Biomarker Reliability, Repeatability, and Measurement Error

Key Takeaways

Reliability describes how well measurements distinguish stable differences between people despite measurement noise. [1]
Repeatability concerns agreement under closely matched conditions; reproducibility asks whether results remain comparable when conditions change. [2]
Observed change can combine true biological change, short-term within-person variation, pre-analytical effects, and analytical error. [3]
A high correlation between repeated measurements does not by itself prove close agreement or a small error in the units being measured. [4]

Who This Is Useful For

This page is useful for readers comparing repeated biomarker tests, interpreting changes in biological age estimates, or evaluating whether a measure is suitable for longitudinal research. These questions require more than asking whether a biomarker predicts an outcome: they require evidence about the stability and uncertainty of the measurement itself. [1] [5]

Three Related but Different Concepts

Concept	Main Question	Common Evidence	Main Limitation
Reliability	Does the measure consistently distinguish people from one another?	Intraclass correlation coefficient	Depends on variation between people in the study sample
Repeatability	How close are repeated results under the same conditions?	Within-subject standard deviation, coefficient of variation, repeatability coefficient	May not describe performance across laboratories or devices
Reproducibility	How close are results when relevant conditions change?	Agreement across laboratories, operators, platforms, or protocols	The conditions included must be stated explicitly
Measurement error	How far can an observed value lie from the underlying value because of the measurement process?	Standard error of measurement, limits of agreement, analytical imprecision	May be difficult to separate from real short-term biology

Terminology varies across fields, but the central distinction is between consistency relative to other people and agreement in the biomarker's original units. Quantitative-imaging guidance, for example, treats repeatability as measurements made under the same conditions and reproducibility as measurements made under changed conditions. [2] [4]

Where Variation Enters a Biomarker Result

A reported biomarker value is shaped by the underlying biological quantity and several additional sources of variation. Within-person biology can fluctuate over hours, days, or seasons. Collection time, posture, recent activity, specimen handling, transport, and storage can add pre-analytical variation, while calibration, reagent lots, instruments, and laboratory imprecision contribute analytical variation. [3] [6]

These components matter differently by biomarker. In one longitudinal study, soluble tumour-necrosis- factor receptors had substantially higher long-term intraclass correlations than C-reactive protein and interleukin-6, showing that one blood draw can represent some inflammatory markers more reliably than others. [7]

Why Correlation Is Not Agreement

Pearson correlation measures whether two sets of values rise and fall together. Two test occasions can therefore correlate strongly even if the second is systematically higher than the first. Agreement methods such as Bland-Altman analysis examine the differences between paired measurements and can reveal fixed bias, variation in error across the measurement range, and limits within which most paired differences are expected to lie. [4]

The intraclass correlation coefficient is useful for relative reliability, but its magnitude depends on between-person heterogeneity. A diverse sample can produce a high ICC even when absolute error is too large for tracking small changes within an individual. Reporting an absolute error statistic alongside the ICC therefore answers a different and necessary question. [2] [8]

Interpreting Change Over Time

The difference between two results is not automatically a true biological shift. In laboratory medicine, a reference change value combines analytical imprecision and within-person biological variation to estimate how large a serial difference must be before random variation is an unlikely explanation at a chosen probability level. Pre-analytical variation may also affect the change observed in routine practice. [3] [9]

Reference change values are analyte- and context-specific rather than universal thresholds. They depend on the quality of the biological-variation data, the assay, the population, distributional assumptions, and the confidence level selected. Recent work also shows that regression toward the population mean can make the expected change asymmetric for results near the edges of a reference interval. [9] [10]

Implications for Ageing Biomarkers

Ageing studies often seek gradual longitudinal changes, so measurement error can attenuate associations, misclassify participants, and make small apparent changes difficult to distinguish from noise. Composite clocks add further dependencies on preprocessing, laboratory platform, and model specification. Validation frameworks for biomarkers of ageing consequently include reliability and responsiveness alongside prediction and biological relevance. [1] [5]

Repeated measurements can estimate a person's typical level more precisely when errors are sufficiently independent, but repetition does not correct systematic bias or a protocol change. Comparability is strongest when collection conditions, assay platform, processing, and analysis are standardized and documented. [1] [6]

Evidence Quality and Interpretation

Confidence is strong that biomarker interpretation requires both relative reliability and absolute-error information; these address different measurement properties. [2] [4]

Confidence is also strong that biological, pre-analytical, and analytical variation all contribute to serial laboratory results. Their relative size, however, differs by analyte, protocol, population, and time interval. [3] [7]

Confidence is more limited when a reliability estimate from one cohort is transferred directly to a different population or testing system, because ICCs and reproducibility estimates are conditional on the sample and measurement conditions studied. [2] [5]

What This Does Not Mean

It does not mean every difference between repeated tests is meaningless; it means the difference should be compared with relevant uncertainty. [3]
It does not mean a highly repeatable biomarker is necessarily valid for ageing or predictive of health outcomes. [1]
It does not mean a lower ICC always reflects a worse assay; restricted differences between study participants can lower the ICC. [2]
It does not mean averaging repeats removes systematic calibration or protocol bias. [4] [6]

Practical Interpretation Examples

If a biological-age score changes slightly: the change is not interpretable without evidence on test-retest error for that assay, platform, and time interval. [5]
If repeated results have a high correlation: inspect agreement and absolute error before concluding that small individual changes can be tracked. [4] [8]
If samples were processed by different laboratories: the comparison includes reproducibility across laboratories, not only repeatability under one protocol. [2]

Summary

A biomarker result is an estimate, not a direct view of an error-free biological state. Reliability describes consistency relative to differences between people; repeatability and reproducibility describe agreement under specified conditions; and absolute-error measures indicate how much apparent change may arise without a true change. Ageing biomarkers are most interpretable when studies report these properties, standardize measurement conditions, and match uncertainty estimates to the intended use. [1] [2] [5]

References

Mayeux, R. (2004). Biomarkers: potential uses and limitations. NeuroRx, 1(2), 182-188. https://pmc.ncbi.nlm.nih.gov/articles/PMC534923/
Obuchowski, N. A., et al. (2022). Statistical considerations for repeatability and reproducibility of quantitative imaging biomarkers. Translational Oncology, 19, 101375. https://pmc.ncbi.nlm.nih.gov/articles/PMC9667479/
Lacher, D. A., Hughes, J. P., & Carroll, M. D. (2005). Estimate of biological variation of laboratory analytes based on the Third National Health and Nutrition Examination Survey. Clinical Chemistry, 51(2), 450-452. https://pubmed.ncbi.nlm.nih.gov/15681571/
Bland, J. M., & Altman, D. G. (1986). Statistical methods for assessing agreement between two methods of clinical measurement. The Lancet, 327(8476), 307-310. https://pubmed.ncbi.nlm.nih.gov/2868172/
Moqri, M., Herzog, C., Poganik, J. R., et al. (2024). Validation of biomarkers of aging. Nature Medicine, 30(2), 360-372. https://www.nature.com/articles/s41591-023-02784-9
Vaught, J. B. (2006). Blood collection, shipment, processing, and storage. Cancer Epidemiology, Biomarkers & Prevention, 15(9), 1582-1584. https://aacrjournals.org/cebp/article/15/9/1582/174993/Blood-Collection-Shipment-Processing-and-Storage
Hardikar, S., et al. (2014). Intraindividual variability over time in plasma biomarkers of inflammation and effects of long-term storage. Cancer Causes & Control, 25(8), 969-976. https://pmc.ncbi.nlm.nih.gov/articles/PMC4169209/
Weir, J. P. (2005). Quantifying test-retest reliability using the intraclass correlation coefficient and the SEM. Journal of Strength and Conditioning Research, 19(1), 231-240. https://pubmed.ncbi.nlm.nih.gov/15705040/
Jones, G. R. D., et al. (2025). Estimating reference change values using routine patient data: a novel pathology database approach. Clinical Chemistry, 71(2), 307-317. https://academic.oup.com/clinchem/article/71/2/307/7874401
Jones, G. R. D., & Aarsand, A. K. (2024). A new concept for reference change values—regression to the population mean. Clinical Chemistry, 70(8), 1076-1084. https://academic.oup.com/clinchem/article/70/8/1076/7679834

Educational Disclaimer

This content is provided for educational purposes only and does not constitute medical advice.