Longitudinal relationships between physical fitness and phase angle as a biomarker of cellular health in Brazilian male adolescents: a Bayesian analysis

Abstract

The phase angle (PhA), derived from bioelectrical impedance analysis (BIA), has gained attention as a non-invasive indicator of cellular health and body composition. However, its relationship with physical fitness development during adolescence remains underexplored. This study aimed to prospectively investigate the longitudinal associations between phase angle and changes in multiple physical fitness components in male adolescents, examining whether PhA serves as a biomarker for fitness development over time. We conducted a repeated-measures study involving 195 male adolescents aged 10–16 years from a school in Campinas, São Paulo, Brazil. Physical fitness and BIA-derived PhA were assessed at three time points across two academic years (2018–2019), yielding 449 observations. Fitness measures included cardiorespiratory fitness (20-m shuttle run), muscular endurance (sit-up), flexibility (sit-and-reach), muscular power (standing long jump, medicine ball throw), speed (20-m sprint), and agility (4-m shuttle run). Phase angle was calculated using single-frequency BIA. Multilevel Bayesian hierarchical models estimated associations between standardized fitness outcomes, age, PhA, and their interaction, incorporating varying intercepts and slopes to account for individual differences. Age showed positive associations with performance in the 20-m shuttle run (, 68% CI = [0.19, 0.31]), sit-up (, 68% CI = [0.25, 0.37]), standing long jump (, 68% CI = [0.30, 0.41]), and medicine ball throw (, 68% CI = [0.50, 0.64]), indicating improvements in aerobic capacity and muscular fitness. Negative associations were observed for the 20-m sprint (, 68% CI = [-0.31, -0.22]) and 4-m shuttle run (, 68% CI = [-0.35, -0.23]), reflecting enhanced speed and agility. Adjustment for body mass and individual variability confirmed these patterns, though the effect of age on medicine ball throw was reduced. Physical fitness improved with age during adolescence, and phase angle demonstrated consistent positive associations with multiple fitness domains over time. These findings suggest that PhA may serve as a practical, non-invasive biomarker for monitoring physical fitness development in adolescents. Given its ease of implementation and cost-effectiveness, PhA could be integrated into school-based health screenings and fitness assessment programs to identify adolescents at risk for poor fitness trajectories and to evaluate the effectiveness of physical activity interventions. Further research should examine these relationships in girls and across diverse populations to establish reference values and intervention thresholds.

Introduction

The assessment of physical fitness in youth is fundamental for promoting health and holistic development during this critical life stage¹. This emphasis on health-related physical fitness is particularly salient when considering the lifestyle habits of children and adolescents², periods marked by rapid growth and development that lay the foundation for future health trajectories^3–5. Tracking body composition and physical fitness during these formative years allows for the early identification of risk factors associated with chronic diseases, thereby informing more effective preventive strategies⁶. Notably, recent evidence suggests that declines in muscular and cardiorespiratory fitness throughout childhood and adolescence may elevate metabolic risks in both these periods and adulthood⁷. The secondary education phase (12-18 years), for example, often coincides with a decrease in physical activity and consequently lower fitness levels, driven by increasing academic demands and the prioritization of study time⁸.

Body composition, defined by the balance of lean tissue and fat mass, is intrinsically linked to physical fitness and crucial for metabolic and cardiovascular health. Among various assessment tools, bioelectrical impedance analysis (BIA) offers an effective, user-friendly, and non-invasive approach⁹. BIA estimates lean body mass, fat mass, and total body water by measuring the body’s resistance to a low-intensity electrical current⁷, and importantly generates the phase angle (PhA). Traditional BIA-derived measures (total body water, intracellular and extracellular water, fat-free mass) rely on population-specific prediction equations that may have limited validity in growing populations^7,10,11. In contrast, PhA—calculated as the arctangent of the reactance-to-resistance ratio—is a raw bioelectrical parameter reflecting cell membrane integrity and the balance between intracellular and extracellular fluid compartments^12,13. This independence from prediction equations makes PhA potentially more universally applicable across ages, ethnicities, and body types^7,14,15.

Phase angle has gained recognition in anthropometric and auxological research as a marker of cellular health and body cell mass quality^16,17. During adolescence, PhA undergoes age- and maturation-related changes, with sex differences emerging during puberty as boys accumulate greater lean tissue mass¹⁸. Reference values have been established in several pediatric populations, though variation across countries and ethnic groups highlights the need for population-specific norms^7,10,11. Importantly, PhA reflects tissue quality—cellular integrity, membrane function, and the ICW/ECW balance—rather than simply tissue quantity^13,19. This distinction is critical: two individuals with identical muscle mass may exhibit different functional capacities depending on cellular health¹⁹. PhA has demonstrated superior predictive validity compared to traditional body composition measures for functional outcomes including muscle strength and physical performance^1,7, with methodological advantages including reduced equation-dependent measurement error and better reproducibility²⁰.

Physical fitness encompasses components such as muscular strength, cardiovascular endurance, flexibility, and agility²¹, often evaluated using methods like the 20-meter shuttle run, standing long jump, and handgrip strength tests²². These health-related fitness components are critical indicators of physical health and functional capacity in youth^22,23. A correlation between PhA and physical fitness has been shown²⁴, particularly upper body strength, in teenage boxers, highlighting PhA’s potential as an indicator of muscle quality¹³ and a tool for sarcopenia diagnosis²⁵. Emerging research further suggests PhA as a reliable predictor of physical performance and muscle function in both younger^7,14 and young adult populations¹⁵.

Within exercise and sport sciences, cross-sectional studies have documented positive associations between PhA and various fitness components across athletic populations, with athletes typically exhibiting higher PhA values than non-athletes²⁶. Longitudinal intervention studies demonstrate that PhA is responsive to training programs^27,28. However, most research employs cross-sectional designs in elite athletic populations, with limited evidence examining PhA-fitness relationships across multiple domains in general youth populations using longitudinal frameworks^18,29.

Despite growing understanding of the interplay between body composition, physical fitness, and PhA, a significant gap remains in longitudinal studies examining the evolution of these factors throughout childhood and adolescence, particularly in school-based populations. Longitudinal analysis is critical for distinguishing between-person differences from within-person developmental changes. Furthermore, it remains unclear whether PhA relationships with physical fitness are domain-specific or reflect general associations with physical function across multiple health-related fitness components. Bayesian multilevel modeling offers distinct advantages for capturing individual developmental trajectories and appropriately quantifying uncertainty^30,31. Therefore, this study aims to prospectively investigate the relationship between PhA and changes in physical fitness—including both health-related components (cardiorespiratory fitness, muscular endurance, flexibility) and performance-related components (muscular power, speed, agility)—among adolescent students, examining whether PhA serves as a biomarker for fitness development over time.

Methods

Study design and participants

This study employed a repeated-measures design, involving 195 male adolescents recruited through convenience sampling from a local school in Campinas, São Paulo, Brazil. Participants were eligible for inclusion if they met the following criteria: (a) active enrollment in the school; (b) regular attendance in Physical Education classes; and (c) age between 10 and 16 years. Adolescents were excluded if they: (a) had physical disabilities (permanent or temporary) that could interfere with participation in study procedures; (b) were taking prescribed medication; or (c) failed to return a signed informed consent form.

Data collection occurred at three time points over two academic years: February–April 2018, November–December 2018, and February–April 2019. Only participants with at least two measurements, including February–April 2018 and February–April 2019, were included in the analysis. Of the total sample, 59 participants completed assessments at all three time points, while 136 participants had two assessments, resulting in a total of 449 measurements. Our Bayesian multilevel modeling approach naturally accommodates this unbalanced design, using all available data without requiring complete cases (see Statistical Analysis).

All participants were enrolled in the same public school serving a middle-income neighborhood in Campinas. As part of the Brazilian educational system, all students participated in mandatory Physical Education classes twice weekly (90 minutes total). Information on organized sports participation outside of school, dietary habits, and socioeconomic indicators was not systematically collected.

Written informed consent was obtained from all participants and their parents or legal guardians after they were informed of the study’s nature. Participation was voluntary, and participants could withdraw at any time. The study adhered to the ethical principles outlined in the Declaration of Helsinki. Ethical approval was granted by the Research Ethics Committee of the Pontifical Catholic University of Campinas (CAAE: 24727119.1.0000.5481).

Procedures

Age and anthropometry

Chronological age was calculated to the nearest 0.1 year by subtracting the birth date from each assessment date. Stature was measured using a vertical portable stadiometer (Sanny, SBC, SP, Brazil) with a precision of 0.1 cm. Body mass was assessed using a calibrated portable scale (Sanny Digital Glass 200 Control, SBC, SP, Brazil) with an accuracy of 0.05 kg. Based on replicated measurements from 20 participants, the technical errors of measurement were 0.29 cm for stature and 0.51 kg for body mass¹⁸.

Phase angle

The PhA was assessed using a single-frequency (50 kHz) bioelectrical impedance analysis (BIA) device, Quantum II (RJL Systems, Detroit, MI, USA). Before the measurement, participants were instructed to remove all metallic objects. They were then positioned barefoot in a supine posture, with legs abducted at a 45 angle, arms away from the trunk, and hands pronated on a table isolated from electrical conductors.

After five minutes of rest, the skin was cleaned with alcohol, and four electrodes were placed according to standard procedures³²: two on the right hand and two on the right foot. The BIA measurement lasted approximately one minute and provided resistance () and reactance () values in ohms (). The phase angle was then calculated using the established equation (12):

All BIA measurements were conducted during morning Physical Education classes (08:00-11:00h) at all three time points to minimize diurnal variation. Participants were instructed to: (1) avoid vigorous physical activity for 24 hours prior, (2) void bladder before assessment, and (3) limit fluid intake in the hour preceding measurement. Measurements were taken at least 2 hours after breakfast. All assessments were conducted in a temperature-controlled environment (22-24C). While these procedures enhanced measurement standardization, we did not enforce strict fasting protocols or systematically record exact timing of food/fluid intake.

The RJL Quantum II single-frequency bioimpedance analyzer has demonstrated excellent validity and reliability in pediatric populations, with test-retest ICC values >0.95 and high concordance with multi-frequency devices^32,33. In our study, the device was calibrated weekly using the manufacturer’s 500-ohm resistor to ensure measurement accuracy. All measurements were performed by a single trained examiner following standardized protocols. Measurement reliability was assessed through replicated measurements in a subsample of 23 participants, yielding technical errors of measurement of 3.54 () for resistance and 0.49 () for reactance, with corresponding coefficients of variation of 0.35% and 0.33%, respectively¹⁸, indicating excellent measurement precision consistent with published standards³⁴.

Physical fitness

Physical fitness was assessed using a battery of tests adapted from Silva et al.³⁵, which has been validated in Brazilian adolescent populations. The battery assessed both health-related fitness and skill-related fitness components.

Health-related components included: cardiorespiratory fitness (20-meter shuttle run test³⁶, similar to FITNESSGRAM PACER), flexibility (sit-and-reach test using Wells’ bench with feet at 23-cm mark), and abdominal muscular endurance (1-minute sit-up test).Skill-related components included: lower limb power (standing long jump), upper limb power (medicine ball throw with 2-kg ball), speed (20-meter sprint), and agility (4-meter shuttle run).

While health-related fitness has the most direct associations with health outcomes²², we included skill-related components because they are standard in Brazilian school fitness assessments, relate to metabolic health in youth³⁷, and provide comprehensive assessment of physical function during adolescent development. All tests followed standardized protocols³⁵.

Before each assessment, the objectives and procedures of the tests were explained to the participants to ensure proper understanding. Additionally, all participants completed a pre-test before performing the physical fitness assessments to familiarize themselves with the procedures.

Statistical analysis

We adopted a Bayesian multilevel modeling framework, which provides several advantages for longitudinal developmental data. First, Bayesian inference enables direct probability statements about parameters through credible intervals, avoiding interpretational difficulties associated with frequentist approaches^30,31. Second, Bayesian hierarchical models naturally implement partial pooling through shrinkage, where individual-level parameters are informed by population-level distributions, improving stability of estimates particularly for individuals with fewer measurements^31,38. Third, Bayesian methods provide full posterior distributions for all parameters, including variance components, enabling more complete uncertainty quantification than approaches that provide only point estimates³¹. Fourth, Bayesian estimation exhibits robust convergence properties for complex hierarchical models with multiple random effects and interactions^39,40. Fifth, Bayesian posterior predictive checks provide comprehensive model validation tools⁴¹. These features are especially important in developmental research, where capturing inter-individual variability in growth patterns and providing appropriate uncertainty quantification are essential.

Our Bayesian approach naturally accommodates unbalanced data and missing observations. Participants with two measurements (n=136) and those with three measurements (n=59) all contributed to both individual-level trajectories and population-level effects under missing at random assumptions. The hierarchical structure allows participants with fewer observations to benefit from partial pooling, where individual estimates are informed by population-level patterns³⁸.

To explore the relationships between each physical fitness outcome and PhA across adolescence, we modeled each individual’s standardized fitness score () as a function of standardized age () and standardized phase angle (). The model incorporated both varying intercepts and varying slopes for individual participants, accounting for inter-individual variability. We standardized the outcomes to ensure computational efficiency and ease of interpretation, by subtracting each outcome from its sample mean and dividing by the standard deviation. The model is specified as follows:

where, indexes the individuals, and indexes the measurement occasions; represents the standardized physical fitness for participant at time ; is the expected value of the standardized physical fitness at time ; is the population-level intercept, and are the population-level parameters for standardized age, PhA, and their interaction; represents individual-level deviations (between-individuals), modeled as a function of varying intercepts () and varying slopes () for age and phase angle. The individual-level effects follow a multivariate normal distribution with a mean vector and covariance matrix , which captures between-subject variability and correlations among individual-level effects. The covariance matrix is specified as:

where, , , and are the variance components for the individual-level intercept, age effect, and PhA effect, respectively. , , and are the correlations between varying intercept and varying slopes, capturing the degree to which individual differences in initial fitness levels (intercept) are associated with differences in how standardized age and PhA affect fitness.

We specified weakly informative priors to ensure regularization of our estimates:

To check whether the models successfully partitioned the influence of body mass (when included in the model), we inspected the residual plots against the body mass to check the homoscedasticity of the residuals. We ran four chains for 2,000 iterations with a warm-up length of 1,000 iterations in each model. Model convergence was assessed through visual inspection of trace plots and quantitative diagnostics. All values were < 1.01, indicating successful convergence, and effective sample sizes exceeded 1,000 for all parameters, ensuring adequate precision of posterior estimates. Posterior predictive checks confirmed good model fit, with simulated data from the posterior predictive distribution closely matching observed distributions across all fitness outcomes⁴¹. We fitted the models in R⁴² using the brms package³⁹, which calls Stan⁴³. The model predictions were extracted using the tidybayes package⁴⁴ and visualized with the ggplot2 package⁴⁵.

Results

Characteristics of the participants at the study baseline, as reference for description, are summarized in Table 1.

Table 1.

Characteristics of the sample at the study baseline.

	Average	Standard deviation
Chronological age, yr.	12.6	1.2
Stature, cm	154.9	11.5
Body mass, kg	49.7	13.5
Phase angle	5.86	0.77
20-m shuttle run test, m	85	38
Sit-and-reach test, cm	20.4	7.5
1-minute sit-up test, #.	33	10
Standing long jump test, cm	152.4	26.6
Medicine ball throw test, m	3.1	0.7
20-meter sprint test, s	4.31	0.57
4-m shuttle run test, s	6.73	0.59

The Bayesian regression models showed that the population-level effect parameters for standardized age were positive for the 20-m shuttle run (; Fig. 1, panel A), sit-up (; Fig. 3, panel A), standing long jump (; Fig. 4, panel A), and medicine ball throw (; Fig. 5, panel A), indicating an improvement in these physical fitness indicators with increasing age.

Fig. 1.

Longitudinal relationships between the 20-m shuttle run test and phase angle (Panel A) in Brazilian male adolescents, adjusted for within- and between-player variations in body size (Panel B). The 68% uncertainty interval for the posterior predictions is comparable to a standard deviation.

Fig. 3.

Longitudinal relationships between the 1-min sit-ups test and phase angle (Panel A) in Brazilian male adolescents, adjusted for within- and between-player variations in body size (Panel B). The 68% uncertainty interval for the posterior predictions is comparable to a standard deviation.

Fig. 4.

Longitudinal relationships between the Standing long jump test and phase angle (Panel A) in Brazilian male adolescents, adjusted for within- and between-player variations in body size (Panel B). The 68% uncertainty interval for the posterior predictions is comparable to a standard deviation.

Fig. 5.

Longitudinal relationships between the 2-kg medicine ball throw test and phase angle (Panel A) in Brazilian male adolescents, adjusted for within- and between-player variations in body size (Panel B). The 68% uncertainty interval for the posterior predictions is comparable to a standard deviation.

In contrast, the population-level effect parameters for standardized age were negative for the 20-m sprint test (; Fig. 6, panel A) and the 4-m shuttle run test (; Fig. 7, panel A). Since lower times indicate better performance in these tests, the negative coefficients suggest that sprint and agility performance improved with age.

Fig. 6.

Longitudinal relationships between the 20-m sprint test and phase angle (Panel A) in Brazilian male adolescents, adjusted for within- and between-player variations in body size (Panel B). The 68% uncertainty interval for the posterior predictions is comparable to a standard deviation.

Fig. 7.

Longitudinal relationships between the 4-m shuttle run test and phase angle (Panel A) in Brazilian male adolescents, adjusted for within- and between-player variations in body size (Panel B). The 68% uncertainty interval for the posterior predictions is comparable to a standard deviation.

With the exception of the medicine ball throw, all population-level effects for standardized age remained similar when adjusting for between-participant variation (see panel B in the respective figures). For the medicine ball throw, however, the magnitude of the standardized age parameter decreased after adjusting for body dimensions (; Fig. 5, panel B).

For the sit-and-reach test, the population-level effect parameter for age was negative (; Fig. 2, panel A), indicating a decline in flexibility with increasing age. However, when body size was included in the model, the standardized age parameter became positive (; Fig. 2, panel B), suggesting that changes in flexibility with age should be interpreted in the context of body size adjustments.

Fig. 2.

Longitudinal relationships between the Sit-and-reach test and phase angle (Panel A) in Brazilian male adolescents, adjusted for within- and between-player variations in body size (Panel B). The 68% uncertainty interval for the posterior predictions is comparable to a standard deviation.

The Bayesian regression models also showed that the population-level effect parameters for standardized PhA were positive for the 20-m shuttle run (; Fig. 1, panel A), sit-and-reach (; Fig. 2, panel A), sit-up (; Fig. 3, panel A), standing long jump (; Fig. 4, panel A), and medicine ball throw (; Fig. 5, panel A).

Since lower times indicate better performance in these tests, the population-level effect parameters for standardized PhA were negative for the 20-m sprint test (; Fig. 6, panel A) and the 4-m shuttle run test (; Fig. 7, panel A).

Overall, the population-level effect parameters for standardized PhA suggest a relationship between improvements in physical fitness and PhA. These patterns remained similar when variation in body size was accounted for in the models.

Discussion

This longitudinal study investigated the relationship between PhA and changes in physical fitness among adolescent students over time. The main finding is a positive association between PhA values and multiple measures of physical fitness, suggesting that higher PhA levels are indicative of more favorable physical performance trajectories in young males. This aligns with previous research identifying PhA as a valuable functional marker in youth⁴⁶.

While earlier studies have linked PhA to cardiometabolic risk indicators across pediatric populations⁴⁶, the present study extends this evidence base by adopting a longitudinal design. This approach enables the tracking of associations over time and represents a methodological advancement in the field. The use of Bayesian methods is particularly well-suited for repeated-measures data, allowing for the modeling of individual variability and the detailed exploration of developmental trajectories.

In assessing the influence of chronological age on physical fitness components, our findings reinforce existing literature showing progressive improvements in muscle strength and cardiorespiratory endurance throughout adolescence^22,26. Interestingly, a decline in flexibility was initially observed; however, this trend became more favorable once body size was accounted for⁴⁷. This pattern likely reflects the biomechanical challenges of rapid stature growth during adolescence: as limb length increases faster than soft tissue extensibility, relative flexibility (not adjusted for size) may appear to decline, even as absolute range of motion improves. This finding underscores the critical importance of appropriately scaling fitness measures to body dimensions during periods of rapid growth⁴⁸, and highlights that apparent developmental declines in some fitness components may be artifacts of failing to account for changing body proportions. This underscores the importance of considering anthropometric variables when interpreting changes in physical performance, especially during periods of rapid growth¹⁸.

Our findings extend across both health-related fitness components (cardiorespiratory fitness, muscular endurance) and performance-related components (power, speed, agility). While health-related fitness has the most direct associations with chronic disease risk^22,37, performance-related fitness components also contribute to overall physical function and may reflect neuromuscular development during adolescence⁴⁹. The consistent positive associations between PhA and multiple fitness domains—both health-related and performance-related—suggest that PhA may serve as a general marker of physical function rather than being specific to particular fitness components.

Our findings have several practical implications for adolescent health assessment. Phase angle measurement via BIA offers a non-invasive, relatively inexpensive, and rapid assessment tool that could be integrated into school-based health screenings. Unlike traditional fitness tests that require considerable space, time, and physical exertion, PhA can be measured in minutes with minimal burden on students and staff. Our results suggest PhA may serve as a general marker of physical function across multiple fitness domains, potentially identifying adolescents with poor cellular health or compromised fitness trajectories who would benefit from targeted interventions. Furthermore, the sensitivity of PhA to training and nutritional status suggests it could be used to monitor responses to school-based physical activity or nutrition programs. However, establishing age- and sex-specific reference values and intervention thresholds for diverse populations remains a priority before widespread implementation.

PhA stands out for being non-invasive, cost-effective, and easy to implement, making it a practical tool for both educational and clinical environments. Recent research has shown that PhA is responsive to physical exercise interventions, particularly those aimed at improving muscular strength and aerobic fitness²⁹. This sensitivity enhances its utility in school-based health promotion programs, where it could support both developmental monitoring and intervention evaluation.

Several important limitations warrant consideration. First and most critically, our sample included only male adolescents from a single school using convenience sampling, substantially limiting generalizability to female adolescents, other age groups, socioeconomic contexts, and diverse populations. Second, we did not systematically collect data on organized sports participation, dietary habits, training regimens, or detailed physical activity patterns beyond mandatory Physical Education classes. Given that PhA is influenced by training status, diet, and hydration⁵⁰, these unmeasured factors may contribute to individual variability in PhA trajectories. However, our Bayesian multilevel models explicitly account for individual heterogeneity through varying intercepts and slopes, capturing this variability in the model structure. Third, while we adjusted for body size (mass and stature) in our analyses, we did not assess biological maturation status (e.g., secondary sexual characteristics, skeletal age), which may moderate relationships between PhA and fitness during puberty¹⁸. Future longitudinal studies should include comprehensive assessments of maturation, sports participation, dietary intake, and other lifestyle factors to better characterize determinants of PhA-fitness relationships across adolescence and to examine whether PhA provides unique information beyond traditional body composition measures.

In conclusion, this study demonstrates consistent positive associations between phase angle and multiple physical fitness components during male adolescence, suggesting PhA may serve as a practical biomarker for tracking physical development. Future research should: (1) examine these relationships in female adolescents, where different pubertal trajectories may produce distinct patterns, (2) include comprehensive assessments of biological maturation, sports participation, and dietary habits to identify determinants of PhA-fitness relationships, (3) conduct intervention studies to assess PhA responsiveness to school-based fitness programs, and (4) establish population-specific reference values and clinical thresholds to facilitate PhA implementation in youth health assessments.

Author contributions

AMdM, RRB, GF and HMC, conceived, designed, and helped to write and revise the manuscript; AMdM, GG-J and HMC, were responsible for coordinating the study, contributed to the intellectual content, and revise the manuscript, AMdM, FCK, GG-J and HMC interpreted the data, helped to write and revise the manuscript. All authors contributed to the study design, critically reviewed the manuscript, and approved the final version.

Funding

FCK was supported by grants from the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior – CAPES (finance code 001). The study was partially supported by the Fundação de Amparo à Pesquisa do Estado de São Paulo (São Paulo Research Foundation).

Data availability

The datasets and codes to reproduce the analysis presented in this study can be found at https://osf.io/grb56/.

Declarations

Competing interests

The authors declare no competing interests.