EUROPEAN FOOTBALL / 2020-2025

The cost of
being absent.

An injury is a record. Its impact is days, matches and histories that accumulate. Explore what 15,603 recorded absences reveal about European football.

Explore the analysis Read the discussion ↗

Víctor Gómez · Physiotherapy × data science

EVENT → TIME LOSS → RECORDED END

52° / 13°

MEDIAN REPORTED ABSENCE18 days

Per record, all absences · not an individual prognosis

Conceptual motion. It does not depict players or current availability.

15.603absence records

4.081recorded player names

145recorded clubs

5leagues · five seasons

Source: Sanan Muzaffarov · Kaggle v2 · CC BY-SA 4.0Source and limits ↗

01 / FREQUENCY AND CONSEQUENCE

Counting is only the beginning.

The most frequent label and the one accounting for the most days answer different questions. A medical team needs to see both.

MOST RECORDS

Hamstring injury

1.256 records

Hamstring injury: a 29-day median and 52,483 accumulated reported days.

MOST ACCUMULATED DAYS

Cruciate ligament tear

56.768 days

Cruciate ligament tear: 226 records and a 230.5-day median. The label does not identify which cruciate ligament or verify the diagnosis.

Injuries only: 39 non-injury labels are excluded (3,093 records, such as coronavirus, illness or lack of fitness), using the same filter as Sports Rehab Intelligence. Cruciate tears account for 8% more days than hamstring injuries with 5.6 times fewer records. Episode sums can repeat days and matches.

02 / BURDEN EXPLORER

Burden has more than one axis.

Frequency, duration and missed matches in one view. Filter and select a bubble to inspect its original source label.

LeagueSource seasonAbsence typeRecord view

Sensitivity keeps one row per matching name, label and date interval. It does not adjudicate the correct club or confirm unique clinical episodes.

Each circle = one source labelArea ∝ missed matches

Both axes use logarithmic scales. Original English labels retained.

View the same results as an accessible table

Absence statistics by original label, with active filters
Label	Records	Median days	Summed days	Summed matches

These are recorded counts and consequences. They are not incidence rates or individual risk estimates. Each filter uses the median calculated from its own underlying records.

03 / NEW ANALYSIS · SEASONS

The pandemic is in the archive too.

With every absence counted, 21/22 looks like the worst season. Once non-injury absences are separated, it records the fewest injuries. Coronavirus and other illnesses explain most of the gap.

InjuriesNon-injury absences

Original records by source season. 20/21 and 21/22 include 965 coronavirus records.

04 / NEW ANALYSIS · BODY REGIONS

The knee weighs more than it appears.

Grouping labels by region, the knee is 14% of injury records but 30% of days. The thigh is the most frequent region and its share of days is smaller.

Share of injury recordsShare of absence days

Regions assigned from keywords in the original label (12,510 injury records), using the same rules as Sports Rehab Intelligence. Generic labels such as “Muscle injury” or “Knock” have no region. Day sums can repeat episodes.

05 /THE RECORDING ALSO COUNTS

Same absence. Another context.

There are no exact whole-row duplicates. Yet 981 groups share the same name, label and dates across different clubs.

Reported days and matches agree within those groups. They may repeat club context; the file alone cannot adjudicate this. Both scenarios remain visible and traceable.

Read the audit ↗

SUM OF REPORTED DAYS−5,5 %

15,603 records563.314

14,611 candidates532.422

992 additional rows removed in the sensitivity scenario. This does not confirm 14,611 independent clinical injuries.

06 /UNEQUAL HISTORIES

A player is not one row.

2,795 names have multiple records. 1,124 have a label that appears again at a later onset. That describes a recorded history; it does not confirm reinjury.

Records per recorded name

2.795 / 4.081

History begins where the file begins.

A first record is not necessarily a first lifetime injury. Names are not verified identifiers. Intervals can overlap, and a gap without a record does not prove availability.

07 /THE AVAILABILITY QUESTION

We see recorded loss. The opportunities are missing.

An availability percentage needs the total eligible opportunities. It cannot be invented from the first and last injury dates.

THE FILE PROVIDESRecorded absences

Dates, reported duration, missed matches and absence labels.

THE FILE DOES NOT PROVIDEEligible opportunities

Registration and transfers, complete fixtures, eligibility and exposure hours.

Relevance to rehabilitation. Surveillance guidance separates frequency, severity and burden. Prospective research links lower injury burden and higher availability with team performance; this file cannot reproduce a causal comparison or its exposure-based rates. Read the sources and discussion ↗

08 /RESEARCH YOU CAN INSPECT

From the question to open code.

Inspect the decisions, download the aggregates and return to the source file. Conclusions should be traceable to their data.

01 / GITHUB

Reproduce the analysis

Python pipeline, tests, instructions and a website without build dependencies.

↗02 / DATOS

Download the aggregates

Statistics by original label, with sample sizes and CC BY-SA 4.0 attribution.

↓03 / EVIDENCIA

Read before concluding

Targeted literature search, critical discussion, references and interpretation limits.

↗

Kaggle release ↗BibTeX references ↓Full methodology ↗

PROVENANCE, AUTHORSHIP AND RIGHTS

Source data. Independent analysis.

Data: Sanan Muzaffarov, European Football Injuries (2020-2025), Kaggle, version 2. Retrieved 26 October 2026. Listed license: CC BY-SA 4.0. File integrity was verified; clinical collection and upstream sources were not independently verified.

Analysis, code and design: Víctor Andrés Gómez López. Original code: MIT. Derived data and figures: CC BY-SA 4.0, with documented transformations. Scientific references retain their authors' rights; findings are cited and paraphrased without republishing articles or third-party figures.

Original dataset ↗CC BY-SA 4.0 ↗Provenance ↓