The associations between internal and external measures of training load and intensity are important in understanding the training process and the validity of specific internal measures. We aimed to provide meta-analytic estimates of the relationships, as determined by a correlation coefficient, between internal and external measures of load and intensity during team-sport training and competition. A further aim was to examine the moderating effects of training mode on these relationships. We searched six electronic databases (Scopus, Web of Science, PubMed, MEDLINE, SPORTDiscus, CINAHL) for original research articles published up to September 2017. A Boolean search phrase was created to include search terms relevant to team-sport athletes (population; 37 keywords), internal load (dependent variable; 35 keywords), and external load (independent variable; 81 keywords). Articles were considered for meta-analysis when a correlation coefficient describing the association between at least one internal and one external measure of session load or intensity, measured in the time or frequency domain, was obtained from team-sport athletes during normal training or match-play (i.e., unstructured observational study). The final data sample included 122 estimates from 13 independent studies describing 15 unique relationships between three internal and nine external measures of load and intensity. This sample included 295 athletes and 10,418 individual session observations. Internal measures were session ratings of perceived exertion (sRPE), sRPE training load (sRPE-TL), and heart-rate-derived training impulse (TRIMP). External measures were total distance (TD), the distance covered at high and very high speeds (HSRD≥13.1-15.0kmh-1 and VHSRD≥16.9-19.8kmh-1, respectively), accelerometer load (AL), and the number of sustained impacts (Impacts>2-5G). Distinct training modes were identified as either mixed (reference condition), skills, metabolic, or neuromuscular. Separate random effects meta-analyses were conducted for each dataset (n=15) to determine the pooled relationships between internal and external measures of load and intensity. The moderating effects of training mode were examined using random-effects meta-regression for datasets with at least ten estimates (n=4). Magnitude-based inferences were used to interpret analyses outcomes. During all training modes combined, the external load relationships for sRPE-TL were possibly very large with TD [r=0.79; 90% confidence interval (CI) 0.74to 0.83], possibly large with AL (r=0.63; 90% CI 0.54to 0.70) and Impacts (r=0.57; 90% CI 0.47to 0.64), and likely moderate with HSRD (r=0.47; 90% CI 0.32to 0.59). The relationship between TRIMP and AL was possibly large (r=0.54; 90% CI 0.40to 0.66). All other relationships were unclear or not possible to infer (r range 0.17-0.74, n=10 datasets). Between-estimate heterogeneity [standard deviations (SDs) representing unexplained variation; τ] in the pooled internal-external relationships were trivial to extremely large for sRPE (τrange =0.00-0.47), small to large for sRPE-TL (τrange =0.07-0.31), and trivial to moderate for TRIMP (τrange=0.00-0.17). The internal-external load relationships during mixed training were possibly very large for sRPE-TL with TD (r=0.82; 90% CI 0.75to 0.87) and AL (r=0.81; 90% CI 0.74to 0.86), and TRIMP with AL (r=0.72; 90% CI 0.55to 0.84), and possibly large for sRPE-TL with HSRD (r=0.65; 90% CI 0.44to 0.80). A reduction in these correlation magnitudes was evident for all other training modes (range of the change in r when compared with mixed training -0.08 to -0.58), with these differences being unclear to possibly large. Training mode explained 24-100% of the between-estimate variance in the internal-external load relationships. Measures of internal load derived from perceived exertion and heart rate show consistently positive associations with running- and accelerometer-derived external loads and intensity during team-sport training and competition, but the magnitude and uncertainty of these relationships are measure and training mode dependent.
Read full abstract