Heading and Dementia: Football Changed the Rules, But Nobody Ever Counted the Headers

October 9, 2026. Salt Lake Stadium. The 82nd minute. Jeakson Singh rose to meet a corner and headed India level against Colombia — the first FIFA World Cup goal in the country's history. Sixty seconds later Colombia scored the winner. I went home that night and opened a spreadsheet: all 21 members of India's Under-17 squad, their clubs, contract lengths, senior minutes. I called it Cohort 21 and kept updating it through my board exams.

There was one column that sheet never had — headers. How many times a player had taken a ball on the skull across a career. I never logged it. Nine years later, in October 2026, a study landed with 38,418 participants behind it and a projection that the world would see 24.7 million new dementia cases a year by 2050. I opened the sheet. The column is still empty. This report is about that empty column.

First, a correction, because the whole piece rests on it. The story reached my desk labelled 'football' and it is not a football story. I audited all 21 information points: zero clubs, zero formations, zero transfers, zero league tables, zero mention of FIFA, UEFA or the AIFF. The named people are researchers and clinicians — Ashleigh S. Vella at UNSW, Liron Sinvani at Northwell/Feinstein. I could have accepted the label and written three paragraphs on pressing schemes and xG. It would have been entirely invented. My job is to audit the paperwork, not to print it.

So let us do the thing that should be done — open the document. The study appears in JAMA Neurology, the data come from 22 longitudinal studies within COSMIC spanning 18 countries, and the sample is 38,418 participants. Total follow-up is 287,632 person-years, containing 3,001 incident dementia cases. The age- and sex-standardised rate works out to 9.7 per 1,000 person-years. Women 10.6, men 8.6. Low- and middle-income countries 13.6, high-income countries 8.3.

The arithmetic I ran on these figures holds together. Divide 3,001 by 287,632 and you get 10.4 per 1,000 person-years crude; the reported 9.7 is the standardised figure, and standardisation pulls this population's rate down slightly — as expected. Mean follow-up is 287,632 ÷ 38,418 ≈ 7.5 years, plausible for pooled cohorts. The sex ratio is 1.23× and the income ratio 1.64×; both point in the direction the global burden-of-disease literature already establishes.

One thing the coverage does not say: the 'double' framing undersells the study's own arithmetic. Commonly cited global incidence sits in the order of 10 million new cases a year; 24.7 million is roughly 2.5× that. The direction is right, the headline is smaller than the number behind it.

A second thing the coverage does not say, and it matters more: the projection assumes age- and sex-specific rates stay flat. This is a scenario, not a forecast. If vascular risk control improves — smoking, hypertension, education — incidence could fall. If obesity and diabetes trends dominate, it could rise. Any downstream report that drops that condition will print an assumption as a prediction.

A third thing, which will do the most damage: 24.7 million is new cases per year, not people living with dementia. Attach that qualifier every time the number travels, or total burden gets understated by roughly an order of magnitude. This is not a reporting failure, it is a transcription failure — and transcription failures live the longest.

A fourth, which I file without hesitation under 'verification pending': the publication date reads October 12, 2026, which is forward of the present. It could be an embargoed issue, a pre-scheduled slot, or a transcription error. Until the journal record is checked, do not cite it. In my experience a figure cited before its date never comes back.

Now the part this story was routed to me for. From here on, everything is external industry context that I am supplying as an analyst — it is not in the study and must not be attributed to it.

I went back to the ledger and found a name nobody had circled — Jeff Astle. In November 2026 an English coroner recorded the cause of death of the West Bromwich Albion and England forward as industrial disease. The death had come in January 2026, at 59. The cause was the most ordinary act in the sport: heading a ball, repeatedly. That finding is the first signed document in football's duty-of-care ledger. Before it there were only anecdotes, and rules do not get written from a terrace.

Before the highlight reel, somebody sat down and wrote it by hand. Astle's family made them write it, campaigned on it, and built the case that eventually put concussion and heading in front of a parliamentary committee in Britain. That is the method: the story that reaches the terrace was typed into a file first.

Pooled retrospective cohort research on former professional footballers has reported materially elevated mortality from neurodegenerative disease against matched population controls — of the order of 3.5×. That figure sits in my 'to be verified' column and should not be printed without a primary source. The order of magnitude matters; the decimal does not. The direction is clear enough that a rounding argument does not change the conclusion.

So what has football done? I keep two columns — announced and completed.

The announced column is full. Associations including England and Scotland have introduced limits on heading in training at younger age groups — outright bans in some cases, phased age-band introduction in others. In 2026 IFAB wrote permanent concussion substitutes into the Laws of the Game, and several competitions adopted them. These are real entries on paper, and paper is not a small thing.

The completed column is not empty either, and calling it empty is how an auditor loses credibility. Concussion substitution is now part of the rulebook, youth heading restrictions are actual regulation, match-day medical protocols are mandatory in several leagues, and player unions now have language they did not have in 2026. That is progress and I will name it out loud.

The problem is not the rules. It is the denominator. No federation, league or data vendor holds a public dataset of headers per 90 minutes. xG exists. PPDA exists. Pass completion exists. Sprint distance exists. A header count does not. Every rule written about a thing nobody counted is a rule written on an estimate.

The minutes were always there; we just never bothered to count them.

This is where epidemiology and football analytics meet — at the denominator. Raw case counts mean nothing without time at risk in the form of person-years. A player's raw goal tally means nothing without per-90. The pairing of 3,001 and 287,632 does the same work as 12 goals and 1,080 minutes. The study's greatest strength is its design, and that design is used daily in my trade — the only difference is that I apply it to goals, not to skulls.

The inequality finding transfers just as cleanly. 13.6 against 8.3 is not a geography gap, it is a capacity gap — diagnostic access, specialist density, follow-up infrastructure. Football carries the same gap between elite clubs and second- and third-tier clubs: the top club has a neurologist and baseline cognitive testing, the lower-tier club has one physio and an ice bag. That difference is not about squad quality. It is about bookkeeping comfort.

Now the reversal.

13.6 is a floor, not a ceiling. Pooled volunteer cohorts like COSMIC systematically under-sample the populations carrying the heaviest risk — less education, less monitoring, less diagnosis. The low- and middle-income countries the paper flags as most exposed are precisely the ones least visible in the data. The true figure has room to move up, not down. Coverage that prints the inequality as a warning will be understating it.

Second reversal: read 10.6 against 8.6 and the first thought is 'women are more affected' — which is exactly where the researchers themselves stopped. The report states plainly that no single cause was established. Differential survival, diagnostic bias, competing cardiovascular mortality in men, hormonal and lifetime-risk factors — all plausible, none proven. Tomorrow's headline will be 'women more affected' and the caveat will be cut. Cutting caveats is such an old habit in this trade that we now call it an editorial policy.

Third reversal, aimed at the rules we file under 'completed'. Concussion substitution handles acute impact on the pitch. It does not handle the load accumulated across sessions, months and seasons. Repetitive, low-intensity heading does not walk through the door of a substitution. The protocol is necessary and the protocol is not sufficient — both statements are true at once, and hiding one behind the other wins an argument without surviving an audit.

And India's own ledger? Honestly: I hold no public document on heading exposure for the AIFF, the ISL or the I-League. What exists is this — after the 2026 Under-17 World Cup the country recorded academies, youth tournaments and hosting attendance; public documents on medical staffing, baseline cognitive testing and long-term player-welfare standards are sparse. That is not an accusation. It is an empty column, and the only cure for an empty column is to say so.

I will not claim heading is a crisis in Indian football — I do not have the data to make that claim. I will say that a country which hosted two Under-17 World Cups, whose youth teams drew record crowds at Salt Lake, ought to be able to answer how many times a week a child in one of its academies takes a ball on the head. That is not a question. It is a register that was never opened.

24.7 million new cases a year by 2050 — that figure can be deferred until the day somebody proves football is not a cause. The argument about whether the game produces dementia settled its ground in 2026, with a coroner's signature. The open question is whether any federation can answer this: how many headers did this player take in his career?

Until that answer exists, every rule rests on an estimate, every estimate rests on a neck, and every neck starts on an academy pitch where a coach is standing who does not know what he is teaching this week.

I am adding a column to my sheet. Against the 21 names in Cohort 21, it will now read 'headers'. Late, but a ledger opens late or not at all.