From Signal to Biological Meaning: A Systematic Review of Construct-Validity Evidence for External Training Load Metrics in Sport
Sports medicine - open · 2026
Only 28% of studies actually validate GPS/wearable 'load' metrics against the biological thing they claim to measure, so treat those numbers as movement data, not proof of fatigue or risk.
Part of: Training load and injury risk: what the evidence says · Injury prevention and screening in sport
The paper
Systematic review (narrative synthesis, no meta-analysis) of 46 primary studies (n=7-71 per study, predominantly male samples) evaluating construct-validity evidence for external training load (ETL) metrics in athletes and physically active people.
What they found
Only 13 of 46 studies (28%) used a direct construct-matched criterion to test whether an external load metric actually reflects the biological construct it is interpreted to represent (fatigue, metabolic demand, tissue stress, etc); 16 (35%) relied on alternative proxies and 11 (24%) on group/condition contrasts instead. Construct attribution was explicit and unqualified in only 14/46 studies, with the rest qualified, implicit, or even reverse-targeted, leading the authors to conclude most ETL metrics are validated only as descriptors of external movement or work, not as direct indicators of biological response.
Study design
A systematic review using evidence mapping and narrative synthesis (explicitly not meta-analysis, due to conceptual and methodological heterogeneity) of 46 studies, with underlying samples as small as 7-71 participants and mostly male. This is a methodological/conceptual review auditing the evidence base itself, not a review pooling a clinical outcome, so it sits lower on the quantitative evidence hierarchy than a standard meta-analysis despite the rigorous appraisal framework.
Methodology
Strengths: a broad five-database search to March 2026 and design-appropriate appraisal tools (COSMIN for measurement properties, PROBAST+AI for prediction models, JBI for other designs), which is unusually rigorous for this kind of conceptual review. Weaknesses: each study was reduced to a single 'page-anchored construct-evidence chain,' a bespoke adjudication method with no mention of dual independent raters or inter-rater agreement in the abstract, which leaves room for reviewer-introduced interpretation bias. More fundamentally, the review can only be as strong as what it reviews, and the primary literature itself is thin (samples of 7-71, mostly male), so a clinician should read the 28% figure as a ceiling on confidence, not a precise estimate.
The appraisal
There's no clinical effect size here, the finding is about evidentiary adequacy, not treatment benefit. The practically important number is that fewer than a third of studies actually tested an ETL metric against a construct-matched biological reference, which means most of the 'this GPS number tells us about fatigue/stress/adaptation' claims clinicians hear are inference, not demonstrated validity. Even where appraisal tools (COSMIN, PROBAST+AI, JBI) were applied, the review flagged common limitations in hypothesis testing and analysis across studies, so the handful of direct-criterion studies shouldn't be read as methodologically bulletproof either. The design supports a claim about the state of the evidence base, not a causal or comparative claim about any single metric or athlete outcome.
The gap
The review doesn't tell you which specific metrics (total distance, PlayerLoad, acceleration counts, etc) are well or poorly validated for which specific construct, so you can't yet use it to decide which number on your monitoring dashboard to trust and for what purpose.
Landmark context
This extends the long-standing internal-versus-external training load distinction in the load-monitoring literature (associated with authors like Impellizzeri and colleagues on GPS/accelerometer validity and reliability) from technical measurement validity into construct validity for biological interpretation, formalising a concern that practitioners have informally raised for years but that hadn't been systematically mapped.
What to do Monday
Keep using external load metrics (GPS distance, PlayerLoad, speed/acceleration counts) to describe what an athlete physically did, but don't let a single external number stand in for fatigue, tissue stress, or metabolic strain; pair it with a construct-matched internal measure (heart rate, RPE, lactate, biomarkers, or wellness tools) before making a load-management decision on that basis.
In practice
Applies to two groups using the same numbers for different decisions: physios tracking GPS/accelerometer load through a return-to-sport progression (ACL, hamstring, bone stress injury), and S&C coaches managing microcycle load in team sport. Nothing changes about collecting the external data, distance, PlayerLoad, high-speed running and acceleration counts remain useful for describing what was actually done and for session design, but stop treating a rise in those numbers alone as proof of higher fatigue, tissue stress, or injury risk. In the RTS pathway, pair external load with a construct-matched internal signal before clearing or escalating: session RPE or HR-based internal load for metabolic strain, and neuromuscular readiness tracked with Countermovement jump or Single leg jump (re-tested against the athlete's own baseline through the week) rather than inferring neuromuscular fatigue from accelerometer load alone. On the performance floor, use the same logic when deciding whether to back off a session: a PlayerLoad spike triggers a check against RPE trend or jump output, not an automatic load cut. The caveat is that this pairing approach is sound construct-validity logic, not a protocol the review has validated as a package, so treat it as good practice reasoning rather than a proven algorithm.
The five papers that matter — every week, free.
We read the whole sports-medicine, rehab & performance literature and appraise the papers worth your time.
Get the free weekly issue