Journal List > Ann Lab Med > v.45(3) > 1516090416

TOOLS
Coskun: Are Your Laboratory Data Reproducible? The Critical Role of Imprecision from Replicate Measurements to Clinical Decision-making
Review Article
Laboratory Informatics

Ann Lab Med 2025;45(3):259-271.

Published online: 21 March 2025

DOI: https://doi.org/10.3343/alm.2024.0569

Are Your Laboratory Data Reproducible? The Critical Role of Imprecision from Replicate Measurements to Clinical Decision-making

Abdurrahman Coskun, M.D.

Department of Medical Biochemistry, School of Medicine, Acibadem Mehmet Ali Aydinlar University, Istanbul, Turkey

Corresponding author: Abdurrahman Coskun, M.D. Department of Medical Biochemistry, School of Medicine, Acibadem Mehmet Ali Aydinlar University, Kayisdagi cad., No: 32, 34752 Atasehir – Istanbul, Turkey E-mail: coskun2002@gmail.com

Received 21 October 2024       Revised 7 January 2025       Accepted 4 March 2025

© Korean Society for Laboratory Medicine

(open-access):

This is an open-access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract

Measurement results of biological samples are not perfect and vary because of numerous factors related to the biological samples themselves and the measurement procedures used to analyze them. The imprecision in patients’ laboratory data arising from the measurement procedure, known as analytical variation, depends on the conditions under which the data are collected. Additionally, the sample type and sampling time significantly affect patients’ laboratory results, particularly in serial measurements using samples collected at different time points. For accurate interpretation of patients’ laboratory data, imprecision—both its analytical and biological components—should be properly evaluated and incorporated into data management. With advancements in measurement technologies, analytical imprecision can be minimized to an insignificant level compared to biological imprecision, which is inherent to all biomolecules because of the dynamic nature of metabolism. This review addresses: (i) the theoretical background of variation, (ii) the statistical and metrological evaluation of measurement variation, (iii) the assessment of variation under different conditions in medical laboratories, (iv) the impact of measurement variation on clinical decisions, (v) the influence of biases on measurement variation, and (vi) the variability of analytes in human metabolism. Collectively, both analytical and biological imprecision are inseparable aspects of all measurements in biological samples, with biological imprecision serving as the foundation of personalized laboratory medicine.

Keywords: Analytical variation, Bias, Biological variation, Imprecision, Measurement uncertainty, Precision

INTRODUCTION

Measurement results of analytes in biological samples are not absolute values but display inherent variability, which is a characteristic of all laboratory-generated data [1]. This variability becomes evident when the same analyte is measured repeatedly, even under standardized conditions within the same laboratory. Elevated variation frequently arises from suboptimal measurement procedures, encompassing both pre-analytical and analytical processes. As variability correlates with uncertainty, increased variation across repeated measurements reduces the reliability of data for clinical decision-making.
Laboratories implement QC procedures to monitor variability in measurement results and to adopt corrective actions when variability exceeds acceptable thresholds. The progression from raw data to clinical decision-making is complex and prolonged. Therefore, QC rules are implemented to minimize variability in measurement procedures to an acceptable or negligible level, thus mitigating their negative impact on clinical decisions [2–6]. A comprehensive understanding of variability in laboratory data is essential for ensuring data quality and enhancing clinical applications.
Variation arises from multiple sources and exhibits a multidimensional nature. Therefore, addressing variation effectively requires detailed root-cause analysis and a systematic approach (Fig. 1). This review addresses: (i) the theoretical principles underlying variation, (ii) the statistical and metrological evaluation of measurement variation, (iii) variation assessment under diverse conditions in medical laboratories, (iv) the implications of measurement variation for clinical decisions, (v) the impact of biases on measurement variation, and (vi) the variability of analytes in human metabolism.

TERMINOLOGY

Variation is a broad term describing deviations in measurement results. Terms such as imprecision, precision, bias, trueness, and accuracy refer to distinct aspects of variation but are often used interchangeably. For example, terms such as accuracy and trueness, or imprecision and precision, are frequently used interchangeably despite having distinct meanings. To eliminate ambiguity and promote consistent terminology, an authoritative international reference document is essential. The International Vocabulary of Metrology (VIM) serves as the most reliable resource in this field. The current edition, VIM 3 [7], is widely utilized, with an updated version, VIM 4, expected to be released soon.

Precision

The VIM [7] and the International Organization for Standardization (ISO) standards ISO 3534 [8] and ISO 5725 [9] define precision as the “closeness of agreement between independent test or measurement results obtained under specified conditions.” Measurement conditions determine the classification of precision into three primary categories: repeatability, intermediate precision, and reproducibility.

Imprecision

Imprecision reflects the degree of inconsistency in repeated measurements or observations of the same quantity and is defined as the extent to which repeated measurements of the same sample under specified conditions yield different results. Precision and imprecision are often used interchangeably in scientific literature despite denoting different concepts. Statistically, calculated parameters typically reflect imprecision rather than precision.

Measurement accuracy

The VIM [7] defines measurement accuracy as the “closeness of agreement between a measured value and a reference value of a measurand.” Measurement accuracy pertains to a single measurement result rather than the mean of repeated measurements. Accuracy represents the difference between an individual single measurement result and a reference or target value. The calculation of accuracy must account for the various types of variation affecting a single measurement result. For a single measurement, deviation from the reference value arises from two primary factors: bias and imprecision. The relationship can be expressed as:
(1)
A=bias+k × imprecision
where k corresponds to the coverage factor of the statistical distribution used to calculate accuracy. For a normal distribution, k=1.96 for a 95% probability. This equation demonstrates that accuracy corresponds to the total error (TE) [10]. As accuracy combines bias and imprecision, these terms should not be conceptually conflated, even when both fall within acceptable limits.

Trueness

The VIM [7] defines trueness as the “closeness of agreement between the average of measured values obtained by replicate measurements and a reference value.” Achieving both accuracy and trueness requires a reference or target point. However, trueness requires repeated measurements of the analyte, whereas accuracy does not.

RANDOMNESS IN REPEATED MEASUREMENT RESULTS

Randomness describes the inherent unpredictability of certain processes or phenomena that can only be described probabilistically. Repeated measurements of the same sample, performed under identical methods, instruments, reagents, and laboratory conditions, typically produce varying results. Analyzing these repeated measurements reveals a random distribution of values within a defined interval. Theoretically, when all components of the measurement procedure (e.g., methods, reagents, instruments, and laboratory conditions) are identical, repeated measurements of the same sample should yield the same results. However, in practice, repeated measurements of the same sample under identical conditions typically yield different results. Analyzing these repeated measurements reveals a random distribution of data within a specified interval. Consistency among repeated measurements is realistic and expected, whereas exact sameness is not achievable.
Randomness in repeated measurement results tends to increase over time, contrary to intuitive expectations that prolonged measurements would reduce randomness. The observed increase in randomness primarily results from the degradation of components within the measurement system. Such degradation arises from complex microscopic interactions, including molecular motions, environmental influences, external interventions, and their cumulative effects. These factors contribute to the increasing entropy of the measurement system [11–14]. According to the second law of thermodynamics, entropy tends to increase over time. Local decreases in entropy are possible but are accompanied by corresponding increases elsewhere. To maintain randomness in repeated measurements within a narrow range, measurement systems are calibrated as needed, and QC materials and procedures are employed to monitor system repeatability. Over time, the repeatability of measurement systems declines, leading to reduced stability [15–17]. This decline in stability is observed not only in measurement systems but also in biological samples.

METROLOGICAL AND STATISTICAL PERSPECTIVES OF IMPRECISION

Measurement conditions

Measurement imprecision depends on several factors, including instruments, reagents, environmental conditions, timing, and laboratory personnel. The specific conditions under which measurements are conducted influence imprecision, making the data collection process critical for calculating the SD of the distribution. Laboratory medicine involves data collection from different sources and at varying time intervals. Recognizing the complex effects of internal and external factors on data dispersion is essential. Therefore, a comprehensive evaluation of all relevant variables is required when calculating the SD of a distribution. Among these variables, time, location, and measurement methods or instruments exert significant influence. Measurement conditions are categorized into three primary types, as detailed in the subsequent sections.

Repeatability conditions

Under repeatability conditions, analyses are conducted within a short time interval, whereas factors such as methods, instruments, reagents, personnel, and location remain constant [18, 19] (Fig. 2A). Variation in measurement results under these conditions is minimal compared with that under other conditions. However, bias, when present, reaches its maximum under repeatability conditions, making its contribution to total variation (TV) most evident. While imprecision is minimized, bias remains significant.

Intermediate precision conditions

Intermediate precision conditions represent an intermediate state between repeatability and reproducibility conditions. These conditions involve longer intervals, such as days or months, with the laboratory location remaining constant. Variability arises from changes in instruments, reagents, or personnel. Imprecision under these conditions exceeds that observed under repeatability conditions. However, bias may occasionally resemble random variation and become challenging to detect, particularly when small biases result from environmental or procedural changes (Fig. 2B). Given the limited applicability of repeatability in routine laboratory practice, experimental designs and QC protocols prioritize intermediate precision by conducting tests across multiple days rather than within a single day.

Reproducibility conditions

Reproducibility conditions involve variations in nearly all factors influencing measurement results, including laboratory locations while maintaining a consistent measurement method. The time interval under reproducibility is long, similar to that under intermediate precision conditions, and may extend over days, months, or seasons. Imprecision under reproducibility conditions is the largest, with bias behaving as a random variable and contributing to random variation.
Under repeatability conditions, repeated measurements are obtained from a single instrument over a short period. Intermediate precision conditions involve multiple instruments over a longer period within a single laboratory. Reproducibility conditions extend this further by involving multiple laboratories over an extended time period.
Reproducibility and repeatability represent the two extremes of imprecision. Reproducibility corresponds to the highest degree of imprecision, whereas repeatability represents the lowest. Intermediate precision includes all measures that fall between reproducibility and repeatability (Fig. 2C) [18–22].

Mathematics of imprecision

Statistics is founded in probability theory and employs advanced mathematical principles, contrary to common misconceptions. A population and the data derived from it or obtained through serial sampling represent distributions [23, 24]. A distribution is characterized by central tendency and dispersion surrounding this central point. Various types of distributions are used in statistical analysis, with the normal distribution being one example (Fig. 3) [24]. The central tendency of a distribution is represented by the mean, the arithmetic average of all data points; the median, the middle value when data are ordered; and the mode, the most frequently occurring data point. In contrast to the central tendency of a distribution, the mathematical representation of data dispersion poses challenges. The dispersion of a distribution is generally expressed in terms of the SD or percentiles and quartiles. The precision of the measurement procedure is inversely proportional to the magnitude of the SD: a large SD indicates low precision, whereas a low SD indicates high precision.
The mean separates the distribution into two parts. Calculating the difference between each data point and the mean and summing these differences results in zero because of the cancellation of positive and negative signs. To address this, the differences are squared before summation, yielding the conventional equation for SD:
(2)
SD=1N∑i=1Nxi−μ2
(3)
where μ is the mean and μ=1N∑i=1Nxi
In personalized laboratory medicine, the dataset is small, and if additional data is unavailable, this equation can be used in SD calculations. However, in conventional laboratory medicine, which is based on population data, a slightly different equation is used to calculate the SD of the distribution:
(4)
SD=1N−1∑i=1Nxi−μ2
The difference between equations 2 and 4 is the N–1 in the denominator of equation 4 versus N in the denominator of equation 2. Using N–1 instead of N, known as Bessel’s correction, reduces bias in population variance estimation [25].
In laboratory medicine, CV, a dimensionless parameter defined as the ratio of the SD to the mean of the analyzed data, is frequently used to express imprecision instead of SD. The equation for CV is given below:
(5)
CV%=SDX¯×100
where X¯ is the mean of repeated measurements.
While the normal distribution is commonly assumed in healthcare data analysis, most laboratory data are asymmetrically distributed and often exhibit skewed distributions around the mean. Similar to the normal distribution, the lognormal distribution is also applied. In this case, the logarithm of the data, rather than the data itself, follows a normal distribution [26–28]. For lognormal distributions, the SD and mean are calculated using the following equations [29, 30]:
(6)
SD=eμ+σ22eσ2−1
(7)
X¯=eμ+σ22
where σ is the SD of ln(X¯).
These equations estimate imprecision under repeatability conditions, which are rare in routine practice. Imprecision under intermediate precision or reproducibility conditions requires alternative approaches [31]. Two primary methods are used:
Simplified approach: Data from multiple instruments, methods, or laboratories are combined after outlier analysis, treating them as a single dataset to estimate imprecision. Equal subgroup sizes (e.g., laboratories or instruments) are essential for accuracy. Unbalanced data require more complex mathematical methods [32–34].
Detailed analysis: Variables such as instruments, laboratories, and days are incorporated into the calculation, with imprecision estimated using ANOVA. This method is particularly useful for assessing differences among sites or instruments. The CLSI EP05 guideline provides detailed instructions for modeling and calculating imprecision using this method [35].

IMPRECISION FROM A LABORATORY PERSPECTIVE

Criteria for the acceptability of imprecision in a measurement procedure

In routine practice, accurate measurement of analytes in patient samples requires high-quality instruments. Therefore, the analytical performance, evaluated through analytical bias and imprecision, is critical. Total allowable error (TEa) is a parameter for analytical performance specifications (APSs) but has been the subject of long-standing debate. TEa is not defined in the VIM [7] and has not resolved any issues in the QC of medical laboratories. Furthermore, it lacks a solid scientific foundation, both mathematically and clinically. The sign of bias (+ or –) is critical in TEa, and TEa cannot be added to measurement results as ± TEa, as is done for measurement uncertainty (MU). Therefore, TEa cannot be used as a substitute for MU. It has been used as an accessory to bias and imprecision rather than as a useful tool, and is not commonly used in modern medical laboratories. Imprecision, a key component of APSs, must be managed effectively to prevent adverse effects on patient test results.
Acceptable criteria for APSs of measurement procedures in clinical laboratories have long been debated, mainly during two international conferences. The first conference, held in Stockholm in 1999, established criteria referred to as the Stockholm Consensus [36]. In this conference, a hierarchical five-tier model was proposed for determining the acceptability criteria for APSs. Priority was assigned to the model evaluating the impact of APS on patient outcomes [37, 38]. When the primary model was unsuitable for a specific analyte, subsequent models in the hierarchy were recommended for the APS of the measurand.
In 2014, the European Federation of Clinical Chemistry and Laboratory Medicine (EFLM) convened a strategic conference in Milan, resulting in the Milan Consensus. The Milan Consensus retained the core approach of the Stockholm Consensus but introduced a simplified and condensed three-tier model [39]:
Model 1. APS based on the effect of analytical performance on clinical outcomes
Model 2. APS based on the components of biological variation (BV) of the measurand
Model 3. APS based on the state-of-the-art
The three models are not independent of each other but are interactive [40]. According to the Milan Consensus, preference should be given to models 1 and 2; however, combining different models can be advantageous in certain situations.
Despite significant progress with model 1 [41, 42], its implementation in routine practice remains challenging. This is because a single analyte often serves multiple purposes across clinical scenarios, complicating the application of a standardized model for acceptable APSs based on diverse clinical outcomes. Although model 1 is theoretically optimal, it is not yet widely implemented in practice. However, studies are ongoing to integrate this model into routine laboratory practice [40–44].
Given the above challenges, model 2, which bases APSs on BV, has been widely adopted in medical laboratories. This model assumes that the BV of an analyte represents natural variation, whereas measurement-derived variations are considered noise. Analytical variation is deemed acceptable when the noise level remains insignificant or tolerable compared to natural BV. This approach is logical, as within-subject BV (CVI) serves as a reference for evaluating measurement procedures, including instruments, reagents, and methods. It provides flexibility for most analytes since natural variation is inherent to biological samples. Therefore, significant resource allocation to minimize imprecision to extremely low levels is unnecessary when such reductions do not significantly enhance test results in clinical practice.
While model 2 is beneficial for estimating APSs, its implementation in routine practice is challenging. First, reliable BV data are essential [45–47]. Since BV is intended for universal use, a reliable and accessible database of commonly requested measurands is crucial for successfully applying model 2. To achieve this, the EFLM BV Committee conducted a multinational project [48], the European Biological Variation Study (EuBIVAS), to estimate BV for numerous measurands [49–54]. The EFLM BV Technical Committee also performed a meta-analysis of published BV data [53, 55, 56] and launched a database containing reliable BV values for most measurands [57].
Second, for certain analytes, particularly those with extremely low CVI, such as sodium [59], or those with high CVI, such as C-reactive protein (CRP), the BV model is not applicable. Analytes with extremely low CVI cannot be implemented because of the limited capacity of measurement procedures. Analytes with a high CVI are not useful in medical practice because of the negative effect of increased analytical imprecision on clinical outcomes, particularly in disease monitoring, clinical classifications, and medical decisions affecting patient outcomes. To address this problem, a three-level model based on BV data, with optimum, desirable, and minimum performance, has been proposed. For the analytical CV (CVA), the reference is the CVI of the analyte, and a cofactor is used to estimate the acceptable quality level.
(8)
CVA=k×CVI
For optimum, desirable, and minimum performance, k=0.25, 0.50, and 0.75, respectively.
The logic behind this approach is the contribution of analytical noise to the TV, which includes both BV and variation from the analytical system. The TV can be estimated as follows:
(9)
TV=CVA2+CVI2
Using equation 9, we can calculate the contribution of CVA to the TV. For k values of 0.25, 0.50, and 0.75, the TV will increase by 3%, 12%, and 25%, respectively.
The CVI varies significantly among different analytes. For example, the CVI of CRP is 33.7%, whereas for sodium, it is 0.5%. A single model applicable to all analytical systems based on CVI is not feasible, and using this triple approach may not be sufficient.
The primary aim is to establish analytical goals and implement assays that meet these objectives [58]. For analytes with high CVI, the cofactor for optimum performance (k=0.25) is recommended, whereas for those with low CVI, the cofactor for minimum performance (k=0.75) is preferred [1]. This approach introduces flexibility into the model but does not resolve challenges associated with analytes exhibiting extremely low or high CVI levels. For example, the CVI for Na is 0.5%, and even when using the cofactor for minimum performance (k=0.75), the CVA would be 0.38%, which is difficult to achieve with state-of-the-art instruments. Similarly, the CVI for CRP is 33.7%, and even with the cofactor for optimum performance (k=0.25), the resulting CVA would be 8.3%, which is high for clinical purposes. Therefore, analytes with extreme CVI values should be excluded from this model, with model 3 being recommended for their routine application [59].

Imprecision and measurement uncertainty

The VIM defines MU as the “parameter characterizing the dispersion of the values being attributed to a measurand, based on the information used.” Measurement results are inherently imperfect [60], as evidenced both theoretically and practically. For instance, expressing the measurement result of an analyte as an a unit, mathematically denoted as with infinite precision (“a.000….∞”), implies a frictionless scenario that is physically unattainable and inherently incomplete. The presence of infinite digits after the decimal point in “a” ensures that MU cannot be reduced to zero theoretically.
Practically, MU also cannot reach zero owing to uncertainties associated with various steps in the analytical process. Nevertheless, MU can be minimized to an acceptable level based on the intended application of the measurement results for the analytes. Therefore, measurement results are not perfect and should always be accompanied by ±MU [61].
MU is intrinsic to all measurements performed in medical laboratories and must be managed correctly to ensure accurate data interpretation and patient safety. Notably, MU is related to the measurement procedure rather than the analyte itself, suggesting that multiple MU values may exist for the same analyte depending on the measurement procedure. To integrate MU into medical laboratories, its definition and role in measurement procedures must be established [31, 61].
While ISO 15189 recommends estimating the MU for each measurand [62], few terms have been debated as much as “uncertainty” in analytical procedures in laboratory medicine [10, 63–71], and it is still not well understood and not implemented in laboratory practice at a desirable level. In various scientific fields, particularly in physics, the concept of MU has been meticulously analyzed and practically implemented [72–76]. However, in medical sciences, it is less well understood and applied, even today [77]. Historically, total allowable error (TEa) has been incorrectly used as a performance criterion in medical laboratories. In 2017, the Bureau International des Poids et Mesures transitioned to using universal constants instead of physical objects for SI unit assessments, aiming to minimize MU associated with SI units [73, 75, 78–81]. Despite this advancement, ISO guidelines for calculating MU in medical laboratories became available only in 2019 [82]. Efforts by the International Federation of Clinical Chemistry and Laboratory Medicine (IFCC) and EFLM to standardize MU implementation in medical laboratories have made progress, but further time is needed.
Accurate MU calculation presents challenges [83]. Two main approaches exist: bottom-up and top-down [83–85]. The bottom-up approach involves a detailed analysis of all factors contributing to MU and incorporates them into calculations [83]. The contributing factors are often illustrated using fishbone diagrams. In contrast, the top-down approach relies on QC data collected in the laboratory [84, 86] over extended periods, typically exceeding three months.
While the bottom-up approach is effective and more accurate for industrial measurements, it is less practical for biological samples owing to their inherent variability. Biological samples are influenced by BV, and even when the imprecision or MU is reduced to zero, the inherent BV cannot be eliminated. Given the numerous factors affecting MU, it is impractical to include most of them or to spend excessive time and laboratory resources in the pursuit of a more precise MU measurement. Rather than meticulously analyzing each component, only the most significant factors should be included in the MU calculation. To manage the MU in measurement procedures for medical laboratories, a pragmatic (preferably, top-down) approach should be adopted.
The ISO guideline for calculating MU in medical laboratories (ISO/TS 20914:2019) is based on a similar approach [82]. According to ISO/TS 20914:2019, the MU of measurands in medical laboratories should be based on imprecision, calibration, and bias, and calculated as follows [82]:
(10)
MU=k×Ucal2+URw2+UBias2
where Rw represents long-term imprecision estimated from internal QC data.
The ISO approach is flexible and reasonable, allowing adaptation to available data. Contrary to common belief in the laboratory community, estimating bias is not straightforward [87, 88], and separating bias as an independent component in long-term observations is challenging and lacks a strong scientific basis. Additionally, calibration uncertainty should be provided by the manufacturer but is often not available. In such instances, the MU can be based solely on imprecision, which is not to be calculated from data obtained under repeatability conditions but from data collected under intermediate precision or reproducibility conditions. Finally, the MU can be calculated as follows:
(11)
MU=k×URw2=k×URw
For 95% probability, k=1.96, which can be rounded to 2.0. Accordingly, the MU can be considered as twice the imprecision of the measurands.
The use of TEa instead of MU has long been debated. It was often speculated that, while MU is scientifically correct, it is too complex, and therefore, TEa should be used instead [63]. However, this argument is now widely regarded as unfounded, as MU is not only scientifically accurate but also easier to calculate and implement in medical laboratories than TEa.

IMPRECISION FROM A CLINICAL PERSPECTIVE

Imprecision and disease diagnosis

Disease diagnosis relies on clinical presentation, physical examination, radiological findings, and various investigations, including laboratory tests. Imprecision contributes to diagnostic uncertainty and must be accounted for in medical decision-making [89–91]. Laboratory data analysis frequently employs specific limits, such as the upper and lower limits (ULa/LLs) of reference intervals (RIs) and decision limits (DLs), for disease diagnosis [23]. However, patient samples are typically measured only once, and because of MU, decisions based on single measurement results can lead to misdiagnosis, particularly when results are near the UL/LL of the RI or the DL. Therefore, repeated measurements using different samples are essential for accurate decision-making. Determining the proximity of a measurement result to the DL, which requires repetition, and the number of repetitions required remain critical considerations.
Because of the CVI of the measurand, repeated measurements of the same sample are not sufficient for an accurate diagnosis. Therefore, as mentioned above, the CVI of the measurand should be included in the calculations. When the higher level of the measurand is diagnostic, the following equations should be used to make the decision:
(12)
X¯=x1+ x2+x3+…+xnn 
(13)
CVT=CVA2+CVI2
(14)
X¯−k×CVTn>DL
When the lower level of the measurand is diagnostic, equation 14 should be modified as follows:
(15)
X¯+k×CVTn<DL
For 95% probability, k=1.65. The imprecision of measurement results is inversely proportional to the square root of the number of samples and repetitions. Decisions based on equations 14 and 15 are more objective than those based on a single measurement result.

Imprecision and disease monitoring

Imprecision is also a critical parameter in disease monitoring, which relies on measurement results from repeated samples collected at different times. Because of the repeated sampling, the CVI of the measurands and imprecision of the measurement procedure should be included in the calculations.
The monitoring of patients or healthy individuals is a long-term process, requiring the analytical imprecision used in reference change value (RCV) calculations to be derived from data collected under intermediate precision or reproducibility conditions rather than repeatability conditions. When samples are analyzed within a single laboratory, the CVA should be calculated from data collected under intermediate precision conditions. However, for samples analyzed across multiple laboratories, the CVA is more appropriately calculated from data collected under reproducibility conditions.
The monitoring of patient’s laboratory results relies entirely on imprecision, including both biological imprecision, referred to as within-subject BV, and analytical imprecision. The combined imprecision can be calculated using the following equation, with necessary modifications (Fig. 4) [92–94].
(16)
RCV=k×2×CVA2+CVI/P2
where k=1.96 or k=1.65 for a two-sided or one-sided 95% probability level, respectively, and CVI/P represents within-subject/person BV.
Equation 16 represents a symmetrical approach to calculating the RCV; however, depending on the data distribution, particularly those used to estimate CVI/P, an asymmetrical approach may be more realistic for monitoring patients’ laboratory data. In this approach, RCVs for increase and decrease are calculated separately using the following formulas [95]:
(17)
SDA, log2=LogeCVA2+1
(18)
SDI/P, log2=LogeCVI/P2+1
(19)
SD*=SDA, log2+SDI/P, log2 
(20)
RCV%=100%×e±Za ×2 ×SD* −1
where SDA,log is the analytical SD, SDI/P,log, is the within-subject SD; and SD* is the combination of SDA,log and SDI/P,log.
Ideally, the RCV should be calculated using CVP, representing within-person variation. However, in cases of insufficient data or inadequate data quality, CVI may be used instead. In both diagnosis and disease monitoring, the decisive factor is CVI/P rather than CVA. As outlined earlier, the desirable CVA should be maintained at <0.5×CVI, ensuring that the contribution of the measurement method’s imprecision to TV remains <12%, with the remaining 88% attributable to CVI, reflecting sample variation. For accurate diagnosis and effective monitoring, the imprecision of the measurement procedure must remain <0.5×CVI, and ideally even lower.

IMPRECISION AND BIAS

In routine practice, imprecision and bias are often considered two separate parameters. However, in reality, these two parameters are intertwined, particularly in long-term observations, and are difficult to separate [31, 96]. This is especially true when estimating imprecision from data collected under intermediate precision and reproducibility conditions.
For a conglomerate of laboratories serving a healthcare group or hospitals, an acceptable CVA for a measurand within each laboratory individually, but an unacceptable CVA when considering all laboratories collectively, may indicate that at least one laboratory measures the same measurand differently from the others. In other words, even when the CVA for each laboratory is acceptable and potentially very low, combining datasets from different laboratories and calculating the overall imprecision as a unified dataset provides a practical approach to identifying inter-laboratory bias.

BV: THE INTERMEDIATE PRECISION OF HUMAN METABOLISM

Imprecision is not exclusive to measurement procedures. In human metabolism, analyte concentrations are influenced by various factors, and measurement results of analytes from samples taken from the same individual, even at the same time on different days, show variation [97, 98]. This variation is observed as fluctuations in the concentration of a biomolecule around a homeostatic set point (HSP), implying that biomolecule concentrations vary within a defined interval. For certain analytes, variation can be attributed to hormonal regulation. Other contributing factors are the production and degradation of biomolecules, their excretion, or their movement from tissues and organs into circulation, followed by degradation and excretion. In any case, variation around the HSP is observed for all biomolecules.
CVI and CVP, which are used to describe physiological variation, essentially refer to the same phenomenon, i.e., the variation in a biomolecule’s concentration or activity around an HSP within a defined interval. The difference between CVI and CVP lies in the data source used to estimate them [23]. CVI represents the variation of a molecule around the HSP of a population, not an individual, and is estimated from population data, making it population-specific. Despite being referred to as “within-subject BV,” it is not specific to a single subject but represents the pooled mean of the variations observed in a group of individuals. In contrast, CVP represents the variation in a biomolecule’s concentration or activity around the HSP of an individual, making it specific to that individual. Thus, any given analyte has a single CVI but multiple CVPs. As CVI represents the within-subject variation of a molecule for a population, the best estimate is the global estimation of CVI. For most analytes, CVIs are provided in the EFLM Biological Variation Database [57]. In contrast, CVP is estimated relatively easily and depends on the availability of steady-state data from an individual. It can be calculated using a four-step procedure:
1. Collect serial samples from the individual at the same time of the day at regular time intervals (e.g., daily, weekly, or monthly).
2. Measure the analyte concentration/activity in the serial samples in duplicate.
3. Calculate SDA from the duplicate measurements and determine SDT and the mean from the entire dataset of measurement results.
4. Calculate SDP as follows:
(21)
SDP=SDT2−SDA2 
SDP can be converted to CVP using the mean of the serial samples. The samples should ideally be measured in a single run. However, for certain parameters, such as complete blood count parameters, this may not always be feasible (e.g., when estimating the BV of complete blood count parameters) [99]. In such cases, reagents and consumables from the same lot should be used for measurements. Additionally, the lack of significant bias during the measurement period must be demonstrated.

PERSPECTIVES

As shown in Fig. 4, imprecision in patient data comprises two primary components: biological imprecision, represented by CVI/P, and instrumental imprecision, represented by CVA. Advances in laboratory instruments and measurement procedures have reduced instrumental imprecision to levels where it is no longer a significant factor compared to biological imprecision.
Biological imprecision is the key factor in personalized laboratory medicine. Accurate interpretation of patient laboratory data requires comparison with reference values derived from the individual’s data [23, 100]. Biological imprecision is critical not only for personalized reference intervals (prRIs) but also for personalized decision limits (prDLs), personalized action limits (prALs), and personalized reference change values (prRCVs) [98, 101, 102]. A novel statistical approach using small sample sizes is essential for estimating these personalized reference values [23].

CONCLUSION

Imprecision is an inherent aspect of repeated measurements and varies depending on the measurement conditions. The assessment of imprecision should reflect the conditions under which measurements are performed in practice. Laboratory data are typically collected using diverse instruments and reagents, by different personnel, and over varying time periods. Therefore, imprecision assessment should include all sources of variability—referred to as intermediate precision conditions—rather than being limited to repeatability conditions, which involve short timeframes and the use of the same instruments and reagents.
When patient data are reported from multiple laboratories in different locations, the total variability must include both intermediate conditions and between-laboratory variability, referred to as reproducibility conditions, in CVA calculations. Variability in laboratory test results is not solely attributable to the measurement procedure; natural fluctuations in human metabolite concentrations, known as within-person BV, must also be considered in imprecision calculations and data interpretation.
In conclusion, both instrumental and biological imprecision are integral to all measurements involving biological samples. Imprecision forms the foundation of personalized laboratory medicine, shaping the future of modern laboratory practices.

ACKNOWLEDGEMENTS

None.

Notes

AUTHOR CONTRIBUTIONS

Coskun A contributed to the literature review, the manuscript writing, editing, proofreading, and approved the final version of the manuscript.

CONFLICTS OF INTEREST

None declared.

RESEARCH FUNDING

None declared.

References

1. Fraser CG. 2001. Biological variation: from principles to practice. AAAC Press;Washington, DC: DOI: 10.1016/s0009-8981(03)00072-x.
2. de la Lastra JMP, Wardell SJT, Pal T, de la Fuente-Nunez C, Pletzer D. 2024; From data to decisions: leveraging artificial intelligence and machine learning in combating antimicrobial resistance - a comprehensive review. J Med Syst. 48:71. DOI: 10.1007/s10916-024-02089-5. PMID: 39088151. PMCID: PMC11294375.
crossref
3. Souza-Pereira L, Ouhbi S, Pombo N. 2021; Quality-in-use characteristics for clinical decision support system assessment. Comput Methods Programs Biomed. 207:106169. DOI: 10.1016/j.cmpb.2021.106169. PMID: 34062492.
crossref
4. Souza-Pereira L, Ouhbi S, Pombo N. 2021; A process model for quality in use evaluation of clinical decision support systems. J Biomed Inform. 123:103917. DOI: 10.1016/j.jbi.2021.103917. PMID: 34563692.
crossref
5. Laker LF, Froehle CM, Windeler JB, Lindsell CJ. 2018; Quality and efficiency of the clinical decision-making process: information overload and emphasis framing. Prod Oper Manag. 27:2213–25. DOI: 10.1111/poms.12777.
crossref
6. Anderson OW, Shields MC. 1982; Quality measurement and control in physician decision making: state of the art. Health Serv Res. 17:125–55.
7. Terminology in analytical measurement: introduction to VIM 3. https://www.eurachem.org/index.php/publications/guides/48-gdtam11. Updated on Oct 2024.
8. ISO 3534-1:2006. Statistics - vocabulary and symbols - Part 1: General statistical terms and terms used in probability. https://www.iso.org/standard/40145.html. Updated on Oct 2024.
9. ISO 5725-1:2023. Accuracy (trueness and precision) of measurement methods and results. Part 1: General principles and definitions. https://www.iso.org/obp/ui/#iso:std:iso:5725:-1:ed-2:v1:en. Updated on Oct 2024.
10. Oosterhuis WP, Theodorsson E. 2016; Total error vs. measurement uncertainty: revolution or evolution? Clin Chem Lab Med. 54:235–9. DOI: 10.1515/cclm-2015-0997. PMID: 26540227.
crossref
11. Rasberry SD. 2009; Entropy in quality systems. Accred Qual Assur. 14:65–6. DOI: 10.1007/s00769-008-0477-4.
crossref
12. Bushuyev SD, Sochnev SV. 1999; Entropy measurement as a project control tool. Int J Proj Manag. 17:343–50. DOI: 10.1016/S0263-7863(98)00049-0.
crossref
13. Kvålseth TO. 2016; On the measurement of randomness (uncertainty): a more informative entropy. Entropy. 18:159. DOI: 10.3390/e18050159. PMID: cf12ac161d314d8f97b7895cdd676bc6.
crossref
14. Jha PK, Jha R, Datt R, Guha SK. 2011; Entropy in good manufacturing system: tool for quality assurance. Eur J Oper Res. 211:658–65. DOI: 10.1016/j.ejor.2011.01.005.
crossref
15. Ramirez BS, Runger GC. 2006; Quantitative techniques to evaluate process stability. Qual Eng. 18:53–68. DOI: 10.1080/08982110500403581.
crossref
16. Schmidt RL, Pearson LN. 2019; Is your assay stable? Using process stability and capability to evaluate assay performance. Clin Chim Acta. 490:28–33. DOI: 10.1016/j.cca.2018.12.015. PMID: 30553861.
crossref
17. Pai FY, Yeh TM, Hung YH. 2015; Analysis on accuracy of bias, linearity and stability of measurement system in ball screw processes by simulation. Sustainability. 7:15464–86. DOI: 10.3390/su71115464.
crossref
18. Ogrodnik P. 2020. Evaluation (validation and verification). Medical device design. 2nd ed. Elsevier;Amsterdam: p. 317–75. DOI: 10.1016/B978-0-12-814962-1.00010-7.
19. ISO 21748:2017. Guidance for the use of repeatability, reproducibility and trueness estimates in measurement uncertainty evaluation. https://www.iso.org/obp/ui/#iso:std:iso:21748:ed-2:v1:en. Updated on Oct 2024.
20. Gundersen OE. 2021; The fundamental principles of reproducibility. Philos Trans A Math Phys Eng Sci. 379:20200210. DOI: 10.1098/rsta.2020.0210. PMID: 33775150.
crossref
21. Mulugeta L, Drach A, Erdemir A, Hunt CA, Horner M, Ku JP, et al. 2018; Credibility, replicability, and reproducibility in simulation for biomedicine and clinical applications in neuroscience. Front Neuroinform. 12:18. DOI: 10.3389/fninf.2018.00018. PMID: 29713272. PMCID: PMC5911506. PMID: 7431a3efc528447a80abff729f20fcb4.
crossref
22. Stevens JR. 2017; Replicability and reproducibility in comparative psychology. Front Psychol. 8:862. DOI: 10.3389/fpsyg.2017.00862. PMID: 28603511. PMCID: PMC5445189. PMID: 3cb0904b3736412e96e75d57c47901a5.
crossref
23. Coskun A. 2024; Diagnosis based on population data versus personalized data: the evolving paradigm in laboratory medicine. Diagnostics (Basel). 14:2135. DOI: 10.3390/diagnostics14192135. PMID: 39410539. PMCID: PMC11475514. PMID: a1d4fd3f8d7c41c680d3751f08061e7e.
crossref
24. Krishnamoorthy K. 2020. Handbook of statistical distributions with applications. 2nd ed. CRC Press;DOI: 10.1201/9781420011371.
25. Radziwill NM. 2017. Statistics (the easier way) with R. 2nd ed. Lapis Lucera.
26. Bebu I, Mathew T. 2008; Comparing the means and variances of a bivariate log-normal distribution. Stat Med. 27:2684–96. DOI: 10.1002/sim.3080. PMID: 17907261.
crossref
27. Fletcher D. 2008; Confidence intervals for the mean of the delta-lognormal distribution. Environ Ecol Stat. 15:175–89. DOI: 10.1007/s10651-007-0046-8.
crossref
28. Zou GY, Taleban J, Huo CY. 2009; Confidence interval estimation for lognormal data with application to health economics. Comput Stat Data Anal. 53:3755–64. DOI: 10.1016/j.csda.2009.03.016.
crossref
29. Dunn WL, Shultis JK. 2023. Some common probability distributions. Exploring Monte Carlo methods. Elsevier;Amsterdam: p. 447–95. DOI: 10.1016/B978-0-12-819739-4.00019-6.
30. Kissell R, Poserina J. 2017. Advanced math and statistics. Optimal sports math, statistics, and fantasy. Elsevier;Amsterdam: p. 103–35. DOI: 10.1016/B978-0-12-805163-4.00004-9. PMID: 28638574. PMCID: PMC5473192.
31. Coskun A, Theodorsson E, Oosterhuis WP, Sandberg S. European Federation of Clinical Chemistry and Laboratory Medicine Task and Finish Group on Practical Approach to Measurement Uncertainty. 2022; Measurement uncertainty for practical use. Clin Chim Acta. 531:352–60. DOI: 10.1016/j.cca.2022.04.1003. PMID: 35513038.
crossref
32. Cummings WB, Gaylor DW. 1974; Variance component testing in unbalanced nested designs. J Am Stat Assoc. 69:765–71. DOI: 10.1080/01621459.1974.10480202.
crossref
33. Gower JC. 1962; Variance component estimation for unbalanced hierarchical classifications. Biometrics. 18:537. DOI: 10.2307/2527900.
crossref
34. Webster R. 2023; Unbalanced analysis of variance. In: Daya Sagar BS, Cheng Q, McKinley J, Agterberg F (eds). Encyclopedia of mathematical Geosciences. Springer. 1579–83. DOI: 10.1007/978-3-030-85040-1.
35. EP05-A3. Evaluating quantitative measurement precision. https://clsi.org/standards/products/method-evaluation/documents/ep05/. Updated on Oct 2024.
36. Kallner A, McQueen M, Heuck C. 1999; The Stockholm Consensus Conference on quality specifications in laboratory medicine, 25-26 April 1999. Scand J Clin Lab Invest. 59:475–6. DOI: 10.1080/00365519950185175. PMID: 10667681.
crossref
37. Fraser CG. 2015; The 1999 Stockholm Consensus Conference on quality specifications in laboratory medicine. Clin Chem Lab Med. 53:837–40. DOI: 10.1515/cclm-2014-0914. PMID: 25720125.
crossref
38. Fraser CG, Kallner A, Kenny D, Petersen PH. 1999; Introduction: strategies to set global quality specifications in laboratory medicine. Scand J Clin Lab Investig. 59:477–8. DOI: 10.1080/00365519950185184.
crossref
39. Sandberg S, Fraser CG, Horvath AR, Jansen R, Jones G, Oosterhuis W, et al. 2015; Defining analytical performance specifications: consensus statement from the 1st Strategic Conference of the European Federation of Clinical Chemistry and Laboratory Medicine. Clin Chem Lab Med. 53:833–5. DOI: 10.1515/cclm-2015-0067. PMID: 25719329.
crossref
40. Jones GRD, Bell KJL, Ceriotti F, Loh TP, Lord S, Sandberg S, et al. 2024; Applying the Milan models to setting analytical performance specifications - considering all the information. Clin Chem Lab Med. 62:1531–7. DOI: 10.1515/cclm-2024-0104. PMID: 38801089.
41. Smith AF, Shinkins B, Hall PS, Hulme CT, Messenger MP. 2019; Toward a framework for outcome-based analytical performance specifications: a methodology review of indirect methods for evaluating the impact of measurement uncertainty on clinical outcomes. Clin Chem. 65:1363–74. DOI: 10.1373/clinchem.2018.300954. PMID: 31444309. PMCID: PMC7055686.
crossref
42. Loh TP, Smith AF, Bell KJL, Lord SJ, Ceriotti F, Jones G, et al. 2021; Setting analytical performance specifications using HbA1c as a model measurand. Clin Chim Acta. 523:407–14. DOI: 10.1016/j.cca.2021.10.016. PMID: 34666026.
crossref
43. Horvath AR, Bossuyt PMM, Sandberg S, John AS, Monaghan PJ, Verhagen-Kamerbeek WDJ, et al. 2015; Setting analytical performance specifications based on outcome studies - is it possible? Clin Chem Lab Med. 53:841–8. DOI: 10.1515/cclm-2015-0214.
crossref
44. Ceriotti F, Fernandez-Calle P, Klee GG, Nordin G, Sandberg S, Streichert T, et al. 2017; Criteria for assigning laboratory measurands to models for analytical performance specifications defined in the 1st EFLM Strategic Conference. Clin Chem Lab Med. 55:189–94. DOI: 10.1515/cclm-2016-0091. PMID: 27506603.
crossref
45. Ricós C, Álvarez V, Perich C, Fernández-Calle P, Minchinela J, Cava F, et al. 2015; Rationale for using data on biological variation. Clin Chem Lab Med. 53:863–70. DOI: 10.1515/cclm-2014-1142. PMID: 25928757.
crossref
46. Carobene A, Braga F, Roraas T, Sandberg S, Bartlett WA. 2013; A systematic review of data on biological variation for alanine aminotransferase, aspartate aminotransferase and γ-glutamyl transferase. Clin Chem Lab Med. 51:1997–2007. DOI: 10.1515/cclm-2013-0096. PMID: 24072574.
crossref
47. Carobene A. 2015; Reliability of biological variation data available in an online database: need for improvement. Clin Chem Lab Med. 53:871–7. DOI: 10.1515/cclm-2014-1133. PMID: 25883200.
crossref
48. Carobene A, Strollo M, Jonker N, Barla G, Bartlett WA, Sandberg S, et al. 2016; Sample collections from healthy volunteers for biological variation estimates' update: a new project undertaken by the Working Group on Biological Variation established by the European Federation of Clinical Chemistry and Laboratory Medicine. Clin Chem Lab Med. 54:1599–608. DOI: 10.1515/cclm-2016-0035. PMID: 27169681.
crossref
49. Carobene A, Aarsand AK, Bartlett WA, Coskun A, Diaz-Garzon J, Fernandez-Calle P, et al. 2022; The European Biological Variation study (EuBIVAS): a summary report. Clin Chem Lab Med. 60:505–17. DOI: 10.1515/cclm-2021-0370. PMID: 34049424.
crossref
50. Coşkun A, Aarsand AK, Sandberg S, Guerra E, Locatelli M, Díaz-Garzón J, et al. 2022; Within- and between-subject biological variation data for tumor markers based on the European Biological Variation study. Clin Chem Lab Med. 60:543–52. DOI: 10.1515/cclm-2021-0283. PMID: 33964202.
crossref
51. Ceriotti F, Díaz-Garzón Marco JDG, Fernández-Calle P, Maregnani A, Aarsand AK, Coskun A, et al. 2020; The European Biological Variation study (EuBIVAS): weekly biological variation of cardiac troponin I estimated by the use of two different high-sensitivity cardiac troponin I assays. Clin Chem Lab Med. 58:1741–7. DOI: 10.1515/cclm-2019-1182. PMID: 32304296.
crossref
52. Bottani M, Aarsand AK, Banfi G, Locatelli M, Coşkun A, Díaz-Garzón J, et al. 2022; European Biological Variation study (EuBIVAS): within- and between-subject biological variation estimates for serum thyroid biomarkers based on weekly samplings from 91 healthy participants. Clin Chem Lab Med. 60:523–32. DOI: 10.1515/cclm-2020-1885. PMID: 33561908.
crossref
53. Díaz-Garzón J, Fernández-Calle P, Minchinela J, Aarsand AK, Bartlett WA, Aslan B, et al. 2019; Biological variation data for lipid cardiovascular risk assessment biomarkers. A systematic review applying the biological variation data critical appraisal checklist (BIVAC). Clin Chim Acta. 495:467–75. DOI: 10.1016/j.cca.2019.05.013. PMID: 31103621.
crossref
54. Aarsand AK, Kristoffersen AH, Sandberg S, Støve B, Coşkun A, Fernandez-Calle P, et al. 2021; The European biological variation study (EuBIVAS): biological variation data for coagulation markers estimated by a Bayesian model. Clin Chem. 67:1259–70. DOI: 10.1093/clinchem/hvab100. PMID: 34387652.
crossref
55. González-Lao E, Corte Z, Simón M, Ricós C, Coskun A, Braga F, et al. 2019; Systematic review of the biological variation data for diabetes related analytes. Clin Chim Acta. 488:61–7. DOI: 10.1016/j.cca.2018.10.031. PMID: 30389455.
crossref
56. Diaz-Garzon J, Fernandez-Calle P, Sandberg S, Özcürümez M, Bartlett WA, Coskun A, et al. 2021; Biological variation of cardiac troponins in health and disease: a systematic review and meta-analysis. Clin Chem. 67:256–64. DOI: 10.1093/clinchem/hvaa261. PMID: 33279972.
crossref
57. EFLM Biological Variation. https://biologicalvariation.eu/. Updated on Aug 2024.
58. Sikaris K. 2008; Analytical quality-what should we be aiming for? Clin Biochem Rev. 29(S1):S5–10.
59. Oosterhuis WP, Coskun A, Sandberg S, Theodorsson E. 2023; Performance specifications for sodium should not be based on biological variation. Clin Chim Acta. 540:117221. DOI: 10.1016/j.cca.2023.117221. PMID: 36640931.
crossref
60. White GH. 2008; Basics of estimating measurement uncertainty. Clin Biochem Rev. 29(S1):S53–60. PMID: 18852859. PMCID: PMC2556585.
61. Coskun A, Berçik İnal B, Serdar M. 2019; Measurement uncertainty in laboratory medicine: the bridge between medical and industrial metrology. Turk J Biochem. 44:121–5. DOI: 10.1515/tjb-2019-0170.
crossref
62. ISO 15189:2022. Medical laboratories-requirements for quality and competence. https://www.iso.org/standard/76677.html. Updated on Oct 2024.
63. Westgard JO. 2018; Error methods are more practical, but uncertainty methods May still be preferred. Clin Chem. 64:636–8. DOI: 10.1373/clinchem.2017.284406. PMID: 29311055.
crossref
64. Farrance I, Badrick T, Sikaris KA. 2016; Uncertainty in measurement and total error - are they so incompatible? Clin Chem Lab Med. 54:1309–11. DOI: 10.1515/cclm-2016-0314. PMID: 27227711.
crossref
65. Oosterhuis WP, Bayat H, Armbruster D, Coskun A, Freeman KP, Kallner A, et al. 2018; The use of error and uncertainty methods in the medical laboratory. Clin Chem Lab Med. 56:209–19. DOI: 10.1515/cclm-2017-0341. PMID: 28796637.
crossref
66. Farrance I, Badrick T, Frenkel R. 2018; Uncertainty in measurement and total error: different roads to the same quality destination? Clin Chem Lab Med. 56:2010–14. DOI: 10.1515/cclm-2018-0421. PMID: 29949508.
crossref
67. Panteghini M, Sandberg S. 2016; Total error vs. measurement uncertainty: the match continues. Clin Chem Lab Med. 54:195–6. DOI: 10.1515/cclm-2015-1036. PMID: 26581072.
crossref
68. Rozet E, Marini RD, Ziemons E, Hubert P, Dewé W, Rudaz S, et al. 2011; Total error and uncertainty: friends or foes? TrAC Trends Anal Chem. 30:797–806. DOI: 10.1016/j.trac.2010.12.009.
crossref
69. Karadağ C, Demirel N. 2023; Total analytical error and measurement uncertainty for analytical performance evaluation and determination of gray zones of glucose critical value limits. Lab Med. 54:153–9. DOI: 10.1093/labmed/lmac077. PMID: 36053235.
crossref
70. Westgard JO, Westgard SA. 2017; Measuring analytical quality: total analytical error versus measurement uncertainty. Clin Lab Med. 37:1–13. DOI: 10.1016/j.cll.2016.09.001. PMID: 28153359.
71. Westgard JO. 2010; Managing quality vs. measuring uncertainty in the medical laboratory. Clin Chem Lab Med. 48:31–40. DOI: 10.1515/CCLM.2010.024. PMID: 19919330.
crossref
72. 2018; Better research through metrology. Nat Methods. 15;15:395. DOI: 10.1038/s41592-018-0035-x. PMID: 29855569.
73. Reich E. 2014; Ampere to get rational redefinition. Nature. 505:273–4. DOI: 10.1038/505273a. PMID: 24429609.
crossref
74. Gibney E. 2015; Kilogram conflict resolved at last. Nature. 526:305–6. DOI: 10.1038/526305a. PMID: 26469019.
crossref
75. 2017; SI units need reform to avoid confusion. Nature. 548:135. DOI: 10.1038/548135b. PMID: 28796224.
76. 2019; Measure for measure. Nat Mater. 18:299. DOI: 10.1038/s41563-019-0333-4. PMID: 30894759.
77. Ceriotti F, Buoro S, Pasotti F. 2024; How clinical laboratories select and use analytical performance specifications (APS) in Italy. Clin Chem Lab Med. 62:1470–3. DOI: 10.1515/cclm-2023-1314. PMID: 38411177.
crossref
78. Jones N. 2009; Metrology: the new and improved kelvin. Nature. 459:902–4. DOI: 10.1038/459902a. PMID: 19536237.
crossref
79. Fischer J, Ullrich J. 2016; The new system of units. Nat Phys. 12:4–7. DOI: 10.1038/nphys3612.
crossref
80. Gibney E. 2017; New definitions of scientific units are on the horizon. Nature. 550:312–3. DOI: 10.1038/550312a. PMID: 29052643.
crossref
81. Schlamminger S, Yang I, Kumar H. 2020; Redefinition of SI units and its implications. Mapan J Metrol Soc India. 35:471–4. DOI: 10.1007/s12647-020-00421-1.
crossref
82. ISO/TS 20914:2019. Medical laboratories - practical guidance for the estimation of measurement uncertainty. https://www.iso.org/obp/ui/#iso:std:iso:ts:20914:ed-1:v1:en. Updated on Oct 2024.
83. JCGM GUM-6:2020 Guide to the expression of uncertainty in measurement-Part 6: Developing and using measurement models Guide pour l'expression de l'incertitude de mesure-Partie 6: Élaboration et utilisation des modèles de mesure. www.bipm.org, 2020 g. Updated on Oct 2024.
84. Martinello F, Snoj N, Skitek M, Jerin A. 2020; The top-down approach to measurement uncertainty: which formula should we use in laboratory medicine? Biochem Med (Zagreb). 30:020101. DOI: 10.11613/BM.2020.020101. PMID: 32292278. PMCID: PMC7138004.
85. Lee JH, Choi JH, Youn JS, Cha YJ, Song W, Park AJ. 2015; Comparison between bottom-up and top-down approaches in the estimation of measurement uncertainty. Clin Chem Lab Med. 53:1025–32. DOI: 10.1515/cclm-2014-0801. PMID: 25539513.
crossref
86. Gu H, Lee J, Hong J, Lee W, Yun YM, Chun S, et al. 2022; Practical considerations for clinical laboratories in top-down approach for assessing the measurement uncertainty of clinical chemistry analytes. Ann Lab Med. 42:630–7. DOI: 10.3343/alm.2022.42.6.630. PMID: 35765871. PMCID: PMC9277045.
crossref
87. Theodorsson E, Magnusson B, Leito I. 2014; Bias in clinical chemistry. Bioanalysis. 6:2855–75. DOI: 10.4155/bio.14.249. PMID: 25486232.
crossref
88. Coskun A. 2024; Bias in laboratory medicine: the dark side of the moon. Ann Lab Med. 44:6–20. DOI: 10.3343/alm.2024.44.1.6. PMID: 37665281. PMCID: PMC10485854.
crossref
89. Lindberg DAB, Watson FR. 1974; Imprecision of laboratory determinations and diagnostic accuracy: theoretical considerations. Methods Inf Med. 13:151–8. DOI: 10.1055/s-0038-1636147. PMID: 4419173.
crossref
90. Haeckel R, Wosniok W. 2011; A new concept to derive permissible limits for analytical imprecision and bias considering diagnostic requirements and technical state-of-the-art. Clin Chem Lab Med. 49:623–35. DOI: 10.1515/CCLM.2011.116. PMID: 21345158.
crossref
91. Loh TP, Markus C, Lim CY. 2024; Impact of analytical imprecision and bias on patient classification. Am J Clin Pathol. 161:4–8. DOI: 10.1093/ajcp/aqad115. PMID: 37769333.
crossref
92. Fraser CG. 2011; Reference change values. Clin Chem Lab Med. 50:807–12. DOI: 10.1515/cclm.2011.733. PMID: 21958344.
crossref
93. Jones GRD, Aarsand AK, Carobene A, Coskun A, Fernandez-Calle P, Bartlett B, et al. 2024; A new concept for reference change values-regression to the population mean. Clin Chem. 70:1076–84. DOI: 10.1093/clinchem/hvae067. PMID: 38776253.
crossref
94. Plebani M, Lippi G. 2012; Biological variation and reference change values: an essential piece of the puzzle of laboratory testing. Clin Chem Lab Med. 50:189–90. DOI: 10.1515/cclm.2011.751. PMID: 22505541.
crossref
95. Røraas T, Støve B, Petersen PH, Sandberg S. 2016; Biological variation: the effect of different distributions on estimated within-person variation and reference change values. Clin Chem. 62:725–36. DOI: 10.1373/clinchem.2015.252296. PMID: 27001492.
crossref
96. Thelen MHM, Van Schrojenstein Lantman M. 2024; When bias becomes part of imprecision: how to use analytical performance specifications to determine acceptability of lot-lot variation and other sources of possibly unacceptable bias. Clin Chem Lab Med. 62:1505–11. DOI: 10.1515/cclm-2023-1303. PMID: 38353157.
crossref
97. Coskun A, Zarepour A, Zarrabi A. 2023; Physiological rhythms and biological variation of biomolecules: the road to personalized laboratory medicine. Int J Mol Sci. 24:6275. DOI: 10.3390/ijms24076275. PMID: 37047252. PMCID: PMC10094461.
crossref
98. Coskun A, Lippi G. 2024; The impact of physiological variations on personalized reference intervals and decision limits: an in-depth analysis. Clin Chem Lab Med. 62:2140–7. DOI: 10.1515/cclm-2024-0009. PMID: 38452477.
crossref
99. Coşkun A, Carobene A, Kilercik M, Serteser M, Sandberg S, Aarsand AK, et al. 2018; Within-subject and between-subject biological variation estimates of 21 hematological parameters in 30 healthy subjects. Clin Chem Lab Med. 56:1309–18. DOI: 10.1515/cclm-2017-1155. PMID: 29605821.
crossref
100. Coskun A, Sandberg S, Unsal I, Serteser M, Aarsand AK. 2022; Personalized reference intervals: from theory to practice. Crit Rev Clin Lab Sci. 59:501–16. DOI: 10.1080/10408363.2022.2070905. PMID: 35579539.
crossref
101. Coskun A, Lippi G. 2024; Personalized laboratory medicine in the digital health era: recent developments and future challenges. Clin Chem Lab Med. 62:402–9. DOI: 10.1515/cclm-2023-0808. PMID: 37768883.
crossref
102. Coskun A, Ertaylan G, Pusparum M, Van Hoof R, Kaya ZZ, Khosravi A, et al. 2024; Advancing personalized medicine: integrating statistical algorithms with omics and nano-omics for enhanced diagnostic accuracy and treatment efficacy. Biochim Biophys Acta Mol Basis Dis. 1870:167339. DOI: 10.1016/j.bbadis.2024.167339. PMID: 38986819.
crossref

Fig. 1

Concentration and activity of biomolecules in human metabolism are influenced by numerous types of variation. Processing variation relates to the measurement system rather than human metabolism. This figure is modified and reprinted with permission from [98] (copyright © 2024 Walter de Gruyter GmbH).

alm-45-3-259-f1.tif
Fig. 2

Sources of analytical variation under different measurement conditions, including repeatability conditions (A), intermediate precision conditions (B), and reproducibility conditions (C).

alm-45-3-259-f2.tif
Fig. 3

Various types of distribution used in statistical analysis: normal distribution (A), truncated normal distribution (B), semicircular distribution (C), and a hypothetical distribution with lower and upper limits (D). For laboratory data, the distribution in (D) and its skewed variations appear to provide a more realistic representation. The figure is reprinted from [23] and used under the CC-BY license.

alm-45-3-259-f3.tif
Fig. 4

Imprecision of laboratory data arises from biological imprecision, referred to as within-subject or within-person biological variation, and analytical imprecision.

alm-45-3-259-f4.tif
TOOLS
Similar articles