A large prospective study of recreational runners has reported an association between the acute:chronic workload ratio and the onset of running-related injury, measured from GPS data on both distance and duration. The headline result is not the one most runners would predict. It was not the dramatic spikes that tracked with injury but modest ones: a fortnightly increase in the ratio in the range of roughly 0.10 to 0.78 was associated with elevated risk. Big jumps were not the signal. Small, steady ones were.
The acute:chronic workload ratio compares recent training, usually a week, against a longer baseline, usually a month. It was imported into distance running from team sports, where it became popular partly because it produces a single number that a coach can look at. The appeal is obvious and so is the problem: a ratio built from two rolling averages is sensitive to how both windows are defined, to what counts as load, and to what a runner was doing before the window opened. Two runners with identical ratios can be in entirely different situations.
There are several readings of the finding that do not require modest increases to be causally dangerous. Runners who are already carrying a niggle tend to trim their training, which depresses the acute window and can produce exactly the sort of small movement the study associates with injury — the injury may precede the ratio change rather than follow it. A GPS-derived measure based on distance and duration also cannot see intensity, terrain or surface, so a fortnight of identical mileage on hills and on flat roads registers the same. And runners who make very large jumps may simply be a self-selecting group who tolerate them.
The study sits alongside a run of 2026 work pointed at the same question from different angles. One time-series analysis examined daily load, accumulated workload and injury occurrence across 74 distance runners and 42,766 daily observations. A separate methodological paper, published in Frontiers in Public Health in August, benchmarked explainable deep learning models on seven-day training-load histories to classify injury-labelled days. That paper's authors were careful about what it establishes: it does not demonstrate that any particular reporting protocol or injury definition improves prediction, and frames itself as setting out design questions for future prospective studies rather than answering them.
For a runner reading this rather than a researcher, the practical conclusion is deflationary and probably useful. The evidence does not support treating any single ratio as a dashboard warning light, and the long-standing systematic review position — that the link between changes in training load and running injury is weaker and messier than the coaching literature implies — still stands. The things with better evidence behind them remain unglamorous: consistency over months rather than weeks, adequate sleep, strength work, and a willingness to back off when something hurts rather than when a number moves.