European Society of Cardiology sport categories capture remodelling gradients but cannot reliably classify individual cardiac magnetic resonance phenotypes in elite athletes
Key result
ESC sport categories link to ~3.7-fold higher RVEDV remodelling odds per category but poorly predict phenotypes.
- OR 3.73
- 95% CI 2.88-4.85
- P<0.001
- n=1,029
Why the study?
The ESC sport classification is based largely on physiological assumptions and echocardiography, and had not been validated against large-scale CMR phenotypes in elite athletes.
Does the European Society of Cardiology (ESC) sport classification accurately reflect cardiac magnetic resonance (CMR)-derived remodelling gradients and reliably predict individual CMR phenotypes in elite athletes?
Key Points
- To evaluate whether European Society of Cardiology sport categories follow expected cardiac remodeling trends on cardiac magnetic resonance imaging and can accurately predict individual athlete phenotypes.
- Cross-sectional CMR study evaluating 1,029 participants (ELITE+ cohort), consisting of 780 elite athletes across 48 sports (mean age 25±6 years, 40% women) and 249 sex-matched controls.
- Athletes were assigned to ESC categories (Skill, Power, Mixed, Endurance) and CMR metrics were converted to Z-scores using sex-matched controls.
- Associations were tested using linear and ordinal logistic regression, with individual phenotype classification accuracy quantified via isotonic regression-derived thresholds.
- Right ventricular end-diastolic volume (RVEDV) had the strongest category association (R²=0.168, P<0.001), with mean Z-scores of 0.54 (Skill), 0.57 (Power), 1.14 (Mixed), and 1.99 (Endurance).
- RV remodeling prevalence increased across categories from 8% to 10%, 29%, and 62% (odds ratio 3.73 per category, 95% CI 2.88–4.85, P<0.001).
- Individual phenotype prediction achieved a balanced accuracy of 39% (95% CI 34–46%) against a 25% chance level, with positive predictive values of 6.4% (Skill), 13.9% (Power), 53.8% (Mixed), and 61.7% (Endurance).
Study Design
Type
Cross-Sectional (n=1,029)
Structured PICO
Does the European Society of Cardiology (ESC) sport classification accurately reflect cardiac magnetic resonance (CMR)-derived remodelling gradients and reliably predict individual CMR phenotypes in elite athletes?
P
Population
1,029 individuals including 780 elite athletes (mean age 25 years, 40% women, across 48 sports) and 249 sex-matched controls from two harmonised cohorts.
E
Exposure
European Society of Cardiology (ESC) sport category assignment per primary sport (Skill, Power, Mixed, Endurance)
C
Comparator
Sex-matched non-athlete controls (n=249) used to convert CMR metrics to Z-scores
O
Outcome
Association between CMR-derived cardiac remodelling metrics (primarily right ventricular end-diastolic volume [RVEDV] Z-score) and ESC sport categories, and balanced accuracy of ESC sport categories for predicting individual CMR phenotypessurrogate
Main Result
Odds Ratio: 3.73 (95% CI 2.88–4.85)
Absolute Event Rate: 62% vs 8%
p-value: p=<0.001
Although ESC sport categories capture population-level gradients in right ventricular remodelling on CMR, they have poor accuracy (39%) for classifying individual athlete phenotypes, supporting a phenotype-led rather than sport category-led evaluation of athletes.
Limitations
- Overlap in RVEDV values across categories, particularly in Skill and Power, limits individual phenotype discrimination.
Key expert perspectives
Captured external expert commentary on this paper, strongest first. Original sources are linked where available.
“The clinician should first look at the heart muscle, the chambers, and the relationships between them, and combine that with symptoms, family history, and training data. Sport type should not take centre stage.”
“In athletes, the question is often whether an abnormal heart shape is caused by exercise or by disease. This study shows that relying solely on sport type to assess an athlete's heart carries risks.”
Implication
RV remodeling tracks ESC categories at group level; leaves open reliable individual phenotyping in athletes.
Abstract
Background The European Society of Cardiology (ESC) ranks sports (Skill, Power, Mixed, Endurance) to reflect presumed increasing cardiovascular load and expected cardiac remodelling (Fig. 1). It is based largely on physiological assumptions and echocardiography, and has not been validated against large-scale cardiac magnetic resonance (CMR) phenotypes in elite athletes. Purpose To evaluate the ESC sport classification by formally testing how closely CMR-derived remodelling follows the expected increasing trend across ESC categories, and how accurately sport categories can predict individual CMR phenotypes. Methods Cross-sectional CMR study in ELITE+, comprising two harmonised elite athlete cohorts. Athletes were assigned ESC categories per primary sport. Indexed left ventricular (LV) mass, biventricular end-diastolic volumes (EDV), LV wall thickness, ejection fractions, and three concentricity indices were converted to Z-scores using sex-matched controls. Associations between each CMR metric and ESC categories were evaluated using linear regression; the metric with the highest explained variance defined the primary CMR remodelling axis (data-driven). We tested whether remodelling on this axis (Z-score ≥1.645) increased across categories using logistic regression with ESC category as an ordinal predictor. For individual phenotype prediction, we derived thresholds using isotonic regression and quantified balanced accuracy (macro-averaged sensitivity; bootstrap 95%CI). Results We included 1,029 individuals: 780 elite athletes (25±6 years, 40% women, 48 sports) and 249 controls. ESC categories were Skill (n=24; 3%), Power (n=78; 10%), Mixed (n=365; 47%), and Endurance (n=313; 40%). RVEDV showed the strongest association (R²=0.168; P<0.001) with ESC categories. Mean RVEDV Z-scores increased monotonically: 0.54 (Skill), 0.57 (Power), 1.14 (Mixed), 1.99 (Endurance) (Fig. 2A). RV remodelling prevalence increased from 8% to 10% to 29% to 62% (odds ratio 3.73 per category; 95% CI 2.88-4.85; P<0.001) (Fig. 2B). However, individual phenotype prediction was limited: using isotonic regression-derived thresholds (Skill: Z<0.55; Power: 0.55≤Z<0.86; Mixed: 0.86≤Z<1.56; Endurance: Z≥1.56), balanced accuracy was 39% (95%CI 34-46%), versus 25% chance-level. Category-specific positive predictive values were 6.4% (Skill), 13.9% (Power), 53.8% (Mixed), and 61.7% (Endurance), consistent with overlap in RVEDV values, particularly in Skill and Power. Conclusion The ESC sport classification captures a remodelling gradient at population level, most strongly reflected by RVEDV on CMR. However, even after data-driven optimisation of RVEDV thresholds, sport category provides limited discrimination of individual CMR phenotypes. Athlete evaluation should therefore be phenotype-led rather than sport category-led, preferably supported by objective training exposure and performance measures.
Discussion
No takes yet. Share an insight, caveat, or question.
Cite This Study
Diepen et al. (2026) conducted a cross-sectional in Cardiac remodelling in elite athletes (n=1,029). ESC sport categories (Skill, Power, Mixed, Endurance) vs. Sex-matched controls and across ordinal ESC sport categories was evaluated on Right ventricular end-diastolic volume (RVEDV) remodelling (Z-score ≥1.645) and individual CMR phenotype prediction accuracy (OR 3.73, 95% CI 2.88-4.85, p=<0.001). ESC sport categories captured an RVEDV remodelling gradient (OR 3.73 per category; 95% CI 2.88-4.85) but showed limited accuracy (39%) for classifying individual CMR phenotypes.