Data Source and Searches

We conducted a systematic literature search of the MEDLINE, Scopus, and Cochrane databases from inception to April 30, 2016. We also included articles published online ahead of print at the time of our search. The primary search used the following keywords: echo or cardiac magnetic resonance (CMR) or CMR imaging, with or without: Doppler, colour, stress, three-dimensional; right heart; right ventricular: dimension, function, size, structure. In addition, the population search used the following items: athletes, training, exercise. Where applicable, keywords were searched as the root and truncation symbol (e.g., ventric*). Additionally, we manually searched references from papers about studies, review articles, and meta-analyses. All searches were confined to human studies. Studies not in English were excluded.

Study Selection

We assessed studies for inclusion by using the following a priori defined criteria: (1) the study explicitly stated that it evaluated competitive athletes, with no history of cardiovascular disease; (2) the study reported at least three parameters of RV dimensions and function, measured by CMR or two-dimensional echocardiography according to current clinical standards ; (3) the mean age of the study cohort was set between 18 and 39 years both for echocardiographic and CMR studies; (4) a measure of statistical variance was reported; (5) year of publication ≧1990 for echocardiographic studies and ≧2000 for CMR studies. Because of the paucity of CMR data referring to female athletes, for CMR studies only male athletes were analyzed. Study arms that reported populations that potentially overlapped with other studies were excluded. Echocardiographic studies with ≤20 athletes were also excluded.

A flowchart showing the derivation of the reference cohort is shown in Figure 1 for echocardiographic studies and in Figure 2 for CMR studies.

Results of the literature search and disposition of echocardiographic articles screened for inclusion. MRI , Magnetic resonance imaging.

Results of the literature search and disposition of CMR articles screened for inclusion.

Data Collection

Each eligible article meeting the inclusion criteria was reviewed by two independent reviewers (M.S. and F.L.) who compared decision-making and discussed disagreements. Discrepancies were resolved by consensus. Each data set was reviewed for units and methods of measurement, range checks were performed to identify and exclude biologically implausible values, and summary statistics were cross-checked against published results, where available. Athletes were classified into strength (such as bodybuilders and weightlifters), endurance (such as long-distance runners, swimmers, and cross-country skiers), and combined (such as rowers, cyclists, and speed skaters) groups according to the intensity level of static and dynamic components. Heterogeneous groups of athletes of different sports classifications were categorized as mixed.

In longitudinal studies investigating training-induced RV remodeling, the RV measurements after the longest exercise exposure were used.

Data Synthesis, Statistical Analysis, and Development of Reference Values

To explore the influence of type of training (endurance, strength, mixed, or combined disciplines), gender, and age on right heart size and function, a meta-analysis was conducted to integrate the results of a set of studies about RV parameters measured by echocardiography and CMR in competitive athletes engaged in different disciplines. Specifically, a meta-regression was used that allowed for between-study heterogeneity. The class of sports activity (i.e., endurance vs nonendurance disciplines) and age were included in the model as moderator variables. Since the levels of the moderators may influence the heterogeneity among the studies, the meta-regression analysis was based on a mixed effects model. Mixed effects models provided unconditional inference about the average effect in the entire population of studies from which the included studies were assumed to be randomly selected. The Knapp-Hartung adjustment to the standard errors of the estimated coefficients was applied in order to adjust the statistics and the confidence intervals of the estimates so that their significance properties are closer to the nominal 5% type I error and the confidence intervals maintain a 95% coverage level. Therefore, the tests on individual coefficients as well as the confidence intervals relied on t distributions with k – 1 degrees of freedom, where k is the number of studies. An underlying heteroskedastic covariance structure was considered in the study-aggregate meta-analysis. The sampling variances were calculated based on large sample approximation. Maximum-likelihood estimation was used when estimating the heterogeneity parameters of meta-analysis. The between-study heterogeneity in the true measurements of the dependent variables was assessed by the parameter τ ² , and its percentage over the total variability in the effect size estimates was also calculated ( I ² ). A test of the residual heterogeneity ( Q statistic) was also applied. For each variable, a sequence of two meta-regressions were implemented, which included the reduced model consisting of the basic random effects model (i.e., a meta-regression without moderators) and the models with the Class moderator. The Bayesian information criterion (BIC) was used for model selection among the above finite set of models. The model with the lowest BIC was preferred. The coefficients from the selected model quantified the direction and magnitude of the relationship between the average “true” outcome in the population of studies and the moderator variable(s) included in the model. The 95% CIs around the estimated average outcome and prediction intervals of new possible data were provided for different levels of the significant moderators. We also reported 99% CIs in order to provide further information. Forest plots gave the graphical overview of the results from the fitted model. Funnel plots were also drawn, and the Egger’s regression test for funnel plot asymmetry indicated whether either publication bias or systematic difference between larger and smaller studies occurred. Significant rejection of the null hypothesis of symmetry therefore would cast doubts on the reliability of the metaregression.

The coefficients from the fitted model estimated the direction and magnitude of the relationship between the average true outcome in the population of studies and the moderator variable included in the model.

RV reference values were defined at the 95 ^th percentile and lower reference values at the 5 ^th percentile. In accordance with earlier recommendations, these values defined 90% of the population as normal, but they allowed for nonnormal distributions.

Reference values were derived in male competitive athletes divided according to sports disciplines for RV size (RV outflow tract [RVOT] parasternal long-axis view [PLAX], RVOT parasternal short-axis view [PSAX], RV end-diastolic and end-systolic area, RV basal and midcavity diameter, RA area and their indexed values, and RV wall thickness) and RV function (fractional area change [FAC], tricuspid annular plane systolic excursion [TAPSE], s’ and e’ velocities, and RV strain). Reference values were established in male competitive athletes divided according to sports disciplines also by CMR for RV end-diastolic and end-systolic volumes, RV stroke volume, and RV ejection fraction. In female competitive athletes only the reduced meta-regression model was considered (i.e., reporting only mean values ± SE). All statistical analysis was carried out with the comprehensive software meta-analysis package for R: metafor 1.9-8 (2015-09-28).