How to Build a Biomarker Panel That Tracks Your Own Training

Most athlete biomarker panels fail for one of two reasons: the blood was drawn at a time that made the result uninterpretable, or the reader compared a single value to a population reference range instead of to their own history. The fix is a small panel measured repeatedly under standardized conditions, analyzed against a personal baseline with an explicit noise model. Roughly a dozen markers, sampled every eight to twelve weeks plus targeted draws around hard training blocks, will tell you more than a 90-analyte panel run once.

This page covers what to include and when to draw. It also covers how to decide whether a change is real and how to model the resulting time series. Interpretation of any abnormal result belongs with a clinician. The analysis below is about measurement quality and inference. It does not cover diagnosis.

What belongs on the panel

Group markers by the question they answer, and accept that most questions need two or three markers rather than one.

Iron status is the highest-yield block for endurance athletes because deficiency is common and correctable under medical supervision. Measure ferritin, transferrin saturation (serum iron plus total iron binding capacity), soluble transferrin receptor, and hemoglobin together. Ferritin alone is unreliable because it rises with inflammation, so pair it with high-sensitivity C-reactive protein (hsCRP) and discard iron interpretation when hsCRP is elevated.

Muscle damage and inflammation are covered by creatine kinase (CK), lactate dehydrogenase, and either hsCRP or interleukin-6 if you can get it. CK is a kinetics marker rather than a level marker. After unfamiliar eccentric work it rises over 24 to 48 hours, and the peak varies by more than an order of magnitude between people with the same training load. A single CK value without a known time since the last session is close to uninformative.

For the endocrine axis, measure morning serum cortisol along with total and free testosterone and sex hormone binding globulin. Add either IGF-1 or thyroid markers if you suspect energy deficiency. Cortisol is the most frequently overinterpreted number on any athlete panel. Its diurnal amplitude and day-to-day variation are large relative to training-induced shifts, which is why single-sample cortisol tracks endurance load poorly and repeated standardized sampling is required to see anything real.¹

Round this out with a metabolic block.

• fasting glucose
• HbA1c
• a lipid panel
• vitamin D
• magnesium
• kidney and liver function

If you wear a continuous glucose monitor, the CGM series replaces fasting glucose as the more informative measurement and adds post-exercise and overnight dynamics that a single draw cannot show.

Pre-analytical control is most of the signal

The largest source of variance in athlete biomarker data is the condition of the athlete at the draw. The laboratory is not the largest source. Fix the protocol and hold it constant.

Draw fasted in the morning within a one-hour window. Stay seated and rested for at least ten minutes beforehand. Take no hard training for 24 hours and no unusual eccentric loading for 72 hours. In the same row as the results, record the exact clock time and hours since the last session. Record session type, sleep duration, and hydration state as well.

Posture and plasma volume matter more than people expect. Moving from supine to standing shifts plasma volume by several percent, which changes every concentration-based measurement, including hemoglobin and ferritin. Endurance training also expands plasma volume over weeks, so a falling hemoglobin concentration during a base block can reflect dilution rather than any change in total hemoglobin mass. If hemoglobin is the number you care about, the mass measurement (via carbon monoxide rebreathing) is the one that answers the question.

Deciding whether a change is real

Use the reference change value rather than the population reference interval. The reference change value combines analytical and biological variation:

RCV = 1.96 * sqrt(2) * sqrt(CVa^2 + CVi^2)

Here CVa is the assay coefficient of variation from your lab and CVi is within-subject biological variation. For ferritin, with CVa near 5% and CVi near 15%, the RCV is about 44%. A move from 45 to 55 ng/mL is noise. A move from 45 to 22 ng/mL is a real change and worth a conversation with a physician.

Ask your lab for its actual analytical CV per assay and, where possible, batch your samples through one laboratory and one platform. Switching vendors mid-series introduces a step change that will look like a physiological event. If you must switch, run a split sample through both labs so you can estimate the offset.

Once you have five or more clean draws, estimate your own CVi directly and stop relying on published values. This is the point at which personal data starts to outperform reference ranges: your own dispersion is usually narrower than the population’s, so smaller deviations become detectable.

Modeling the series

Keep results in one tidy CSV with one row per draw and one column per analyte, plus the covariate columns from your draw protocol. Store the unit and the assay method alongside each value, because unit conventions differ between labs and silent conversions are a common failure mode.

For analysis, we would fit a per-analyte mixed model in R with lme4 or in Python with statsmodels, using training load as a fixed effect and week as a smooth term. With shorter series, a simpler approach works well: compute a rolling personal z-score against your own mean and standard deviation, then flag any value beyond the RCV. Change-point detection with the ruptures package is useful for finding the week a trend turned, which is often before you noticed symptoms.

A Bayesian personal-baseline approach is the strongest option and is exactly what anti-doping programs use for longitudinal hematology. Each new observation updates a posterior for your individual mean, and the flag threshold is defined on that posterior rather than on a population range. PyMC handles this in a few dozen lines.

Markers that track specific adaptations

Some markers answer a narrow, useful question about whether a specific intervention is working.

For altitude or normobaric hypoxic training, the relevant question is whether erythropoiesis is responding. Erythropoietin, reticulocyte count, and iron availability at camp entry are the candidate predictors of who gains hemoglobin mass. Pre-camp iron status is the single most actionable one.² For heat acclimation, the interesting result is that short protocols can improve cardiovascular and thermoregulatory function without moving kidney stress markers. A normal creatinine or injury-marker panel after a heat block is therefore expected rather than reassuring evidence of nothing happening.³

Two research-stage measurements are worth watching. Cell-free circulating mitochondrial DNA rises with exercise stress and is being evaluated as a marker of physiological load and adaptation in athletes.⁴ Untargeted metabolomics by LC-MS or NMR gives a broad readout of substrate use and recovery, though annotation confidence and panel standardization remain the limiting factors for individual interpretation.⁵ Treat both as exploratory rather than as decision inputs.

Sex-specific timing

For athletes who menstruate, cycle phase is a covariate you must record. It is not an inconvenience. Markers of muscle damage and inflammation vary across phases, and studies that ignore phase produce conflicting results for that reason alone.⁶ Cycle length itself is variable in trained populations and correlates with other physiological measures, so a single assumed 28-day template will misclassify phases.⁷

The practical rule is to log cycle day at every draw and, once you have enough data, either model phase as a factor or hold it constant by always drawing in the same phase. The second option is simpler and costs nothing.

What genetics adds

A whole-genome sequence is a fixed backdrop rather than a training signal. Common variants associated with trainability and exercise response explain small fractions of variance. No genotype panel currently predicts an individual training outcome well enough to change a program.⁸ What the genome does provide is context for blood results. Variants affecting iron handling, lipid metabolism, or drug and caffeine metabolism change how you read a marker. Pharmacogenetic findings in particular are for a clinician to act on. They are not for self-directed adjustment.

Questions people also ask

How often should I test? Every eight to twelve weeks for the standing panel, with extra draws before and after a training camp or altitude block. Higher frequency without protocol control adds noise rather than resolution.

Is a single elevated CK a problem? On its own it means very little without the time since your last session and the type of loading. Persistently high CK with dark urine or severe weakness is a reason to seek medical care promptly.

Can cortisol tell me if I am overtrained? Not from one sample. Cortisol’s variability is large relative to training effects, so only repeated standardized morning measurements interpreted alongside performance and sleep data carry any signal.¹

Should I use saliva or dried blood spots instead of venipuncture? They improve compliance and are reasonable for frequent hormone or metabolite sampling, but analytical CVs are generally higher. Keep one method per series so your baseline stays comparable.

What does a CGM add for an athlete? Overnight and post-session glucose dynamics, carbohydrate tolerance under fatigue, and fueling response during long sessions. None of that is visible in a single fasting glucose value.

Footnotes

1. Luiza Stadnik, Alicja Kozłowska, Karolina Przybysz, et al. Cortisol as a Blood Biomarker for Monitoring Endurance Training - A Narrative Review. Quality in Sport, 2026. https://doi.org/10.12775/qs.2026.65.74045 ↩ ↩²

2. Simon Klügel, Svenja Nolte, Sebastian Hacker, et al. Biomarkers predicting the effectiveness of normobaric altitude training. European Journal of Applied Physiology, 2025. https://doi.org/10.1007/s00421-025-06088-3 ↩

3. Daniel Snape, Barney Wainwright, Iain T. Parsons, et al. Five days of heat acclimation improves cardiovascular and thermoregulatory responses without altering renal stress biomarkers in endurance athletes. PLOS One, 2026. https://doi.org/10.1371/journal.pone.0345346 ↩

4. Arpan Chattopadhyay, Patrick Diel, Hemanth Naick B. Cell-Free Circulating Mitochondrial DNA: A Biomarker of Physiological Stress and Adaptation in Athletes. International Journal of Sports Medicine, 2026. https://doi.org/10.1055/a-2929-9682 ↩

5. Donghai Lin, Yifen Chen, Caihua Huang. Exercise and Sports Metabolomics: Analytical Platforms, Metabolite Annotation, Pathway-Level Interpretation, and Biomarker-Panel Readiness. Metabolites, 2026. https://doi.org/10.3390/metabo16080530 ↩

6. Farwa Baber, Jana Jaklová Dytrtová, Michal Jakl, et al. Biomarkers of exercise-induced muscle damage and inflammation across menstrual cycle phases: a critical review. Frontiers in Sports and Active Living, 2026. https://doi.org/10.3389/fspor.2026.1878668 ↩

7. Ragini Adhikari, Chayan Kundu, Magali Giacomoni, et al. Menstrual Cycle Length Variability and Its Physiological Correlates in Indian Combat Athletes: An Observational Study. Women in Sport and Physical Activity Journal, 2026. https://doi.org/10.1123/wspaj.2026-0061 ↩

8. Dhurata Bozo, Suela Xhufi. Biomarkers and Genetic Factors Related to Physical Activity and Its Health Benefits. Albanian Journal of Natural and Technical Sciences, 2022. https://doi.org/10.67537/ajnts2702021 ↩