Blood Test – Based Age Acceleration Is Inversely Associated with High-Volume Sports Activity

JUHÁSZ, V., A. ORSZÁG, D. BALLA, L. SZABÓ, N. SYDÓ, O. KISS, E. CSULAK, M. BABITY, Z. DOHY, R. SKODA, D. BECKER, B. MERKELY, A. BENCZÚR, H. VÁGÓ, and C. KEREPESI. Blood Test – Based Age Acceleration Is Inversely Associated with High-Volume Sports Activity. Med. Sci. Sports Exerc. , Vol. 56, No. 5, pp. 868 - 875, 2024. Purpose: We develop blood test – based aging clocks and examine how these clocks reflect high-volume sports activity. Methods: We use blood tests and body metrics data of 421 Hungarian athletes and 283 age-matched controls (mean age, 24.1 and 23.9 yr, respectively), the latter selected from a group of healthy Caucasians of the National Health and Nutrition Examination Survey (NHANES) to represent the general population ( n = 11,412). We train two age prediction models (i.e., aging clocks) using the NHANES dataset: the first model relies on blood test parameters only, whereas the second one additionally incorporates body measurements and sex. Results: We findlower age accelerationamong athletes compared with the age-matchedcontrols with a me-dianvalue of − 1.7 and 1.4yr, P < 0.0001.BMI is positively associated with age acceleration among the age-matched controls ( r = 0.17, P< 0.01) and theunrestrictedNHANESpopulation( r =0.11, P< 0.001).WefindnoassociationbetweenBMIandageaccelerationwithintheathletedataset.Instead, age acceleration is positively associated with body fat percentage ( r = 0.21, P < 0.05) and negatively associated with skeletal muscle mass (Pearson r = − 0.18, P< 0.05)amongathletes.Themostimportantbloodtestfeaturesinagepredictionswereserumferritin,meancellvolume,bloodureanitrogen, and albumin levels. Conclusions: We develop and apply blood test – based aging clocks to adult athletes and healthy controls. The data suggest that high-volume sports activity is associated with slowed biological aging. Here, we propose an alternative, promising application of routine blood tests

Over the recent years, various methods have emerged to assess the biological age of young athletes.McClean et al. (15) used hand-wrist imaging to predict biological age in pediatric athletes.They found that ECG interpretation governed by biological age provided excellent diagnostic accuracy and performed better in detecting an underlying cardiac condition than chronological age.Another application involves measuring the telomere length, which seems to carry clinical importance as early as childhood.Skilton et al. (16) found that arterial wall thickness, a marker of vascular disease risk, was associated with decreased telomere length, a marker of aging, already at the age of 8 yr.Spólnicka et al. (17) examined DNA methylation of young elite athletes (mean age 24.8 yr) and matched controls (mean age 24.2 yr) by using epigenetic clocks based on five CpG sites.They found an increased epigenetic age acceleration in athletes, which contradicts epidemiological data showing longer life span of athletes.The age predictions were based on only two genes, TRIM59 and KLF14, which are linked to anticancer and anti-inflammatory effects potentially contributing to the increased life expectancy of athletes.The authors concluded that intense physical training might have a complex influence on the aging process.
Here, we aim to investigate whether standard blood tests are suitable for determining the biological age of athletes and the general population.Standard blood tests offer advantages because of their cost-effectiveness, regular availability, and widespread accessibility in significant quantities.Blood test-based age prediction models were developed using deep learning, and the age acceleration of these clocks was associated with all-cause mortality and smoking status (18)(19)(20).A recent study using the National Health and Nutrition Examination Survey (NHANES) database for training and testing discovered that factors such as being male, having a lower socioeconomic status, exposure to tobacco, leading a sedentary lifestyle, experiencing obesity, or having a systemic disease were linked to a higher deviation in personalized physiological age (PPA) (21).
In this study, we train and test aging clocks based on blood tests and body measurements using the publicly available NHANES database.Additionally, we apply the clocks on a dataset comprising 421 Hungarian athletes previously collected in another study (22).

METHODS
Participants and measurements.We studied 421 Hungarian athletes of Caucasian ethnicity, ranging in age from 14 to 56 yr (hereafter referred to as the athlete dataset).The athletes included individuals engaged in leisure (n = 56), competitive (n = 211), and elite-level (n = 154) sports activities (categorized according to the 2020 ESC guidelines [23]), and their blood samples were collected at a tertiary cardiovascular center.The recruitment strategy did not involve any active solicitation or intervention.Professional athletes visited the clinic for mandatory testing before being permitted to return to play after SARS-CoV-2 infection, whereas leisure athletes requested testing on their own initiative.The laboratory testing was carried out approximately 24 d after the athletic subjects' SARS-CoV-2 infection.The detailed results of these tests have been published by Juhász et al. (22).Note that written informed consent was obtained from the participants beforehand.
From the NHANES database, we selected Caucasian individuals who self-reported being in good, very good, or excellent health status.We further refined our selection to individuals for whom all 36 blood measurements, along with height, weight, and BMI data, were available.For better modeling accuracy, we only included records with age information provided in months.After applying these criteria, our search strategy yielded a total of 11,412 patient records from the period spanning 2001 to 2010 with an age range from 12 to 79.2 yr.This dataset is referred to as the NHANES full dataset.Details of the measurements used in the study are shown in Table 1.
Machine learning-based age prediction.In constructing aging clocks based on blood tests, we used machine learning algorithms to train a regression model.In this setup, chronological age served as the outcome, whereas predictors included blood test parameters, body measurements, and sex.First, we split the 11,412 samples of the NHANES full dataset into a training set of 9243 samples, a validation set of 1027 samples, and a test set of 1142 samples, hereafter referred to as the NHANES training, validation, and test datasets.We refrained from employing imputation methods; rather, we kept missing values intact in the data.Missing value distributions of the datasets are available in Supplemental Figure 1 (Supplemental Digital Content 1, The distribution of missing values in the different datasets, http://links.lww.com/MSS/C991).
We deployed LightGBM, an efficient implementation of the gradient boosting decision tree algorithm (24).We trained two age predictors by using the NHANES training dataset.The first predictor solely encompassed variables related to blood test parameters (referred to as clock 1), whereas the second one incorporated body measurements (height, weight, and BMI) and sex in addition (referred to as clock 2).We performed model parameter optimization using the NHANES validation set.We selected the models with the lowest mean absolute error (MAE) obtained between chronological age and predicted age.For both clocks, the optimal model was constructed by using depth 5 trees with 10 leaves, a learning rate of 0.2, and 100 boosting iterations.
To  Statistical analysis.We used the Python package SciPy (26) for statistical calculations.We calculated two-sided t-tests to compare two independent groups.A P value less than 0.05 was considered significant.For comparing repeated test results, a one-sample t-test was calculated for the differences (popmean = 0) after we adjusted the predicted age of the second measurement by the elapsed time.We evaluated correlations by the Pearson correlation coefficient (r).We used the following notation for P values: ns for P > 0.05, * 0.05 > P ≥ 0.01, ** 0.01 > P ≥ 0.001, *** 0.001 > P ≥ 0.0001, and **** 0.0001 ≥ P.

RESULTS
Performance of blood test-based age prediction is comparable for athletes and controls.To assess the biological age of athletes, we developed two aging clocks (clocks 1 and 2) using a large cohort of healthy Caucasian people from the NHANES database.We trained clock 1 on only blood tests, whereas clock 2 was trained using blood tests, body measurements (height, weight, BMI), and sex.We tested both clocks on three datasets: (i) the NHANES test dataset, (ii) the age-matched NHANES test dataset, and (iii) the athlete dataset (Table 2, Fig. 1).The prediction performance for the athlete dataset was comparable with the age-matched NHANES test dataset by both clocks (Table 2).The athlete dataset comprised 421 subjects, delineated as follows: 154 elite athletes (37%), 211 competitive athletes (50%), and 56 leisure athletes (13%).Among these athletes, 88% engaged in a minimum of 6 h of weekly training volume, contributing to an average weekly training duration of 13.2 h.The majority were involved in mixed sports disciplines (n = 322, 76%), followed by endurance (n = 53, 13%), power (n = 38, 9%), and skill (n = 8, 2%) types.
We also examined the associations between BMI and blood test-based age acceleration.We observed that BMI was positively associated with clock 1 age acceleration for both the NHANES test dataset (r = 0.113, P < 0.001) and the age-matched NHANES test dataset (r = 0.168, P < 0.01), Figures 2C and 2D.By contrast, within the athlete dataset, we observed no association between BMI and age acceleration; however, age acceleration based on both aging clocks was positively associated with body fat percentage (r = 0.208 and P < 0.05 for clock 1, as well as r = 0.259 and P < 0.01 for clock 2), Figures 2E and 2F.Accordingly, SMM is negatively associated with the clock 1 age acceleration within the athlete dataset (r = −0.182,P < 0.05), Figure 2E.In summary, BMI is positively associated with blood test-based age acceleration in the general population; however, athletes exhibit distinct patterns.In their case, body fat percentage may be a more appropriate indicator of aging.
We further investigated the possible sex differences in blood test-based age acceleration.Age acceleration was higher in female athletes compared with male athletes (median = 2.9, IQR = [−2.8,9.4], vs median = −5.9,IQR = [−12.1,2.3] yr, P < 0.0001), Figure 2I.By contrast, in the age-matched  We hypothesized that the observed sex differences influence the decreased age acceleration in athletes, as depicted in Figures 2A and 2B.Therefore, we separated the athlete dataset and the age-matched NHANES dataset by sex and found that the effect remained in males but did not remain in females (Supplemental Fig. 2, Supplemental Digital Content 1, http:// links.lww.com/MSS/C991).In other words, the age acceleration of female athletes was not lower than that of female controls.
Feature importance analysis.We applied the Shapley Additive Explanation method (27) to examine the most important features used by clocks 1 and 2. We investigated the effect of the 10 most important features on the predicted age when clocks 1 and 2 were applied separately to the athlete dataset and the age-matched NHANES test dataset (Fig. 3).Serum ferritin levels, mean cell volume, blood urea nitrogen, and serum albumin are among the essential variables in the age prediction in the NHANES dataset, with the first three positively and the fourth negatively associated with laboratory testbased age.Another observation was that for clock 2, BMI is a more important variable in the age-matched NHANES dataset than in the athlete dataset.Sex was the sixth most important feature when testing in the general population, but it was not among the 10 most important ones (14th) for athletes.

APPLIED SCIENCES
Downloaded from http://journals.lww.com/acsm-msse by BhDMf5ePHKav1zEoum1tQfN4a+kJLhEZgbsIHo4XMi0hCyw CX1AWnYQp/IlQrHD3i3D0OdRyi7TvSFl4Cf3VC1y0abggQZXdgGj2MwlZLeI= on 05/08/2024 Repeated test analysis.In our experiments, we use the laboratory testing data from athletes for a secondary purpose, primarily aimed at investigating asymptomatic, mild, or moderately symptomatic SARS-CoV-2 infections.Subsequently, for certain individuals, a follow-up examination was conducted after their recovery.Among the athlete dataset, complete repeated blood tests were available for 20 individuals, with a median time difference of 107 d between the samplings.To evaluate the potential effect of recent SARS-CoV-2 infection on blood test-based age, we analyzed the variance in predicted age between the two measurements, accounting for the elapsed time.The difference was not significant by using one-sample t-tests (see Methods).We also examined the consistency of age acceleration between the two measurements.The age acceleration of the first measurement was strongly associated with the age acceleration of the second measurement for both clock 1 (r = 0.7231, P = 0.00031) and clock 2 (r = 0.6184, P = 0.00365).

DISCUSSION
We presented two aging clocks trained on blood tests, body measurements, and sex data in different setups and demonstrated a stable performance of the clocks across the examined general and athletic populations.After matching the ethnicity (considering only Caucasians), health status (considering only reportedly healthy individuals), and age between the NHANES database and the Hungarian athlete datasets, we observed no considerable difference in the accuracy of the clocks across the groups (Table 2).Interestingly, we observed a decrease in performance (a remarkable reduction of r and a slight increase in MAE for both clocks) after narrowing the age range of the NHANES test dataset by age matching to the athlete dataset.This observation indicates that age distribution may influence age prediction performance evaluations.
A previous study examined the performance of blood testbased age prediction models across different populations (18).When the models were trained and tested on the same population, the best performance was r = 0.84 and MAE = 6.25 (East Europe).However, the performance decreased remarkably as the models were trained and tested across different populations.For example, in the case when a model was trained on blood tests and sex variables of the Canadian population and tested on the East European population, the performance degraded to r = 0.52 and MAE = 9.68.Here, we reached similar performances when we trained on the Caucasian American population and tested on the Caucasian Hungarian population.
Blood test-based age acceleration was significantly lower in the athlete dataset for both clocks, which underlines the beneficial effect of regular physical activity.An intriguing finding is that age acceleration is lower in athletic males compared with athletic females (Fig. 2G), but this phenomenon is reversed in nonathletes (Figs.2H and 2I).In other words, performing high-volume physical activity has a greater effect on biological age in young adult males than in females.The sex-separated analysis (Supplemental Fig. 2, Supplemental Digital Content 1, http://links.lww.com/MSS/C991) also supports this idea.
In the general population, BMI was positively associated with increased biological aging measures.By contrast, when the same clock was applied to athletes, BMI was not associated with age acceleration.This suggests that using BMI is more appropriate in nonathletes, but other measures are necessary in athletes to elucidate associations between biological age and body composition.Indeed, in a cohort of 2435 middle-aged, nonathlete subjects, an increased accelerometer-measured physical activity was associated with lower epigenetic age, albeit BMI attenuated this effect (28).On the other hand, Garrido-Chamorro et al. (29,30) found that BMI alone is unsuitable for reporting the quality of body composition in athletes despite showing a good association with body fat content.
By contrast, in athletes, SMM is inversely associated with age acceleration, indicating a beneficial effect on aging.Additionally, body fat percentage is positively associated with age acceleration, suggesting an adverse effect on aging in athletes.These two findings again underscore the importance of applying advanced body composition parameters in athletes within and beyond clinical environments.
Being highly active and engaging in regular physical activity can lead to a wide range of physiological changes in the body, some of which are reflected in laboratory markers (31).These changes may have implications not only in the context of daily diagnostics for clinicians working with athletes but also in the estimation of biological age.Our feature importance analysis revealed several clinically meaningful laboratory parameters (such as serum ferritin, blood urea nitrogen, mean cell volume, and albumin levels) that are important in age prediction by our blood test-based clocks.
The estimation of biological age has become an important tool in the assessment of athletes and physically active individuals in the last decade.In young adolescent athletes, biological maturation has been mainly measured using noninvasive methods, including skeletal age, pubertal status, and peak height velocity (32,33).Sex differences in the rate of decline in physical activity observed during adolescence and the biological age measured using peak height velocity were closely associated with the commencement of the reduction in exercise volume (34).
During adulthood, biological age has other meaningful implications in sports.The optimal or "best" age for different sports and even positions within one discipline (e.g., goalkeeper vs striker in soccer) is not readily definable.At the same time, determining when to retire from a professional athletic career is highly dependent on the individual's performance and, intuitively, biological age.Chronological age may also lack appropriateness in predicting the beginning of a decline in physical performance in athletes approaching the end of their professional careers (35).
In contemporary times, assessing the health status of competitive athletes gains increasing attention, which involves performing comprehensive and complex sports medicine investigations.Laboratory tests are widely available and performed in a variety of clinical settings.Our results suggest that it is practical to define and exploit the potential added value of blood tests in biological age estimation, as long as epigenetic testing is not routinely performed and remains expensive.In addition to its recognized prognostic significance, using the degree and direction of changes in biological age as an outcome variable could emerge as a promising marker for enhancing physical activity-based lifestyle interventions and extending its applicability beyond.Nevertheless, blood test-based biological age measuring models need further improvement and research to increase their accuracy and define applicability in athletes and other populations.
Limitations.Laboratory blood testing in the athlete dataset was carried out after recovery from an asymptomatic, mild, or moderately symptomatic SARS-CoV-2 infection.Although none of the subjects was hospitalized before due to COVID-19 and had no limiting symptoms upon presenting for blood testing, the recent infection might have caused some alterations in their results.On the other hand, we did not find a significant difference in the predicted age in those subjects who returned to a repeated laboratory testing median 107 d later.Furthermore, age accelerations showed a strong association between the two examined time points with both clocks.
In the absence of directly comparable training volumerelated variables, we were not able to contrast the physical activity of the NHANES dataset and the athlete dataset.The lack of a high-volume control group from the same geographical area limited the direct comparison of the Hungarian athlete data to a local control group.We relied on data from the NHANES database in the United States, which provided a comparable set of matched variables, such as ethnicity, health status, and age.Although we achieved the same age prediction performance for the two matched datasets, there still can be some batch effect influencing the results.In addition, the training and testing data contain only Caucasians limiting usage and applicability to other ethnic groups.

CONCLUSIONS
We developed two distinct blood test-based aging clocks demonstrating stable performance and applied them to athletes and healthy controls.Our findings suggest that high-volume sports activity may slow down the aging process.Because laboratory tests are readily accessible, blood test-based aging clocks offer a convenient method for predicting the biological age of diverse populations including athletes.This approach also carries the potential to become an instrument for adjusting training programs and clinical treatments to biological age instead of chronological age.

FIGURE 2 -
FIGURE 2-Inverse association of blood test-based age acceleration and high-volume sports activity.A-B.Differences in age acceleration between the athlete dataset (athlete population) and the age-matched NHANES dataset (general population) evaluated by clocks 1 and 2, respectively.C-F.Pearson correlation coefficient of age, predicted age, delta age, and age acceleration against training volume, SMM, fat percentage, and BMI over the different datasets.G-I.Sex difference in age acceleration over the three datasets.BMI and sex difference are analyzed only for clock 1, as clock 2 used both as training variables.Pred, predicted; Acc, acceleration; vol, volume.

FIGURE 3 -
FIGURE 3-The most important features of the two aging clocks evaluated on the age-matched NHANES test and the athlete datasets.Each dot indicates a test sample.Dots concentrating on the left or right side of the panels indicate a negative or positive effect on the predicted age, respectively.Blue dots indicate low, and red dots indicate high values of the given variable for the given test sample (gray dots indicate missing values).

Table 1 .
Out of the NHANES test set, we selected a control group whose age distribution matched the age distribution of the athlete dataset, which we hereafter refer to as the age-matched NHANES test dataset.By age matching, we sampled 283 of the 1142 individuals in the NHANES test dataset, see Predicted age, delta age, and age acceleration.We examined three distinct aging-related metrics that have the potential to capture the difference between chronological and biological age.These metrics encompass predicted age, delta age, and age acceleration.The predicted age is the output of the age prediction models (clock 1 and clock 2) based on blood test parameters.We hypothesize that predicted age is a better estimate of biological age than chronological age.We calculated delta age (or age gap) by subtracting the chronological age from the predicted age of each individual (i.e., delta age = predicted age − chronological age).This measurement represents the difference between biological age and chronological age.A higher delta age indicates accelerated aging when comparing two individuals with the same chronological age.
Age acceleration is defined byThompson et al. (25)as the deviation of the individual's biological age from the trend.Age acceleration is calculated for an individual as the residual

TABLE 1 .
Descriptive statistics of the measurements used in the athlete dataset, the age-matched NHANES test dataset, and the NHANES full dataset.The first 10 rows contain age, sex, body metrics, and weekly training hours, followed by 36 laboratory blood test measurements.afterfitting a regression line between predicted and chronological age.In our experiments, we relied on the regression line of the NHANES test dataset as a reference for both the NHANES test dataset and the athlete dataset.In our statistical analysis, we preferred age acceleration over delta age because age acceleration is independent of age by definition.In essence, our investigation focuses on the deviation from the usual aging pattern, irrespective of an individual's chronological age.This approach is deemed a more dependable measure for comparing aging-related effects across diverse age groups and populations.

TABLE 2 .
Performance of the blood test-based age prediction models.
FIGURE 1-Chronological and predicted age over the age-matched NHANES test and the athlete dataset.A. Age prediction of clock 1 (blood test only).B.Age prediction of clock 2 (blood test, body metrics, and sex).SPORTS DECELERATE BIOLOGICAL AGING Medicine & Science in Sports & Exercise ® 871APPLIED SCIENCESNHANES test database, we observed that the mean age acceleration of females was lower compared with males (median = 0.