

• HRV drops for 24–72 hours after very hard sessions.
• RHR elevates for 24–48 hours.
• Sleep is sometimes disrupted the night following very hard late-day sessions.
• Subjective fatigue is elevated.

Published support. Foundational training-load literature (Banister, Hellard, Vermeire and colleagues on fitness-fatigue models). The phenomenon is central to how ACWR, monotony, and strain are computed. See what ACWR is and why it matters.

How it mimics low EA. A 2-week block of heavy training produces a sustained autonomic depression from training load alone. Without any under-fueling, the wearable signature looks like early-stage low EA. The distinction is that with adequate fueling and adequate recovery days, the autonomic signals recover; with insufficient fueling, they do not.

How to handle it.

• Personally: Know your training load. A heavy week is expected to suppress recovery markers. A heavy month is expected to cumulatively suppress recovery markers further. This is normal and planned. What’s not normal is continued suppression during a deload week.
• Surveillance tools: Compute weekly TRIMP z-score relative to the user’s baseline. When z-score is 2.0 or greater — a clear training spike — suppress the autonomic-only EA rule. Let the weight-trajectory rule continue to evaluate, since weight is less directly confounded by training load. Restore autonomic flag eligibility when training returns to baseline and signals haven’t recovered.

This is the training-load z-score gate baked into Omnio’s EA surveillance design. It’s a non-negotiable part of the flagging logic, because without it, every athlete in a build phase would receive false flags.

For the fuller picture of training load interpretation, see adaptive training intelligence and how wearables measure stress and strain.

Confounder 5: Travel and Jet Lag

What happens. Crossing time zones produces circadian desynchronization that affects multiple biomarkers.

• HRV is suppressed for several days to a week depending on direction and number of time zones crossed.
• Sleep is disrupted for the same window.
• RHR is modestly elevated.
• Subjective fatigue and malaise.

Published support. Broadly documented in circadian biology and jet-lag literature.

How it mimics low EA. Post-travel autonomic depression lasting 5–10 days can look like a low-EA onset to a naive detector. Frequent travelers can chronically appear to be in a low-grade autonomic depression.

How to handle it.

• Personally: Know when you traveled. Expect data impact for 1–7 days depending on trip.
• Surveillance tools: Suppress autonomic EA flags for ~7 days following detected travel (via GPS, calendar, or user-logged events). Count flight days as elevated-confounder days.

Confounder 6: Sleep Debt

What happens. Accumulated sleep deprivation lowers HRV, elevates RHR, and dominates subjective fatigue independently of energy availability.

• Single night of 4 hours sleep: next-morning HRV 10–20 percent below baseline.
• Multi-day sleep restriction: progressive HRV decline, RHR elevation, cognitive impairment.

Published support. Well-documented in sleep and autonomic literature.

How it mimics low EA. Sleep restriction produces the same signature. Someone who’s been averaging 6 hours of sleep for two weeks due to work pressure will look autonomically identical to someone with low EA.

How to handle it.

• Personally: If your sleep has been poor for an independent reason (work stress, new baby, travel), the data signals reflect that, not EA.
• Surveillance tools: Compute sleep debt as a rolling deficit from personalized sleep duration baseline. When sleep debt is substantial, suppress or heavily discount autonomic EA flags. The logic has to distinguish sleep debt caused by low EA (a consequence) from sleep debt that is causing the autonomic shift (a confounder) — the direction is different, but in practice both patterns warrant suppressing a confident flag.

Confounder 7: Psychological Stress

What happens. Chronic or acute psychological stress lowers HRV, elevates RHR, disrupts sleep, and produces subjective fatigue. The autonomic effect of a difficult period at work or in a relationship is real and measurable.

Published support. Extensive psychophysiology literature on autonomic effects of stress.

How it mimics low EA. Chronic stress produces an autonomic signature indistinguishable from mild low EA at wearable resolution.

How to handle it.

• Personally: Reflect on life stressors when reading your data. A bad patch at work is a sufficient explanation for a few weeks of autonomic depression.
• Surveillance tools: Subjective stress can be logged as a tracker input. When self-reported stress is elevated, weight the autonomic signal lower. This is a softer gate than illness or alcohol because self-reports of stress are less categorical, but the direction of handling is the same.

Confounder Layering

The real world usually has more than one confounder active. A luteal-phase week with a hard training block and a Saturday with drinks is the norm, not the exception. Each of these individually would explain some autonomic depression; together, they explain a lot.

A well-designed surveillance tool handles confounder layering explicitly. If two or more confounders are active, the threshold for firing an autonomic flag should be higher. The flag should surface not just the pattern but the confounders-to-rule-out list, so the user can quickly see what might be explaining the signal.

A poorly designed tool just fires on threshold. A user sees “low EA suspected” every three weeks for a year and stops listening.

How Omnio’s Gating Works, Concretely

The gating logic in Omnio’s EA surveillance (currently in shadow mode) applies in layered fashion:

1. Compute the raw compound signal — the multi-biomarker trajectory over 2–3 weeks, relative to personalized (and phase-matched for women) baselines.
2. Check illness tracker — if illness logged in past 5 days, suppress autonomic rule.
3. Check alcohol tracker — if drinking logged in past 48 hours above threshold, suppress autonomic rule.
4. Check training-load z-score — if weekly TRIMP z ≥ 2.0, suppress autonomic rule.
5. Check travel — if recent travel detected, suppress autonomic rule.
6. Check sleep debt — if substantial sleep deficit accumulated, suppress or discount autonomic rule.
7. Weight stall rule evaluates independently — unintentional weight decline over 3+ weeks during training is less confounder-vulnerable and is evaluated separately.
8. If the rule fires after gating, the flag is presented with the confounders-already-considered list and a confounder-dismiss option that asks the user to confirm whether the flag is already explained.
9. Dismissed flags with confounder reasons feed back into the calibration: they count toward the empirical false-positive rate and are used to tune the thresholds.

The full shadow-mode regime stays in place until the calibration shows that flags, once they fire, are meaningful a sufficient fraction of the time. This is not a product-launch schedule; it’s a correctness gate.

What You Can Do as a User

Log your confounders. If you track nothing else, track illness, alcohol, and travel. Those three logs eliminate the majority of false-positive autonomic signals.

Know your cycle. If you menstruate, keep cycle tracking current. Read autonomic data through the lens of phase.

Know your training load. Awareness of when you’re in a heavy block versus a deload week disambiguates most transient autonomic depressions.

Check your sleep debt. Before ascribing a data pattern to EA, check whether you’ve been sleeping 7+ hours consistently. If not, sleep is probably doing most of the work.

Use rolling averages. A 7-day or 14-day rolling trend is vastly more informative than any single day.

Take the flag as a starting point. If a pattern survives all your confounder checks and still looks like underfueling, it’s worth talking to a clinician. That’s the intended use.

Putting It Together

The biomarker signatures of sustained low EA are real. So are the signatures of luteal phase, illness, alcohol, heavy training, travel, and sleep debt. They look nearly identical in wearable data.

A responsible surveillance tool accounts for all of them explicitly. Without confounder gating, false-positive rates are too high for the tool to be trustworthy. With gating, the tool’s role becomes narrower but more defensible: flagging compound patterns that survive confounder analysis, framed as worth a clinical review rather than as diagnoses.

For the broader clinical context, see the energy availability and RED-S pillar and what is RED-S. For the biomarker-level detail, see biomarker signatures of underfueling. For response options, see refeed protocols. For broader training-load context, see adaptive training intelligence.