Clinical trial lay summaries, explained.

A clinical trial lay summary explained: telling the people who took part what was found

One psoriasis study, from the consent form to the public record

1. The promise at consentVLT-302's participants are told, from March 2023 on, that a summary will appear in the EU database.
2. The trial ends and the clock startsJanuary 14, 2026: the last participant in the last EU country finishes. The sponsor has 15 days to notify, and one year to file results.
3. The study report is finishedOctober 2, 2026: the final report arrives, more than eight months into the year.
4. A writer turns it into plain wordsTen sections, every number checked, every sentence graded, reviewers signed. The "7 out of 10" is caught here, or published.
5. Translated, filed and publicEach language checked against the English, PDFs filed in CTIS, the EU's trial database, by January 14, 2027. Anyone can read it the same day.

Step 4 is the subject of this guide. The study, people and figures in this example are fictional.

Why a good reading score proves little, and three more lay summary surprises

What lay summary, CTIS, Annex V and clinical study report mean

Fourteen terms are defined in the one-page summary.

What a lay summary must contain: the ten Annex V elements, and where each one goes wrong

In shortAnnex V of the Regulation lists ten elements and Article 37 adds one quality test: the summary must be written "in a manner that is understandable to laypersons". Sponsors may combine or reorder the elements. Good Lay Summary Practice then explains what each one means in practice, and it is guidance, with no force of law on its own.

Element wording from Annex V; explanations from Good Lay Summary Practice.1,5

Lay summary, technical summary and study report: three documents from one trial

Regulation Article 37(4).1

EU CTR Article 37, the UK's 2026 rules and the NIH draft: who must publish a plain language summary

In shortThe EU has required lay summaries since the Regulation applied in 2022, and every trial in the EU has been under it since January 31, 2025. The UK added its own version on April 28, 2026. The US requires results on ClinicalTrials.gov, its public register, and no plain-language version, though the National Institutes of Health (NIH) proposed one for the research it funds on August 27, 2026. ICH E6(R3), the international good clinical practice guideline, asks every sponsor to consider telling participants.

CTIS results posting deadlines: 12, 6 and 30 months, and what actually starts the clock

In shortThe clock starts at the end of the trial, as the protocol defines it, in the last EU country, and that date must be notified within 15 days. Trials in adults: 12 months. Trials in children: 6 months. Most summaries become public the day they are filed, and adult-only early trials are held back until month 30. The study report's arrival date, set by nobody's rule, decides how much of the year is left.

When a lay summary becomes public in CTIS, by trial category

EMA/263067/2023, publication table.4 Two 30-month figures are in circulation: the 2021 guidance's submission allowance and the current rules' publication date. EMA's handbook gives one year to submit and names no 30-month allowance, so a calendar still using 30 months to file can be 18 months late.5,3

Why the VLT-302 writer has 99 days out of a 365-day year

Guidance names this risk directly: "if the creation of the LS [lay summary] awaits the final CSR [study report] or full set of structured trial results, the authoring and translations of the LS may become time critical activities". It suggests developing the summary in parallel with the source documents, and allows a two-step translation: first from a draft master, then adjusted to the final one.5 For a trial in children, starting at the final report is rarely an option.

Early termination, interim analyses and the only scientific excuse

Writing a lay summary from the clinical study report: which count, which events, which table

In shortA clinical study report holds several correct versions of most facts: people screened, randomized (assigned to a group by chance), treated and analyzed; adverse events and adverse reactions. A lay summary needs one consistent choice, stated plainly. The report counts every medical problem during the trial, while Annex V asks for reactions a doctor thought the medicine might have caused, and copying the wrong table is an easy mistake to make.

PASI 90 is the report's measure of skin at least 90% better.

Adverse events in the report, adverse reactions in the summary

An adverse event is any medical problem during a trial, whatever caused it. An adverse reaction is one where the investigator, the person responsible for the trial at a site, saw "at least a reasonable possibility" that the medicine caused it. The technical summary reports events, and Annex V asks the lay summary for reactions. Guidance calls this an intentional difference and says the benefit is a text that is easier to follow, without bulky tables of unrelated events and without asking readers to weigh rates against the placebo group themselves.5

Choosing one participant count and sticking to it

Screened, randomized, treated and analyzed can be four different correct numbers. Guidance asks that differences between randomized and treated be "presented clearly to avoid confusion" and that the population numbers align with the results section as far as possible. In a rare disease, a breakdown fine enough to identify one participant has gone too far.5

Lay summary reading level: the Flesch-Kincaid grade worked out on two sentences, and what it misses

In shortThe Flesch-Kincaid formula turns sentence length and syllables per word into a US school grade. EU guidance calls a 6th grade level ideal and says a good summary should normally be readable from about age 12. The formula is quick and useful and cannot tell whether a sentence means anything, so guidance says to use it only as a supplement, and it is built for English.

Our hand count on the guide's fictional case, counting each number as one syllable. Tools count numbers, abbreviations and syllables differently, so another tool can land a point away. Compare a grade only with one from the same tool. A hazard ratio compares how often an event happens over time in two groups; section 07 turns this one into plain numbers.

The target: grade 6 as the ideal, grades 6 to 8 for patient material

The Regulation sets no reading level. The Expert Group's Annex 3 says "an ideal reading grade level is 6th grade" and "strive to get as close to this as possible", and asks for text suited to literacy proficiency levels 2 to 3 on a five-level scale, where level 3 roughly matches finishing high school. It also says a well-written summary "would normally be accessible by young people from the age of 12 years upwards".6

The American Medical Association and NIH recommend grades six to eight for patient material, according to one 2025 review.22 Registered trial text falls far short of that: the same review scored 17,175 urologic cancer trial descriptions on ClinicalTrials.gov at a mean grade of 18.7.

Who the summary is for: a quarter of adults read at the lowest levels

The same survey puts 25% of adults at the same low band in numeracy on the OECD average, which is one reason a bare percentage can fail its reader.23

Four things a Flesch-Kincaid grade cannot see

Turning hazard ratios and percentages into plain numbers without changing the result

In shortUse whole numbers, keep denominators (the "out of" number) the same, do the arithmetic for the reader, and prefer absolute counts to relative ones. A relative reduction such as "34% fewer" looks bigger and persuades more than the same result given as "10 fewer out of 100". The count is usually in the study report, in the same table as the statistic it replaces.

One depression result written five ways

A hazard ratio below 1 means the event happened less often in the treated group than in the comparison group, and the confidence interval is the statistician's range around that figure. Neither tells a reader how many people were helped.

47 ÷ 246 = 19.1%. 71 ÷ 244 = 29.1%. 29.1 − 19.1 = 10.0 points. 10.0 ÷ 29.1 = 34%. Fictional example.

A Cochrane review, a formal pooling of published research, combined 35 studies on this. People understood natural frequencies better than probabilities. They understood relative and absolute reductions about equally, yet relative ones "may be perceived to be larger" and are more likely to persuade. The evidence was rated moderate quality and no comparison measured behavior.25 The Expert Group's advice follows: present absolute numbers and consider percentages, "rather than relative risks, odds ratios etc."6

When rounding 64.7% to "7 out of 10" changes the result

Table 3.3 of the guidance.5 The Recommendations add a common form: "x out of xx people [xx%]".6

Non-promotional language in a lay summary: words to watch and how to rewrite a claim

In shortA lay summary can look like an advertisement without anyone meaning it to. Guidance says to keep the tone factual, report the good and the bad, avoid inferences, and be careful with words like "safe", "effective" and "better" unless a number stands behind them. The risk peaks in the comments section, often written by someone who has spent years on the trial.

The drafts and rewrites are this guide's fictional examples, built on the guidance's rules. 5,6

What lay summary guidance says to do, avoid and handle with care

Lay summary translation and back translation: what each check catches

In shortThe Regulation does not ask for translations, and guidance expects them in each country's local language as a minimum. Good practice translates from an approved English master, the final text every translation starts from, with a locked glossary of agreed terms. The ideal process uses two translators and a third reviewer. Numbers drift and meaning shifts in ways only a check against the source can catch.

Which languages a lay summary needs, and by when

A study of consent documents in Zambia shows the kinds of problems that surface. Translators found no local equivalent for terms such as "pre-eclampsia", and words like "research", "benefits" and "risks" were understood differently. A back translation exposed a reference to "funerals" that the original did not contain.

The authors note that back translation is "not always required by ethical review bodies".28 It is a small qualitative study in another setting, and the same kinds of failure can occur anywhere. No public figure gives the size of the problem in EU languages.

Fewer than half of the first CTIS trials reported on time, and what the audit could not see

In shortThe first study of results reporting in CTIS counted 234 Phase II to IV drug trials due to report: 116 did so on time, 20 late and 98 not fully. The gaps were mostly in the scientific document, with only 4 trials missing the lay summary alone. The study is a preprint, posted before peer review, and it checked whether a lay summary existed without reading it, so nobody has yet published how good CTIS lay summaries are.

Some documents said results had never been analyzed or were still being analyzed. Others held only vague narrative or redacted data.30 The law treats the two summaries as one filing, so a sponsor should plan them as one job.

Return of results: what the public wants and what past participants recall

In the 2025 wave of the same survey, 81% of respondents were interested in their individual results and 78% in a summary of study results.32

What a late or misleading lay summary can cost a trial sponsor

In shortPenalties are national: German law allows a fine of up to €25,000, and we found no published case of one. The larger exposures are public: the end-of-trial date is visible, so anyone can see a missed deadline, and EU regulators said in July 2026 that they have developed monitoring by Member State. A misleading summary damages something else: the trust of the people it was written for.

AI lay summaries: easier to read by the numbers, and what regulators expect

In shortLarge language models, the kind of AI that writes text, can bring a trial summary's reading grade down fast. Studies show that this does not prove people understand it, or that the facts survived. No regulator has written a rule for AI in lay summaries, and the sponsor stays responsible for every word. Software suits work with a clear rule, such as comparing numbers. People keep the judgment.

What FDA, EMA and the EU AI Act say about AI in lay summaries

How disclosure teams produce a lay summary today: who does what, and where the weeks go

In shortA medical writer or an agency drafts from the study report, a clinician and a legal reviewer check it, patients read it where the budget allows, a coordinator runs translations, and a disclosure lead files it. Translation and review take most of the calendar. No regulator or study publishes how long this takes, so the figures below are planning estimates to replace with your own.

Medical writer to disclosure lead: who touches a lay summary

Six hand-offs between the study report and CTIS

Where the working days go on a hand-written lay summary, by our estimate

The post quoting that 20-hour figure gives no dataset behind it.47 We found no public figure for agency turnaround or translation time either. The workshop report calls the steps "complex and potentially onerous", and says inefficient compliance could make the process costly enough to threaten the aim.27

In-house writers, a specialist agency or plain language summary software

A psoriasis summary and a depression summary: three flags to decide, one hazard ratio to rewrite, one signature

In shortThis is the example from the top of the page, inside the product. On Wednesday, October 7, 2026, Lena Ortiz, a medical writer, has seven summaries on her desk. She sees each sentence beside its source and its translations, decides the flags on VLT-302 (sentences the software marks for a person to settle), rewrites a hazard ratio on NXL-125, and signs as author with the meaning of her signature on screen.

1

The report, the plain English and two translations, one row per sentence

VLT-302 opens as columns: the study report passage, the lay English, German, Polish. Each English sentence carries its reading grade and a tick that its fact matches the report. Each translated sentence carries its numbers read back.

The check ribbon along the top reads 28 of 29 facts matching, reading grade 8.9 against a target of 8, 5 of 8 questions answered by the reader test (software that reads only the summary), and 3 sentences to decide. The deadline is January 14, 2027, 99 days away.

2

Report jargon left in, grade 11 to grade 7

She filters to "To decide". The first flag: "treatment-emergent adverse events" is still in the draft. The plain-language checker proposes "at least one medical problem", adds that not all of them were caused by the medicine, and shows the grade dropping from 11 to 7. It cannot change the text itself.

In a live summary, this section would report adverse reactions as the investigator judged them; the sample sentence reports all events, with that caveat.5

3

"About 7 out of 10" against 277 of 428

The fact checker marks the results sentence as different from Table 11.4 on page 52: 277 of 428 people, 64.7%. Its rule flags any rounding that moves a number by more than 2 points, and this one moves it by 5. It proposes "about 65 out of 100". The reader test got the same question wrong from the draft.

The third flag is the comments section's "worked very well and is a safe treatment", with a neutral rewrite that says this is one study.

4

A correct hazard ratio at grade 14

On NXL-125, a depression study of Talvoren, there is one flag: "The risk of relapse was reduced with Talvoren versus placebo during the randomized withdrawal phase (hazard ratio 0.64; 95% confidence interval 0.44 to 0.92)." It matches Table 11.3 and reads at grade 14 on this screen.

The table also holds the counts: 47 of 246 on Talvoren (19.1%), 71 of 244 on placebo (29.1%). The checker proposes: "About one in three fewer people on Talvoren had their depression come back." That is a relative figure. A team following the Expert Group's advice would edit the rewrite to give the counts as well: about 19 out of 100 and 29 out of 100.6,25

5

Accepted: the ribbon moves and both translations re-run

Lena accepts. The sentence reads at grade 7 with its tick, "19.1 vs 29.1 out of 100, 34% fewer, Table 11.3", and is marked as accepted by her. The German and Polish rows re-translate and read back with the same meaning. NXL-125's ribbon now reads 24 of 24 facts, reading grade 7.9 (from 9.2) and 8 of 8 questions (from 7). The button changes to "Approve and sign".

6

She signs as author, with what the signature means

The signing window repeats what she is signing: 24 of 24 statements match, reading grade 7.9, reader test 8 of 8. The meaning reads "I am the author and confirm every fact matches the final study report". Next, Priya Ranvelle's patient panel reads it, then Dr. Maya Chen and Dana Okafor sign, and translations into the other languages start from this signed English.

The solution's design target is 2.4 days from a final study report to a checked first draft. Measure it on your own reports.

Twelve more screens, from the participant's leaflet to the disclosure lead's posting runway

In shortOne writer's day is the story. These screens show the rest: the summary as a participant reads it, its reading grade, the reader test, the languages, a new study report being drafted, the portfolio against its deadlines, the shared glossary, and the rules a team sets once.

Six agents read, draft and check a lay summary, and six named people sign it

In shortEach agent, a piece of AI software, has one job and limits it does not cross. None can change a sentence once it is flagged, approve a language or post anything. The medical writer decides every flag and signs as author, and five more people each sign their own step. Posting stays blocked while any fact is unchecked or any flag is open.

Study report reader to translator: the six lay summary agents

The writer decides each flag, and five people sign after her

Signatures record the name, the meaning, and the time; each is an e-signature that moves the summary to its next stage. Each agent step, accepted fix, dismissed flag and signature is on the audit trail from the day the study report arrives, as controls that support your compliance with your own quality system. Botminds holds a SOC 2 Type II audit and manages information security to ISO 27001.

A final study report in; a checked, signed lay summary in every language out, ready for CTIS

In shortThe solution reads the documents your team already has, writes to your house style, and leaves a signed package for each language with a source behind each sentence. It sits alongside the systems you keep.

Document systems, translation memory and registers around a lay summary

Built for a portfolio: a shared plain-word glossary across every summary and language, each trial on one runway, and the same rules applied every time.

What a hand-checked lay summary costs you in writer days, translation weeks and public misses

In shortStart from your own costs. Nobody publishes a cost per summary, an error rate or an agency turnaround, so the numbers that matter are yours. A few published figures frame them: how many trials miss the deadline, what share of readers struggle with text and numbers, and how often participants hear back.

Timing three lay summaries to find your own baseline

Participants who never hear back, and what a promotional line can cost

What faster lay summaries return when you count writer hours, translation rework and days back on the clock

In shortPut in your own numbers. The calculator values the hours saved on writing and checking and the translation rework you avoid, then sets them against what the tool and your testing cost a year. It leaves out what matters most, on-time filing and fewer public errors, because those have no honest dollar figure.

Fact pulling, number checks and translation rework: where a lay summary saves

Summaries a year, languages and your hourly cost

A sponsor filing five summaries a year may gain more from a single good template, a locked glossary and one trained writer than from software. Volume, languages and six-month trials are what change the answer.

How ready your disclosure team is: study volume, source reports, house style, languages and sign-off

In shortPick the answer closest to your team today. Eight questions score 0 to 2; the last is for reflection. The meter adds up your score as you go, and each question says why it matters and what a strong answer sounds like.

A low score means settling your CTIS calendar and house style first

What to ask a plain language summary tool or agency about sources, readers and back translation

In shortAsk every vendor the same questions, us included. The answers show how a summary is built: where each fact comes from, how numbers are written, how plainness is measured, how translations are checked, who decides, and how you would test it on your own reports.

We found no published independent comparison of lay summary vendors, as of November 2026. A planted-error test on your own reports is the fairest measure, and it measures your current process too.

Test lay summary software on your own study reports with planted errors and real readers

In shortMeasure first, then test on your own reports with errors you planted, then decide with four numbers. Keep your people signing and your patient readers reading throughout. None of these steps commits you to a vendor.

1. Audit ten filed summaries against their reports, sentence by sentence, and log each error type. That is your error baseline.
2. Time your next three summaries by step, including translations and review rounds.
3. Gather 12 recent final reports with your house style and glossary. Copy some drafts and plant known errors: a bad rounding, a wrong group, a promise in the comments, a changed number in a translation.
4. Run in parallel on live trials for eight weeks, with your writers signing every summary and readers reading them.
5. Decide with four numbers: catch rate by error type, false alarms on clean copies, writer hours to a signed summary, and how many key questions real readers answer correctly.

Scoring a lay summary pilot: catch rate, false alarms, hours and reader answers

Number fidelity is the check behind the catch rate; the other four are the numbers you decide on.

Lay summary questions answered: Annex V, CTIS deadlines, reading level, translation and AI

In shortQuestions writers, disclosure teams and buyers search for, each answered on its own.