A toxicology study run by a contract lab ends in a draft report several hundred pages long. Before the lab's study director signs it, a toxicologist at the drug company that paid for the study reads it, page against page. That reader checks that the report says the same thing everywhere and matches the plan both sides agreed.
This guide is for sponsor toxicologists and study monitors, toxicology leads, toxicologic pathologists, nonclinical writers, quality assurance, and anyone deciding whether software should help them read contract lab drafts. No background needed. It stays with documents, discrepancies and the reviewer's decision: how to run a study, how to make a test article and what any result means for a dose are all outside it.
Imagine you are the study monitor for Northwind Bio, a small biotech: the toxicologist who reads every report its contract labs send back. On Friday, October 2, 2026, a contract lab, Quillhaven Preclinical, sends Draft v1 of a 13-week rat study on your company's fictional compound, Dravenzo. It runs to 412 pages. Your comments are due on October 16, and two other drafts are open on your desk.
Page 6 is the summary, the page most readers will ever open. It says a finding in one organ was seen in one male out of ten in the top dose group. You turn to the pathology table on page 231.
Summary · page 6
Top dose, males
1 of 10
Animals named
none
Table 22 · page 231
Top dose, males
2 of 10
Animals named
4107, 4112
1 of 10 in the text2 of 10 in the tableSame study, same finding, 225 pages apart.
Appendix 9, page 356, lists both animals, which backs the table. Why the summary says one is for the lab to explain. You write one numbered comment naming both pages, and the lab's study director decides how to correct it before signing.
A careful review catches the sentence and the table that disagree while the fix is still a redraft, before it has to become a signed amendment.
Reading a contract lab's draft: the protocol, the tables, the program, QA and the signatures
What the sponsor's reader checks, who answers for what, the rules with their dates, what goes wrong and how teams do it today, explained without selling anything. If you have never opened a GLP report, start here.
01 · In plain words
A second reader between the lab's draft and the lab's signature
Regulators ask drug makers to test a new compound in laboratory studies before it goes further. Many companies do not own the animal facilities, so they hire a contract lab, usually called a contract research organization or CRO. The drug maker is the sponsor. It agrees a written plan with the lab, called the protocol, and the lab does the work under a set of rules called Good Laboratory Practice, or GLP. One scientist at the lab, the study director, is in charge of the study from start to end.
When the work ends, the lab writes it up. A GLP report is long because it holds everything: a short summary, the methods, dozens of tables, a report from the pathologist who examined the tissues, a report from the scientist who measured the compound in blood, listings for every animal, the protocol and its changes. The fictional report in this guide runs to 412 pages. The summary on page 6 is the part most people read, and the tables behind it are the part that holds the record.
Before the study director signs, the lab usually sends the sponsor a draft. A toxicologist at the sponsor, often called the study monitor, reads it within an agreed window. The reader looks for places where the report disagrees with itself, with its protocol or with the sponsor's other studies, and sends the lab a numbered list of comments. The study director answers each one and decides what to change.
Most of the reading is plain comparison: this sentence against that table, this table against that listing. Finding a disagreement takes hours. Deciding whether it matters takes a scientist's judgment and a few minutes. A report also keeps being used for years, copied into later summaries and into the application a regulator reads, so a wrong number on page 6 tends to travel. That is why careful sponsors do this reading on every draft, and why they would like to spend less of it on the finding.
One draft's route from the lab's desk to the sponsor's file
The lab finishes its draftQuillhaven Preclinical assembles 412 pages from its contributors' reports and sends Draft v1 on October 2.
The sponsor's toxicologist reads itReport against protocol, tables against text, this study against the others, QA and compliance statements.
The comments go backOne numbered letter to the study director, with a location and a request for each comment.
The lab answers and revisesThe study director decides each comment, QA audits the report, and the next draft comes back to be compared.
The study director signsAfter that, a correction needs a signed amendment that says what changed and why.
Step 2 is the subject of this guide. The company, lab and study in this example are fictional.
Why a sponsor reads a report it did not write, and three more surprises
The sponsor signs nothing in the lab's report
The study director signs the final report, and the lab's quality assurance (QA) unit signs its own statement.1 The sponsor's toxicologist signs an internal review. The sponsor still answers to regulators for the package of reports it files.2
Every comment leaves a trail at the lab
OECD recommends the test facility keep the correspondence or the draft versions, to show how far the sponsor commented.3 Write each comment as if an inspector will read it.
A clean audit does not prove a clean summary
QA's review asks whether the report reflects the raw data.4 First on the list of top findings in studies that FDA's veterinary center presented in 2023 is a final report that "does not fully and accurately reflect the raw data".5 Reports with a QA statement still carry errors.
Nobody publishes how long this takes
We found no regulator, journal or trade body that measures the hours per sponsor review, the errors per draft or the comments acted on. Your own count is the only baseline you can defend.
Study monitor, study director, test article, NOAEL
Study monitor
The sponsor's toxicologist who follows a study at the lab and reads its drafts. The GLP rules do not define the title.
Study director
The lab's scientist with overall responsibility for the study, its report and its conclusions: "the single point of study control".6
Test article
The compound the study is about. In this guide's example it is fictional.
NOAEL
No-observed-adverse-effect level: the study's stated conclusion about which dose group showed no harmful effect. To a reviewer it is a line that must read the same wherever it appears.
What the sponsor's toxicologist decides, and what stays with the study director
In shortThe lab owns its report and its conclusions. QA owns its statement. The sponsor owns its comments, its own review record and the package it files with a regulator. A reviewer who keeps those lines clear writes better comments and has an easier inspection.
The US rule is plain. The study director "has overall responsibility for the technical conduct of the study, as well as for the interpretation, analysis, documentation and reporting of results".6 The international rule says the same: OECD's sponsor document says that even if a sponsor hires subcontractors itself, "the responsibility for the whole study remains with the Study Director, including the validity of the raw data and the report".2
OECD's 2020 paper adds that the study director has "the ultimate responsibility for the scientific validity of a study", and it strongly recommends that the study director and the sponsor's representative are different people.3
The sponsor has real duties too. Under 21 CFR 58.10 a sponsor must tell a contractor that the service is part of a GLP study.7 When several studies go to a regulator as one package, "the responsibility for the integrity of the assembled package of unaltered final reports lies with the sponsor".2 An Investigational New Drug application (IND) must carry, for each study under Part 58, a statement that it complied or a brief statement of the reason it did not.8
A sponsor that files a report containing a wrong summary sentence has filed a wrong summary sentence, whoever typed it.
Four parties touch a draft, and only one signs for its conclusions
Our reading of the US rule and the OECD documents
No roleReads or asksContributesAnswers for it
Study directorLab QA unitStudy pathologistSponsor's toxicologistConclusions of the studyAnswers for itNo roleContributesReads or asksThe pathology reportReads or asksNo roleAnswers for itReads or asksThe QA statementReads or asksAnswers for itNo roleNo roleProtocol changes recordedAnswers for itReads or asksNo roleContributesComments on the draftReads or asksNo roleNo roleAnswers for itThe package filed with a regulatorNo roleNo roleNo roleAnswers for it
Three things a sponsor's reader does with every line of a draft
Most lines
Confirm
The sentence matches its table and its protocol. No comment needed. Most statements in a draft end here.
Some lines
Ask
The sentence and its source disagree, or a required item is missing. The reader writes one comment with both locations, and asks the study director to correct or explain.
A few lines
Take it further
The sponsor's scientist reads a conclusion differently. The reader records the sponsor's view, asks for the study director's reasoning, and raises it with toxicology leadership. The study director still writes the conclusion.
Two different signatures. In this guide "the toxicologist signs" always means the sponsor's internal review. The signature that makes a GLP report final is the study director's, required by 21 CFR 58.185(b).1
03 · The rules and their dates
GLP rules a draft is read against: 21 CFR Part 58, OECD and FDA guidance, as of November 2026
In shortFDA's report rule, 21 CFR 58.185, was last amended in 1987. FDA's 2016 rewrite is still a proposal with no final rule found. The newer thinking sits in OECD papers from 2020 to 2022 and in FDA's 2021 peer review guidance. All of it describes the lab's duties first; the sponsor's reader uses them as a checklist.
The report rule is decades old, and the guidance around it is recent
198721 CFR 58.185 last amended
The list of what a final report holds.
1998OECD No. 11 on sponsors
Responsibility stays with the study director.
Aug 2016FDA proposes a GLP rewrite
No final rule found on November 3, 2026.
Apr 2020OECD No. 21 on sponsor influence
Draft review named as a scenario; keep the correspondence.
Dec 2021FDA peer review guidance
Final. Peer-review statement in the final report.
2022OECD No. 23 on QA
What the QA statement should do.
Oct 2026Study Data guide v6.3
SEND goes with a draft report.
Feb 2027FDA rule on nonclinical testing terms
A direct final rule, effective February 4, 2027 unless withdrawn. No Part 58 change found.
The 14 items a final report must hold, the study director's signature, and corrections only by amendment.1
In force. Read in the CFR 2025 edition (revised as of April 1, 2025), the newest on govinfo on November 3, 2026.
21 CFR 58.35(b)(6) and (7)
QA reviews the report against methods, standard operating procedures (SOPs) and raw data, and signs a statement with inspection dates.4
In force.
21 CFR 58.120 and 58.130
The protocol's contents, signed and dated changes with reasons, and a study conducted according to the protocol.16
In force.
FDA GLP proposal, 81 FR 58341
A quality system rewrite. It keeps individual scientists' reports, restating FDA's 1987 view that combined reports "would obscure the individual scientist's accountability".12
Proposal. No final rule found. FDA's April 30, 2026 notice renews the current requirements.13
OECD No. 21
Sponsor review of a draft as a named scenario; correspondence kept; QA audit after sponsor comments; review deadlines in contracts.3
Current OECD position paper, addressed to test facilities and useful to sponsors.
OECD No. 23
What a QA statement should say and confirm, and that amendments are inspected too.10
Current OECD advisory document.
FDA peer review Q&A
Pathology peer review, a second pathologist checking the study pathologist's diagnoses: named in the protocol, a signed statement in the final report, freedom from undue influence.9
Final, December 2021. Nonbinding.
EU Directives 2004/9/EC and 2004/10/EC
The European GLP framework. EMA coordinates GLP inspections; the competent authority responsible for the facility carries them out.17
In force.
04 · What a draft holds
Inside a 400-page toxicology draft: where each of the 14 required items sits and how to check it
In short21 CFR 58.185 lists 14 items. A reader can check some by looking for them, some against the protocol, and some only by comparing pages. The comparisons take the time, and they are where the sample draft's errors sit.
The list is short enough to keep beside you. A report must name the facility and the dates, give the objectives and procedures from the approved protocol "including any changes", the statistical methods, the test article, its stability, the methods, the test system, the dosage and duration, "all circumstances that may have affected the quality or integrity of the data", the people, the calculations and conclusions, the signed reports of each scientist, where everything is stored, and QA's signed statement.1
How you check it
Items in 58.185(a)
What a miss looks like
Is it there?
(1) facility and dates · (10) names · (12) signed scientists' reports · (13) storage locations · (14) QA statement
A contributor's report left unsigned. EPA's GLP program lists a lack of other scientists' names among its 2025 findings.18
Does it match the protocol?
(2) objectives, procedures and changes · (3) statistics · (4) test article · (5) stability · (6) methods · (7) test system · (8) dosage and duration
An amendment changed a schedule and the methods section kept the old one.
Does it agree with itself?
(9) circumstances affecting the data · (11) transformations, summary, analysis and conclusions
A table shows it; the summary or the deviation list leaves it out.
Where discrepancies sit in the sample's drafts, lab by lab
In this guide's fictional program, the sponsor's team counted where 42 discrepancies sat across nine reports from three fictional labs. These are invented numbers, shown to make the pattern visible: for two labs the summary carries the most, and for the third the toxicokinetic section does. Count your own for a quarter and you will learn where to read first for each lab you use.
Summaries carry the most errors at two sample labs; the third slips in toxicokinetics
Share of each lab's discrepancies, by report section
SummaryClinical pathologyPathology tablesToxicokineticsMethods and protocol
Quillhaven38%14%21%16%11%
Tarnwick36%16%21%12%15%
Hollowmere26%12%16%34%12%
Fictional example from this guide
Show the numbers
Summary
Clinical pathology
Pathology tables
Toxicokinetics
Methods and protocol
Quillhaven
38%
14%
21%
16%
11%
Tarnwick
36%
16%
21%
12%
15%
Hollowmere
26%
12%
16%
34%
12%
The pages that are separately signed
A GLP report contains reports. The pathologist's report and the toxicokinetic or bioanalytical report (blood levels of the compound) are usually written and signed by their own scientists, and 58.185(a)(12) requires those signed reports in the final report.1 In its 2016 proposal FDA kept those separate reports, restating its 1987 view that reports combining several scientists' work "would obscure the individual scientist's accountability for accurate reporting".12
So a summary written by the study director can disagree with a signed contributor report in the same binding. The reader's working assumption: when a summary and a signed contributor report disagree, ask which is right. Do not assume.
05 · Report against protocol
Checking a report against its protocol, its amendments and its deviations
In shortProtocol changes need a signed, dated amendment with a reason. Anything that departed from the protocol belongs in the report. A reader lines up the protocol, each amendment and the methods section, and looks for an out-of-range result with no matching deviation.
The study has to be run according to its protocol, and all changes "and the reasons therefore" must be documented, signed by the study director, dated and kept with the protocol.16 The report must then give the procedures in the approved protocol "including any changes in the original protocol".1 So a reader holds three things side by side: the protocol, the latest amendment, and what the methods section says happened.
FDA's FY2024 list of study director themes reads like a reviewer's checklist: protocol deviations left out of the final report, unforeseen circumstances not noted, changes to the protocol not properly documented, and a final report that "did not include a description of all circumstances that may have affected the quality or integrity of the data".19
A result outside its range should leave the same mark in four places
How one out-of-range result should appear through a report, on our reading of the rules
1Protocol or amendment sets the range
The acceptance criterion and what happens outside it.
Sponsor and study directorBefore the phase
2The appendix shows the result
The value in its table, on its page.
Lab scientistDuring the study
3A deviation is recorded
With the study director's assessment of its effect.
Study directorWhen it occurs
4The report's statement agrees
Deviation list, compliance statement and summary all say so.
The acceptance criterion and what happens outside it.
Sponsor and study director
Before the phase
The appendix shows the result
The value in its table, on its page.
Lab scientist
During the study
A deviation is recorded
With the study director's assessment of its effect.
Study director
When it occurs
The report's statement agrees
Deviation list, compliance statement and summary all say so.
Study director
In the draft
The sample's unrecorded result
In the fictional draft, protocol amendment 2, section 8.3, on page 402, says results outside 90 to 110% of nominal "will be reported as protocol deviations and their impact assessed by the Study Director". Appendix 3 on page 41 shows one result at 87.0% for Week 6, and then says no deviations were recorded. Each page reads fine alone, and only the two side by side show the gap. The comment asks for the deviation and the study director's assessment of what it means. Writing that assessment is the study director's job.
What to look for in a protocol check
Amendments counted
Every amendment the sponsor approved is listed, in order, and the last one's changes reached the methods.
Schedules
The weeks and days named in the summary, the methods and the protocol are the same. A second fictional report, from another lab, gives ECG weeks two ways.
Group sizes
The summary and the group allocation table give the same numbers. That second report says 2 per sex in one and 3 per sex in the other.
Deviations listed
Every out-of-range value and missed step in the appendices appears in the deviation list, with an assessment.
06 · Tables against text
When a table disagrees with its text: six discrepancy types in one sample program
In shortMost discrepancies are small and exact. A count differs by one. A sentence says no change and a table marks one. A header names the wrong unit. Each takes seconds to confirm once someone has put the two pages together, and hours to find without help.
In the sample, the largest group is a summary sentence against a table. A summary can be written once while its tables keep changing, and nothing forces anyone to re-read the sentence. Journals show the same pattern at a larger scale: Pitkin and colleagues checked abstracts in six medical journals in the 1990s against the same article's body, and found 18% to 68% of abstracts, depending on the journal, held data that were inconsistent with the body or absent from it.20
Those are journal articles, and no comparable count exists for GLP reports. The mechanism is the reason to look.
Summary against table is the sample's largest category, then incidence counts
42 discrepancies across nine reports in the guide's fictional program
Summary against table16
Incidence counts9
Toxicokinetic values6
Unreported deviations5
Target organs across studies4
Labels and terminology2
Fictional example from this guide
Show the numbers
Item
Value
Summary against table
16
Incidence counts
9
Toxicokinetic values
6
Unreported deviations
5
Target organs across studies
4
Labels and terminology
2
Type
Fictional example, in the draft's words
Why a reviewer cares
Sentence against table
Summary, page 6: "no test article-related changes" in clinical chemistry at the mid dose. Table 14, page 188: one value marked as statistically significant at that dose.
The summary is what later documents copy. The comment asks the lab to revise it or to justify it.
Count against table
Summary: 1 of 10. Table 22: 2 of 10. Appendix 9 names both animals.
Three places hold the number, and the listing settles which is right.
Value from the wrong row
Summary names Day 91 and gives the Day 1 number. See the next section.
The sentence looks right, because the number exists in the report.
Appendix against protocol
87.0% with a 90 to 110% criterion and "no deviations were recorded".
58.185(a)(9) asks for all circumstances that may have affected the data.1
List against table
"Target organs were the liver and the testis." Table 22 and an earlier study also point to the thyroid.
Target organ lists get carried into cross-study tables.
Label against values
Table 18's header says "% body weight"; the footnote and values are in grams.
Minor and easy to fix, yet it can mislead whoever copies the column.
Why a second NOAEL in one report needs a comment of its own
The sample's second report names the mid dose group as its NOAEL on page 5 and the high dose group in the discussion on page 29. It takes one line to show the disagreement. The sponsor's comment asks the study director to align the two and give the rationale.
The Society of Toxicologic Pathology's working group recommends that adverse findings be defined with a justification, and that study NOAELs "be established at the level of the overall study report", combining the expertise of all contributing disciplines.21 A reader who finds two has found a report that has not finished deciding, and that decision belongs to the study director.
07 · Reading the right row
A toxicokinetic value in the summary: which day, which sex, which dose group
In shortA toxicokinetic table has one value for each day, sex and group. A sentence names one of them. A reader checks that the number printed is the one in the row the sentence names. Quoting a neighboring row produces a real-looking number that survives a casual read.
Toxicokinetics, or TK, is the measurement of how much of a compound reached the bloodstream during the study. A TK table is usually a small grid: rows for the sampling day and sex, columns for the dose groups. Reports summarize it in a sentence that names a day, a sex and a group. AUC, the number most often quoted, stands for area under the curve, a measure of exposure over a set time.
In the fictional draft, the summary says the male AUC on Day 91 at the NOAEL was 48,200 ng·h/mL. The TK report's Table 3 on page 389 prints 42,800 in that row, and 48,200 one group of rows up, against Day 1.
The summary's number exists in the table, one row from where the sentence points
AUC at the stated NOAEL group, TK Table 3, page 389 of the fictional draft, ng·h/mL
Day 1, males (what the summary printed)48,200
Day 1, females66,900
Day 91, males (what the sentence names)42,800
Day 91, females61,500
Fictional example from this guide
Show the numbers
Item
Value
Day 1, males (what the summary printed)
48,200
Day 1, females
66,900
Day 91, males (what the sentence names)
42,800
Day 91, females
61,500
Four questions for any exposure value in a summary
Which day does the sentence name? Find that row. A number from another day still looks like a real result.
Which sex? Reports may give both sexes, or one. A sentence that gives one should say which.
Which group? The group named in the sentence and the column in the table must be the same.
Does anything else use it? Other sections, later summaries or the program's cross-study table may repeat the number. List where it went, so one correction reaches every copy.
08 · Across studies
Consistency across a program's studies: target organs, terminology and the 2.6.7 tables
In shortEach report is correct or incorrect on its own, but a sponsor reads a program. Two labs may name one finding two ways, or one report may leave out an organ that an earlier study listed. The reader asks about the gap. The study director may have a good reason.
A program's studies are usually run by several labs over several years. Sponsors keep a running table of them, which feeds the Investigator's Brochure and the toxicology tables of the common technical document, Module 2.6.7 under ICH M4S, the international guideline for those summaries.22 When a new draft arrives, the reader asks one extra question: does this report fit the others? The answer is often yes, and sometimes there is a reason it should differ. The question is cheap to ask and costly to skip, because the sponsor answers for the assembled package.2
The thyroid shows up in two rat studies, and the newer summary leaves it out
The sample's program as each report states it. Study T-013's text lists liver and testis; its Table 22 and study T-011 also point to the thyroid.
ClearAdaptiveAdverseSets the NOAEL
T-011 rat 4 wkT-012 dog 4 wkT-013 rat 13 wkT-014 dog 13 wkLiverAdaptiveAdverseSets the NOAELAdverseThyroidAdaptiveClearAdaptiveClearTestisClearClearSets the NOAELClearGI tractClearAdverseClearAdverseKidneyClearClearClearClear
Fictional example from this guide
Show the numbers
T-011 rat 4 wk
T-012 dog 4 wk
T-013 rat 13 wk
T-014 dog 13 wk
Liver
Adaptive
Adverse
Sets the NOAEL
Adverse
Thyroid
Adaptive
Clear
Adaptive
Clear
Testis
Clear
Clear
Sets the NOAEL
Clear
GI tract
Clear
Adverse
Clear
Adverse
Kidney
Clear
Clear
Clear
Clear
Three program checks worth making on every draft
Target organs
Same list, or a stated reason
An organ named by an earlier study and absent from this one's list gets one question. Species can differ for good reasons, and the study director may say so.
Terminology
One finding, one name
Pathologists at different labs may word a finding differently. SEND, FDA's standard format for nonclinical datasets, comes with controlled terminology that gives shared terms, and FDA's data guide asks that the report and the datasets be consistent.11
Datasets
The PDF and SEND tell one story
FDA's data guide asks that results in the report and in the SEND datasets be traceable and align. SEND goes in with a draft report, and an updated reviewer's guide is recommended with the final and with amended reports.11
Why draft values get a label in a program table
FDA's 1995 Phase 1 guidance lets an IND's integrated summary rest on "unaudited draft toxicologic reports", with final, fully quality-assured reports available within 120 days of the start of the human study and an update identifying any differences.23 So a value from a draft may already be in a filed document.
A program table should mark each value's source and status, so that a change in the next draft leads straight to every place the value went. A draft value with no such mark is the common way an old number survives into a filing.
09 · QA and signatures
Checking the QA statement, the compliance statement and the peer-review statement in a GLP draft
In shortThree statements and several signatures make a report final. A reader checks that each is present, specific and consistent with the rest. Timing matters: OECD recommends QA inspect the report after the sponsor's comments are addressed, so a statement on an early draft may be a placeholder.
The QA statement is required: QA must "prepare and sign a statement to be included with the final study report which shall specify the dates inspections were made and findings reported to management and to the study director".4 OECD's 2022 document asks for more: types of inspections, phases and dates, and it says the statement "would also serve to confirm that the final study report reflects the raw data". QA should make sure no later change requires another inspection, and an incomplete statement "could lead to the rejection of the study".10
FDA's veterinary center lists "the quality assurance unit (QAU) inspection of the FSR not present on the QA statement" among its top findings: an inspection of the report that the statement forgets to mention.5 A reader can catch that with the statement open beside the study calendar.
The sponsor's comments come before QA's last look and the study director's signature
Usual order for a contract study, from OECD's recommendations
1Contributors sign
Pathology and toxicokinetic reports, with any peer-review statement.
Lab scientistsBefore or during the draft
2Sponsor comments
Numbered comments on the draft, kept on file.
Sponsor's toxicologistComment window
3Study director answers
Each comment is accepted, or answered with a reason.
Study directorNext draft
4QA inspects the report
After sponsor comments are addressed, as OECD recommends.
Lab QA unitBefore signature
5QA signs its statement
Inspections, phases, dates; report reflects the raw data.
Lab QA unitAt finalization
6Study director signs
Final report signed and dated. Later changes are amendments.
If the protocol calls for peer review, is a statement signed, dated and in the final report? Were differences resolved by the protocol's process, and did any change reach the pathology report by amendment?
Corrections or additions to a final report must be made "in the form of an amendment by the study director", identifying what is added or corrected and why, signed and dated by the person responsible.1 OECD says amendments are inspected by QA, and a revised or additional QA statement follows.10
FDA's FY2024 themes include corrections "not made in the form of an amendment", and its veterinary center lists amendments that "do not clearly identify which part(s) of the FSR were changed and why".19,5 Any filed copy then needs its own update: the data guide recommends an updated reviewer's guide for amended reports.11 This is the practical case for catching things in the draft.
10 · Writing the comment
How a sponsor comments on a GLP draft without steering its conclusions
In shortOECD treats sponsor review as normal and warns about undue influence, intentional or not. A good comment points at two locations and asks a question. The study director decides and answers. Keep every comment, and agree the review deadline in the contract.
OECD's 2020 paper describes the tension without blame: the contract relationship "could potentially lead to a sponsor exerting pressure on the test facility", and it could also lead to "a desire by the CRO to meet the customer's needs".3
FDA's peer review guidance asks sponsors and those doing the work to "have processes in place to ensure that the studies are transparent and free from undue influence that could impact the conclusions of the studies".9 A sponsor can press for a good report and still keep clear of influence: comment on facts, in writing, and leave the conclusions to the person the rules name.
Instead of
Write
Why it reads better on file
Change the NOAEL to the mid dose.
The summary, page 5, names one dose group as the NOAEL; the discussion, page 29, names another. Please align them and give the rationale.
Points to the disagreement and leaves the decision where the rule puts it.
This finding isn't related to the compound.
The summary reports no related change at this dose; Table 14 marks one. Please revise the summary or give the reason it is unrelated.
Offers both outcomes. Either is the study director's to choose.
Remove the sentence.
Page 6 gives Day 91 and the value in TK Table 3 for that row is different. Please confirm which is correct.
A request to remove content is what an inspector looks for first.
A phone call about the pathology.
The question in the letter, with a note of any call kept in the file.
OECD expects the correspondence to allow "full verification of all study decisions and the input received from the sponsor".3
Comment windows and what happens when a sponsor is late
OECD also names the opposite risk: a sponsor who delays the draft review, which holds up reporting and archiving. It advises test facilities to set timelines in the study plan or contract, for example a maximum review time, after which "the final report will be issued by the study director with or without comments".3
FDA's 2016 proposal took up the same concern about sponsors delaying finalization.12 We found no published standard length for a comment window. Sample labs in this guide use 10 or 15 days, which are fictional settings. Check your own contracts, because your window may close on its own.
Keep the review and the lab's duties separate on paper
Direct talks between a sponsor and a principal investigator that leave out the study director "must be avoided", in OECD's words.3 A sponsor's own audit of the lab should remain independent of the test facility's systems and should not influence the outcome of the study.3 An internal review record that says what was read, what was asked and who decided what on the sponsor's side protects everyone. It also gives the sponsor's own QA something to check.
11 · What goes wrong
What FDA and EPA cite when a final report falls short, and how many GLP inspections end in a 483
In shortNearly half of FDA's FY2024 GLP inspections ended with a Form 483, the list of observations an investigator leaves at the end of an inspection. Recent warning letters to testing facilities center on final reports that left things out. Published numbers cover inspections of labs. None measures errors in reports, or how many sponsors catch them.
46%
of FDA's 37 GLP inspections in fiscal 2024 ended with a Form 483, 17 inspections in all.19
FDA, FY 2024 GLP 483 trends
18,075
GLP final reports a year in FDA's paperwork estimate, at 27.65 hours each. An estimate for burden purposes, carried over with no adjustments.13
FDA notice, April 30, 2026
20 of 34
animals in one study had abnormal cageside observations in the raw data that the final report did not fully include, in FDA's October 2026 letter to a testing facility.25
FDA warning letter, October 6, 2026
Voluntary action findings matched no-action findings in fiscal 2024
GLP inspections, final classified, all FDA centers. Remote assessments are not included.
No action indicatedVoluntary action indicatedOfficial action indicated
FDA's two fiscal 2024 documents count differently: the 483 trends slides give 37 GLP inspections,19 and the inspection metrics give 44 final classified inspections across centers.26 Each figure on this page is used with its own source.
Three recent warning letters about what a final report left out or got wrong
Letter
What FDA says
What a sponsor's reader could look for
Lovelace Biomedical Research Institute, October 6, 2026
Cited under 58.35(b)(6): the QAU's review of the final report. For 20 of 34 animals, and for all 48 in a second protocol, the final report did not include all the abnormal cageside observations in the raw data; recorded dosing issues were left out of both. The revised final reports "did not completely address the individual dosing issues".25
A summary with fewer observations than the individual listings. Revised drafts compared against every earlier comment.
Vedic Lifesciences, March 9, 2026
Final reports signed by people who were not responsible for the studies; the testing facility and storage location wrongly named.24
Facility, study director and storage locations against the contract and protocol.
SAMM Solutions dba BTS Research, October 20, 2023
QAU failed to assure that the final report reflected the raw data. Items cited include additional dosing not reflected in an appendix, wrong dosing dates, and death dates and circumstances that did not match the record.27
Dates and animal counts traced between the listings and the text.
These letters are FDA's account of the facility inspected. A sponsor reading a draft sees only the report, and most listings would need the raw data to prove a gap. The check catches what the report itself shows.
EPA and a CDER case study name the same report gaps
EPA's GLP program listed miscalculation and misreporting and "final report does not reflect raw data" among its 2025 findings, and told an industry audience: "EPA reviewers only see derived data in the final report".18 A CDER case study of translated reports lists wrong tables, omissions and compliance statements that differ from the original.28 If you use labs that report in another language, read the translation against the original on a sample of tables.
The market behind these reports, and how little is measured
Market research firms size preclinical contract research at about $6.8 billion for 2025 (IMARC, $6.82 billion; Towards Healthcare, $6.8 billion).29,30 IMARC gives toxicology testing 51.6% of that and biopharmaceutical companies 81.0% of the demand.29 Both are summaries of paid reports whose methods are not public, so read them as rough scale. No public figure exists for hours per sponsor review, errors per draft, comments acted on or the cost of an amendment, which is why Part C asks you to measure.
Toxicology testing is about half of the preclinical contract research market, in IMARC's sizing
51.6%
of the preclinical contract research market goes to toxicology testing, in IMARC's 2025 summary
of the preclinical contract research market goes to toxicology testing, in IMARC's 2025 summary
51.6 of 100
12 · AI in report review
Software reading GLP drafts for a sponsor: the 2026 regulatory texts and the closest published test
In shortNo regulator has written a rule for software that helps a sponsor read a draft. Several texts touch it, and one published paper comes close. Extraction from reports has one strong published result, from one company. We found none on catching errors in a contract lab's draft. People stay accountable under every text.
Little of it binds, and the GMP and device texts stay out of a GLP review
Our reading of each text as of November 2026. Ask counsel about your own uses.
NoIndirectlyIn partYes
Binding todayReaches a sponsor's reading toolFDA AI draft, January 2025NoIndirectlyEMA reflection paper, 2024NoIn partFDA and EMA principles, January 2026NoIndirectlyOECD Nos. 17 and 22IndirectlyIndirectlyDraft EU GMP Annex 22NoNoFDA software assurance, 2026NoNoEU AI Act, as amended 2026YesIndirectly
Still a draft. It "does not address" AI used for operational efficiencies, such as internal workflows or drafting, that do not impact "the reliability of results from a nonclinical or clinical study".31 A tool whose output changes what goes into a submission deserves a closer look at that clause.
EMA reflection paper, 2024
Section 2.3.2 says SOPs should extend to AI in non-clinical development, and that OECD GLP Nos. 17 and 22 should be considered.32
FDA and EMA guiding principles, January 14, 2026
Shared principles of good AI practice across the medicine's life cycle, without binding detail.33
OECD No. 17 and No. 22
Written for test facilities' computerized systems and data integrity. No. 22 says audit trail review need not include every system activity and points to designed and validated system reports. They shape what you ask of a lab, and of your own procedures.37,14
Draft EU GMP Annex 22
A manufacturing text and still a draft: EudraLex Volume 4 listed no Annex 22 on November 3, 2026. Its restriction on generative AI and LLMs concerns critical GMP uses, so it does not reach a GLP report review.34
FDA Computer Software Assurance, February 2026
Scope is device production and quality system software. It does not cover this work.35
EU AI Act, as amended by Regulation (EU) 2026/1744
In force July 27, 2026. Annex III high-risk rules from December 2, 2027; Annex I product rules from August 2, 2028. Ask counsel whether any of your uses fall in scope.36
What the closest published paper measured
Doktorova and colleagues built a pipeline that turns toxicology study reports into structured data, tested on 200 reports. Sensitivity was typically above 95% and precision frequently above 97% for histopathology, organ weight and NOAEL-related endpoints, lower for route of administration and substance identifiers. Targeted human curation was part of the method, and all authors are current or former employees of the company whose reports were used.38 It measures reading a report's content, and that is the first half of review. It does not measure catching a report's errors.
We found no study of software checking a contract lab's draft against its own tables. Ask any vendor to show you measured catches and false alarms on your reports.
What AI changes and what stays with people
Good fit
Finding and comparing
Splitting a report, reading tables, keeping pages, comparing a sentence with its table, listing what repeats a value. A fault here shows up when you open the page.
With care
Drafting a comment
Useful as a first version. A person checks both locations and the wording, because the comment goes on file at the lab.
People only
Deciding and signing
Whether a difference matters, whether to send it, what the sponsor thinks of a conclusion. The study director still answers for the report, and your reviewer for the review.
13 · How teams do it today
A sponsor's review desk today: study monitors, PDFs, a spreadsheet and three drafts at once
In shortOne toxicologist reads the draft, helped by a pathologist and a writer, usually from a PDF and a personal spreadsheet. Most of the time goes on finding, and the hand-offs fall inside a fixed comment window. Teams work in house, through contract toxicologists, or with data tools that cover part of the job.
Who reads a contract lab's draft at a sponsor
SM
Study monitor
Reads the draft, writes the comments, signs the sponsor's review.
Reads the pathology report and forms the sponsor's view. The study's pathology report stays the lab pathologist's responsibility.9
NW
Nonclinical writer
Carries values into later summaries and needs to know which are still draft.
SQ
Sponsor's QA
Audits labs and checks that filed values trace to the signed reports.
Most hand-offs happen inside one comment window
A typical path through the sample's team. There is no published workflow study.
1Draft arrives
Portal or email, with protocol and amendments somewhere else.
Study monitorDay 0
2Read and compare
Summary, tables, listings, TK report, protocol.
Study monitorDays 1 to 3
3Pathology read
Disputed calls and terms.
Sponsor's pathologistDays 3 to 5
4Program check
Earlier studies and the running table.
Study monitor, writerDays 4 to 6
5Comment letter
Numbered comments agreed and sent.
Study monitor, leadBy the due date
6Next draft
Each comment checked against the revision.
Study monitorAfter the lab replies
Fictional example from this guide
Show the numbers
Step
What happens
Who
Time
Draft arrives
Portal or email, with protocol and amendments somewhere else.
Study monitor
Day 0
Read and compare
Summary, tables, listings, TK report, protocol.
Study monitor
Days 1 to 3
Pathology read
Disputed calls and terms.
Sponsor's pathologist
Days 3 to 5
Program check
Earlier studies and the running table.
Study monitor, writer
Days 4 to 6
Comment letter
Numbered comments agreed and sent.
Study monitor, lead
By the due date
Next draft
Each comment checked against the revision.
Study monitor
After the lab replies
Where a fictional draft's 22.4 hours go by hand
In this guide's fictional example, the sponsor's team puts a by-hand review of one 13-week draft at 22.4 hours. No published source gives such a figure. The split below is an estimate for this example only: 19 of the hours go on finding and comparing, and 3.4 on deciding, writing the letter and signing. Time your own.
Finding and comparing take 19 of a by-hand review's 22.4 hours
Hours for one 13-week draft in the fictional example; the split is an estimate
Finding sectionsSummary against tablesToxicokinetic valuesProtocol, deviations, other studiesDeciding, letter, signing
By hand47353.4
Fictional example from this guide
Show the numbers
Finding sections
Summary against tables
Toxicokinetic values
Protocol, deviations, other studies
Deciding, letter, signing
By hand
4
7
3
5
3.4
Three ways teams cover this work
Way
What it gives
Where it stops
In house, by hand
Your scientists read everything and know your program.
Time. Several drafts land at once, and the reading gets thinner as the window closes.
Contract toxicologists
Extra experienced readers when volume peaks, or for a kind of study you see rarely.
Another hand-off, and program knowledge has to be passed over each time.
Software
SEND visualization tools show the data; document tools extract from PDFs; checkers compare report sections.
A SEND view does not read the study director's sentences. Check what any tool compares, and how it keeps the page.
Part B · See it working
Toxicology Report Review in Botminds: a fictional sponsor's 13-week rat draft, read, compared and signed
Part A described the work without naming a product. This part shows how Botminds does it, at a fictional biotech reading drafts from three fictional labs. Every company, person, compound, number and document in the screens is sample data, and the screens here keep to documents and discrepancies.
14 · One draft, start to finish
Draft v1 of study CRV-T-013: 412 pages read, six discrepancies decided, one letter and a signed sponsor review
In shortThis is the example from the top of the page, inside the product. Six agents read the draft as it arrives. Dr. Priya Ranvelle, the study monitor, works through six discrepancies with both pages side by side, decides each one and signs the sponsor's review. The comment letter leaves only after her signature, and the lab's study director answers it.
1
Fourteen contract lab reports, sorted by what needs her
Dr. Ranvelle's list holds 14 reports from three labs across three programs, each with its stage, its discrepancies and its comment due date. Four are ready for her. Study CRV-T-013, a 13-week rat study from Quillhaven Preclinical, sits on top: Draft v1, 412 pages, five major and one minor discrepancy, comments due October 16.
Others wait at other stages: three being read, one with the lab, and six signed.
The reports list. Every draft, amendment and final report from every lab, with its stage, discrepancies and due date.
2
Six agents read the draft before she opens it
The report reader split the 412 pages into sections and tables and matched them to the protocol and its amendments in 41 seconds. The findings extractor pulled 214 findings, seven of them test article-related, with SEND terms mapped. The exposure reader kept the day, sex and dose group of each toxicokinetic value with its page. The consistency checker ran its seven checks and found six discrepancies. A cross-study check compared the draft with the seven other Dravenzo studies and flagged the thyroid.
The whole run took 3 minutes 14 seconds, and it ends on the words "Study monitor decides".
The agents at work. Each step with what it found and how long it took, ending with the person who decides.
3
The draft on the left, the discrepancies on the right
She opens the report. The summary sits on the left with its sentences highlighted, the section list shows where the issues are, and the panel on the right lists each discrepancy with its two sources. D1: the summary says no related changes at the mid dose, and Table 14 on page 188 marks one. D2: 1 of 10 in the summary, 2 of 10 in Table 22.
The source report stays unchanged, and the Original PDF button opens the whole file.
Report review. The report with highlights, its sections, and each discrepancy with both sources. In the sample, values are fictional.
4
Six comments go into one letter to the lab's study director
She adds all six to the letter, editing any wording she wants to change. A dismissal would need a reason on the record. Each comment names its location, says what disagrees and asks for a correction or an explanation. The sample's wording follows Part A: comment 1 asks the lab to revise the summary, or justify why the change is unrelated.
The letter goes to the study director at the lab, by role in these screens. The lab's reply is asked for within 10 working days, by October 21.
The comment letter. One numbered comment per accepted discrepancy. The study director decides what each one changes.
5
She signs the sponsor's review, and the letter leaves
The signing screen shows what she is signing: six comments to the lab, and the seven findings as extracted. She chooses the meaning, "Reviewed, comments sent to the CRO", and signs with her password as study monitor. Her signature, its meaning, the time and the report version go on the audit trail.
This signature records the sponsor's review. It is separate from the study director's signature on the final report and from QA's statement, both of which come later, at the lab.
Sign the sponsor review. The meaning of the signature, the signer and what is being signed.
6
The whole review on one trail
The activity trail lists every step with its time: the agents' reading on October 2, her decisions and signature on October 7, the letter delivered through the lab portal, the program table refreshed. The next draft will be checked against every comment, so each one closes on evidence.
That trail is also what you would show your own QA, or your counterpart at the lab if anyone asks how a comment came about.
The activity trail. Every agent step, decision, signature and letter delivery, with its time.
15 · Each check, both pages
Compare view for toxicology discrepancies: count, exposure row, deviation, target organ, second NOAEL
In shortEach discrepancy opens as two pages side by side, with the two values marked. The reader confirms in seconds what would have taken a page hunt. Nothing changes in the source report, and the checker proposes while the person decides.
A count against the pathology table
D2 opens the summary beside Table 22. The panel says 1 of 10 and 2 of 10, and the individual listing in Appendix 9 names both animals. The table is from the pathology report, marked as peer reviewed, so the reader asks the lab to correct the summary and its narrative.
The summary against Table 22. Both pages, both values highlighted, with the discrepancy named above them.
The value from the wrong row
D3 opens the summary beside TK Table 3. The sentence names Day 91 in males and prints 48,200 ng·h/mL. Table 3 prints 42,800 in that row and 48,200 in the Day 1 row. The comment asks the lab to correct the summary.
The summary against TK Table 3. The same number appears in the wrong row. Sample values.
A result with no deviation behind it
D4 opens Appendix 3 beside protocol amendment 2. Week 6 reads 87.0% for the top group; the amendment says anything outside 90 to 110% will be reported as a deviation, and the appendix says none was recorded. The comment asks for the deviation and the study director's assessment.
The appendix against the protocol amendment. Two documents written months apart, read together. Sample values.
An organ missing from the list
D5 reads "Target organs were the liver and the testis" beside Table 22, where the thyroid also shows a finding, and an earlier 4-week study lists it as a target organ. The sample's comment proposes adding it, with the adaptive interpretation if appropriate. That is for the study director to judge.
The summary list against the table and an earlier study. The cross-study check brings in the sponsor's other reports. Sample values.
Two NOAELs in one report
The second report, from Tarnwick Laboratories, names the mid dose group as its NOAEL in the summary on page 5 and the high dose group in the discussion on page 29. The checker flags it major and says only that two different values in one report would carry into later documents. It also finds ECG weeks given two ways and recovery group sizes that differ.
A summary against its own discussion. Three discrepancies in the second sample report.
A finding keeps its page
The findings extractor keeps the report's own term beside its SEND term, the count in each dose group, and a numbered link to the page. It records what the report states. The sponsor's pathologist reads the same page and forms the sponsor's view, and any disagreement goes to the study director as a question in the letter.
One finding with its counts. Sample values; the number links to the cited page.
16 · The program view
A program's toxicology studies on one page: target organs, a cited cross-study table, the lab scorecard and your rules
In shortThe same reading builds a program view. Organs run down the side, studies across, every cell the report's own statement, and draft values marked. Writers export the table for later documents, and QA checks each value against the signed report. What the reviewer checks and who signs are your settings.
The target-organ view flags what does not fit
Each cell shows what that report states for that organ. A red dot marks what the consistency checker flagged, and the thyroid and testis cells for the 13-week rat study carry one. The side panel lists the studies behind the view. The note at the bottom says draft values are labeled as draft, and the signed GLP report stays the official record.
The target-organ view. Every cell opens its source page. Sample program.
A cross-study table with every value cited
The table lists species, duration, route and target organs for each study, each value opening its source page. The two 13-week reports carry a "Draft v1, may change" label, so a writer sees which values may still move. It exports with source references for the Investigator's Brochure and Module 2.6.7, and QA checks each value against the signed report before anything is filed.
The cross-study table. Working copy. The signed reports stay the record.
What the sample dashboard counts
Review hours per draft, what the checker finds, where discrepancies sit by section and a scorecard for each lab. On this sample data the median review is 6.4 hours against a 22.4-hour by-hand baseline, 4.7 discrepancies are found per report, and 93% of comments led to a change in the next draft. All are sample values that illustrate design targets. Read the last one as a record of how a report changed: the study director decides each comment, so it is a poor score to push a lab on.
The dashboard. Sample data. Lab names are fictional.
On the sample data, finding and comparing shrink while deciding hardly moves
Hours for one 13-week draft: the sample's by-hand baseline against its design target. The split by step is an estimate for this example.
By hand, 22.4 hoursDesign target, 6.4 hours
Finding sections4 hours0.5 hours
Summary against tables7 hours1.5 hours
Toxicokinetic values3 hours0.5 hours
Protocol, deviations, other studies5 hours1 hour
Deciding, letter, signing3.4 hours2.9 hours
Fictional example from this guide
Show the numbers
Step
By hand, 22.4 hours
Design target, 6.4 hours
Finding sections
4 hours
0.5 hours
Summary against tables
7 hours
1.5 hours
Toxicokinetic values
3 hours
0.5 hours
Protocol, deviations, other studies
5 hours
1 hour
Deciding, letter, signing
3.4 hours
2.9 hours
Your checks, in your terms
Seven checks can each be turned on or off: summary against tables, incidence against listings, TK report against summary, report against protocol and amendments, dose formulation and deviations, target organs across the program, and terminology against SEND. You set how a mismatch is graded. Every change to a setting goes on the audit trail.
The checks the reviewer runs. Each check on or off, with its severity.
Who signs, and each lab's window
You name who signs each program's review, who forms the sponsor's view on adverse and adaptive calls, and who checks exports. Each lab has its own comment window, and the SEND terminology release and grading scale are settings too.
Sign-off and comment windows. Named people for each decision, and a window per lab.
17 · Who signs what
Six agents read and compare; the study monitor decides and signs the sponsor's review
In shortThe agents read, extract, compare and draft. Each has a line it does not cross. A named toxicologist decides every comment, and the lab's study director decides the report. Source reports are read only, and every step is on the audit trail.
Report reader to letter drafter: the six agents behind a draft review
Report reader
Splits each report into summary, methods, results, tables, listings, pathology, TK and appendices, keeps page numbers, matches it to its protocol, amendments and earlier versions, and flags missing GLP or QA statements.
Never alters the source PDF. Every section keeps its page range.
Findings extractor
Pulls dose groups, findings by organ with SEND terms, incidence by sex and group, severity, reversibility and the stated NOAEL, each with its page.
Never gives a value without a page citation. Marks adverse or adaptive only as the report states it.
Exposure reader
Reads the toxicokinetic tables and keeps the day, sex and dose group of each value with its page, so a summary sentence can be checked against the row it names.
Never quotes a value without its row.
Consistency checker
Compares summary with tables, tables with listings, the TK report with the summary, the report with its protocol and amendments, and the study with the rest of the program.
Never decides. It shows both sources and proposes. Severity rules are yours.
Cross-study table builder
Keeps the program table and target-organ view up to date from confirmed findings, with every cell cited.
Never hides a draft value's status. Draft values are labeled, and exports carry source references.
CRO letter drafter
Turns accepted discrepancies into a numbered comment table for the study director, with the location, what disagrees and the request.
Never sends before the study monitor signs. One request per comment.
Who decides what on a contract lab's draft
Fictional, from the sample data
PR
Dr. Priya Ranvelle · study monitor, toxicology
Accepts or dismisses each discrepancy, with a reason on the record for a dismissal, and signs the sponsor's review.
"I spend two days finding the problems in a draft and an hour deciding them."
OH
Dr. Omar Haddad · toxicology lead
Signs the reviews for his own program and takes disputed readings further.
SP
Dr. Sam Patel · sponsor's toxicologic pathologist
Forms the sponsor's view of adverse and adaptive calls, before the review is signed. The study's pathology report remains the lab pathologist's.9
LO
Lena Ortiz · nonclinical writer
Takes the cross-study table into later documents, with draft values marked.
DO
Dana Okafor · quality assurance
Checks exported values against the signed reports.
"If it isn't cited to a page of the signed report, it isn't in the submission."
Draft reports go in; a cited discrepancy list and a signed review come out
From the lab
GLP and non-GLP study reports: draft, amendment and final
Toxicokinetic reports
Pathology reports and individual animal listings
Protocols and amendments
From your team
Your review procedure and severity rules
Your earlier signed reports for the program
Who signs each program, and each lab's comment window
Agents read, extract and compareEvery page of every report, against itself, its protocol and your program
The toxicologist signs each review
Every value linked to its page
Outputs
Cited findings and stated NOAELs for each study
A discrepancy list with both sources
A numbered comment letter to the study director
A signed sponsor review, with its audit trail
A cross-study table for the Investigator's Brochure and Module 2.6.7
What changes for the team
Discrepancies to decide, with two pages open
The next draft checked against each comment
Draft values marked wherever they went
How contract lab reports come in and comment letters go out
Contract lab portalsReports in, numbered comment letters out after signature
SEND terminologyControlled terms, by release, for mapping
Roles, records and the controls behind every review
Reports stay read only
No agent changes a source report. The signed GLP report stays the official record.
Maker and checker
Agents propose, and named people decide. QA checks exports against signed pages.
Audit trail
Each agent step, decision, dismissal reason, signature and setting change, with who and when, and ready to export.
Credentials
Audited for SOC 2 Type II; information security managed to ISO 27001. The controls support your compliance with Part 1139 and GLP expectations. Your own procedures decide fitness for use.
Part C · Ready to automate?
Toxicology report review software for sponsors: pricing a late error, working out your return, testing on closed drafts
Written to you, the person deciding. It starts with what errors and delays cost, gives you a way to work out your own return, and ends with questions for any vendor, including us.
18 · What late errors cost
What a missed discrepancy costs a sponsor: reviewer hours, amendments and a filing that needs correcting
In shortNo one publishes the cost of a missed error in a GLP report. What you can price is your reviewers' hours, the amendments and re-filings a late fix sets off, and the delay if a key report slips. A week of counting gets you most of the baseline.
$105.60/hr
FDA's own rate for sponsor, QA and executive management time in its 2016 analysis of the GLP rule. It is a 2016 figure, so your current rate is probably higher.40
FDA regulatory impact analysis, 2016
27.65 hrs
per final report in FDA's paperwork estimate for 58.185, the facility's reporting work. It measures burden for the paperwork law and is unrelated to review time.13
FDA notice, April 30, 2026
$500K–$800K
lost sales per day of delay in Tufts' two published figures from one dataset, plus about $40,000 a day in direct Phase II and III trial cost. These are clinical trial figures, shown for scale only.41
Tufts CSDD, 2024
A late fix touches more people than the same fix made in a draft
Hours across lab and sponsor for one correction. Assumptions only: replace with your own.
Caught in the draftCaught after signature
Lab corrects the report1 hour6 hours
QA looks again0.5 hours4 hours
Sponsor's reviewer follows up0.5 hours3 hours
Filed documents updated0 hours8 hours
Illustrative · replace with your own numbers
Show the numbers
Caught in the draft
Caught after signature
Lab corrects the report
1 hour
6 hours
QA looks again
0.5 hours
4 hours
Sponsor's reviewer follows up
0.5 hours
3 hours
Filed documents updated
0 hours
8 hours
Finding your own draft review baseline in a week
Cost
Where it hides
How to find your number
Reviewer hours per draft
Inside toxicologists' weeks; rarely timed
Time three drafts of different lengths, split into finding, deciding and writing.
Comments per draft, and their kind
Comment letters in email or on the lab portal
Sort last year's letters by the six types in section 06, by lab.
Errors that survived to the final
Amendments and QC notes
Count amendments per final report last year, and read why each was issued.
Filed documents touched by a late change
Writers' change logs
Pick one late change, and list every summary and table it reached.
Missed comment windows
Lab reminders; reports issued with no sponsor comment
Ask your labs how many drafts passed the agreed time, and check your contracts for what happens then.3
FDA's estimate puts a final report near a protocol, and a QA statement at one hour
Average hours per record or report, for labs' paperwork. It says nothing about a sponsor's review.
Costs that sit outside the calculator: the late amendment, the thin read, the lost reviewer
The number that travelsA wrong summary value copied into a brochure, a Module 2 table and a slide lasts longer than the draft it came from.
The thin read at the deadlineThree drafts in one week means the third gets the least attention, whatever it holds.
An inspector's questionA comment that reads like an instruction takes time to explain years later.
What leaves with a reviewerWhich lab tends to slip where, and why the program's table reads as it does.
Your toxicologists' judgmentHours spent hunting pages are hours lost to the conclusions only they can read.
19 · Your return
What faster draft reviews return: reviewer hours released and fewer late corrections
In shortThe return comes mostly from reviewer time on finding, then from errors caught in the draft. Work it out with your own counts. Only the hourly rate below comes from a published source, and it is FDA's 2016 figure.
Reading, comparing, follow-up: where a sponsor's review saves
Reviewer time
Discrepancies to decide, with pages open
Sections, tables and values arrive found; the reviewer spends the hours deciding.
drafts × hours saved × hourly cost
Late corrections
Caught while it is still a redraft
A check on each draft should leave fewer errors to surface after signature.
late corrections × share caught earlier × hours per correction × hourly cost
Next drafts
Each comment closed on evidence
The next draft is checked against the letter, so follow-up reading shrinks.
revised drafts × minutes saved × hourly cost
Program tables
Values that know where they went
Writers find what changed without re-reading, because draft values carry their status and page.
documents updated × hours saved × hourly cost
Your hours back on drafts a year, at your hourly cost
Only the hourly cost comes from a published figure: FDA's 2016 rate for sponsor time.40 The by-hand hours and the target are the solution's sample baseline and design target. Every other input is an assumption of ours. Replace them all.
Your volumeHours per draftLate correctionsYour cost
Illustrative yearly value
Draft review time today, valued$0
Reviewer time released, valued$0
Late corrections avoided, valued$0
After your yearly cost$0
Reviewer hours released a year0
Drafts a year to cover your cost0
Formulas, per year: Time today = drafts × hours today × hourly cost Time released = drafts × (hours today − hours with automation) × hourly cost Late corrections avoided = corrections × share caught ÷ 100 × hours per correction × hourly cost After your cost = time released + corrections avoided − yearly cost Break-even drafts = yearly cost ÷ ((hours today − hours with automation) × hourly cost) The result leaves out what errors in filed documents cost you, and any delay. Released hours are value only when you decide what they are for: deeper reading, more programs per reviewer, or less overtime at the deadline.
The hours saved per draft decide the break-even volume
Drafts a year to cover $50,000 of yearly cost, at $105.60 an hour
Saves 16 hours a draft30 drafts
Saves 8 hours a draft59 drafts
Saves 4 hours a draft118 drafts
Illustrative · replace with your own numbers
Show the numbers
Hours saved
Drafts a year
Saves 16 hours a draft
30 drafts
Saves 8 hours a draft
59 drafts
Saves 4 hours a draft
118 drafts
A sponsor reviewing 40 long repeat-dose drafts a year clears the first line. A sponsor with six short in vitro reports a year, tests on cells in a dish, does not, and should weigh that before buying anything.
Start here20 · Readiness check
Scoring your sponsor review desk: drafts, labs, baseline, records and who signs
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.
1
Drafts a year
How many contract lab drafts, including amendments, does your team review in a year?
Why it mattersVolume turns hours per draft into capacity. Peaks matter as much as the total.
A strong answer"About 40, a third of them long repeat-dose studies, with three or four open at once."
2
Your labs
How many contract labs write your reports, and do you know how each one's reports are laid out?
Why it mattersEach lab has its own template. Several labs mean more comparison across studies and more value from one view.
A strong answer"Three labs, and we have two recent signed reports from each to test on."
3
Your baseline
Do you know how long a draft review takes, and how many comments one produces?
Why it mattersNo published benchmark exists, so a pilot can only be judged against your own numbers.
A strong answer"We timed three drafts by step and sorted last year's comments by type."
4
Protocols on hand
When a draft arrives, can you put your hands on its signed protocol and every amendment?
Why it mattersThe protocol check needs them, and a tool can only compare what it can read.
A strong answer"They sit in one document system, filed by study, with amendments in order."
5
Your program table
Is there one cross-study table per program, with a source and a status for each value?
Why it mattersThis is where draft values travel from. Without status marks, a change in a draft is hard to chase.
A strong answer"One table per program, each value cited to a report page and marked draft or final."
6
Comments on file
Do you have a written way of commenting on a draft, and could you show every comment on a study?
Why it mattersOECD expects correspondence to be retained at the facility.3 Your record should match the lab's.
A strong answer"An SOP says comments are written, numbered and factual, and we keep each letter by study."
7
Who decides
Is it clear who signs each sponsor review, who forms the sponsor's view on pathology, and who takes a dispute further?
Why it mattersSoftware routes decisions to named people, so the names must exist. The study director's role has to stay clear in all of this.
A strong answer"Named per program, with deputies, and our SOP says the study director decides the report."
8
Your quality unit
Has QA said what it needs before a tool reads reports and proposes comments?
Why it mattersYour own procedures decide how a tool is assessed for its use. Nobody can buy that as a certificate.
A strong answer"QA has seen the plan, wants a risk assessment and an audit trail export, and agrees the signed report stays the record."
9
Vision · for reflection
If a regulator asked you tomorrow where a number in your Module 2 came from and how the sponsor commented on its report, what would it take to answer in under an hour?
Why it mattersYour answer lists the gaps. Each gap is a candidate job for software, or for a procedure.
A strong answerA short, specific list, for example: "every value cited to a signed page, every comment on file, and one table per program."
A low score means filing protocols and timing three drafts first
0–6
Explore
Build the baseline first: time three drafts, sort last year's comments by type, file protocols by study and write down how you comment. Then start with one check, such as summary against tables.
7–11
Ready to pilot
Run a measured pilot on closed drafts from your own labs, with errors planted in copies. Compare catches, false alarms and reviewer minutes with your current process, and keep your people signing every review.
12–16
Ready to implement
Plan the full desk: labs, portals, programs, signers, severity rules, audit trail exports and how your QA samples the output. Measure the first ten live drafts as carefully as the pilot.
21 · Test it on closed drafts
A blind trial on your own closed drafts: planted errors, vendor questions and four numbers to decide on
In shortAsk each vendor the same questions, then test on drafts whose answers you already know. Plant errors in copies of them, on your own systems. Decide on catches, false alarms, reviewer minutes and whether your QA would accept the record.
Eight questions that test a draft-review tool on pages, protocols and the study director's role
A
Reading contract lab PDFs and keeping pages
A1
Can you show every value beside its page, on a report from a lab you have never seen?
Why it mattersLayouts differ by lab, and a citation you cannot open is worth little.
Strong answer · warning sign
A live run on your file, each value opening its page.
A demonstration on the vendor's own report only.
A2
What happens to a table your tool cannot read cleanly?
Why it mattersA missed row looks like agreement.
Strong answer · warning sign
It says so, and sends the table to a person with a confidence mark.
"It reads everything."
B
Catching disagreements in a GLP draft
B1
What are your catch rate and false alarm rate on planted errors, and on whose reports?
Why it mattersWe found no published figure for this task, so yours is the only evidence.
Strong answer · warning sign
Measured on your closed drafts, by error type.
One accuracy percentage with no method.
B2
Do you compare against the protocol, each amendment and earlier studies, or only within a report?
Why it mattersUnrecorded deviations and missing target organs live across documents.
Strong answer · warning sign
Shows both documents for each flag.
Compares summary and tables only.
B3
Does the next draft get checked against each earlier comment?
Why it mattersFDA's October 2026 letter says the revised final reports "did not completely address the individual dosing issues".25
Strong answer · warning sign
Each comment closes against a changed page, or stays open.
You re-run the whole check and compare by eye.
C
People, records and the study director's role
C1
Can anything reach a lab without a person's signature?
Why it mattersEvery comment becomes correspondence on the study file.3
Strong answer · warning sign
Nothing leaves until a named person signs, and this is on the audit trail.
Auto-send for minor comments.
C2
Does a proposed comment ask, or does it instruct on a conclusion?
Why it mattersWording that steers a conclusion is what OECD and FDA warn about.3,9
Strong answer · warning sign
Asks with two locations; your SOP's wording is a setting.
Drafts "change the NOAEL to".
C3
What exactly do you claim about GxP and Part 11?
Why it mattersA product cannot be certified for GxP. Fitness for use is your procedure's decision.
Strong answer · warning sign
Named controls that support your compliance, and the evidence for each.
"Fully compliant" or "certified" with no scope.
Twelve weeks on closed drafts, then a decision
Time three drafts, split by finding, deciding and writing. That is your baseline.
Pick 15 to 20 closed drafts from your own labs, with their signed finals, protocols and your original comments. Your earlier comments are your answer key.
Plant errors in copies: a changed count, a value from a neighboring row, a missing deviation, a dropped target organ. Keep the list away from the vendor.
Run the test set blind, then score catches, false alarms and what a reviewer needs to confirm each flag.
Have QA read the record, then decide on the four numbers below.
A twelve-week pilot on closed drafts ends with a decision on four numbers
Your teamVendor
Time three drafts; sort last year's comments
Choose closed drafts; plant errors in copies
Load procedures, severity rules and signers
Run the test set blind
Parallel run on one live draft a lab
Score catches, false alarms, minutes
QA reads the record; decide
Week 04812
Illustrative · replace with your own numbers
Show the numbers
Activity
From
To
Time three drafts; sort last year's comments
0
2
Choose closed drafts; plant errors in copies
1
4
Load procedures, severity rules and signers
2
5
Run the test set blind
4
6
Parallel run on one live draft a lab
5
10
Score catches, false alarms, minutes
9
11
QA reads the record; decide
10
12
Catches, false alarms, reviewer minutes and QA's verdict
Measure
How
Why
Catches, by type
Planted errors and your own past comments found, out of the total
Whether it finds what you find
False alarms per draft
Flags a reviewer dismissed, with the reason
Too many and people stop reading them
Reviewer minutes per draft
Against your timed baseline, on drafts of similar length
The value lever
QA's verdict on the record
Would QA accept the trail and the citations in an audit?
The condition for going live
Working-session questions for toxicology, pathology, writing and QA
Bring the people who read, write from and audit these reports into one room.
D
Your own review habits
D1
Which comment types do your labs accept without debate, and which lead to a discussion?
Why it mattersFactual comments are the first to automate.
Strong answer · warning sign
Sorted last year's letters by kind.
"It depends on the lab."
D2
What does your reviewer do when a comment window closes on an unread section?
Why it mattersOECD notes reports may issue with or without comments.3
Strong answer · warning sign
Written rule for escalating or extending.
It has never been discussed.
D3
Who updates filed documents when a final differs from the draft they used?
Why it mattersFDA's 1995 guidance asks for an update that identifies differences.23
Strong answer · warning sign
A named owner and a list of each value's copies.
"The writer will notice."
22 · Frequently asked questions
Toxicology report review FAQ: comment windows, QA statements, amendments, NOAELs, peer review, SEND and AI
In shortThe questions toxicologists and buyers search for most, each answered on its own.
What is toxicology report review?
Toxicology report review is a sponsor scientist reading a contract lab's study report, usually a GLP draft, before the lab's study director signs. The reviewer checks the report against its protocol, its tables against its text, the study against the program's other studies, and the QA and compliance statements. The result is a list of numbered comments to the lab and an internal review record.
Can a sponsor change a GLP study report?
A sponsor can comment on a draft, and OECD names this as a normal scenario, but the study director decides what the report says. OECD recommends the test facility keep the correspondence or draft versions, and that QA audit the report after sponsor comments are integrated. Once signed, changes need an amendment from the study director.
What must a GLP final report contain?
21 CFR 58.185 lists 14 items, including the facility and dates, objectives and procedures with protocol changes, methods, dosage, all circumstances that may have affected the data, people involved, conclusions, the signed reports of individual scientists, storage locations and the QA statement. The study director signs and dates the report. This text has been unchanged since 1987.
What is the QA statement in a GLP report?
The QA statement is a signed statement from the lab's quality assurance unit, required by 21 CFR 58.35(b)(7), that specifies the dates of inspections and the dates findings were reported to management and the study director. OECD's 2022 guidance adds types of inspections and phases, and says the statement would also serve to confirm that the report reflects the raw data. An incomplete statement could lead to rejection of the study.
What happens if an error is found after a GLP report is signed?
Corrections must be made as an amendment by the study director, identifying the part corrected and the reason, signed and dated by the person responsible. OECD says QA inspects amendments, and a revised or additional QA statement follows. Any document that used the old value then needs checking too, and FDA's data guide recommends an updated reviewer's guide for amended reports.
Who sets the NOAEL in a contract lab's toxicology report?
NOAEL stands for no-observed-adverse-effect level. It is a conclusion stated in the study report, and it belongs to the study director, drawing on the contributing scientists. The Society of Toxicologic Pathology's working group recommends establishing it at the level of the overall report and justifying the decisions behind it. A sponsor's reviewer checks that it reads the same throughout and asks about any disagreement.
How long does a sponsor get to comment on a GLP draft?
No rule sets a period, and we found no published standard. OECD advises test facilities to define timelines in the study plan or contract, and gives the example of a maximum review time after which the report is issued with or without the sponsor's comments. Check your own contracts, because your window may close on its own.
What is pathology peer review, and does it go in the report?
Pathology peer review is a second pathologist's review of the study pathologist's diagnoses and interpretations. FDA's December 2021 guidance says it should be described in the protocol or an amendment, with a signed and dated peer-review statement included in the final study report. The pathology report remains the study pathologist's responsibility, and changes after a retrospective review are documented in an amendment.
Does SEND replace the PDF toxicology report?
No. SEND is the standard dataset format FDA requires for many nonclinical studies, and FDA's data guide asks that results in the report and the datasets be traceable. SEND goes in with a draft report when one is submitted. The signed report holds the study director's sentences and conclusions, so a reviewer still reads it, and checks that it agrees with the data.
Can AI review a toxicology report?
Software can read a report, extract its values with their pages, compare sections and draft comments, and one 2026 paper reports sensitivity typically above 95% for several endpoints when extracting from 200 reports at one company. We found no published study of catching errors in contract lab drafts. A named toxicologist still decides each comment, and the study director still decides the report.
Does report review software have to be validated for GLP?
OECD's computerized systems guidance is written for test facilities, and a sponsor's reading tool does not create the lab's raw data. Your quality unit should still decide, in your own procedures, how to assess a tool for its use. Treat any vendor claim of being certified or compliant for GxP with care: a product can offer controls that support your compliance, and the assessment stays yours.
What is the difference between report review and nonclinical summaries?
Report review checks one lab's study report before it is final. Nonclinical summaries, Modules 2.4 and 2.6 of the common technical document, come later and condense many final reports for a regulator. A clean review makes those summaries easier, because the values they copy have already been checked against their tables.
23 · One-page summary
Toxicology report review on one page: the six checks, November 2026 status, readiness and terms
Print this page
In shortThe whole guide on one screen: the checks, the rules as of November 2026, the readiness scale and the terms.
The checks
Report against protocol: every amendment reached the methods; every out-of-range result has a deviation.
Summary against tables: each sentence against its table, each count against its listing.
Each exposure value from the row its sentence names: day, sex, group.
This study against the program: target organs, terms, SEND.
Statements: QA, compliance, peer review, signatures, facility and storage.
One numbered letter: two locations, a question, on file. The study director decides.
Readiness: score 0–2 on eight questions: 0–6 explore · 7–11 pilot · 12–16 implement.
As of November 2026: 21 CFR 58.185 unchanged since 1987 · FDA's 2016 GLP proposal has no final rule found · OECD No. 21 (2020) and No. 23 (2022) current · FDA peer review guidance final, December 2021 · Study Data guide v6.3, October 2026 · FDA's AI guidance a draft · Annex 22 a draft, and a GMP text.
13 terms
Sponsor
The company that commissions a study and may file it.
Contract lab, CRO
The test facility hired to run the study.
GLP
Good Laboratory Practice: the rules for planning, running, recording and reporting a nonclinical study.
Study director
The lab's single point of study control; signs the final report.
Study monitor
The sponsor's toxicologist who follows the study and reads drafts.
Protocol
The study's approved written plan; changes need signed, dated reasons.
Deviation
A departure from the protocol or an SOP, documented with its assessment.
QA statement
QA's signed statement of its inspections, included with the final report.
Peer-review statement
A second pathologist's signed statement, included in the final report.
Amendment
The only way to change a signed final report: what changed and why.
NOAEL
No-observed-adverse-effect level: a conclusion stated in the report.
Toxicokinetics, AUC
Measures of how much compound reached the bloodstream in a study.
SEND
The standard format for nonclinical datasets submitted to FDA.
24 · Sources
Sources
Accessed November 2026. Regulatory status statements are as of November 2026.
21 CFR 58.185, Reporting of nonclinical laboratory study results, CFR 2025 edition (revised as of April 1, 2025). govinfo.gov
OECD, Series on Principles of GLP No. 11, The Role and Responsibilities of the Sponsor in the Application of the Principles of GLP, 1998, English and Japanese copy published by the Japan Society of Quality Assurance. jsqa.com
OECD, Series on Principles of GLP No. 21, Position Paper Regarding Possible Influence of Sponsors on Conclusions of GLP Studies, ENV/JM/MONO(2020)5, declassified April 30, 2020, as published by Germany's GLP authority. bfr.bund.de
21 CFR 58.35, Quality assurance unit, CFR 2025 edition. govinfo.gov
FDA Center for Veterinary Medicine, QASR Hot Topics, presentation, March 16, 2023, top findings in studies. fda.gov
21 CFR 58.33, Study director, CFR 2025 edition. govinfo.gov
21 CFR 58.10, Applicability to studies performed under grants and contracts, CFR 2025 edition. govinfo.gov
21 CFR 312.23(a)(8), IND content and format, pharmacology and toxicology information, CFR 2025 edition. govinfo.gov
FDA, Pathology Peer Review in Nonclinical Toxicology Studies: Questions and Answers, final guidance, December 2021, docket FDA-2019-D-2330. Nonbinding. fda.gov
OECD, Series on Principles of GLP No. 23, Advisory Document on Quality Assurance and GLP, ENV/CBC/MONO(2022)20, as published by Germany's GLP authority. bfr.bund.de
FDA, Study Data Technical Conformance Guide, version 6.3, October 2026, sections on the nonclinical reviewer's guide, SEND scope and resubmission, Appendix I. Nonbinding. fda.gov
FDA, Good Laboratory Practice for Nonclinical Laboratory Studies, proposed rule, 81 FR 58341, August 24, 2016, docket FDA-2010-N-0548. No final rule found on November 3, 2026. govinfo.gov
FDA, Agency Information Collection Activities; Good Laboratory Practice Requirements for Nonclinical Laboratory Studies, notice, FR Doc. 2026-08371, docket FDA-2026-N-3532, scheduled for April 30, 2026, Tables 1 and 2. federalregister.gov
OECD, Series on Principles of GLP No. 22, Advisory Document on GLP Data Integrity, ENV/CBC/MONO(2021)26, September 20, 2021, section 7.2, as published by Germany's GLP authority. bfr.bund.de
FDA, Nonclinical Testing Terminology, direct final rule, Docket FDA-2026-N-5347, FR Doc. 2026-19350, Federal Register Vol. 91, No. 182, September 22, 2026. govinfo.gov
21 CFR 58.120, Protocol, and 58.130, Conduct of a nonclinical laboratory study, CFR 2025 edition. govinfo.gov
EMA, Good laboratory practice compliance page: Directives 2004/9/EC and 2004/10/EC; inspections by the competent authority. ema.europa.eu
US EPA, Good Laboratory Practice Program Regulatory Update, presentation to the National Alliance of Independent Crop Consultants, January 28, 2026, 2025 inspection findings. naicc.org
FDA, FY 2024 Good Laboratory Practice FDA 483 Observation Trends. fda.gov
Pitkin, Branagan and Burmeister, Accuracy of data in abstracts of published research articles, JAMA 281(12):1110-1, 1999. PMID 10188662. pubmed.ncbi.nlm.nih.gov
Kerlin et al., Recommended ("Best") Practices for Determining, Communicating, and Using Adverse Effect Data from Nonclinical Studies, Toxicol Pathol 44(2):147-62, 2016. PMID 26704930. pubmed.ncbi.nlm.nih.gov
ICH, M4S(R2) The Common Technical Document: Safety, December 20, 2002. database.ich.org
FDA, Content and Format of Investigational New Drug Applications (INDs) for Phase 1 Studies, guidance, November 1995. fda.gov
FDA, warning letter to Vedic Lifesciences Pvt. Ltd., March 9, 2026. fda.gov
FDA, warning letter to Lovelace Biomedical Research Institute, MARCS-CMS 725136, October 6, 2026, as the testing facility for two nonclinical studies (sponsors redacted). fda.gov
FDA, warning letter to SAMM Solutions, Inc. dba BTS Research, October 20, 2023. fda.gov
Lanning and Chen, FDA CDER, Good Laboratory Practice (GLP) Case Studies, presentation, 2022, Case 2 on translated reports. fda.gov
IMARC Group, Preclinical CRO Market Size, Share, Trends, and Forecast, read November 3, 2026. A paid-report summary; its method is not public. imarcgroup.com
Towards Healthcare, Preclinical CRO Market Sizing, read November 3, 2026. A paid-report summary; its method is not public. towardshealthcare.com
FDA, Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products, draft guidance, January 2025. fda.gov
EMA, Reflection paper on the use of Artificial Intelligence in the medicinal product lifecycle, EMA/CHMP/CVMP/83833/2023, September 9, 2024, section 2.3.2. ema.europa.eu
FDA and EMA, Guiding Principles of Good AI Practice in Drug Development, content current January 14, 2026. fda.gov
European Commission and PIC/S, draft EU GMP Annex 22, consultation July 7 to October 7, 2025. EudraLex Volume 4 listed no Annex 22 on November 3, 2026. EMA report EMA/156789/2026 on the June 30, 2026 workshop. picscheme.org
FDA, Computer Software Assurance for Production and Quality Management System Software, final guidance, February 2026. fda.gov
Regulation (EU) 2026/1744 (Digital Omnibus on AI), Official Journal of the European Union, L series, July 24, 2026, in force on the third day after publication (July 27, 2026): Annex III high-risk rules from December 2, 2027, Annex I rules from August 2, 2028; the European Commission's AI Act page states the same dates. eur-lex.europa.eu
OECD, Series on Principles of GLP No. 17, Application of GLP Principles to Computerised Systems, ENV/JM/MONO(2016)13, 2016, reissued November 9, 2022, paragraphs 12 and 22 to 27, as published by Germany's GLP authority. bfr.bund.de
Doktorova et al., Transforming animal study toxicology reports into structured, harmonized data using large language models, Arch Toxicol 100(9):4105-4134, 2026. PMID 42133063. pubmed.ncbi.nlm.nih.gov
21 CFR 11.10, 11.50 and 11.70, electronic records and signatures, CFR 2025 edition. govinfo.gov
FDA, Good Laboratory Practice for Nonclinical Laboratory Studies, proposed rule, Regulatory Impact Analysis, 2016, wage notes to its cost tables. fda.gov
Smith, DiMasi and Getz, Tufts CSDD, journal article in Ther Innov Regul Sci 58(5):855-862, 2024 (PMID 38773058), and white paper, Quantifying the Value of a Day of Delay in Drug Development, August 2024. csdd.tufts.edu
Next step
See it on your own contract lab drafts.
Explore the solution, or bring one closed draft with its protocol, amendments and final report. We will check it with you, every discrepancy shown beside both pages, and your toxicologist decides and signs.
We use essential cookies to make this site work, and cookieless, anonymised analytics to understand how it's used. "Accept all" also turns on Google Analytics, which sets cookies. Privacy policy