Long-term within- and between-subject biological variation of 29 routine laboratory measurands in athletes

Study population

A flowchart depicting the timeline of recruitment and inclusion of study subjects and sample collections is provided in Figure 1. Thirty subjects (15 women, median age 36 years, range 19–53 years) were recruited from three triathlon clubs in Madrid, Spain and fulfilled the following inclusion criteria: Healthy, age>18 years, >13 h of training per week (including running, cycling and/or swimming), normal standard medical examination and normal results for a stress test supervised by a physician specialist in sports medicine. The latter included electrocardiogram, spirometry, and dynamometry with measured maximum oxygen intake (VO₂max), anaerobic and aerobic threshold, maximum heart rate (HR) and maximum power (P_max).

Figure 1:

Flow chart of the study.

Time-line of the sample collection and complementary tests.

The study was approved by the host Institution Research Ethical Committee and Informed Consent was signed by all the participants.

Sample collections

Following the Røraas simulation [11] to assure adequate study power i.e. a confidence interval (CI) width lower than 1/3 * (CV_I), 11 serum samples per subject were collected monthly (January to November) on Saturday mornings between 8 and 10 am after 10 h fasting. Subjects were instructed to avoid a heavy dinner, to be out of competition for one week and to avoid high-intensity exercise the 24 h prior to sampling. High-intensity exercise was defined by a heart rate (HR) above 80% of the maximum HR. In connection with each sampling, the subjects completed a questionnaire about fasting, vitamin supplementation and diet changes, health status and use of medication, date of last competition, details on last training (time, type and intensity) and in females, date of last menstrual period.

All aspects of the preanalytical phase were standardized [10, 12]. Venous blood samples were collected after 15 min of seated resting, and always by the same phlebotomist. In the first, 4th and 9th collection points, samples were also drawn to verify the health status of the athletes and to evaluate the results against the exclusion criteria (Supplementary File and Table 1). The laboratory results excluded any chronic or acute disease that could influence the BV estimates.

The blood samples were centrifuged at 2,000g for 10 min, then serum samples were aliquoted and stored at −80 °C, within 2 h after the blood collection.

Sample analysis

All samples were analyzed in duplicate at the end of the collection period, within 2 h (after defrosting at room temperature and vortexing), on an Advia 2400 chemistry analyzer (Siemens Healthineers) for alkaline phosphatase (ALP), alanine aminotransferase (ALT), amylase, aspartate aminotransferase (AST), lactate dehydrogenase (LDH), creatinekinase (CK), total and direct bilirubin, gamma-glutamyltransferase (GGT), glucose, triglycerides, total cholesterol, HDL-cholesterol (HDL-C), LDL cholesterol (LDL-C), iron, ferritin, transferrin, magnesium, calcium, phosphate, potassium, chloride, sodium, total protein, albumin, urate, creatinine, urea and C-reactive protein (CRP). Information on analytical principles are provided in Supplementary Table 2. All samples from each subject were measured in the same run and with the same reagent lot. We used Liquid Assayed Multiqual (Bio-Rad) for internal quality control at two concentrations and participated in EQA programs for all measurands. All quality control results were within set limits.

Statistical analysis

Trend analysis was performed for each individual by plotting the results of each subject against the collection dates, as well as for the overall population. Also, the total result from all subjects was plotted to identify transversal trends related to preanalytical or analytical error, lot to lot shifts, seasonal variations etc. If the equation had an r coefficient greater than 0.75 and the 95% confidence interval of the slope did not include 0, the trend was considered significant. If a significant trend was observed for a measurand in a particular subject, the data was excluded.

Outlier analysis was performed between replicates from the same sample and between the samples from the same subject, using the Cochran C test to assure the homoscedasticity for CV_A and CV_I [1, 13]. Reed’s criterion was applied to assess differences between the athletes’ mean values [14]. Thereafter, a CV-ANOVA was performed to deliver CV_I estimates and after log-transforming the data, an ANOVA was applied to deliver CV_G estimates [15]. For the calculations, we used RStudio Desktop 1.3.1093 and Microsoft Excel 2010.

We compared the BV estimates delivered by this study with previously published estimates from the general population and considered differences statistically significant when the 95% CI did not overlap.

To explore the potential influence of different factors on the BV, we evaluated several parameters, as detailed below, by a linear mixed model (LMM) [16] by including these variables as fixed effects and all laboratory results from each individual. The BV estimates (CV_I and CV_G) and the CVA were the results from the LMM (random effects). We calculated the CV_I with and without including the variable that was assessed for, following the script:

1. Model without variable: log (measurand’ results) ∼ (1|Subject) + (1| Subject:Visit)

2. Model including the variable: log (measurand’ results) ∼ Variable + (1| Subject) + (1| Subject:Visit)

The following fixed effects variables were used

1. Health status: symptoms of disease registered in the week before sampling (yes/no).

2. Sampling interval: time since the previous blood collection (number of days).

3. Moderate intensity exercise performed in the 24 h before sampling (yes/no).

4. The number of hours of endurance activity during the week before sampling (hours/week).

5. Variables for duration: number of hours of endurance activity per week (h/w).

6. Variables related to the intensity of exercise the week before sampling:

   1. (1) Average HR during all the training session classified by the percentage of the maximum HR, from T1 to T5: T1 (<65%); T2 (65–80%); T3 (80–90%); T4 (95–98%); T5 (100%) [17].

   2. (2) Average of subjective perception of effort, recorded by the Borg scale: 0–4 sleep or inactive; 5–7 absence of effort; 8–9 very, very light; 10 very light; 11–12 fairly light; 13–14 somewhat hard, 15–16 hard; 17–19 very hard; 20 maximum effort [18].

A difference of more than 5% in the CV_I, after including the different variables in the LMM, was considered as significant. We performed an ANOVA between both models (including and without including the variable) to verify whether the differences were significant (p<0.05).

To calculate APS [1], RCV) [19], and the index of individuality (II) [20] the following formula were applied.

1. CV_A = 0.5·(CV_I)

2. Bias = 0.5·(CV_I ² + CV_G ²)^0.5

3. RCV = 100%·(exp (±Z·2^1/2·((CV_LnA ² + CV_LnI ²))^1/2)−1);

1. II = CV_I/CV_G

BV estimates in athletes

For 22 of the measurands, the percentage of outliers eliminated to fulfill statistical criteria (0.3–10.1%) (Table 2) was comparable to those identified in previous studies in a healthy population for enzymes (1.2–14.2%) [28] and substrates (0–5.6%) [29, 30]. For total and direct bilirubin, CK, ferritin, GGT, iron and CRP, on the other hand, a high percentage of outliers, from 21-84%, had to be removed to reach homoscedasticity, a requirement for ANOVA analysis. This highly variable behavior observed for these measurands, as illustrated in Figure 2, may indicate that in an athlete population, there is not a clear steady-state situation, but that factors related to their high-exertion exercise regime may influence the concentration of these markers. For CRP, results corresponding to the visits in which the subjects reported having been ill or having had an event that could elevate the acute phase reactants during the seven days before drawing blood were excluded from the data set. Nevertheless, if estimating CV_I on the raw data set without removing outliers identified by statistical tests, this estimate was 10 times higher (data not shown) than those observed in the recently published EuBIVAS study in the general population [5] and twice as high as in other published studies in athletes. The higher variation in athletes and the lack of homogeneity of the data may be related to e.g. inflammatory processes linked to muscle damage during physical activity, distorting the steady-state situation. Therefore, BV estimates were not reported for total and direct bilirubin, CK, ferritin, GGT, iron and CRP.

Regarding differences between gender, urea and ALP were associated with higher CV_I estimates in female than male athletes. This observation confirms the previous finding in the general population for urea [29]. For ALP and other enzymes, no differences between males and females have been reported for the general population [28], previous data on athletes are lacking.

BV estimates in athletes vs. the general population

Different BV estimates may be expected in athletes considering the continuous high-endurance exercise, compared to a general, more sedate population. Since a control group was not available for our study, estimates from studies from healthy subjects, derived from the EFLM BV database [31] and another athlete-specific study [8] are presented in Table 4 for comparison. However, differences should be interpreted with caution, considering that different factors related to study design and execution may influence the BV estimates derived from different studies. Of note, our study covered a full season and thus has a longer observation period than most of the other studies available for comparison. Regarding CV_G, no differences between athletes and the general population were observed for any measurand [31].

Reference change value, index of individuality and analytical performance specifications

The enzymes AST, amylase, ALP and LDH have a low II (<0.6), indicating that they are good candidates for the application of RCV when monitoring in individual with laboratory test results over time. A high individuality was also observed in the measurands related to lipid metabolism (total cholesterol, HDL-C, LDL-C), which is why the use of the RCV for monitoring lipids in athletes is highly recommended (Table 3). For the application of the RCV, each laboratory should apply its own CVA to be applied in athletes tested outside of acute exertion. APS for these measurands in athletes would be, generally, more permissive, in particular for enzymes (AST and ALT). BV estimates derived from athletes in this study could help to set APS in both clinical laboratories providing health services to athletes, the accredited laboratories of the World Anti-doping Agency (WADA) network and the organizations that audit these laboratories [42].

Study limitations

In our study, the athletes varied in their exercise regimen. Ideally, a more standardized approach would be preferable, where all athletes followed the same setup, this was, however, practically difficult to achieve. Furthermore, an inclusion criterion was that athletes should train at least 13 h per week. Our data indicate that this was not achieved, however, a number of athletes reported not registering all exercise session throughout the study. The ANOVA models depend on variances being homogeneously distributed for the resulting mean BV estimates to be representative of the population. In our study, this lacked for seven measurands and BV data were therefore not reported for these. It may be argued that the derivation of BV in athletes may be problematic, due to the difficulty to reliably differentiate physiology from pathophysiology. Further study of our data, for these in homogeneously distributed as well as for the other measurands, by e.g. Bayesian models, which provide individual within-participant BV estimates, would be helpful to better understand the effect of exercise in athletes and its potential effect on the derived BV estimates [26]. We have compared our data derived from our athlete population with meta-analysis derived global estimates published in the EFLM BV Database. The inclusion of a control group in our study, consisting of non-athletes, would have provided better grounds for assessing differences between athletes and the general population.