Nineteen Incidents In 2018. Ninety-One In 2020. The Rule Changed In 2019.
Two epidemiologists recoded all 618 entries in a national register against the definition in force before April 2019. On that definition the 2020 total is 24. Forty-one facilities have reported nothing in ten years.
A months-long reporting project. Documents cited below are held in The Vault and available to readers.
MANCHESTER, England — Recoded on the definition that was in force in 2018, the national register's total for 2020 is 24 incidents. The register publishes 91.
The gap is not an error and nobody appears to have been misled on purpose. It is a rule change, dated April 2019, and it is one sentence long.
The number in three submissions
The register records reportable incidents at 214 registered facilities. It holds 618 entries between 2016 and 2025, obtained by this newspaper under freedom of information in four tranches over eleven months.
Nineteen entries are dated 2018. Ninety-one are dated 2020. That is a rise of 379 per cent in two years, and the figure appears in three submissions to a parliamentary committee this year — twice sourced to the register, once sourced to the first submission.
What we did with the 618
We paid two epidemiologists, who have not met and who worked without sight of each other's files, to code every entry twice: once against the definition that applied before April 2019, once against the one that replaced it.
Both codings run in full with this story, entry by entry, including the 44 entries the two of them assigned differently. Where they disagreed, we counted the entry as reportable under the old definition — the choice that makes the old series highest and our own finding weakest.
One sentence, page 34
Until April 2019, a reportable incident was a release, a loss of containment, or an exposure.
The amendment adds three things to Annex C in a single sentence on page 34: a near-miss in which no release occurred, a spill contained within secondary containment, and any breach of a glove.
The change-control file shows the sentence drafted in November 2018, consulted on for six weeks, and answered by four respondents, three of whom asked for it to go further. Auriol Dacre-Nwosu, the committee secretary who drafted it, told us by telephone that she remembers arguing for the word "any".
Held constant
On the old definition, the ten annual totals run 21, 18, 19, 23, 24, 20, 22, 19, 17, 18.
The line drifts down rather than up, and the interval around the slope comfortably includes no change at all — which is the careful way of saying that ten years of this register show nothing much happening.
Of the 618 entries, 201 would have been reportable in 2018. The other 417 became reportable in April 2019 because somebody widened a definition deliberately, in public, and wrote down that they had done it.
The number nobody has quoted
Forty-one of the 214 facilities have reported nothing in ten years.
We wrote to all 214 with two questions: how many incidents have you reported, and how many have you recorded internally. A hundred and seventy-three answered.
Twenty-two of the 41 silent facilities answered. Six of those gave an internal count above zero. The largest was 31.
Eleven of the 41 run containment level 3 work continuously, on their own description of themselves.
Zero reported is not zero occurred, and it is not evidence of concealment either. Nineteen of the 41 did not reply at all, and for those we could not establish operating hours, staff numbers, or whether the building is in use.
The nearest comparable series in this newspaper's vault, PX-2179, is a police force's wildlife log that tripled in nine weeks when somebody added one option to a dropdown menu. Different subject, same shape.
What this desk takes from it
The spike is a definition. That took nine weeks to establish and is, we think, now closed.
Professor Rosalind Achebe-Thorne, who wrote one of the three submissions, has asked the committee clerk to correct hers. Her reply is printed below in full, and the strongest part of it is not the concession. She argues that holding the 2018 definition constant reinstates a blindness the 2019 sentence was written to fix, and that a flat line drawn with a rule nobody defends is not reassurance. This desk has no answer to that beyond the chart's own caption.
One further caution about this register, which we set out separately: its narrative field stops at 250 characters, and 561 of the 618 entries were coded from a sentence that breaks off.
And every number in this story is a count of events. None of it is a count of anyone becoming ill. The register has no field for that, and never had.
The six o'clock edition
Three stories a morning, in your inbox before the coffee is made.
Sources & Method
We did not audit a laboratory. We audited a definition: 618 entries recoded twice, by two epidemiologists who have not met and were paid the same fee in advance, against the guidance as it stood before and after April 2019, with both codings published entry by entry. We then wrote to all 214 registered facilities and asked them the two questions the register cannot answer about itself.
Who we spoke to
- National laboratory-incident register, 2016 to 2025, 618 entries across 214 registered facilities, with facility identifiers replaced by codes. Obtained under freedom of information in four tranches over eleven months, after one internal review September 2025 to July 2026 Released as a database export. The narrative field in every entry stops at 250 characters.
- Change-control file for the reporting guidance and its annexes, Drafting history, consultation responses and the superseded text of Annex C. Obtained under freedom of information; the consultation responses were released unredacted May 2026 Auriol Dacre-Nwosu, the committee secretary who drafted the April 2019 sentence, confirmed the sequence by telephone and remembers arguing for the word 'any'.
- Dr. Hester Balogun-Vane, Epidemiologist; first recoding, blind. Commissioned to code all 618 entries twice, against both definitions, with no sight of the second coder's work March to June 2026 Her coding sheet is published entire. Recodes 2020 as 26 under the old definition.
- Dr. Marek Prosser-Adeyemi, Epidemiologist; second recoding, blind. Given the identical file, the identical instructions, and no contact of any kind with the first coder March to June 2026 Recodes 2020 as 22. The published figure of 24 is the reconciliation rule agreed before either began.
- Written answers from 173 of the 214 registered facilities, Two questions: how many incidents reported, how many recorded internally. Letters sent to every registered facility; the answers published in full, and the facilities that did not answer listed by code April to July 2026 Six facilities reported an internal count higher than their reported count. The largest gap was 31.
Documents
- PX-2179 — Thirty years of one police force's wildlife-report log, 1996–2026, with the 1980 capture file accepted
What we could not confirm
- Whether 214 is all of them. The register covers premises registered under the notification scheme. Work done under a university's own permissions, or at a site whose registration lapsed mid-year, would not appear in it, and we could not build an independent list of laboratories to check the register against.
- The recoding of the 561 entries whose narratives stop at exactly 250 characters. Where the text breaks before an outcome is stated, both coders were instructed to take the more serious reading. That instruction raises the old-definition count in every year, and we cannot say by how much or whether it raises some years more than others.
- The 41 facilities that have reported nothing. Nineteen of them did not answer our letter and we did not visit any of them. Zero reported is not zero occurred; it is also not proof that anything was hidden, and a small dormant unit reporting nothing for ten years is exactly what a well-behaved register should look like.
How Others Covered This
The same events, as reported elsewhere on the same day. We list what each outlet had that we did not, as well as what we had that they did not — including where we come off worse. Why we print this.
-
The Meridian TelegraphLAB ACCIDENTS UP 379 PER CENT — AND NOBODY TOLD PARLIAMENT
Took the register's own annual totals as a series and did not ask whether the thing being counted had changed.
Had that we did not
The totals, accurately transcribed. Their table is correct and ours is built from the same file.
Left out
The April 2019 amendment, which is one sentence on page 34 of the annex their own disclosure enclosed.
-
Signal & LedgerReporting Uplift At 214 Sites: What The 2019 Annex Changed For Compliance Teams
Read the amendment as a cost story for readers who run laboratories, and read it early and accurately.
Had that we did not
The clearest published account of what became reportable in April 2019 — clearer than ours, and eleven months before it.
Left out
The 41 facilities that have reported nothing, whose compliance cost, on this framing, is zero and looks like efficiency.
-
They Buriedthis newspaperNineteen Incidents In 2018. Ninety-One In 2020. The Rule Changed In 2019.
Paid two epidemiologists to recode all 618 entries blind against both definitions, and published both codings entire.
Had that we did not
Both codings, the 44 disagreements, the change-control file, and the written answers of 173 facilities.
Left out
Our headline sells the spike we spent nine weeks disproving, and the 41 silent facilities — the finding that actually frightened this desk — arrive in the fifth section, behind a chart. — V. Ashcombe-Doyle, standards editor
Right of Reply
They Buried contacted Professor Rosalind Achebe-Thorne, author of one of the three parliamentary submissions on 6 July 2026 in writing, with the full draft, both coding sheets, the change-control file and no deadline. Replied 4 August 2026, asking that it run entire. Printed unedited, including the parts about us.
Your arithmetic is right and I have written to the committee clerk asking that my submission be corrected. It should not have taken a newspaper eleven months of disclosure requests to do that, and I am not going to spend four paragraphs explaining why I did not check the annex. I did not check the annex.
Now let me tell you what I think you have done.
You have taken a definition that was abandoned because it could not see anything until it was too late, applied it to nine years of data, discovered that it still cannot see anything, and printed the flat line it produces as though flatness were reassurance. That is not holding a variable constant. That is reinstating the blindness and calling the result stability.
The old rule counted releases, losses of containment and exposures. Every one of those is an event that has already happened to a person or to the air. The April 2019 sentence added the near-miss, the spill the cabinet caught, and the glove. Those are the events that occur before the ones you can count, and the entire reason a modern reporting system exists is to move the eye earlier. Your chart takes the eye back to 2018 and reports what it can see from there. Of course it is flat. It was flat in 1994 too.
I will give you the honest version of the counter-argument, because you gave me your coding sheets and I owe you that. It is true that a widened definition inflates a series, that three submissions including mine used the inflated series as though it measured deterioration, and that this was sloppy and consequential. Committees allocate inspectors on numbers like that. I helped point inspectors at the sites that were writing things down.
But you should understand what your headline will be used for. It will be quoted, without the seventh section, by people who would like a reason not to fund an inspectorate. I know this because I have watched it happen to a null result before. Your paper's habit of putting the deflation in the headline is admirable when the subject is a legend about a nurse. When the subject is whether anyone comes to look at a cabinet, it has a cost, and the cost does not fall on you.
The strongest thing in your story is the forty-one, and you have buried it under a chart. Eleven sites running containment level 3 continuously, for ten years, with nothing at all to say. I reported a glove breach in 2003 and was asked, not unkindly, whether I really wanted to make the work. I said yes and I filled in the form and nothing whatever happened to me, which is the point: the cost of reporting is a conversation, and forty-one places appear to have found even that too expensive.
Write that story next. Put it at the top. And when you do, do not hold anything constant.
Published unedited under our right-of-reply guarantee.
How was this story?
We publish the result, whatever it is. Reader verdicts appear on the front page and in our newsroom metrics.
Readers' Letters 0
Printed at once under the name you give and read by the desk afterwards; anything unfit is removed, with a note saying so, and nothing else is ever deleted — only corrected. Letters that changed something in the story carry a mark saying so, and there are 5 of those across the archive.