unique participants across running and walking records
2,506 JSON files · 1,832 running task recordsSPORTS REHABILITATION · HUMAN PERFORMANCE
When a runner is injured, what does their movement tell us?
A careful look at running biomechanics and current injury status, built to show what the data can support, and where it stops.
Cross-sectional phenotyping · not future-injury prediction
MOTION STUDY / 01FUKUCHI 2017 · 2.5 m/s
ILLUSTRATIVE PHASEStance
Concept illustration only · no participant waveform or event is shown
RUNNING INJURY CLINIC KINEMATIC DATASET · FIGSHARE v12026
THE DATASET
A large public archive; a deliberately small technical pilot.
1,798people in the source datasetFerber et al., 2024
1,402with running observationsmetadata audit
30people in the technical pilot15 healthy · 15 injured
EVENT VALIDATION PENDING
A complete usable-cycle count has not been established.
01 / QUESTIONFROM MOVEMENT TO PHENOTYPE
THE RESEARCH QUESTION
Which running movement patterns are associated with current injury status?
The project organizes reported injury labels and treadmill kinematics into a transparent, participant-level research workflow. It starts with source-provided summaries and a 30-person engineering pilot, then asks whether robust, diagnosis-aware comparisons are feasible.
This dataset is cross-sectional. It does not show that a movement pattern caused an injury, predicts a future injury, or diagnoses an individual runner.
DESIGNCross-sectional association
UNIT OF ANALYSISOne selected record per person
OUTCOMEConservative current-status label
KEY COMPARISONInjury status, with speed in view
THE SELECTION, NOT THE WHOLE ARCHIVE
A cohort with rules you can inspect.
The candidate counts are metadata-based and provisional. They are not the final biomechanically usable sample.
earliest eligible records, one per participant, with resolved status and usable metadata
353 healthy904 injured15 + 15 files for source integrity, feature coverage and engineering quality checks
cycle/event gate still openCounts are reproduced from the pinned Figshare v1 metadata and the documented conservative label rule. Candidate ≠ final inclusion.
A DESIGN CHOICE THAT MATTERS
Speed changes the question.
Before comparing biomechanics, account for how fast each runner moved. In the candidate cohort, mean treadmill speed differed between current-status groups. A small experimental study also found speed-related changes in lower-limb joint kinematics and kinetics. The gap is a reason to adjust and test overlap, not evidence that speed caused the group difference.
Read the speed studyCANDIDATE COHORT · MEAN SPEEDm/s
Healthy
2.74Injured
2.550123.0
Metadata balance check only · no biomechanical group test has been run.
02 / NEW ANALYSISBALANCE BEFORE BIOMECHANICS
NEW ANALYSIS · COVARIATE BALANCE
Speed and age separate the groups more than body size or experience.
Standardized mean difference (injured minus healthy) in the 1,257-person candidate cohort. Observational studies often treat |SMD| below 0.1 as balanced. This is a design check, not a biomechanical result.
Healthy higherInjured higher
Dashed lines: ±0.1. Years running is missing for 286 candidates (242 healthy and 729 injured reported it).
03 / NEW ANALYSISONE PERSON, SEVERAL RECORDS
NEW ANALYSIS · LABEL STABILITY
A current label is a snapshot.
209 of 1,402 runners (14.9%) have more than one running record. Comparing each person's earliest and latest record, 20 of them (9.6%) carry a different status. Order follows metadata timestamps, not verified clinical follow-up.
Earliest record ↓Latest record →
04 / PROCESSSOURCE TO AUDIT TRAIL
HOW THE WORK IS BUILT
Keep every transformation visible.
The source archive holds marker trajectories, static calibration markers, metadata and precomputed discrete variables. The pilot inspects these inputs one file at a time and preserves the upstream values, signs, units and file lineage.
Pin the source
Versioned Figshare metadata, archive manifests, checksums and a fixed participant-file manifest make the local input auditable.
SOURCE · SHA-256Define the cohort
Keep person and file distinct. Apply conservative injury-status rules and surface unknown or discordant labels for review.
PERSON-LEVEL · SENSITIVITYScreen the signals
Check marker geometry, finite trajectories, sampling rate and summary consistency without silently correcting the source.
STRUCTURE · ENGINEERING QCValidate events and cycles
Re-run the supplied MATLAB workflow and compare touchdown/toe-off events, retained normalized cycles and regenerated summaries before widening the cohort.
PENDING MATLAB RUNPILOT ENGINEERING CHECKS · N = 30SOURCE SUMMARIES
Files passing structural checks30 / 30
Finite bilateral source values in selected registry780 / 780
Independently validated event/cycle records0
Structural checks do not establish correct event timing or clinical validity. The 780 values are source-provided summaries.
Open aggregate pilot descriptives ↗Read the repeated-record and label audit ↗
WHAT THE LITERATURE SAYS
Why phenotype carefully?
Evidence is mixed, definitions vary and study populations matter. The strongest case for this project is not a universal injury marker. It is a transparent way to ask more specific, reproducible questions.
No single gait measure explains running injury.
Lopes and colleagues found only peak hip adduction associated in their pooled analysis; the result was influenced by a small subset of studies. They also highlighted inconsistent measurement definitions and study methods.
Lopes et al., 2023Findings do not transfer automatically.
A prospective-study review reported different patterns by sex and competition level, with limited evidence for particular associations in female recreational runners.
Vannatta et al., 2020Diagnosis and timing shape interpretation.
In 224 injury-free recreational runners, some baseline factors were associated with medically diagnosed injuries over follow-up. This design answers a different question from current-status comparisons.
52-week prospective cohortCompare like with like.
Experimental changes in running speed alter peak kinematics and kinetics at the hip, knee and ankle. Speed therefore belongs in the design and interpretation.
Orendurff et al., 2018Small studies are useful when labeled honestly.
A 2025 sprint study paired IMU and force-velocity measures with surveillance, but only two prospective injuries occurred. The authors frame it as feasibility work, not a predictive result.
Wolski et al., 2025A classifier on this source is already published.
Xiao et al. (2026) reported a random-forest ROC-AUC of 0.790, with high sensitivity and limited specificity. Our complementary angle audits repeated-record labels and cohort timing before future modeling.
Xiao et al., 2026↳
Project implication: report diagnosis definitions, speed support, missingness and event-quality gates before any association estimate. Treat exploratory subgroup differences as hypotheses, not screening advice.
Full literature review and search notes →INTERPRETATION GUARDRAILS
What this project can and cannot say.
+
Can support
- An auditable description of candidate cohort composition and data quality.
- Carefully defined, speed-aware associations with current injury labels after event validation.
- A reproducible starting point for diagnosis-specific research hypotheses.
−
Cannot establish
- That a movement pattern caused injury or predicts who will be injured later.
- A validated screening cut-off, clinical diagnosis or treatment recommendation.
- Independent external validation by a related dataset without a verified participant crosswalk.
OPEN METHODS · SOURCE ATTRIBUTION
Made to be checked, cited and extended.
The raw archive is large, so the repository uses bounded, checksum-verified acquisition and a fixed pilot manifest. No raw participant trajectories are copied into this GitHub repository. Start with the source and reproduction guide.
Two licenses, two kinds of work.
The source dataset is CC BY 4.0; its supplied MATLAB scripts are MIT. This project's code is MIT. Derived pilot values retain source attribution and an explicit adaptation notice.
REPRODUCTION PATHPYTHON · MATLAB GATE
01
python scripts/acquire_data.py --pilot-manifest …02
python scripts/run_pilot.py03
validate_official_events(projectRoot)PENDING04
python scripts/verify_pilot_cache.pyRAW INPUTS · IMMUTABLECHECKSUMS · MANIFEST · SOURCE CODE
Do not expand the cohort until the official event/cycle runner has been executed and reviewed.