Daily load spikes did not hold up as a next-day injury signal in elite women’s football
In elite women’s football, staff often read a player’s latest training-load and wellness scores as an early warning system. This analysis put that habit to a stricter test and found no stable evidence that within-player deviations in acute load, or acute load paired with wellness changes, predicted next-day self-reported lower-limb pain onset.
The study used the public SoccerMon dataset, a rare open monitoring resource drawn from 50 players on two Norwegian elite women’s teams across the 2020 and 2021 seasons. Daily subjective data included session load, fatigue, readiness, soreness, mood, stress, sleep quality, and sleep duration, along with injury reports captured through a smartphone-based monitoring system. The outcome was a new lower-limb pain or injury episode beginning on day t after a 7-day washout, framed as an index player-day.
The investigators focused on within-player change rather than absolute values. In sport science, that distinction matters because a player’s “normal” training tolerance is not the same as another player’s, and the same load can mean different things depending on recent recovery, symptom burden, and prior exposure. Biologically, a spike in acute load could plausibly increase tissue strain, while soreness or fatigue might signal incomplete recovery or a state in which another training bout becomes harder to absorb. The question was whether those day-to-day deviations carried reproducible information about the next day’s lower-limb pain risk.
They did not.
Sparse events and missing reporting limited how far the signal could travel
The primary risk set contained 42 onset-days from 15 players. A complete-case soreness interaction analysis was smaller still, with 28 onset-days from nine event players. Those counts are not large enough to support confident inference when the exposure definitions are complex and the same athletes contribute repeated observations across two seasons.
The study’s central estimate came from retrospective full-period within-player standardization and Firth logistic regression. Under that approach, the focal acute-load × soreness interaction produced an odds ratio of 0.63 with a 95% confidence interval of 0.45 to 0.97. On paper, that suggests a lower odds of next-day lower-limb pain onset under the modeled interaction. In practice, the estimate was fragile. When within-player standardization was recalculated using only observations available before each day, the interaction moved to 0.97 with a 95% confidence interval of 0.73 to 1.25, a result that is far closer to null.
That shift is the most important finding in the paper. It shows how sensitive the apparent association was to the way the longitudinal exposure was built. In a monitoring dataset with incomplete reporting, event clustering, and repeated measures from a small number of players, retrospective standardization can make a signal look cleaner than it is. Reconstructing the exposure using only prior information is more faithful to the daily decision environment, and under that approach the apparent association largely disappeared.
What the monitoring numbers can and cannot tell a coaching staff
The study was not designed to build a prediction tool or establish a universal training threshold. It tested a registered within-player hypothesis in a public dataset and then stress-tested the result against alternative analytical choices, including recurrent-event risk sets, corrected wellness polarity, and sensitivity analyses around exposure construction, team and season instability, and model-space multiplicity.
Those design choices matter because athlete-monitoring data are easy to overread. A higher-than-usual soreness score may reflect recent musculoskeletal loading, a transient symptom flare, or a reporting artifact. A higher-than-usual acute load may reflect a normal training progression, a congested match schedule, or an off-day recovery pattern captured imperfectly by subjective reporting. Without enough outcome events, those distinctions blur quickly, and apparent associations can be driven as much by missingness and scale handling as by physiology.
The final interpretation is restrained and clinically useful. Daily load and wellness signals should be treated as contextual longitudinal information, not as stand-alone next-day risk thresholds. In a team setting, that means a single fatigue or soreness score should not be read as an injury forecast in isolation. The more defensible use is as one piece of a broader conversation about recovery, workload tolerance, symptom evolution, and whether a player’s current pattern differs meaningfully from her own recent baseline.
Reproducibility was a strength; certainty about risk prediction was not
The analysis was built from public data and accompanied by reproducibility materials, including code, derived non-identifiable inputs, model outputs, logs, and statistical analysis records. Raw files were checked, hashes were recorded, and the dataset was rebuilt against the analytical version before modeling. That level of transparency strengthens confidence in the mechanics of the analysis even when the biological signal remains weak.
The participant description itself was limited by the open dataset. Because player metadata were removed, age, height, body mass, position, and other personal characteristics were unavailable and were not inferred. The final analysis therefore speaks to a specific and narrow population: anonymous players from two Norwegian elite women’s teams monitored during two seasons. Team and season imbalance, combined with sparse events and missing monitoring data, limited precision and made the estimates vulnerable to specification choices.
The practical implication is straightforward. Monitoring systems can still be valuable for understanding how a player is trending, but this dataset does not justify treating acute-load or load-wellness interactions as reliable next-day injury alarms. The result is a caution against turning subjective athlete-monitoring scores into rigid thresholds when the underlying event rate is low and the data stream is incomplete. Any future work that seeks a stronger answer will need more events, more complete reporting, and a design that can better separate true physiological change from the noise of daily sport.