Predicted versus measured thoracic gas volume in estimating body fat percentage among male college athletes and sedentary controls: A cross-sectional observational study

INTRODUCTION

Body composition assessment is essential in sports science, and accurately measuring body fat percentage (BF%) is crucial for optimising athletic performance, training plans and monitoring health. The fat-to-lean mass ratio in athletic populations directly correlates with performance outcomes, though optimal ratios vary considerably across sports disciplines.^[1,2] In competitive sports, accurate body composition measurement is particularly important, as even minor measurement errors can significantly impact training decisions, performance evaluation and athlete classification, especially when athletes are near critical thresholds for weight categories or team selection.

Air displacement plethysmography (ADP) via the Body Composition Pod (BOD POD) has become a widely adopted body composition assessment tool because it is non-invasive, rapid, and demonstrates high test-retest reliability compared with conventional methods such as hydrostatic weighing and skinfold measurements.^[3,4] The BOD POD determines body composition by measuring body volume through air displacement in a sealed chamber, calculating body density and subsequently estimating BF%. However, the accuracy of these calculations critically depends on thoracic gas volume (TGV), which represents the air volume within the lungs and airways during measurement. TGV can be either predicted using demographic equations (TGVp) based on age, sex, height and weight or directly measured (TGVm) through spirometric assessment during controlled breathing manoeuvres against an occluded airway.

Previous studies have consistently demonstrated substantial discrepancies between these methods, with TGVm exceeding TGVp, particularly in athletic populations. Miller et al.^[5] reported that male college athletes had significantly higher TGVm than TGVp (4.59 ± 0.88 L vs. 4.11 ± 0.45 L, p < 0.001), leading to systematic BF% underestimation of 1.2 ± 1.8% when using predicted values. Similarly, Ducharme et al.^[6] found that TGVp underestimated TGVm by 0.49 ± 0.7 L in trained men, resulting in BF% underestimation of 1.3 ± 1.8%. These findings suggest that prediction equations developed on general populations may not adequately account for training-induced respiratory adaptations in athletes.

Despite these established discrepancies, critical gaps remain in understanding how different training modalities influence TGV prediction accuracy. Most existing studies have either examined heterogeneous athletic populations without distinguishing between training types or focused on single sports.^[6-9] This limitation is significant because different training modalities induce distinct physiological adaptations. Anaerobic training, encompassing sports such as sprinting, weightlifting and throwing events, primarily promotes muscular hypertrophy, phosphocreatine system development and neuromuscular power adaptations.^[10,11] In contrast, intermittent sports requiring repeated high-intensity efforts with recovery periods, such as football, basketball and hockey, induce mixed adaptations including enhanced aerobic capacity, anaerobic power and specific respiratory muscle endurance.^[12] These differential adaptations likely result in varying patterns of lung volume changes that could affect prediction equation accuracy differently across athlete types.

Furthermore, no studies have systematically examined whether measurement errors exhibit constant bias (consistent underestimation across all values) or proportional bias (errors that vary with measurement magnitude) across different training modalities. This distinction has important practical implications: constant bias might be correctable with simple adjustment factors, while proportional bias would require measurement-specific corrections or mandate direct TGV measurement for accurate assessment.

Therefore, this study aimed to determine whether TGVp provides accurate BF% estimation compared to TGVm in sedentary individuals, anaerobic athletes and intermittent sport athletes using BOD POD measurements. Specifically, we examined: (1) The magnitude of difference between TGVp and TGVm across groups; (2) The resulting impact on BF% estimation accuracy; and (3) Whether measurement errors demonstrate constant or proportional bias patterns. We hypothesised that TGVm would significantly exceed TGVp in both athletic groups, with both the magnitude and pattern of difference varying by training modality. We further hypothesised that athletic populations would exhibit proportional bias, where prediction equation accuracy deteriorates as physiological adaptations increasingly deviate from population norms. These findings will provide evidence-based guidance for optimising body composition assessment protocols and identifying when direct TGV measurement is essential versus when predicted values might suffice for different populations.

MATERIALS AND METHODS

RESULTS

DISCUSSION

This study examined the accuracy of TGVm versus TGVp in body composition assessment across different training modalities. Our primary finding was that TGVp systematically underestimated TGVm in all groups, leading to a clinically significant BF% underestimation of 2.0–2.5%. However, the relationship between training modality and prediction accuracy proved more complex than anticipated, with distinct patterns of constant versus proportional bias emerging across groups. These findings have important implications for body composition assessment protocols in athletic populations.

A particularly striking discovery was the presence of proportional bias in athletic groups but not in sedentary individuals. The sedentary group exhibited constant bias across all measurement ranges (β = 0.334, p = 0.070), suggesting that a simple correction factor might suffice for this population. In contrast, the anaerobic group demonstrated significant proportional bias for TGV (β = 0.158, p = 0.005, R² = 0.335), indicating that prediction error increased with larger lung volumes. This pattern suggests that as anaerobic athletes develop greater lung capacity, potentially through training-induced respiratory muscle strengthening, the prediction equations become progressively less accurate.

Most remarkably, the intermittent sport group showed complex bias patterns: negative proportional bias for TGV (β = −0.191, p = 0.045) but strong positive proportional bias for BF% (β = 0.510, p < 0.001, R² = 0.839). The negative TGV coefficient represents an unexpected phenomenon where prediction accuracy actually improved for athletes with larger lung volumes. We hypothesise this may reflect sport-specific adaptations in intermittent athletes that coincidentally align with prediction model assumptions at higher lung volumes. However, the strong proportional bias in BF% indicates that, despite this apparent TGV improvement, the error in final body composition estimates varies dramatically across the body fat spectrum. Athletes with lower body fat show different prediction errors than those with higher body fat, potentially leading to misclassification near critical thresholds for team selection or weight categories.

The differential bias patterns across training modalities likely reflect distinct physiological adaptations. Anaerobic training induces adaptations that may influence TGV measurement accuracy through mechanisms that remain incompletely understood.^[15] The proportional bias observed in anaerobic athletes (β = 0.158, p = 0.005) suggests that prediction errors increase with larger lung volumes, possibly reflecting training-induced changes in respiratory muscle function or breathing patterns not accounted for in standard prediction equations. However, the specific physiological mechanisms underlying this bias require further investigation. These changes may affect the pressure–volume relationships assumed by prediction equations, which were developed using general populations. The proportional nature of the bias suggests these effects amplify with greater muscular development.

Intermittent sport athletes demonstrated the highest absolute TGVm values (6.27 ± 1.10 L) among the three groups. This finding aligns with research showing that sports requiring repeated high-intensity efforts can induce substantial respiratory adaptations, including enhanced respiratory muscle endurance and increased lung volumes.^[12] The mixed metabolic demands of sports such as basketball, football and hockey, combining sustained moderate activity with repeated sprints, may stimulate unique respiratory adaptations not captured by traditional aerobic-anaerobic classifications. The large standard deviation in this group (±1.10 L) suggests considerable heterogeneity, possibly reflecting positional differences, varying playing time or individual training emphasis.

Our findings both confirm and extend previous research. The overall magnitude of TGV underestimation (0.29–0.54 L) and resulting BF% errors (2.0–2.5%) align with previous findings.^[5,6] However, our training-modality-specific analysis reveals previously unrecognised complexity. Wagner et al.^[9] noted that prediction errors could vary substantially among individuals, our study is the first to demonstrate systematic proportional bias patterns that differ by training type.

The finding that all groups, including sedentary individuals, showed significant BF% underestimation challenges the assumption that prediction equations work adequately for non-athletic populations. Even our sedentary group, presumably most similar to the populations used to develop prediction equations, showed very large effect sizes (Cohen’s d = −3.64 for TGV, d = −4.86 for BF%). This suggests fundamental limitations in current prediction models that extend beyond athletic adaptations.

The presence of proportional bias has critical implications for practitioners. For anaerobic athletes and especially intermittent sport athletes, no single correction factor can address prediction errors across all individuals. The strong proportional bias in BF% for intermittent athletes (R² = 0.839) means errors vary predictably but dramatically with body composition level. An athlete with 8% body fat might have a different magnitude of error than a teammate with 15% body fat, despite identical measurement protocols.

These findings strongly support universal adoption of measured TGV for individual athlete assessment. The magnitude of error we observed (2.0–2.5% BF% underestimation) exceeds the typical measurement precision of BOD POD itself (±1–2%), effectively doubling the potential error in body composition assessment. This degree of inaccuracy has substantial practical implications across multiple domains of sports medicine and performance management. In clinical settings, such errors could lead to misclassification of athletes relative to body composition thresholds, potentially missing cases of relative energy deficiency in sport or inappropriately clearing athletes with suboptimal body composition. For weight-class sports, where athletes often manipulate body composition to compete in specific categories, a 2–2.5% underestimation could mask dangerous weight-cutting practices or result in suboptimal category selection. The impact extends to rehabilitation contexts, where accurate body composition monitoring is essential for tracking muscle mass preservation during injury recovery and determining readiness for return to play. Furthermore, coaches and sports scientists relying on body composition trends to evaluate training effectiveness may draw incorrect conclusions about program efficacy, potentially leading to unnecessary or counterproductive training modifications. The consistency of BF% underestimation across all groups, despite varying mechanisms of error as confirmed by our Kruskal–Wallis analysis (H[2] = 3.39, p = 0.184), underscores that this is not merely a concern for elite athletes but represents a systematic measurement issue affecting all populations assessed with predicted TGV.

Our use of Welch’s ANOVA and Games–Howell post hoc tests to address heterogeneous variances strengthens our conclusions. The violation of homogeneity assumptions itself provides valuable information-the greater variance in TGV measures among athletic groups (particularly intermittent athletes) suggests that training induces not just mean changes but also increased individual variability in lung adaptations. This heterogeneity further supports individualised measurement rather than population-based predictions.^[3]

The lack of significant correlation between TGV errors and BF% errors within groups (all r < 0.23, p > 0.31) indicates that the relationship between lung volume prediction error and body composition error is complex and likely mediated by body density calculations. This finding underscores that even small TGV errors can have unpredictable effects on final BF% estimates.

Several limitations of the paper should also be discussed. First, our sample comprised only college-aged males, limiting generalisability to females, younger athletes or elite competitors. Sex differences in body composition and respiratory function may yield different bias patterns. Second, while all athletes met minimum training criteria (≥2 years, 8–12 h/week), we lacked detailed training history data that might explain within-group variability. Third, our cross-sectional design cannot determine whether bias patterns change with training progression or detraining.

In addition, while we identified proportional bias, we cannot definitively explain the physiological mechanisms. The negative proportional bias in intermittent athletes’ TGV remains particularly puzzling and warrants further investigation. Finally, we compared predicted versus measured TGV but did not validate against a four-compartment model, meaning our ‘accuracy’ assessments are relative rather than absolute.

Our findings suggest several research priorities. The development of training-specific prediction equations could improve accuracy, though our proportional bias results suggest simple adjustments may be insufficient. Machine learning approaches incorporating multiple anthropometric and training variables might better capture the complex relationships we observed. Longitudinal studies tracking athletes through training cycles could identify when prediction equations begin failing and whether bias patterns are stable within individuals.

Investigation of sex differences in training-related bias patterns is essential for extending these findings to female athletes. Similarly, examining elite athletes would determine whether bias patterns intensify with higher training volumes or reach a plateau. Understanding the physiological basis for negative proportional bias in intermittent athletes could reveal important insights about respiratory adaptations in team sports.

CONCLUSION

This study demonstrates that the relationship between training modality and TGV prediction accuracy is more complex than previously recognised, with distinct patterns of constant and proportional bias emerging across different athletic populations. While TGVp consistently underestimated TGVm by 0.29–0.54 L across all groups, leading to clinically significant BF% errors of 2.0–2.5%, the nature of these errors varied substantially. Sedentary individuals showed constant bias, anaerobic athletes demonstrated proportional bias in TGV, and intermittent sport athletes exhibited strong proportional bias in BF% (R² = 0.839). These findings indicate that predicted TGV should not be used for individual body composition assessment in any population, but particularly not in athletes, where proportional bias makes corrections more complicated. Direct measurement of TGV is essential for accurate body composition assessment, regardless of training status or apparent prediction accuracy.