Clinical protocol authoring, explained.

What a clinical trial protocol is, and why one document ties a whole trial together

One Phase 3 protocol, from a clinical team's idea to the first site visit

1. The clinical team agrees the ideaVeltrimab worked in Phase 2. The next trial needs 660 participants, about 120 sites and 14 countries.
2. A synopsis is writtenNine pages: design, who may join, endpoints (the measures that judge the drug), dose, visits and sample size. Version 1.0 goes to medical writing on October 6.
3. A medical writer drafts the protocolFourteen sections in the sponsor's template, adapted from past protocols and built around the schedule table. This is where a missing Week 16 is caught, or carried forward.
4. Reviewers comment and the sponsor signsStatistics, clinical, operations and the medical monitor each read their sections. The head of medical writing signs the final version.
5. Regulators, ethics committees and sites receive itFrom here, any change is an amendment, with its own review, its own wait and sites working from different versions.

Step 3 is the subject of this guide. The people, drugs, studies and figures in this example are fictional.

Why a protocol is hard to get right: four findings that surprise newcomers

Synopsis, schedule of activities, estimand, amendment: the words to know

Fourteen terms are defined in the one-page summary.

The ICH M11 protocol template: fourteen sections, and which of them a writer owns

In shortICH M11 gives every interventional drug trial, one that assigns people to a treatment, the same table of contents: eleven main sections and three appendices, with the synopsis, the trial diagram and the schedule at the very front. It comes as three documents with three jobs. Most sponsors will map their own template onto it, work that consists largely of moving and splitting text.

M11's full name is the Clinical Electronic Structured Harmonised Protocol, shortened to CeSHarP. It applies to interventional trials of medicinal products "across all phases and therapeutic areas".3 Its reason for existing is plain: "Variability in format and core content among sponsors contributes to inefficiencies and difficulties in searching, reviewing, and assessing protocols."3

Guideline, template and technical specification do three different jobs

The fourteen sections, and the people who answer for each

Section numbers are those of the ICH M11 template, with titles shortened.4 Your sponsor's own numbers can differ at the lower levels.

ICH M11, E6(R3), 21 CFR 312 and the UK's 2026 rules: what binds a protocol, and since when

In shortBinding content rules sit in law: 21 CFR 312, FDA's rules on trials of new drugs, in the US, and Regulation 536/2014 in the EU. ICH guidelines set the shared practice. In the past two years M11 became final in the EU and the US, E6(R3), the guideline on good clinical practice for running trials, took effect, and the UK rewrote its clinical trial regulations. EU reform is still a proposal.

How to build a schedule of activities, the table every other protocol section quotes

In shortThe schedule of activities lists visits across the top and procedures down the side, with a window on each visit. It decides which tests exist, when each endpoint is read and how long each day at the site runs. Teams who build it first and check the endpoints against it catch the mistake that opens this guide: a main result with no visit to collect it.

The M11 template is short and exact about this table. The schedule "must capture the procedures that will be accomplished at each trial visit, and all contact with participants". That includes tests used for eligibility, randomization or stratification, and for decisions on stopping treatment. "Allowable windows should be stated for all visits and procedures. A tabular format is recommended."4 Where sampling is extensive, such as serial pharmacokinetic (PK) sampling, repeated blood samples that track the drug's level in the body, a separate table may be added.

Week 16 in a table copied from a Week 12 study

Here is part of the VLT-302 table as a first adaptation of VLT-201 might leave it. Skin scores are PASI (the Psoriasis Area and Severity Index), sPGA (a doctor's overall rating) and BSA (the share of body surface affected).

Fictional example. The synopsis says both main results are read at Week 16, and the schedule collects other data that week, so the visit looks complete at a glance.

Four checks to run on every schedule of assessments before anyone reviews the prose

The fictional VLT-302 has 11 visits, which is lighter than the Phase 3 mean. Benchmarks describe past practice, and a schedule has no need to reach them.

Objectives, endpoints and estimands: a worked psoriasis example of what section 3 must say

In shortAn objective is the question in plain words. An estimand makes it exact: population, treatment, endpoint, summary and how to handle events that interrupt a participant's course. An endpoint, the measure that judges the drug, is one part of that. M11's section 3 asks for all of it in a table, and the writer needs the biostatistician to fill it.

ICH E9(R1) introduced the estimand so that a trial's design and its analysis answer the same question.15 M11 turns it into a table under each objective and adds an instruction that every writer should keep in view: "Ensure alignment with every other section of the protocol."4 The table has four named parts, plus a row for each intercurrent event, an event such as stopping treatment that changes what a result means.

The writing trap is timing. Estimands take conversation, so they tend to arrive late, while the rest of the protocol keeps moving. Teams who treat the estimand table as new text, agreed with the statistician before the first review round, avoid rewriting sections 4, 7 and 10 around it. Teams who paste objectives from a past study often find that its estimand no longer matches its endpoints.

Protocol consistency checks: the facts a protocol repeats, and where they drift apart

In shortM11's design removes repetition "wherever possible", yet a working protocol still states its key facts many times. Each repeat is a chance to disagree. A short list of facts accounts for most of the checking, and it can be walked through by hand or by software in the same order.

The template's authors built in the fix they could: "Unnecessary repetition is eliminated wherever possible."3 The synopsis, however, must summarize the rest. Section 3 must name what section 8 measures. Section 10 must analyze what section 3 defines. So reviewers still read 100 pages looking for one number that did not move. The fictional VLT-302 draft shows three kinds of drift.

Seven protocol facts, and the sections that have to agree on each

Protocol complexity benchmarks: how heavy a Phase 3 trial is, and how fast it grew

In shortTufts benchmarks put an average Phase 3 protocol at 34.5 distinct procedures performed 266 times, across 21 visits, 82 sites and about 3.45 million data points. Procedures grew faster than endpoints. About a quarter of procedures support only supplementary or exploratory endpoints, which is where a design team has room to cut.

From Tufts CSDD benchmarks: 48 Phase 1, 72 Phase 2 and 67 Phase 3 protocols.1 These are means of a voluntary company sample. Spread is wide.

The non-core share: procedures that hold up no primary or key secondary endpoint

Non-core procedures rose from 21.2% in 2013-2016 to 24.7% in 2017-2020 across Phase 2 and 3.1 No regulator has a formula for protocol complexity, and there is no agreed protocol complexity score. Published measures count the same things: endpoints, criteria, procedures, visits and how long a visit lasts. A score earns its keep only when someone can act on it. A protocol-level number says that a protocol is heavy, and leaves open which visit to lighten.

How to calculate patient burden per visit, and where the 550 mL blood draw limit comes from

In shortTufts and the consulting firm ZS built a burden score from patients' own ratings of procedures, and it tracks trial duration, screen failure (people tested who turn out ineligible) and amendments. A simpler per-visit score, built from minutes, milliliters and needle sticks, tells a team which visit to fix. The 550 mL figure people quote sets a route for ethics review, for healthy adults. A Phase 3 trial's blood limit comes from its sponsor and its ethics committee.

What happened when patients rated 60 trial procedures

In 2019, 591 US patients rated the burden of 60 common trial procedures. A second survey of 2,680 patients worldwide added travel, location and lifestyle limits. The later version of the algorithm was reported as "highly associated with, and predictive of" trial duration, and linked to screen failure rate and the number of substantial amendments.21

In a workshop where eight companies scored 66 past protocols, labs and blood work carried the largest share of burden, and the score did not correlate significantly with dropout.21 That last result is worth saying aloud to anyone who promises a burden score will keep more participants in a trial.

A 2025 analysis of 156 Phase 2 and 3 protocols from 14 companies found burden "rising steadily since 2011". Questionnaires, lab and blood work and routine procedures contributed most. Visit length grew "in large part due to the volume of procedures performed per visit".22

A per-visit burden score, worked through on Week 4

A simple team instrument weighs what a schedule already holds. The weights below are one fictional sponsor's choice, used throughout this guide. They are unrelated to the Tufts algorithm and carry no authority beyond that sponsor's own agreement to use them.

The blood draw limits for research, and what each one governs

A participant with psoriasis falls outside the "healthy adults" line of that 1998 list, and a Phase 3 drug trial normally gets full board review. So the 550 mL in a schedule tool is a sponsor's chosen reference point, useful for spotting a heavy schedule, with no legal force.

Protocol amendment vs deviation: what each is, who approves it, and how often it happens

In shortAn amendment changes the protocol, on purpose and before the fact. A deviation is a departure from the protocol at a site, found afterwards. Three in four protocols are now amended, on average 3.3 times. The often quoted "45% avoidable" dates from 2016. The 2024 study calls 77% unavoidable, which leaves about 23%. Either share is real money on a large trial.

The same protocol change under US, EU and UK rules

Add a liver panel to a draft and it is an edit. Add the same panel after the trial starts and, under 21 CFR 312.30, it is the textbook protocol amendment.17

How often protocols are amended, and how long a change takes to land

Avoidable protocol amendments: the 45% from 2016 and the 23% or so from 2024

Both numbers rest on sponsors' own judgment after the fact. Read together, they say most amendments have causes outside the writing, and a sizable minority are preventable. Multiply a 23% share by a mean of 3.3 amendments and a typical protocol carries roughly 0.8 of an avoidable amendment. That is our arithmetic from the two published figures, and a rough one.2

Deviations: the same table, read from the site's side

In that study, deviations touched 32.8% of patients in Phase 3 protocols.27 A later Tufts report, which we read only as quoted in a WCG presentation, puts Phase 3 protocols with a primary completion date in 2023-2025 at an average of 296.3 deviations and 3.4 substantial amendments.28

FDA's December 2024 draft guidance adopts the ICH E3(R1) definition of a deviation, "any change, divergence, or departure from the study design or procedures defined in the protocol", and of an important deviation. It was still a draft in every source read.29 M11 gives deviations a home at section 11.10.4 A pattern of the same deviation at many sites often means the protocol itself needs changing.

Nine common protocol writing errors, why they happen and where they surface

In shortThese errors share one cause: a good old document adapted under time pressure, with several people deciding in parallel. They surface at review, at a site or during an inspection. No public source counts how often each occurs, so count your own from your amendment log.

What FDA investigators write down about following the investigational plan

FDA's Bioresearch Monitoring metrics for fiscal 2024 list "Failure to follow the investigational plan; protocol deviations" among the common observations at clinical investigator inspections, beside recordkeeping and informed consent issues.30 The list gives no percentages. It is a reminder that a protocol is the yardstick inspectors use. Ambiguity in the text becomes a site's finding.

A heavy schedule falls on sick people's working days, and a missed endpoint visit can leave data that cannot answer the question participants joined the trial to answer. E8(R1) asks that study procedures be necessary "from a scientific viewpoint and do not place undue burden on study participants", and it encourages consulting patients early in design.14

What AI can do in clinical trial protocol writing, what regulators expect, and where digital protocols stand

In shortLarge language models, the AI behind chat assistants, write fluent protocol text, and alone they score poorly on clinical logic. Working from retrieved documents roughly doubles that score. No published study measures writer hours saved. FDA's draft AI guidance sets aside drafting that does not affect study reliability, and regulators' other papers ask for the sponsor to answer for what it uses.

What two studies measured when AI wrote protocol text

The same study's limits are the useful part. It scored each section alone, so it could not measure whether endpoints agreed with eligibility: the failure this guide spends most time on.31 A 2026 study on 900 protocol and consent form pairs is closer in spirit.

Handling schedule tables as structured data raised factual accuracy, and having experts verify those tables raised it further. People checking a table gave the strongest result in that study.32

What FDA, EMA and the EU expect of AI in protocol writing

The principles of data integrity known as ALCOA+ (records attributable, legible, contemporaneous, original, accurate, and more) apply to a protocol's audit trail as they do to other trial records.

Structured protocols: USDM, the Common Protocol Template and FHIR exchange

A protocol that exists only as a Word file has to be retyped for each system downstream: data capture, the trial registry, the site's visit plan. The M11 technical specification points toward content that moves as data.3 The industry group TransCelerate says its Common Protocol Template, version 11, is aligned with M11 and with the Unified Study Definitions Model (USDM) from the data standards body CDISC.43

The health data standards body HL7 publishes a Clinical Study Protocol guide on its FHIR exchange format, built for exchanging M11-aligned protocols between sponsors and regulators. It is a ballot version, still on trial use, and mainly exchanges narrative with some structured content so far.44 So digital exchange is arriving in steps, and a team's nearest gain lies in a structured schedule and reusable content.

Who writes a clinical trial protocol today: from synopsis to signature, by hand

In shortA medical writer owns the document, and five or more specialists own its decisions. Hand-offs run through comments and email. The split of writer days below is illustrative, since no public figure exists. The published National Cancer Institute timelines show that review and waiting dominate the calendar.

The people a protocol passes through before it is signed

Where a medical writer's 38 days might go in a by-hand Phase 3 draft

The whole path is far longer than the draft. When the National Cancer Institute reformed its trial development, the historical median for Phase 1-2 studies from concept to activation was 541 days, falling to 442 under firm deadlines, an 18.3% decrease. Phase 3 studies showed a 45.7% decrease on a smaller sample of 25.45 That figure includes review, negotiation and queues as well as writing. Its lesson travels beyond oncology: deadlines and tracking moved the median.

In house, outsourced, structured software or AI-assisted: four ways to run protocol writing

From a nine-page synopsis to Week 4 back under the limit: the schedule of assessments at work

In shortThis is the example from the top of the page, inside the product. Lena Ortiz, senior medical writer, opens the protocol on its schedule. The missing Week 16 is flagged at once. She fixes it with one click, opens the hardest visit, adds the medical monitor's liver panel, and sees Week 4 go from 53 to 70, over the limit. A suggestion brings it to 57, and the protocol text follows.

1

The protocol opens on its schedule, with Week 16 flagged

Synopsis version 1.0 arrives at 8:12 am on October 6. At 8:14 the first draft exists: 14 sections in sponsor template v3.2, 58% of it approved language. Lena's screen is Table 1: 11 visits across, 15 procedures down, and a burden bar over every visit against a dashed line at 60.

The consistency checker has already found the problem: "Week 16 is the co-primary endpoint visit, but PASI, sPGA and BSA are not scheduled there." Co-primary means both main results are read at that visit. The table came from VLT-201, whose primary visit was Week 12. Empty dashed cells mark the gap.

2

One click adds the skin scores at Week 16

Lena clicks "Add them at Week 16". The three cells fill, the shaded column turns green, Section 1.3 updates and the draft is saved as version 0.2. Week 16's score moves from 52 to 57, near the limit and under it.

The change appears under Latest changes, with its time, and the checker's finding closes with Lena's name on it.

3

Day 1, the hardest visit, with the arithmetic shown

She clicks the Day 1 bar. Eligibility, four ratings, vital signs, laboratory tests, a pregnancy test, a PK sample (blood that tracks the drug's level), antibodies and the first injection, with 15 minutes of check-in and waiting: 1 hour 55 minutes, 21 mL of blood, two needle sticks.

0.25 × 115 + 0.6 × 21 + 9 × 2 = 59, against a limit of 60. Draws taken together count as one stick. The weights and limit are settings the sponsor owns.

4

The medical monitor's liver panel goes in

Dr. Priya Ranvelle, the medical monitor, asks for a liver panel at Weeks 2, 4 and 8. Lena chooses "+ Add assessment". The source is shown before anything happens: CRV-201 §8.3, approved language with the weeks adapted. Week 2 is marked as a new visit.

The panel is on the list with its origin and its own draw. The screen also offers two other assessments to show that any request can come in this way.

5

Week 4 jumps to 70, and a suggestion appears

A new Week 2 visit scores 26. Week 4 now means 1 hour 55 minutes, 24 mL and three sticks, and its bar turns red at 70. The patient burden scorer proposes the smallest change that brings it back: take the panel from the pre-dose PK draw, the same needle, with two more tubes for the central lab. Weeks 2 and 8 stay as their own draw.

It names who decides: Lena or Omar Haddad, the clinical operations lead. A suggestion is never applied by itself, and the scorer may never drop an assessment an endpoint or a safety rule depends on.

6

Back to 57, with the protocol text following

Lena clicks "Combine with the PK draw". One needle stick and 15 minutes fewer: 0.25 × 100 + 0.6 × 24 + 9 × 2 = 57. Sections 1.3 and 8.3 are rewritten, and the draft is saved as version 0.4.

Across the trial, blood is now 146 mL, up from 122. The most in any 8 weeks is 68 mL, up from 52, against the 550 mL the sponsor chose as a reference point (Part A explains what that figure means).

Select any screen to open it larger. The sample's dose and dosing weeks are left off the screens. A first draft in two minutes is what this sample shows on screen; the solution's four-day figure is the design target for a checked draft, with questions answered and every check decided.

Reused protocol language with its source, consistency checks and one question per decision

In shortBehind the schedule sits the document workspace. Every passage shows whether it was reused, adapted or written new, and opens at its source page. Three findings remain after Week 16, each with both passages. Four decisions the synopsis left open go to named people. The writer approves each section.

1

The synopsis, read into 31 fields

The synopsis reader takes design, population, objectives, endpoints, dosing, visits and sample size as 31 fields, each with its page, and highlights the passages it used. It quotes endpoint wording and never infers a timepoint the synopsis does not state.

The closest approved protocols are ranked beside it: VLT-201 at 92% match, VLT-115 at 71% and CRV-201 at 54%.

2

Section 8, with each passage marked

Two passages are reused word for word from VLT-201, Amendment 1, version 3.0, pages 41 and 42. One is adapted from CRV-201, page 49, with its changed words marked. One is new, from the synopsis. Each opens at its page.

Section 8.3 now ends: "At Week 4 it is taken from the same venipuncture as the pre-dose PK sample." These subsection numbers are this sponsor's. In the M11 template, efficacy is 8.3 and safety is 8.4.4 The liver rule on screen is sample text.

3

Every changed word is marked

In Section 5, ten passages are reused and two adapted. The rule for moderate-to-severe disease changed from IGA to sPGA, two doctors' overall ratings of the skin. The source panel says what changed and why: "IGA ≥3 became sPGA ≥3 to match the synopsis". A clinical lead who wants to know what moved since the last study reads only the marks.

4

Mismatches found, both passages shown

The Week 16 finding shows as fixed, with the name and the version. Three remain. The contraception period is 20 weeks in Section 5.1 and 16 in Section 9.4, reused from VLT-201. Section 8.1 allows ±3 days through Week 16 and the schedule footnote allows ±7. And sPGA appears in Section 1.1 before it is spelled out.

Each comes with one suggested fix, to accept or to dismiss with a reason. The checker never applies a fix itself.

5

One question per decision, to the right person

What the synopsis does not settle becomes a question for a named person. Dr. Maya Chen, on whether prior biologic use should be capped. Sam Patel, on the testing order for the key secondary endpoints. Dr. Priya Ranvelle, on whether a data monitoring committee is needed. Dr. Chen again, on whether the Week 16 switch is blinded.

Each answer is written back into its section with name and time. Until then, the draft keeps the safe default and says so: "The draft keeps the cap of two."

6

The writer approves each section, and the head of writing signs

Lena approves section by section, and each section then goes to its reviewer: Sam for objectives and estimands, Maya for population and intervention, Omar for schedule and procedures, Priya for adverse events. When all 14 are approved, a full recheck runs, the team reviews, and the draft goes to Dana Okafor, head of medical writing, for an electronic signature with a reason. The version is then locked.

On the screen, a sister study, VLT-212, has reached that point: 14 of 14 approved, no open checks, ready to send. Every version is kept and nothing is overwritten.

The protocol list, the library of final protocols, the dashboard and your burden settings

In shortAround one protocol sit the screens a medical writing team uses all year: every open protocol with its stage, a library that is the only source of reusable language, a dashboard of the work, and settings that hold your template, reviewers, reuse rules and checks.

Six protocol writing agents and the medical writer who approves every section

In shortSix specialist agents read, reuse, write, check and score. None of them can approve a section or apply a suggestion. The writer approves every section, named reviewers follow, and the head of medical writing signs the final version with a reason.

What each protocol agent does, and where it stops

Who approves a section, who decides a burden suggestion and who signs a protocol

The people in the VLT-302 sample, in their own words

From one synopsis and your final protocols to a checked draft and a scored schedule

Records and systems a protocol authoring rollout connects to

Audit trail, signatures and credentials a quality team will ask about

Each fix accepted, check dismissed with its reason, question answered and section approved is recorded with name, time and version. Drafts stay in your environment, and models do not train on them. The final signature shows signer, time and meaning, which is what 21 CFR 11.50 looks for in a signed record.42 These are controls that support your compliance with Part 11, and validating them for your use stays with you.

Botminds' information security is managed to ISO 27001 and audited under SOC 2 Type II. No product can be certified for good clinical practice or Part 11, and we make no such claim. The solution runs at enterprise scale across many protocols, with roles, maker-checker approvals and an exportable audit trail.

What protocol amendments, heavy visits and slow drafting cost a sponsor

In shortStart from your own costs. Published figures frame the scale: what one amendment costs to implement, how long it takes, and what a delay day costs in a running trial. No study publishes writer hours per protocol or the share of amendments traced to an internal mismatch. You can measure both in a week from your own records.

Seven protocol costs to find in a week, starting with your last 20 amendments

Wary sites, tired writers and a third needle: what the totals leave out

What faster protocol drafts return: writer days released, plus a share of avoidable amendments

In shortPut in your own numbers. The calculator values writer days released and, separately, what stopping a small share of avoidable amendments is worth. Defaults use published figures where they exist and illustrative ones elsewhere. The hourly cost is an illustrative loaded cost, meaning pay plus overhead: replace it with your own.

Copying, cross-checking, amendments, burden: where protocol automation saves

Protocols a year, writer days and avoidable amendments on your own figures

How ready your medical writing team is: template, final protocols, schedule and sign-off

In shortPick the answer closest to your team today. Eight questions score 0 to 2, and 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.

Twelve questions for any protocol authoring software vendor, us included

In shortAlmost every tool demonstrates well on a clean sample. These questions test what a demonstration hides: whether it holds your template, whether a sentence keeps its source, whether the checks catch planted errors, and who can approve what. Ask us too.

Testing protocol software in ten weeks: past protocols, planted errors, one live synopsis

In shortStart with protocols you already signed, because you know how they turned out. Plant errors to measure the checks, score last year's schedules for burden, then draft one live synopsis in parallel with your usual process. Decide on three numbers: writer days, planted errors caught and reuse that survives review.

Writer days, planted errors caught and reuse that holds up in review

Protocol authoring questions answered: ICH M11, amendments, burden, blood limits and AI

In shortThe questions medical writers and buyers search for most, each answered on its own.