Week 1 · Module 1
Manuscript diagnostic
Twenty-four questions across authorship, structure, claim discipline and submission — and a personal study order for the next four weeks
Your scenario
A manuscript comes back from a journal with a decision letter you were not expecting: reject without review. The science was sound. The editor's note runs to two sentences — the study question was not clear from the abstract, and the conclusions did not follow from the design. Both are writing failures, not science failures, and both were fixable in an afternoon.
What this course is
Most manuscript training teaches the IMRAD structure and stops. Structure is the easy part and you probably already know it. What separates a manuscript that gets reviewed from one that gets desk-rejected is narrower and more specific: a single defensible message, a Results section that reports without interpreting, and a Discussion that claims exactly what the design supports.
This course spends half its modules on Results and Discussion for that reason. Introduction and Methods are compressed into Week 4, and the Introduction is deliberately taught last — because it should be written last.
Take the diagnostic honestly
Do not look anything up. A wrong answer is worth more than a right one here, because it is the only way to find the gap. Guessing correctly and skipping the module is how people arrive at a submission confident and wrong.
—
—
Your twelve areas — study the red ones first
Before you move on
Write down your three weakest areas. Return here after the capstone and note what changed.
Next module
1.2 — Authorship and the publication plan
Week 1 · Module 2
Authorship and the publication plan
Who goes on the paper, who does not, and the decisions that must be settled before a word is written — because settling them afterwards is where collaborations break
Your scenario
The draft is finished. At the final review meeting a department head who has not appeared in any prior discussion asks to be added as second author, and a junior registrar who collected every data point is not on the list at all. Nobody wrote the authorship down at the start. The next four weeks are spent on this rather than on the science, and one collaborator leaves the group.
The four ICMJE criteria
An author must satisfy all four. Meeting three makes someone a contributor to be acknowledged, not an author.
| # | Criterion | What it excludes |
|---|---|---|
| 1 | Substantial contribution to conception or design, or to acquisition, analysis or interpretation of data | Providing patients without design input; running a routine assay to order |
| 2 | Drafting the work or revising it critically for important intellectual content | Reading and saying it looks fine; correcting typographical errors |
| 3 | Final approval of the version to be published | Being unavailable at sign-off, however senior |
| 4 | Agreement to be accountable for all aspects of the work | Anyone unwilling to answer questions about accuracy or integrity |
The three named abuses
Guest (honorary) authorship — adding someone senior who does not meet the criteria, for prestige or politics.
Gift authorship — adding a colleague as a favour, or by reciprocal arrangement.
Ghost authorship — a substantive contributor, often a professional writer or a sponsor statistician, who is neither an author nor disclosed.
All three are publication ethics breaches. All three are common. Journals increasingly require a signed contributorship statement precisely because of it.
Gift authorship — adding a colleague as a favour, or by reciprocal arrangement.
Ghost authorship — a substantive contributor, often a professional writer or a sponsor statistician, who is neither an author nor disclosed.
All three are publication ethics breaches. All three are common. Journals increasingly require a signed contributorship statement precisely because of it.
Roles that get acknowledged, not authored
Acquisition of funding · general supervision of a research group · provision of patients or materials · routine technical assistance · language editing · administrative support · department heads by virtue of position.
Each deserves a named acknowledgement with the contribution described. That is not a consolation prize — it is the correct credit.
Each deserves a named acknowledgement with the contribution described. That is not a consolation prize — it is the correct credit.
Professional writing support and GPP 2022
A professional medical writer may draft a manuscript entirely, under the direction of the named authors, without becoming an author. What is not permitted is doing so invisibly.
| Requirement | What it looks like in the manuscript |
|---|---|
| Disclose the writing support | Named in the acknowledgements with the role stated: Writing support was provided by [name, affiliation]. |
| Disclose who paid for it | …funded by [sponsor]. Never omitted, even when the sponsor had no editorial control. |
| Authors direct and approve | The named authors determine the content, review every draft, and approve the final version. |
| Authors control the data | Authors have access to the data and are accountable for its integrity. A writer never adjusts data. |
| No editorial veto by the funder | Stated explicitly where relevant: the sponsor did not have the right to suppress publication. |
Where the line sits
Writing a manuscript for investigators, with disclosure, is legitimate professional practice. Being listed as an author in exchange for writing it is not. Nor is writing a paper for a client who did not do the study, arranging its acceptance, or adjusting data to make an argument work.
The distinction is simple and worth being able to state cleanly: the investigators own the science and the authorship; the writer owns the prose and is acknowledged for it.
The distinction is simple and worth being able to state cleanly: the investigators own the science and the authorship; the writer owns the prose and is acknowledged for it.
The publication plan
Settle all of this before drafting. Ten minutes at the start; four weeks of argument avoided at the end.
| Decision | Why it must come first |
|---|---|
| Author list and order | Order is negotiated, not discovered. First author does the most work; last is usually senior/guarantor. |
| Corresponding author | Carries correspondence, revisions and post-publication queries. A real commitment, not a title. |
| Contributorship statement | Increasingly mandatory. Write it early and the authorship question answers itself. |
| Target journal, plus two fallbacks | Determines word limits, structure, reference style and reporting checklist. |
| Reporting guideline | CONSORT for trials, STROBE for observational, PRISMA for systematic reviews, CARE for case reports. |
| Registration and data sharing | Prospective registration is required for trials by ICMJE. A data sharing statement is required by most journals. |
| Conflicts of interest | Collect from every author at the start; chasing declarations at submission causes most delays. |
| Timeline and review rounds | Two full author review rounds is realistic. One is optimistic; three means nobody decided anything. |
Drills
Drill 1 of 3
A statistician performed all analyses, wrote the statistical methods text, and approved the final draft. Author or acknowledgement?
AAuthor — all four criteria are met
BAcknowledgement — statistical work is technical support
CNeither — statisticians are listed separately
DAuthor only if externally employed
Drill 2 of 3
A department head who provided laboratory space and general encouragement asks to be added as senior author. What is the correct response?
AAdd them — seniority conventionally warrants inclusion
BAcknowledge them specifically for facilities and support, and explain the ICMJE criteria
CAdd them as a middle author, which matters less
DSay nothing and submit without them
Drill 3 of 3
A sponsor-funded manuscript was drafted by an agency medical writer. Which acknowledgement is compliant?
A“The authors thank the writing team for editorial assistance.”
BNo acknowledgement — the authors approved the final version
C“Medical writing support was provided by [name], [agency], and was funded by [sponsor]. The authors directed the content and approved the final manuscript.”
DList the writer as an author to be transparent
Production drill
Draft the contributorship statement, acknowledgements and disclosures for this project.
The team:
· Dr A — conceived the study, designed it, reviewed and approved the draft
· Dr B — enrolled 60% of patients, contributed to interpretation, revised the draft critically, approved it
· Dr C — provided 40% of patients from her clinic; no other involvement; has not read the draft
· Mr D — performed all statistical analyses, wrote the statistical methods, approved the draft
· Prof E — head of department; secured the departmental grant; asked to be listed
· Ms F — professional medical writer, agency-employed, drafted the manuscript from an outline agreed by Dr A and Dr B; funded by the grant
· Mr G — research coordinator, managed the database and scheduled visits
· Dr A — conceived the study, designed it, reviewed and approved the draft
· Dr B — enrolled 60% of patients, contributed to interpretation, revised the draft critically, approved it
· Dr C — provided 40% of patients from her clinic; no other involvement; has not read the draft
· Mr D — performed all statistical analyses, wrote the statistical methods, approved the draft
· Prof E — head of department; secured the departmental grant; asked to be listed
· Ms F — professional medical writer, agency-employed, drafted the manuscript from an outline agreed by Dr A and Dr B; funded by the grant
· Mr G — research coordinator, managed the database and scheduled visits
Authors: Dr A, Dr B, Mr D.
Contributorship statement: “Dr A conceived and designed the study. Dr B contributed to data acquisition and interpretation. Mr D performed the statistical analyses and drafted the statistical methods. All authors critically revised the manuscript for important intellectual content, approved the final version, and agree to be accountable for all aspects of the work.”
Acknowledgements: “The authors thank Dr C for providing access to patients from her clinic, Prof E for departmental support and for securing funding, and Mr G for database management and study coordination. Medical writing support was provided by Ms F, [agency], funded by [grant]. The authors directed the content of the manuscript and approved the final version.”
The four judgements:
1. Dr C is not an author. Providing patients alone fails criterion 1, and not having read the draft fails 2, 3 and 4. She contributed 40% of the cohort, which feels like a lot — contribution volume is not the test.
2. Prof E is not an author. Funding acquisition and departmental leadership are explicitly acknowledgement-level. This is the conversation to have in week one, not at submission.
3. Mr D is an author. Analysis plus drafting plus approval meets all four. Statisticians are frequently and wrongly relegated to acknowledgements.
4. Ms F is not an author but must be disclosed. She drafted the whole manuscript under author direction, which is legitimate. Omitting her name would be ghostwriting.
The uncomfortable part: two people who expect to be authors are not. That conversation is far easier in week one, framed around a published international standard, than in week twelve framed around your judgement of their contribution.
Contributorship statement: “Dr A conceived and designed the study. Dr B contributed to data acquisition and interpretation. Mr D performed the statistical analyses and drafted the statistical methods. All authors critically revised the manuscript for important intellectual content, approved the final version, and agree to be accountable for all aspects of the work.”
Acknowledgements: “The authors thank Dr C for providing access to patients from her clinic, Prof E for departmental support and for securing funding, and Mr G for database management and study coordination. Medical writing support was provided by Ms F, [agency], funded by [grant]. The authors directed the content of the manuscript and approved the final version.”
The four judgements:
1. Dr C is not an author. Providing patients alone fails criterion 1, and not having read the draft fails 2, 3 and 4. She contributed 40% of the cohort, which feels like a lot — contribution volume is not the test.
2. Prof E is not an author. Funding acquisition and departmental leadership are explicitly acknowledgement-level. This is the conversation to have in week one, not at submission.
3. Mr D is an author. Analysis plus drafting plus approval meets all four. Statisticians are frequently and wrongly relegated to acknowledgements.
4. Ms F is not an author but must be disclosed. She drafted the whole manuscript under author direction, which is legitimate. Omitting her name would be ghostwriting.
The uncomfortable part: two people who expect to be authors are not. That conversation is far easier in week one, framed around a published international standard, than in week twelve framed around your judgement of their contribution.
Reflection
For a manuscript you are working on now: is the author list written down and agreed by everyone on it? If not, that is this week's task.
Authorship disputes are the most common reason collaborations end, and almost all of them trace to a conversation that never happened at the start.
Next module
1.3 — From source documents to a data narrative
Week 1 · Module 3
From source documents to a data narrative
The protocol, the SAP and the output pack contain everything the manuscript needs and far more than it can hold. Extracting the right 5% is the first real skill
Your scenario
You are given a protocol, a statistical analysis plan, an output pack and four days. On day three you realise the median you quoted in the abstract came from a figure generated at an earlier data cut than the table you cited in Results. Both numbers are in your draft. They differ by eight-tenths of a month, and a reviewer will find it.
What each source is for
| Source | Answers | Do not use it for |
|---|---|---|
| Protocol | Design, eligibility, interventions, endpoint definitions, ethics and registration | Results. It describes what was planned, not what happened. |
| Statistical analysis plan | Analysis populations, the testing hierarchy, estimand, handling of missing data, prespecified subgroups | Numbers. It is the rulebook, not the scoreboard. |
| Output pack (TFLs) | Every number you will quote | Interpretation. Outputs do not tell you what matters. |
| Clinical study report | Narrative context, deviations, disposition detail | Definitive figures — it is derived text and may predate the current data cut. |
| Investigator's brochure | Background pharmacology and prior safety for the Introduction | Anything about this study. |
The rule that prevents most errors
Every number in your draft traces to one output, and every output has a data cut-off date. Before you quote anything, check the footer for the cut-off and the population. A figure and a table generated at different cuts will disagree, and the disagreement will look like carelessness rather than what it is.
Build the number sheet first
Before writing a sentence, build a single page listing every number you intend to quote, with its source. It takes about ninety minutes and it is the difference between a draft that can be checked and one that cannot.
| Column | Why |
|---|---|
| The number | As it will appear, with its decimals already decided |
| Table or figure reference | So you, QC and the statistician can verify it in seconds |
| Analysis population and n | Catches the randomised-versus-treated error before it reaches the page |
| Data cut-off date | Catches mixed-cut figures |
| Where it will appear | Abstract, Results text, table, figure legend — so the same value is consistent everywhere |
Never resolve a discrepancy yourself
If two outputs disagree, do not choose the one that looks right, do not recalculate, and do not quietly use the one that suits the narrative.
Raise it in writing with both references and both values. Usually there is an explanation — a different cut, a different population, a footnote you missed. Occasionally you have found a real error, and you will have found it before a reviewer or a regulator did.
Raise it in writing with both references and both values. Usually there is an explanation — a different cut, a different population, a footnote you missed. Occasionally you have found a real error, and you will have found it before a reviewer or a regulator did.
From outputs to a narrative
An output pack has no argument in it. Turning it into one requires three passes.
| Pass | Question | Output |
|---|---|---|
| One | What was asked and what was found? | The primary endpoint result, in one sentence, with its effect size and interval |
| Two | What supports or qualifies it? | Key secondary endpoints, consistency across subgroups, sensitivity analyses, safety |
| Three | What would a sceptical reader attack? | Your limitations paragraph, drafted before you write anything else |
Drafting limitations third, not last
Most writers leave limitations until the end, when they are tired and want to finish. Drafting them in the third pass, before the Discussion exists, changes what you write everywhere else — because you now know which claims you will have to defend.
Drills
Drill 1 of 3
You need to state which patients were included in the primary efficacy analysis. Which source is definitive?
AThe protocol
BThe clinical study report narrative
CWhatever the lead author remembers
DThe statistical analysis plan, with the n taken from the disposition output
Drill 2 of 3
A table footer reads “data cut-off 15 January”; the corresponding figure footer reads “data cut-off 30 September”. What do you do?
ARequest a regenerated figure at the current cut before using either
BUse both as supplied and note the difference in the legend
CUse the table, since tables are more precise
DAverage the two values
Drill 3 of 3
The output pack contains 40 subgroup analyses. How many belong in the manuscript?
AAll 40, for completeness
BThose prespecified in the SAP, with the rest labelled exploratory if reported at all
CThe ones with significant results
DNone — subgroups are always exploratory
Production drill
Build the number sheet and list every query you would raise before drafting.
SAP: primary endpoint PFS in ITT; hierarchy PFS → OS → ORR; safety in treated population; 6 prespecified subgroups.
Table 14.1.1 Disposition: 402 randomised (201/201); 389 treated (196/193); 371 with ≥1 post-baseline scan.
Table 14.2.1 PFS (ITT, n = 402): median 8.9 vs 5.2 months; HR 0.66 (95% CI 0.53–0.82). Cut-off 15 Jan.
Figure 14.2.2 KM curve: crosses 50% at ~8.1 months. Cut-off 30 Sep.
Table 14.2.3 OS: not reached vs 21.7 months; HR 0.84 (95% CI 0.62–1.14); p = 0.26.
Table 14.2.4 ORR: 44% vs 29%; p = 0.003. Population not stated in header.
Table 14.3.1 TEAE: n = 402; grade ≥3 in 147 (36.6%).
Table 14.2.7 Subgroups: 11 subgroups presented.
Table 14.1.1 Disposition: 402 randomised (201/201); 389 treated (196/193); 371 with ≥1 post-baseline scan.
Table 14.2.1 PFS (ITT, n = 402): median 8.9 vs 5.2 months; HR 0.66 (95% CI 0.53–0.82). Cut-off 15 Jan.
Figure 14.2.2 KM curve: crosses 50% at ~8.1 months. Cut-off 30 Sep.
Table 14.2.3 OS: not reached vs 21.7 months; HR 0.84 (95% CI 0.62–1.14); p = 0.26.
Table 14.2.4 ORR: 44% vs 29%; p = 0.003. Population not stated in header.
Table 14.3.1 TEAE: n = 402; grade ≥3 in 147 (36.6%).
Table 14.2.7 Subgroups: 11 subgroups presented.
Number sheet (extract):
· Median PFS 8.9 vs 5.2 months — T14.2.1 — ITT n=402 — cut 15 Jan — abstract, Results, Fig legend
· HR 0.66 (95% CI 0.53–0.82) — T14.2.1 — ITT — cut 15 Jan — abstract, Results
· Median OS NR vs 21.7 months — T14.2.3 — ITT — cut 15 Jan — Results
· ORR 44% vs 29% — T14.2.4 — population TBC — Results
· Grade ≥3 TEAE — T14.3.1 — denominator TBC — Results, safety table
Five queries, in order of seriousness:
1. Table 14.3.1 uses the wrong safety denominator. It shows n = 402, the randomised population. Safety must use the treated population, n = 389 (196/193). Recomputed, 147/389 is 37.8%, not 36.6%. This is a real error that would reach publication.
2. Figure 14.2.2 is at a different data cut from Table 14.2.1, which fully explains the 8.9 versus 8.1 discrepancy. Request a regenerated figure at the 15 January cut. Until then, quote nothing from it.
3. Table 14.2.4 does not state its population. ORR is usually analysed in patients with a post-baseline assessment, which would be 371, not 402. Confirm before quoting the percentages.
4. Eleven subgroups are presented but the SAP prespecified six. Identify which six, and label the remaining five exploratory or drop them.
5. ORR sits below OS in the hierarchy, and OS was not significant (p = 0.26). ORR therefore cannot be described as significant — report it with a nominal p value and state that formal testing stopped at OS.
What the drill is really testing: not whether you can copy numbers, but whether you check them before they become sentences. Query 1 is the one that matters — every safety percentage in the paper would otherwise be understated.
· Median PFS 8.9 vs 5.2 months — T14.2.1 — ITT n=402 — cut 15 Jan — abstract, Results, Fig legend
· HR 0.66 (95% CI 0.53–0.82) — T14.2.1 — ITT — cut 15 Jan — abstract, Results
· Median OS NR vs 21.7 months — T14.2.3 — ITT — cut 15 Jan — Results
· ORR 44% vs 29% — T14.2.4 — population TBC — Results
· Grade ≥3 TEAE — T14.3.1 — denominator TBC — Results, safety table
Five queries, in order of seriousness:
1. Table 14.3.1 uses the wrong safety denominator. It shows n = 402, the randomised population. Safety must use the treated population, n = 389 (196/193). Recomputed, 147/389 is 37.8%, not 36.6%. This is a real error that would reach publication.
2. Figure 14.2.2 is at a different data cut from Table 14.2.1, which fully explains the 8.9 versus 8.1 discrepancy. Request a regenerated figure at the 15 January cut. Until then, quote nothing from it.
3. Table 14.2.4 does not state its population. ORR is usually analysed in patients with a post-baseline assessment, which would be 371, not 402. Confirm before quoting the percentages.
4. Eleven subgroups are presented but the SAP prespecified six. Identify which six, and label the remaining five exploratory or drop them.
5. ORR sits below OS in the hierarchy, and OS was not significant (p = 0.26). ORR therefore cannot be described as significant — report it with a nominal p value and state that formal testing stopped at OS.
What the drill is really testing: not whether you can copy numbers, but whether you check them before they become sentences. Query 1 is the one that matters — every safety percentage in the paper would otherwise be understated.
Reflection
On your next manuscript, build the number sheet before writing. How long did it take, and how many discrepancies did it surface?
Next module
1.4 — The core message and the storyboard
Week 1 · Module 4
The core message and the storyboard
One sentence, agreed before drafting, that determines the title, the lead result, the figures and the argument. Manuscripts without one read as a list of things that happened
Your scenario
Three co-authors review your draft. One wants the biomarker subgroup to lead. One wants safety foregrounded. One wants the health-economic implication in the conclusion. Each request is reasonable in isolation. Together they produce a paper about nothing. The disagreement is not about writing — it is that nobody ever agreed what the paper is for.
Definition
Core message
One sentence, defensible from the data, that a reader should retain if they remember nothing else.
It is agreed by all authors before drafting begins and it governs everything downstream: which result leads the abstract, what the title says, which four figures survive, and what the Discussion argues.
A core message is not the objective (what you set out to do) and not the conclusion (the final paragraph). It is the claim the paper exists to make.
It is agreed by all authors before drafting begins and it governs everything downstream: which result leads the abstract, what the title says, which four figures survive, and what the Discussion argues.
A core message is not the objective (what you set out to do) and not the conclusion (the final paragraph). It is the claim the paper exists to make.
Not a core message
A core message
“We studied regimen X in advanced disease.”that is the objective
“Regimen X improved progression-free survival over placebo, with a safety profile consistent with its class.”
“This paper reports the results of the trial.”contentless
“The benefit of regimen X was confined to biomarker-positive patients, supporting biomarker-directed selection.”
“Regimen X is effective, well tolerated, cost-effective and improves quality of life.”four messages is none
“Regimen X extended progression-free survival without an increase in grade 3 or higher toxicity.”
“Regimen X should become standard of care.”not defensible from one trial
“Regimen X warrants evaluation as a first-line option in a confirmatory trial.”
The two tests
1. Is it defensible? Read it beside the primary endpoint result and the study design. If the design cannot support the verb, the message is wrong, not the data.
2. Is it one thing? If it contains and more than once, you have several messages. Pick the one the primary endpoint supports; the others become supporting results or a second paper.
2. Is it one thing? If it contains and more than once, you have several messages. Pick the one the primary endpoint supports; the others become supporting results or a second paper.
The storyboard
Once the message is fixed, plan the paper on one page before writing prose. A storyboard is faster to argue about than a draft, and co-authors will argue.
| Element | Decide now | Because |
|---|---|---|
| Core message | One sentence | Everything else follows from it |
| Target journal | One, plus two fallbacks | Sets word limit, structure, display item cap |
| Display items | Usually 4–6, each named | Journals cap them; choosing late means cutting good work |
| The lead result | Which number opens Results | Prevents the paper drifting toward whatever a co-author finds interesting |
| Section word budget | Rough allocation | Discussions bloat unless budgeted |
| Limitations | Three, named | Drafted early, they discipline every claim you make |
| Conclusion sentence | Written now, in draft | If you cannot write it from the message, the message is not defensible |
Circulate the storyboard, not the draft
Send co-authors the one-page storyboard for approval before writing. Objections at this stage cost an email. The same objections after a full draft cost a rewrite, and by then people are attached to sentences rather than arguing about structure.
This single habit removes most of the pain from multi-author manuscripts.
This single habit removes most of the pain from multi-author manuscripts.
Drills
Drill 1 of 3
A randomised trial met its primary PFS endpoint (HR 0.66). OS was immature and not significant. Which is the best core message?
A“Regimen X improves survival in advanced disease.”
B“We evaluated regimen X in a randomised Phase III trial.”
C“Regimen X improved progression-free survival; overall survival data remain immature.”
D“Regimen X is effective, safe, tolerable and cost-effective.”
Drill 2 of 3
A co-author wants a striking subgroup finding (interaction p = 0.19) to become the core message. How do you respond?
AAgree — they are senior and it is the most interesting result
BUse both as dual core messages
CDrop the subgroup analysis entirely
DExplain that the interaction test does not support a differential effect; report it as exploratory
Drill 3 of 3
The target journal permits five display items. You have material for nine. What is the right approach?
AAt storyboard stage, choose the five that carry the message; move the rest to supplementary
BDraft all nine and cut at the end
CCombine items to fit the cap
DChange journal to one allowing nine
Production drill
Write the core message and the full storyboard for this study.
Design: randomised, double-blind, placebo-controlled Phase III; 402 patients; advanced disease, second line.
Primary: PFS — median 8.9 vs 5.2 months; HR 0.66 (95% CI 0.53–0.82); p < 0.001.
Key secondary: OS — not reached vs 21.7 months; HR 0.84 (0.62–1.14); p = 0.26; immature.
ORR: 44% vs 29% (below hierarchy break; nominal p = 0.003).
Safety: grade ≥3 TEAE 37.8% vs 31.2%; discontinuation for AE 8.0% vs 5.7%.
Subgroups: 6 prespecified; only biomarker-positive status shows interaction (p = 0.03).
Target: a high-impact specialty journal — 3,000 words, 5 display items.
Primary: PFS — median 8.9 vs 5.2 months; HR 0.66 (95% CI 0.53–0.82); p < 0.001.
Key secondary: OS — not reached vs 21.7 months; HR 0.84 (0.62–1.14); p = 0.26; immature.
ORR: 44% vs 29% (below hierarchy break; nominal p = 0.003).
Safety: grade ≥3 TEAE 37.8% vs 31.2%; discontinuation for AE 8.0% vs 5.7%.
Subgroups: 6 prespecified; only biomarker-positive status shows interaction (p = 0.03).
Target: a high-impact specialty journal — 3,000 words, 5 display items.
Core message: “In previously treated advanced disease, regimen X significantly prolonged progression-free survival compared with placebo, with a manageable safety profile; overall survival data remain immature.”
Storyboard:
Journal: target as stated; fallbacks two tiers down. 3,000 words, 5 display items.
Display items: (1) participant flow; (2) baseline characteristics; (3) PFS Kaplan–Meier; (4) forest plot of the six prespecified subgroups; (5) grade ≥3 adverse events by preferred term. OS curve and ORR waterfall to supplementary.
Lead result: median PFS with HR and interval. Not ORR, despite its smaller p value.
Word budget: Introduction 450 · Methods 700 · Results 900 · Discussion 850 · Conclusion 100.
Limitations, drafted now: (1) OS immature at this cut-off; (2) the biomarker interaction was one of six prespecified analyses and the trial was not powered for it; (3) the population was second-line only, limiting generalisability to earlier lines.
Conclusion sentence, in draft: “Regimen X significantly improved progression-free survival in previously treated advanced disease; continued follow-up is required to determine its effect on overall survival.”
Three judgements worth noticing:
1. ORR does not lead, even though nominal p = 0.003 is smaller than the primary endpoint's. It sits below the hierarchy break and is not formally tested.
2. The biomarker interaction does not become the message, despite p = 0.03. It is one of six prespecified analyses in a trial not powered for interaction. It earns a place in the forest plot and a hedged sentence in the Discussion — not the title.
3. OS immaturity is in the core message. Putting it there rather than burying it in limitations is what makes the message defensible, and it pre-empts the first question any reviewer will ask.
Storyboard:
Journal: target as stated; fallbacks two tiers down. 3,000 words, 5 display items.
Display items: (1) participant flow; (2) baseline characteristics; (3) PFS Kaplan–Meier; (4) forest plot of the six prespecified subgroups; (5) grade ≥3 adverse events by preferred term. OS curve and ORR waterfall to supplementary.
Lead result: median PFS with HR and interval. Not ORR, despite its smaller p value.
Word budget: Introduction 450 · Methods 700 · Results 900 · Discussion 850 · Conclusion 100.
Limitations, drafted now: (1) OS immature at this cut-off; (2) the biomarker interaction was one of six prespecified analyses and the trial was not powered for it; (3) the population was second-line only, limiting generalisability to earlier lines.
Conclusion sentence, in draft: “Regimen X significantly improved progression-free survival in previously treated advanced disease; continued follow-up is required to determine its effect on overall survival.”
Three judgements worth noticing:
1. ORR does not lead, even though nominal p = 0.003 is smaller than the primary endpoint's. It sits below the hierarchy break and is not formally tested.
2. The biomarker interaction does not become the message, despite p = 0.03. It is one of six prespecified analyses in a trial not powered for interaction. It earns a place in the forest plot and a hedged sentence in the Discussion — not the title.
3. OS immaturity is in the core message. Putting it there rather than burying it in limitations is what makes the message defensible, and it pre-empts the first question any reviewer will ask.
Reflection
Take a manuscript you are working on. Write its core message in one sentence. If you cannot, that is the reason the draft is difficult.
If the sentence needs two ands, you have more than one paper.
Next week
2.1 — Results architecture
Week 2 · Module 1
Results architecture
What goes in what order, and the single discipline that defines the section: Results report, they do not interpret
Your scenario
A reviewer writes: “The Results section contains substantial interpretation that belongs in the Discussion, and the Discussion repeats results already presented.” You read it again and they are right — you have written the same material twice, in the wrong two places, and the paper is a thousand words longer than it needs to be.
The sequence
The order is not stylistic. The reader cannot interpret what happened until they know who it happened to, so participant flow and baseline characteristics always come first.
Figure 2.1 — The Results sequence. Each stage answers one question and hands the reader what they need for the next. The primary endpoint sits third, not first, because a result is uninterpretable until the reader knows the population it came from.
Report, do not interpret
Interpretation — belongs in Discussion
Reporting — belongs in Results
These data suggest the regimen benefits younger patients.
In patients aged under 65, the hazard ratio was 0.68 (95% CI 0.51–0.91).
Importantly, the safety profile was favourable.
Grade ≥3 events occurred in 147/389 (37.8%) and 121/388 (31.2%).
Consistent with earlier trials, response was high.
Objective response rate was 44% versus 29%.
Surprisingly, overall survival did not differ.
Median overall survival was not reached versus 21.7 months (HR 0.84, 95% CI 0.62–1.14).
The words that give it away
suggests · indicates · importantly · notably · surprisingly · consistent with · in line with · as expected · encouragingly · favourable · promising
Search your Results section for these before submitting. Every hit is either an interpretation to move to the Discussion or an editorialising adverb to delete. This search takes two minutes and removes the most common structural criticism reviewers make.
Search your Results section for these before submitting. Every hit is either an interpretation to move to the Discussion or an editorialising adverb to delete. This search takes two minutes and removes the most common structural criticism reviewers make.
The one permitted exception
You may state that a prespecified analysis was performed and give its result, including whether it met a threshold — that is reporting, not interpreting. “The interaction test was not significant (p = 0.19)” belongs in Results. “The effect therefore appears consistent across subgroups” belongs in the Discussion.
Drills
Drill 1 of 3
Which sentence belongs in Results as written?
A“Encouragingly, response rate was 44% versus 29%.”
B“Objective response rate was 176/402 (44%) versus 117/402 (29%); nominal p = 0.003.”
C“Response rate was 44%, consistent with earlier reports in this setting.”
D“The response rate difference of 15 points is clinically meaningful.”
Drill 2 of 3
Where should participant flow appear?
AIn the Methods
BLast in Results, after safety
CFirst in Results, usually as a flow diagram
DSupplementary material only
Drill 3 of 3
A prespecified secondary endpoint was not significant. What does the Results section do?
AOmit it and address it in the Discussion
BReport it as having failed to reach significance
CMove it to supplementary material
DReport it with the same detail as any other endpoint
Multi-Therapeutic Mastery Structuring Primary & Key Secondary Results Across TAs
Different therapeutic areas follow distinct reporting conventions for effect sizes, baseline adjustments, and precision intervals. Master these 4 foundational therapeutic patterns:
| Therapeutic Area | Trial Design & Primary Endpoint | Statistical Metric Reported | Model Output Text (Audit-Proof Standard) |
|---|---|---|---|
| Cardiology (CVOT) | Time-to-first 3-point MACE (CV Death, Non-fatal MI, Non-fatal Stroke) | Stratified Cox Proportional Hazards: Hazard Ratio (HR), 95% CI, p-value | "A primary endpoint event occurred in 214 of 3,120 patients (6.9%) in the active group vs. 298 of 3,118 (9.6%) in the placebo group (HR 0.71; 95% CI, 0.59 to 0.86; P < 0.001 for superiority)." |
| Endocrinology & Metabolism | Change in HbA1c from Baseline to Week 26/52 in T2D | ANCOVA / MMRM: Least-Squares Mean (LSM) Difference, 95% CI, p-value | "Mean baseline HbA1c was 8.4%. At Week 26, the estimated LSM change from baseline was -1.42% with active drug vs. -0.28% with placebo (treatment difference -1.14%; 95% CI, -1.31 to -0.97; P < 0.001)." |
| Pulmonology & Respiratory | Annualized Asthma Exacerbation Rate (AAER) over 52 Weeks | Negative Binomial Regression: Rate Ratio (RR), 95% CI, % Reduction | "The annualized rate of severe asthma exacerbations was 0.38 (95% CI, 0.31 to 0.47) in the biologic group vs. 0.82 (95% CI, 0.70 to 0.96) in the placebo group (rate ratio 0.46; 95% CI, 0.36 to 0.59; P < 0.001; 54% reduction)." |
| Neurology (Multiple Sclerosis) | Annualized Relapse Rate (ARR) at 96 Weeks in RMS | Poisson / Negative Binomial: ARR, Rate Ratio, Time to Confirmed Disability Progression (CDP) | "The adjusted ARR was 0.11 vs. 0.24 (rate ratio 0.46; 95% CI, 0.37 to 0.58; P < 0.001). The proportion of patients with 12-week CDP was 9.3% vs. 15.0% (HR 0.60; 95% CI, 0.45 to 0.81; P = 0.001)." |
Production drill
Rewrite this Results paragraph. Remove every interpretation and fix the sequence. There are eight problems.
“Encouragingly, the primary endpoint was met, with a median PFS of 8.9 versus 5.2 months (HR 0.66), which is consistent with the mechanism of action. Baseline characteristics were well balanced. Importantly, safety was favourable, with grade 3 or higher events in 36.6%. Surprisingly, overall survival did not differ (p = 0.26), although the data are immature. A total of 402 patients were randomised and 389 received treatment. These findings suggest a meaningful benefit in this population.”
Corrected:
“A total of 402 patients were randomised and 389 received at least one dose. Baseline characteristics were balanced between arms (Table 1). Median progression-free survival was 8.9 months (95% CI 7.4–10.3) versus 5.2 months (95% CI 4.4–6.1); HR 0.66 (95% CI 0.53–0.82), p < 0.001. Median overall survival was not reached versus 21.7 months (HR 0.84, 95% CI 0.62–1.14; p = 0.26). Grade 3 or higher treatment-emergent adverse events occurred in 147/389 patients (37.8%).”
The eight problems:
1. Sequence inverted — disposition and baseline appeared at the end. They open the section.
2. Encouragingly — editorialising adverb, deleted.
3. consistent with the mechanism of action — interpretation, moved to Discussion.
4. Importantly / safety was favourable — both judgements, deleted; the numbers speak.
5. Surprisingly — a judgement about expectation, deleted.
6. although the data are immature — an explanation, which is a Discussion move.
7. These findings suggest a meaningful benefit — the sentence is pure Discussion.
8. 36.6% with no denominator, on the wrong population — safety uses the treated population, so 147/389 (37.8%).
Note what happened to the length: the paragraph got shorter and carried more information. Everything removed was either a duplicate of the Discussion or a judgement the reader can make themselves.
“A total of 402 patients were randomised and 389 received at least one dose. Baseline characteristics were balanced between arms (Table 1). Median progression-free survival was 8.9 months (95% CI 7.4–10.3) versus 5.2 months (95% CI 4.4–6.1); HR 0.66 (95% CI 0.53–0.82), p < 0.001. Median overall survival was not reached versus 21.7 months (HR 0.84, 95% CI 0.62–1.14; p = 0.26). Grade 3 or higher treatment-emergent adverse events occurred in 147/389 patients (37.8%).”
The eight problems:
1. Sequence inverted — disposition and baseline appeared at the end. They open the section.
2. Encouragingly — editorialising adverb, deleted.
3. consistent with the mechanism of action — interpretation, moved to Discussion.
4. Importantly / safety was favourable — both judgements, deleted; the numbers speak.
5. Surprisingly — a judgement about expectation, deleted.
6. although the data are immature — an explanation, which is a Discussion move.
7. These findings suggest a meaningful benefit — the sentence is pure Discussion.
8. 36.6% with no denominator, on the wrong population — safety uses the treated population, so 147/389 (37.8%).
Note what happened to the length: the paragraph got shorter and carried more information. Everything removed was either a duplicate of the Discussion or a judgement the reader can make themselves.
Reflection
Search the Results section of your last manuscript for the eleven interpretation words. How many hits?
Next module
2.2 — Tables, figures and text
Week 2 · Module 2
Tables, figures and text
Three ways to present a finding and one right answer for each — plus the flow diagram every trial and cohort study is required to carry
Your scenario
Your paper has seven tables. The journal permits five display items. Two of your tables have a single row. One duplicates a figure. The baseline table runs to two pages and lists thirty-one variables, of which the reader needs eight. You will now cut under time pressure and lose something that mattered.
The decision rule
| Use | When | Example |
|---|---|---|
| Text | Fewer than about four numbers, and they carry the argument | The primary endpoint result; the response rate difference |
| Table | Many precise values the reader may want to look up individually | Baseline characteristics; adverse events by preferred term |
| Figure | A pattern, relationship or shape that a table would hide | Survival curves; forest plots; change over time |
| Flow diagram | Always, for trials and cohort studies | CONSORT or STROBE participant flow |
| Supplementary | Needed for completeness or reproducibility, not for the argument | Full adverse event listings; sensitivity analyses; protocol amendments |
Four common faults
The one-row table. If it fits in a sentence, write the sentence.
The duplicate. A figure and a table showing the same data spends two of your five items on one finding.
The exhaustive baseline table. Thirty-one variables when eight are relevant. Move the rest to supplementary.
Text that recites a table. The text should highlight two or three values and point to the table for the rest, not walk through every row.
The duplicate. A figure and a table showing the same data spends two of your five items on one finding.
The exhaustive baseline table. Thirty-one variables when eight are relevant. Move the rest to supplementary.
Text that recites a table. The text should highlight two or three values and point to the table for the rest, not walk through every row.
The participant flow diagram
Figure 2.2 — Participant flow (CONSORT). Every denominator used anywhere in the paper appears here. Note the final row: efficacy is analysed by arm randomised (201/201) and safety by treatment received (196/193). Reviewers check this diagram against every percentage in the Results, so build it before you write the section, not after.
Self-standing legends
Display items get extracted — into slides, into other papers, into press coverage. A legend must let the item be understood alone. It needs: what is shown, the population and n, units, what error bars represent, any statistical test, and every abbreviation expanded.
Test: cover the article and read only the legend. If you cannot tell what you are looking at, it is not finished.
Test: cover the article and read only the legend. If you cannot tell what you are looking at, it is not finished.
Drills
Drill 1 of 3
You have one number: the proportion of patients who discontinued for adverse events, in each arm. How should it be presented?
AIn the text, with counts and denominators
BA dedicated table
CA bar chart
DSupplementary material
Drill 2 of 3
Which legend is adequate?
A“Figure 2. Kaplan–Meier curves.”
B“Figure 2. Kaplan–Meier estimates of progression-free survival (PFS) in the intention-to-treat population (n = 402). Tick marks indicate censored observations. Hazard ratio from a stratified Cox model; p value from a stratified log-rank test.”
C“Figure 2. PFS was significantly longer with the study regimen, as described in the Results.”
D“Figure 2. See Table 2 for details.”
Drill 3 of 3
Your baseline table lists 31 variables; 8 are prognostically relevant. What is the right action?
AKeep all 31 for completeness
BKeep only the variables that were balanced
CKeep the relevant and stratification variables; move the rest to supplementary
DDelete the baseline table
Production drill
You have five display item slots. Allocate them, and write the legend for your primary endpoint figure.
Available material:
1. Participant flow · 2. Baseline characteristics (31 variables) · 3. PFS Kaplan–Meier · 4. PFS by prespecified subgroup (forest plot, 6 subgroups) · 5. OS Kaplan–Meier (immature) · 6. Waterfall plot of best response · 7. Grade ≥3 adverse events by preferred term · 8. All adverse events by system organ class · 9. Table of PFS sensitivity analyses
Journal: 5 display items, unlimited supplementary.
1. Participant flow · 2. Baseline characteristics (31 variables) · 3. PFS Kaplan–Meier · 4. PFS by prespecified subgroup (forest plot, 6 subgroups) · 5. OS Kaplan–Meier (immature) · 6. Waterfall plot of best response · 7. Grade ≥3 adverse events by preferred term · 8. All adverse events by system organ class · 9. Table of PFS sensitivity analyses
Journal: 5 display items, unlimited supplementary.
The five:
Figure 1 — Participant flow. Required by CONSORT; carries every denominator.
Table 1 — Baseline characteristics, trimmed to prognostic and stratification variables. Full 31 to supplementary.
Figure 2 — PFS Kaplan–Meier. The primary endpoint; a curve shows shape a median cannot.
Figure 3 — Forest plot of the six prespecified subgroups. Consistency of effect, in one item.
Table 2 — Grade ≥3 adverse events by preferred term. The safety information a clinician acts on.
To supplementary: OS curve (immature, and the numbers sit in the text), waterfall plot (duplicates response data), full AE table by system organ class, sensitivity analyses table.
Figure 2 legend:
“Figure 2. Kaplan–Meier estimates of progression-free survival (PFS) in the intention-to-treat population (n = 402). Tick marks indicate censored observations. The number of patients at risk is shown below the x-axis. The hazard ratio and 95% confidence interval were estimated from a Cox proportional hazards model stratified by region and number of prior lines of therapy; the p value is from a stratified log-rank test. CI, confidence interval; HR, hazard ratio.”
The two judgement calls:
1. OS goes to supplementary even though it is a key secondary endpoint. It is immature and non-significant, so the curve shows little; the numbers belong in the Results text where they are still fully reported. This is not selective reporting — the result is stated in full.
2. The waterfall plot is cut despite being the most visually appealing item on the list. It presents response data already given as a percentage. Attractiveness is not a selection criterion.
Figure 1 — Participant flow. Required by CONSORT; carries every denominator.
Table 1 — Baseline characteristics, trimmed to prognostic and stratification variables. Full 31 to supplementary.
Figure 2 — PFS Kaplan–Meier. The primary endpoint; a curve shows shape a median cannot.
Figure 3 — Forest plot of the six prespecified subgroups. Consistency of effect, in one item.
Table 2 — Grade ≥3 adverse events by preferred term. The safety information a clinician acts on.
To supplementary: OS curve (immature, and the numbers sit in the text), waterfall plot (duplicates response data), full AE table by system organ class, sensitivity analyses table.
Figure 2 legend:
“Figure 2. Kaplan–Meier estimates of progression-free survival (PFS) in the intention-to-treat population (n = 402). Tick marks indicate censored observations. The number of patients at risk is shown below the x-axis. The hazard ratio and 95% confidence interval were estimated from a Cox proportional hazards model stratified by region and number of prior lines of therapy; the p value is from a stratified log-rank test. CI, confidence interval; HR, hazard ratio.”
The two judgement calls:
1. OS goes to supplementary even though it is a key secondary endpoint. It is immature and non-significant, so the curve shows little; the numbers belong in the Results text where they are still fully reported. This is not selective reporting — the result is stated in full.
2. The waterfall plot is cut despite being the most visually appealing item on the list. It presents response data already given as a percentage. Attractiveness is not a selection criterion.
Reflection
Take the last paper you wrote. Could any of its tables have been a sentence? Could any two have been one?
Next module
2.3 — Writing the numbers
Week 2 · Module 3
Writing the numbers
Baseline, efficacy and safety — what to state, what to omit, and the reporting conventions that decide whether a draft passes quality control first time
Your scenario
Quality control returns fourteen findings on your Results section. None concerns the science. A percentage without a denominator; decimals that do not match between arms; a hazard ratio with no confidence interval; and p = 0.000, which is not a possible value. Two days of rework on things that take five seconds each to get right.
The baseline paragraph
Three or four sentences. Its job is to let the reader judge who was studied and whether the arms were comparable — not to recite Table 1.
| Include | Omit |
|---|---|
| Numbers randomised, treated and analysed | Every variable in the table |
| Age, sex, and the two or three strongest prognostic factors | Variables with no prognostic relevance |
| Stratification factors | P values comparing arms — CONSORT discourages them |
| Any visible imbalance in a prognostic factor | The phrase well balanced without evidence |
| Median follow-up with its IQR | Restating what Table 1 already shows |
Baseline p values
In a randomised trial, any imbalance is by definition due to chance, so testing for it answers a question with a known answer. CONSORT explicitly discourages the practice.
Judge imbalance by clinical importance instead. If a strong prognostic factor is visibly imbalanced, the response is a prespecified adjusted analysis, not a significance test.
Judge imbalance by clinical importance instead. If a strong prognostic factor is visibly imbalanced, the response is a prespecified adjusted analysis, not a significance test.
The efficacy paragraph
| Element | Form |
|---|---|
| Median follow-up first | At a median follow-up of 19.8 months (IQR 16.2–24.1)… |
| Both arms, matched precision | 8.9 months versus 5.2 months — never 8.9 versus 5 |
| Effect estimate with interval | HR 0.66 (95% CI 0.53–0.82) — CONSORT item 17a |
| Exact p value | p = 0.009; p < 0.001 below that; never p = 0.000 |
| Test named where not obvious | stratified log-rank |
| Hierarchy position where relevant | nominal p = 0.003; formal testing stopped at overall survival |
The safety paragraph
| Rule | Why |
|---|---|
| Use the treated population as denominator | You are attributing events to actual exposure, not to randomisation |
| Give exposure duration | Rates are uninterpretable without time on treatment |
| Any-grade, then grade ≥3, then serious, then deaths, then discontinuations | Standard order; reviewers look for it |
| Counts with denominators throughout | 147/389 (37.8%) |
| Name the specific events driving the difference | “Higher rate of grade 3 events” without saying which is not actionable |
| Report deaths explicitly, with attribution | Never buried in a table alone |
The convention checklist
Run this over every Results section before it leaves your desk:
· Every percentage has a denominator
· Every estimate has an interval
· Decimal precision matches across arms
· Leading zeros present (
· Confidence level stated (
· En dash in ranges; the word to when a limit is negative
· Safety denominator is the treated population
· No
· Median follow-up reported
Nine checks, about five minutes, and it removes most quality control findings.
· Every percentage has a denominator
· Every estimate has an interval
· Decimal precision matches across arms
· Leading zeros present (
0.66, not .66)· Confidence level stated (
95% CI, not CI)· En dash in ranges; the word to when a limit is negative
· Safety denominator is the treated population
· No
p = 0.000· Median follow-up reported
Nine checks, about five minutes, and it removes most quality control findings.
Drills
Drill 1 of 3
Which baseline sentence is best?
A“Baseline characteristics were well balanced between arms (all p > 0.05).”
B“Baseline characteristics are shown in Table 1: age, sex, ECOG, stage, histology, smoking status, prior lines…”
C“See Table 1 for baseline characteristics.”
D“Median age was 62 years (IQR 55–69) and 236/402 patients (58.7%) were male. Biomarker-positive status, a stratification factor, was present in 252/402 (62.7%). Median follow-up was 19.8 months (IQR 16.2–24.1).”
Drill 2 of 3
Which safety sentence is correct?
B0“Grade ≥3 events occurred in 147/402 patients (36.6%).”
B“Grade ≥3 treatment-emergent adverse events occurred in 147/389 patients (37.8%) versus 121/388 (31.2%), most commonly neutropenia (12.1% vs 4.6%) and fatigue (8.5% vs 6.4%).”
C“Safety was favourable, with grade ≥3 events in 37.8%.”
D“Grade ≥3 events were higher in the treatment arm (37.8% vs 31.2%).”
Drill 3 of 3
Which efficacy sentence passes quality control?
A“Median PFS was 8.9 vs 5 months (HR .66, CI 0.53-0.82, p=0.000).”
C“Median PFS was 8.9 months versus 5.2 months (HR 0.66, 95% CI 0.53–0.82; stratified log-rank p < 0.001).”
B“PFS was significantly longer with the study regimen (p < 0.05).”
D“Median PFS was 8.9 versus 5.2 months (HR 0.6612, 95% CI 0.5341–0.8187).”
Production drill
Fix every reporting error. There are twelve.
“Baseline characteristics were well balanced (all p > 0.05). 402 patients were randomised. Median PFS was 8.9 vs 5 months (HR .66, CI 0.53-0.82, p=0.000). Objective response was achieved by 46.75% of patients. Grade 3 or higher events occurred in 36.6% of the 402 randomised patients. The mean change in score was 3.1374 (95% CI -4.4-10.6). Discontinuation occurred in 31 patients. The result was significant at p<0.05.”
Denominators: 402 randomised, 389 treated. Median follow-up 19.8 months (IQR 16.2–24.1).
Denominators: 402 randomised, 389 treated. Median follow-up 19.8 months (IQR 16.2–24.1).
Corrected:
“In total, 402 patients were randomised and 389 received at least one dose. Baseline characteristics were comparable between arms (Table 1). At a median follow-up of 19.8 months (IQR 16.2–24.1), median PFS was 8.9 months versus 5.2 months (HR 0.66, 95% CI 0.53–0.82; p < 0.001). Objective response was achieved by 188/402 patients (46.8%). Grade 3 or higher events occurred in 147/389 patients (37.8%). The mean change in score was 3.1 (95% CI −4.4 to 10.6). Discontinuation occurred in 31/389 patients (8.0%).”
The twelve:
1. Baseline p values removed — discouraged by CONSORT.
2. Well balanced softened to comparable, with the table cited.
3. Sentence opened on a numeral — recast.
4. Median follow-up was missing entirely — added, and placed first.
5. 5 months → 5.2 months, matching decimal precision.
6. .66 → 0.66.
7. CI → 95% CI.
8. Hyphen → en dash in the interval.
9. p=0.000 → p < 0.001, italic, with spaces.
10. 46.75% → one decimal, with count and denominator.
11. Safety denominator corrected from 402 to 389, changing 36.6% to 37.8%.
12. 3.1374 → 3.1; negative interval limit takes to, not a dash; discontinuations given with denominator; the redundant significant at p<0.05 deleted.
Item 11 is the one that matters. The others are conventions. That one is a factual error that understates the adverse event rate in a published paper.
“In total, 402 patients were randomised and 389 received at least one dose. Baseline characteristics were comparable between arms (Table 1). At a median follow-up of 19.8 months (IQR 16.2–24.1), median PFS was 8.9 months versus 5.2 months (HR 0.66, 95% CI 0.53–0.82; p < 0.001). Objective response was achieved by 188/402 patients (46.8%). Grade 3 or higher events occurred in 147/389 patients (37.8%). The mean change in score was 3.1 (95% CI −4.4 to 10.6). Discontinuation occurred in 31/389 patients (8.0%).”
The twelve:
1. Baseline p values removed — discouraged by CONSORT.
2. Well balanced softened to comparable, with the table cited.
3. Sentence opened on a numeral — recast.
4. Median follow-up was missing entirely — added, and placed first.
5. 5 months → 5.2 months, matching decimal precision.
6. .66 → 0.66.
7. CI → 95% CI.
8. Hyphen → en dash in the interval.
9. p=0.000 → p < 0.001, italic, with spaces.
10. 46.75% → one decimal, with count and denominator.
11. Safety denominator corrected from 402 to 389, changing 36.6% to 37.8%.
12. 3.1374 → 3.1; negative interval limit takes to, not a dash; discontinuations given with denominator; the redundant significant at p<0.05 deleted.
Item 11 is the one that matters. The others are conventions. That one is a factual error that understates the adverse event rate in a published paper.
Reflection
Run the nine-point convention checklist over your last Results section. How many findings?
Next module
2.4 — The Results clinic
Week 2 · Module 4
The Results clinic
A full output pack and a blank page. Everything from Week 2 applied at once, then scored against a rubric
Your scenario
This is the task an agency assessment sets. Not questions about Results sections — an output pack and forty-five minutes.
How to do this properly
Set a timer for 45 minutes. Write the full section as you would submit it. Do not annotate, do not list decisions, do not read the rubric first. Reading the rubric first turns this into a comprehension exercise and measures nothing.
The output pack
Everything you need is here. Every number in your section should trace to a line below.
Design: randomised, double-blind, placebo-controlled Phase III, second-line advanced disease. Stratified by region and prior lines. SAP hierarchy: PFS → OS → ORR → PRO. Safety in treated population.
Disposition: 486 screened; 84 excluded (61 criteria, 23 declined); 402 randomised (201/201); 389 treated (196/193); discontinued 42 vs 68.
Baseline: median age 62 (IQR 55–69); male 236/402 (58.7%); ECOG 0 in 161 (40.0%), 1 in 215 (53.5%), 2 in 26 (6.5%); biomarker-positive 252/402 (62.7%) — a stratification factor.
Follow-up: median 19.8 months (IQR 16.2–24.1).
PFS (ITT, n = 402): median 8.9 (95% CI 7.4–10.3) vs 5.2 (95% CI 4.4–6.1) months; HR 0.66 (95% CI 0.53–0.82); stratified log-rank p < 0.001. Schoenfeld PH test p = 0.44.
OS (ITT): not reached vs 21.7 months; HR 0.84 (95% CI 0.62–1.14); p = 0.26. At risk at 24 months: 41 vs 22.
ORR (n = 371 with post-baseline scan): 44% vs 29%; p = 0.003.
PRO change: −5.2 vs −1.8; p = 0.02. Published MCID = 8 points.
Subgroups (6 prespecified): only biomarker status shows interaction (p = 0.03); HR 0.58 (0.44–0.77) positive, 0.94 (0.66–1.34) negative.
Safety (n = 389): any-grade TEAE 371 (95.4%) vs 352 (90.7%); grade ≥3 147 (37.8%) vs 121 (31.2%); serious 68 (17.5%) vs 54 (13.9%); deaths on treatment 4 (1.0%) vs 3 (0.8%), none attributed to study drug; discontinuation for AE 31 (8.0%) vs 22 (5.7%). Commonest grade ≥3: neutropenia 12.1% vs 4.6%; fatigue 8.5% vs 6.4%.
Disposition: 486 screened; 84 excluded (61 criteria, 23 declined); 402 randomised (201/201); 389 treated (196/193); discontinued 42 vs 68.
Baseline: median age 62 (IQR 55–69); male 236/402 (58.7%); ECOG 0 in 161 (40.0%), 1 in 215 (53.5%), 2 in 26 (6.5%); biomarker-positive 252/402 (62.7%) — a stratification factor.
Follow-up: median 19.8 months (IQR 16.2–24.1).
PFS (ITT, n = 402): median 8.9 (95% CI 7.4–10.3) vs 5.2 (95% CI 4.4–6.1) months; HR 0.66 (95% CI 0.53–0.82); stratified log-rank p < 0.001. Schoenfeld PH test p = 0.44.
OS (ITT): not reached vs 21.7 months; HR 0.84 (95% CI 0.62–1.14); p = 0.26. At risk at 24 months: 41 vs 22.
ORR (n = 371 with post-baseline scan): 44% vs 29%; p = 0.003.
PRO change: −5.2 vs −1.8; p = 0.02. Published MCID = 8 points.
Subgroups (6 prespecified): only biomarker status shows interaction (p = 0.03); HR 0.58 (0.44–0.77) positive, 0.94 (0.66–1.34) negative.
Safety (n = 389): any-grade TEAE 371 (95.4%) vs 352 (90.7%); grade ≥3 147 (37.8%) vs 121 (31.2%); serious 68 (17.5%) vs 54 (13.9%); deaths on treatment 4 (1.0%) vs 3 (0.8%), none attributed to study drug; discontinuation for AE 31 (8.0%) vs 22 (5.7%). Commonest grade ≥3: neutropenia 12.1% vs 4.6%; fatigue 8.5% vs 6.4%.
Model Results section
“Of 486 patients screened, 84 were excluded and 402 were randomised, 201 to each arm; 389 received at least one dose (196 and 193). Forty-two patients in the study regimen arm and 68 in the placebo arm discontinued treatment (Figure 1). Median age was 62 years (IQR 55–69), 236/402 patients (58.7%) were male, and 252/402 (62.7%) were biomarker-positive, a stratification factor. Baseline characteristics were comparable between arms (Table 1).
At a median follow-up of 19.8 months (IQR 16.2–24.1), median progression-free survival was 8.9 months (95% CI 7.4–10.3) with the study regimen and 5.2 months (95% CI 4.4–6.1) with placebo (HR 0.66, 95% CI 0.53–0.82; stratified log-rank p < 0.001; Figure 2). The proportional hazards assumption was supported by testing of Schoenfeld residuals (p = 0.44).
Median overall survival was not reached with the study regimen and was 21.7 months with placebo (HR 0.84, 95% CI 0.62–1.14; p = 0.26), with 41 and 22 patients at risk at 24 months. As overall survival did not meet the prespecified significance threshold, formal testing stopped at this endpoint and subsequent endpoints are reported descriptively. Objective response rate, assessed in the 371 patients with at least one post-baseline assessment, was 44% versus 29% (nominal p = 0.003). The mean change from baseline in patient-reported outcome score was −5.2 versus −1.8 (nominal p = 0.02); the between-group difference of 3.4 points was below the published minimal clinically important difference of 8 points.
Among the six prespecified subgroups, a significant interaction was observed by biomarker status (interaction p = 0.03): the hazard ratio was 0.58 (95% CI 0.44–0.77) in biomarker-positive patients and 0.94 (95% CI 0.66–1.34) in biomarker-negative patients (Figure 3).
In the safety population (n = 389), treatment-emergent adverse events of any grade occurred in 371/389 patients (95.4%) versus 352/388 (90.7%), and grade 3 or higher events in 147/389 (37.8%) versus 121/388 (31.2%), most commonly neutropenia (12.1% vs 4.6%) and fatigue (8.5% vs 6.4%). Serious adverse events occurred in 68/389 (17.5%) versus 54/388 (13.9%). Four deaths (1.0%) and three deaths (0.8%) occurred on treatment; none was attributed to study drug. Adverse events led to discontinuation in 31/389 patients (8.0%) versus 22/388 (5.7%) (Table 2).”
Rubric — 18 points, one each
Sequence and structure (4)
1. Participant flow opens the section
2. Baseline second, selective rather than exhaustive
3. Primary endpoint third
4. Safety last
Discipline (4)
5. No interpretation anywhere — no suggests, importantly, encouragingly
6. No comparison with prior literature
7. No baseline p values
8. Every display item cross-referenced
Statistical handling (5)
9. Median follow-up stated, and stated first
10. Effect estimates carry 95% confidence intervals
11. Hierarchy break at OS stated; ORR and PRO marked nominal
12. PRO difference compared against the MCID of 8 points
13. Proportional hazards assumption addressed
Denominators and conventions (5)
14. Safety uses n = 389/388, not 402
15. ORR uses n = 371, with the reason stated
16. Every percentage carries its count and denominator
17. Decimal precision matched; leading zeros; en dashes; no p = 0.000
18. Deaths reported explicitly with attribution
Scoring
15–18 — publishable. You would pass an agency Results assessment.
11–14 — close. Identify which of the four blocks your misses cluster in and redo that module.
Under 11 — work through 2.1 to 2.3 again before retrying. Do not repeat this immediately.
The four that separate people: items 11, 12, 14 and 15. Everyone reports the numbers. Far fewer notice that the hierarchy broke at OS, that a statistically significant PRO result sits below its MCID, and that ORR and safety each need a different denominator from the primary analysis.
“Of 486 patients screened, 84 were excluded and 402 were randomised, 201 to each arm; 389 received at least one dose (196 and 193). Forty-two patients in the study regimen arm and 68 in the placebo arm discontinued treatment (Figure 1). Median age was 62 years (IQR 55–69), 236/402 patients (58.7%) were male, and 252/402 (62.7%) were biomarker-positive, a stratification factor. Baseline characteristics were comparable between arms (Table 1).
At a median follow-up of 19.8 months (IQR 16.2–24.1), median progression-free survival was 8.9 months (95% CI 7.4–10.3) with the study regimen and 5.2 months (95% CI 4.4–6.1) with placebo (HR 0.66, 95% CI 0.53–0.82; stratified log-rank p < 0.001; Figure 2). The proportional hazards assumption was supported by testing of Schoenfeld residuals (p = 0.44).
Median overall survival was not reached with the study regimen and was 21.7 months with placebo (HR 0.84, 95% CI 0.62–1.14; p = 0.26), with 41 and 22 patients at risk at 24 months. As overall survival did not meet the prespecified significance threshold, formal testing stopped at this endpoint and subsequent endpoints are reported descriptively. Objective response rate, assessed in the 371 patients with at least one post-baseline assessment, was 44% versus 29% (nominal p = 0.003). The mean change from baseline in patient-reported outcome score was −5.2 versus −1.8 (nominal p = 0.02); the between-group difference of 3.4 points was below the published minimal clinically important difference of 8 points.
Among the six prespecified subgroups, a significant interaction was observed by biomarker status (interaction p = 0.03): the hazard ratio was 0.58 (95% CI 0.44–0.77) in biomarker-positive patients and 0.94 (95% CI 0.66–1.34) in biomarker-negative patients (Figure 3).
In the safety population (n = 389), treatment-emergent adverse events of any grade occurred in 371/389 patients (95.4%) versus 352/388 (90.7%), and grade 3 or higher events in 147/389 (37.8%) versus 121/388 (31.2%), most commonly neutropenia (12.1% vs 4.6%) and fatigue (8.5% vs 6.4%). Serious adverse events occurred in 68/389 (17.5%) versus 54/388 (13.9%). Four deaths (1.0%) and three deaths (0.8%) occurred on treatment; none was attributed to study drug. Adverse events led to discontinuation in 31/389 patients (8.0%) versus 22/388 (5.7%) (Table 2).”
Rubric — 18 points, one each
Sequence and structure (4)
1. Participant flow opens the section
2. Baseline second, selective rather than exhaustive
3. Primary endpoint third
4. Safety last
Discipline (4)
5. No interpretation anywhere — no suggests, importantly, encouragingly
6. No comparison with prior literature
7. No baseline p values
8. Every display item cross-referenced
Statistical handling (5)
9. Median follow-up stated, and stated first
10. Effect estimates carry 95% confidence intervals
11. Hierarchy break at OS stated; ORR and PRO marked nominal
12. PRO difference compared against the MCID of 8 points
13. Proportional hazards assumption addressed
Denominators and conventions (5)
14. Safety uses n = 389/388, not 402
15. ORR uses n = 371, with the reason stated
16. Every percentage carries its count and denominator
17. Decimal precision matched; leading zeros; en dashes; no p = 0.000
18. Deaths reported explicitly with attribution
Scoring
15–18 — publishable. You would pass an agency Results assessment.
11–14 — close. Identify which of the four blocks your misses cluster in and redo that module.
Under 11 — work through 2.1 to 2.3 again before retrying. Do not repeat this immediately.
The four that separate people: items 11, 12, 14 and 15. Everyone reports the numbers. Far fewer notice that the hierarchy broke at OS, that a statistically significant PRO result sits below its MCID, and that ORR and safety each need a different denominator from the primary analysis.
Reflection
Which of the four rubric blocks did you lose most points in?
Next week
3.1 — Discussion architecture
Week 3 · Module 1
Discussion architecture
Five moves in a fixed order. Reviewers read for them, and a Discussion that omits one is the most common reason a sound paper reads as unconvincing
Your scenario
Your Discussion runs to 1,400 words. It opens with three paragraphs reviewing the literature, restates every result from the Results section, mentions limitations in a single sentence near the end, and concludes that further research is needed. A reviewer describes it as unfocused. Every individual sentence is defensible. The section as a whole argues nothing.
The five moves
Figure 3.1 — The five moves. The order is not stylistic preference; it is what reviewers scan for. Move 1 missing is the most common structural fault: a Discussion that opens with prior literature leaves the reader unsure what the paper actually found.
Move 1 is the one people skip
Open the Discussion with what you found, in one or two sentences, in the past tense, without citations. Not a restatement of the objective. Not a paragraph of context. Not the literature.
“In this randomised trial of previously treated advanced disease, the study regimen prolonged progression-free survival compared with placebo; overall survival data were immature at this analysis.”
That sentence orients the reader for everything that follows. Its absence is why so many Discussions feel like they are circling something.
“In this randomised trial of previously treated advanced disease, the study regimen prolonged progression-free survival compared with placebo; overall survival data were immature at this analysis.”
That sentence orients the reader for everything that follows. Its absence is why so many Discussions feel like they are circling something.
Three things that do not belong
New results. Anything appearing for the first time in the Discussion will be flagged. All numbers belong in Results.
A full restatement of Results. Refer to findings; do not re-report them with their intervals and p values.
A literature review. Move 2 compares your finding with the two or three most relevant studies. It is not a survey of the field — that belongs in the Introduction, briefly, or in a review article.
A full restatement of Results. Refer to findings; do not re-report them with their intervals and p values.
A literature review. Move 2 compares your finding with the two or three most relevant studies. It is not a survey of the field — that belongs in the Introduction, briefly, or in a review article.
Drills
Drill 1 of 3
Which sentence correctly opens a Discussion?
A“Advanced disease remains a leading cause of mortality worldwide, and treatment options are limited.”
B“The aim of this study was to evaluate the study regimen in second-line advanced disease.”
C“In this randomised trial, the study regimen prolonged progression-free survival compared with placebo; overall survival data were immature.”
D“This study has several limitations that should be considered.”
Drill 2 of 3
Where do limitations belong?
AFirst, to demonstrate objectivity
BLast, as the final paragraph
CIn supplementary material
DFourth, after interpretation and before the conclusion
Drill 3 of 3
Your Discussion is 1,400 words against a target of 750. Where do you cut first?
AThe literature comparison, and any restatement of Results
BLimitations
CThe principal finding
DThe conclusion
Production drill
This Discussion opening has three structural faults. Identify them and rewrite the first paragraph.
“Advanced disease affects over two million people annually and remains a leading cause of cancer death. Standard second-line therapy has changed little in a decade, and several classes have been investigated with mixed results. Smith et al. reported a response rate of 31% with a comparable agent, while Kumar et al. found no benefit in an unselected population. Against this background, our finding of a median PFS of 8.9 months versus 5.2 months (HR 0.66, 95% CI 0.53–0.82, p < 0.001) is notable. This study has some limitations.”
The three faults:
1. Move 1 is missing. Three sentences of context and prior literature arrive before the reader learns what this study found. The finding is buried in sentence four.
2. Results are restated in full. The median, hazard ratio, interval and p value all reappear. The Discussion refers to findings; it does not re-report them.
3. Limitations are announced generically and immediately, in a single contentless sentence, at the wrong point in the sequence.
Rewritten:
“In this randomised trial of previously treated advanced disease, the study regimen significantly prolonged progression-free survival compared with placebo, with a safety profile consistent with its class; overall survival data were immature at this analysis.
This finding is consistent with the response rate reported by Smith et al. in a comparable agent, and contrasts with the absence of benefit reported by Kumar et al., whose population was unselected for biomarker status — a difference our subgroup analysis may help explain.”
What changed: the finding now opens the section, the numbers are gone (they are in Results), and the two prior studies are used to make an argument rather than to demonstrate reading. The word notable has also gone; if a finding is notable, the reader will decide that.
1. Move 1 is missing. Three sentences of context and prior literature arrive before the reader learns what this study found. The finding is buried in sentence four.
2. Results are restated in full. The median, hazard ratio, interval and p value all reappear. The Discussion refers to findings; it does not re-report them.
3. Limitations are announced generically and immediately, in a single contentless sentence, at the wrong point in the sequence.
Rewritten:
“In this randomised trial of previously treated advanced disease, the study regimen significantly prolonged progression-free survival compared with placebo, with a safety profile consistent with its class; overall survival data were immature at this analysis.
This finding is consistent with the response rate reported by Smith et al. in a comparable agent, and contrasts with the absence of benefit reported by Kumar et al., whose population was unselected for biomarker status — a difference our subgroup analysis may help explain.”
What changed: the finding now opens the section, the numbers are gone (they are in Results), and the two prior studies are used to make an argument rather than to demonstrate reading. The word notable has also gone; if a finding is notable, the reader will decide that.
Reflection
Open the Discussion of your last paper. Does the first sentence state what you found?
In most drafts it does not. It is the single highest-return fix in the whole section.
Next module
3.2 — Interpretation without overclaim
Week 3 · Module 2
Interpretation without overclaim
The design determines which verbs are available to you. Choosing one your data cannot support is the most consequential error in the whole manuscript
Your scenario
A reviewer writes: “The authors appear to present a hypothesis-generating subgroup observation as a basis for changing clinical practice.” You never intended that. But your Discussion said the biomarker identifies patients who should receive the regimen, and those two verbs made a recommendation the design cannot support.
What each design licenses
| Design | Strongest available verb | Never available |
|---|---|---|
| Adequately powered RCT, prespecified primary endpoint met | demonstrated · improved · reduced | proved · confirmed from one trial |
| RCT, secondary or underpowered endpoint | indicated · suggested | Causal verbs stated without qualification |
| Single-arm study | showed activity | superior · better than — there is no comparator |
| Prospective observational cohort | was associated with | reduced · prevented · caused |
| Retrospective cohort or database study | was associated with, plus a residual confounding statement | Any causal verb; any practice recommendation |
| Subgroup analysis, significant interaction | may · appeared to differ | identifies · should receive · predicts |
| Subgroup analysis, no significant interaction | Report consistency only | Any claim of a differential effect |
Association is not causation, and the verb must say so
Causal — requires randomisation: reduced, caused, led to, resulted in, improved, prevented, protected against.
Associational — safe for observational data: was associated with, correlated with, occurred more frequently in, was accompanied by.
Writing “the regimen reduced mortality” from a retrospective cohort is not a stylistic slip. It is a claim the data cannot support, and it is the most common reason promotional and regulatory material is rejected at review.
Associational — safe for observational data: was associated with, correlated with, occurred more frequently in, was accompanied by.
Writing “the regimen reduced mortality” from a retrospective cohort is not a stylistic slip. It is a claim the data cannot support, and it is the most common reason promotional and regulatory material is rejected at review.
The four hedging tools
Precise writing combines these rather than stacking one repeatedly.
| Tool | Examples | What it qualifies |
|---|---|---|
| Modal verbs | may, might, could | Whether the claim holds at all |
| Reporting verbs | suggests, indicates, appears to | Strength of inference from data to conclusion |
| Scope limiters | in this cohort, in second-line disease, at 24 weeks | Population, setting or timepoint covered |
| Quantifiers | numerically higher, a proportion of | How much and how many |
You can also under-claim
Calibration works in both directions. If an adequately powered randomised trial met its prespecified primary endpoint, writing “these data may suggest a possible benefit” understates evidence that clinicians need in order to act.
Excessive hedging also reads as a lack of confidence, and in an assessment it suggests you cannot tell strong evidence from weak. Match the language to the design. That is the entire skill.
Excessive hedging also reads as a lack of confidence, and in an assessment it suggests you cannot tell strong evidence from weak. Match the language to the design. That is the entire skill.
Drills
Drill 1 of 3
A subgroup analysis shows a significant interaction (p = 0.03) in one of six prespecified subgroups. Which sentence is defensible?
A“Biomarker status identifies patients who should receive the regimen.”
B“A significant interaction by biomarker status was observed; the trial was not powered for subgroup comparisons, and this finding may warrant prospective evaluation.”
C“Biomarker status predicts response to the regimen.”
DOmit the subgroup analysis
Drill 2 of 3
A large, double-blind randomised trial met its prespecified primary endpoint of overall survival. Which is correctly calibrated?
A“The regimen may be associated with a possible survival benefit.”
B“This trial proves the regimen prolongs life.”
C“The regimen improved overall survival; these data demonstrate a benefit in previously treated advanced disease.”
D“The regimen was associated with longer survival.”
Drill 3 of 3
A statistically significant patient-reported outcome difference of 3.4 points sits below the published MCID of 8. What does the Discussion say?
A“Quality of life was significantly improved.”
BOmit it, since it is not clinically meaningful
C“There was a trend towards improved quality of life.”
D“Although the between-group difference reached statistical significance, it was below the published minimal clinically important difference of 8 points.”
Production drill
Recalibrate every claim. The study is a retrospective cohort of 118 patients, unadjusted for treatment-selection bias.
“Our analysis proves that early initiation definitely reduces mortality and improves quality of life. These results confirm that early treatment should be adopted as standard of care. Early treatment prevents disease progression and protects against hospitalisation. We have demonstrated that this benefit is causal, and these findings establish early initiation as the preferred strategy for all patients.”
Model rewrite:
“In this retrospective analysis of 118 patients, earlier initiation of therapy was associated with lower mortality and with higher patient-reported quality-of-life scores. Earlier treatment was also associated with a lower rate of progression and with fewer hospitalisations. Because the analysis was retrospective and unadjusted for treatment-selection bias, a causal relationship cannot be inferred. These findings suggest that the approach may warrant prospective evaluation; confirmation in a randomised setting is required before any change in practice.”
Every change:
1. proves → was associated with. A retrospective design licenses association language only. This is the fix that matters.
2. definitely reduces → causal verb and intensifier both removed.
3. improves quality of life → tied to what was actually measured, patient-reported scores.
4. confirm … standard of care → suggest … may warrant prospective evaluation. A chart review cannot change standard of care.
5. prevents and protects against → both causal; replaced.
6. demonstrated that this benefit is causal → replaced with the correct opposite statement, naming the specific bias.
7. for all patients → removed; scope limited to the analysed cohort.
Score yourself: if you softened the tone but left proves, prevents or protects standing, you fixed the cosmetic problem and left the substantive one. Those three verbs are the test.
“In this retrospective analysis of 118 patients, earlier initiation of therapy was associated with lower mortality and with higher patient-reported quality-of-life scores. Earlier treatment was also associated with a lower rate of progression and with fewer hospitalisations. Because the analysis was retrospective and unadjusted for treatment-selection bias, a causal relationship cannot be inferred. These findings suggest that the approach may warrant prospective evaluation; confirmation in a randomised setting is required before any change in practice.”
Every change:
1. proves → was associated with. A retrospective design licenses association language only. This is the fix that matters.
2. definitely reduces → causal verb and intensifier both removed.
3. improves quality of life → tied to what was actually measured, patient-reported scores.
4. confirm … standard of care → suggest … may warrant prospective evaluation. A chart review cannot change standard of care.
5. prevents and protects against → both causal; replaced.
6. demonstrated that this benefit is causal → replaced with the correct opposite statement, naming the specific bias.
7. for all patients → removed; scope limited to the analysed cohort.
Score yourself: if you softened the tone but left proves, prevents or protects standing, you fixed the cosmetic problem and left the substantive one. Those three verbs are the test.
Reflection
Find a published abstract in your field whose conclusion overclaims relative to its design. Name the design, quote the verb, and rewrite it.
Learning to see this in published work is what makes you useful in a review meeting, and it is a standard interview question for senior roles.
Next module
3.3 — Limitations that strengthen
Week 3 · Module 3
Limitations that strengthen
A limitations paragraph is not an apology and not a disclaimer. Done properly it pre-empts the reviewer, and it is one of the few places where candour makes a paper more persuasive
Your scenario
Reviewer 2 devotes four paragraphs to the possibility that unmeasured performance status confounded your result, and recommends rejection. You knew about it. Your limitations paragraph said “this study has some limitations, including its retrospective design.” Had you named the specific variable and its likely direction, that reviewer would have been arguing with your analysis rather than discovering it.
The three-part structure
Every limitation worth writing has three components. A limitation missing any of them reads as boilerplate.
| Part | Question it answers | Example |
|---|---|---|
| The threat | What specifically is wrong? | Performance status was not recorded in the database. |
| The consequence | What does it do to the result, and in which direction? | Fitter patients may preferentially have received the study regimen, biasing the estimate away from the null. |
| The response | What did you do, or what should be done? | A sensitivity analysis restricted to patients with recorded ECOG gave consistent results; randomised confirmation remains necessary. |
Boilerplate
Useful
This study has some limitations.names nothing
Performance status was unrecorded and may have influenced treatment selection.
The sample size was small.no consequence given
With 118 patients, the confidence interval does not exclude a clinically relevant difference in either direction.
Limitations are inherent to all research.contentless
The single-centre design limits generalisability to settings with different referral patterns.
Further research is needed.true of everything
A randomised trial powered for overall survival is required before practice changes.
Name three, and name them specifically
Three well-constructed limitations are worth more than eight vague ones. Choose the three a sceptical reader would raise first, and address those.
The most persuasive move available to you is naming the specific unmeasured variable rather than writing a generic residual confounding sentence. A reviewer who sees you have identified the same problem they were about to raise reads the rest of the paper differently.
The most persuasive move available to you is naming the specific unmeasured variable rather than writing a generic residual confounding sentence. A reviewer who sees you have identified the same problem they were about to raise reads the rest of the paper differently.
Do not over-correct
A limitations paragraph that undermines the study is as unhelpful as one that hides its flaws. If your randomised trial met its primary endpoint, the limitations do not include the results may not be real.
State the genuine threats, their direction, and what mitigates them. Then stop. The conclusion follows the limitations, and it should still be able to say something.
State the genuine threats, their direction, and what mitigates them. Then stop. The conclusion follows the limitations, and it should still be able to say something.
Drills
Drill 1 of 3
Which limitation is written well?
A“Performance status was not captured in the database and may have influenced treatment selection; fitter patients may preferentially have received the study regimen, biasing the estimate away from the null.”
B“This study has several limitations.”
C“The sample size was relatively small.”
D“As with all observational research, causality cannot be established.”
Drill 2 of 3
You know a variable that likely confounded your result was not measured. What is the strongest approach?
AOmit it — reviewers may not notice
BName the variable, state its likely direction of effect, and describe any sensitivity analysis
CMention residual confounding in general terms
DArgue that it would not have changed the conclusion
Drill 3 of 3
How many limitations should a typical Discussion carry?
AOne, to avoid weakening the paper
BEight or more, for completeness
CAbout three, each fully constructed
DNone, if the study was well conducted
Production drill
Write the limitations paragraph for this study, three limitations, each fully constructed.
Study: randomised, double-blind, placebo-controlled Phase III; 402 patients; second-line advanced disease.
Findings: primary PFS endpoint met (HR 0.66). OS immature and non-significant (HR 0.84, p = 0.26); 41 and 22 patients at risk at 24 months. Significant biomarker interaction (p = 0.03) among six prespecified subgroups; trial not powered for interaction.
Also: enrolment restricted to ECOG 0–2; median follow-up 19.8 months; 12% of patients had no evaluable biomarker sample.
Findings: primary PFS endpoint met (HR 0.66). OS immature and non-significant (HR 0.84, p = 0.26); 41 and 22 patients at risk at 24 months. Significant biomarker interaction (p = 0.03) among six prespecified subgroups; trial not powered for interaction.
Also: enrolment restricted to ECOG 0–2; median follow-up 19.8 months; 12% of patients had no evaluable biomarker sample.
Model paragraph:
“This study has three principal limitations. First, overall survival data were immature at this analysis, with 41 and 22 patients remaining at risk at 24 months; the confidence interval around the hazard ratio does not exclude either a clinically relevant survival benefit or none, and longer follow-up is required before the effect on survival can be characterised. Second, the interaction observed by biomarker status was one of six prespecified subgroup analyses in a trial powered for the overall population rather than for subgroup comparisons; while the interaction was statistically significant and biologically plausible, it should be regarded as hypothesis-generating and requires prospective confirmation. Third, enrolment was restricted to patients with ECOG performance status 0 to 2, and 12% of randomised patients had no evaluable biomarker sample; the findings may therefore not generalise to patients with poorer performance status, and the biomarker analysis applies to a subset of the trial population.”
Check each against the three parts:
Limitation 1 — threat: OS immature, with the numbers at risk quoted. Consequence: interval excludes nothing in either direction. Response: longer follow-up.
Limitation 2 — threat: one of six subgroups, underpowered for interaction. Consequence: hypothesis-generating only. Response: prospective confirmation.
Limitation 3 — threat: restricted eligibility plus 12% missing biomarker data. Consequence: limited generalisability, and the biomarker analysis covers a subset. Response: implicit in the scope statement.
What this paragraph does not do: it does not apologise, it does not say the result may not be real, and it does not end on further research is needed. The primary endpoint was met in a well-conducted randomised trial, and the limitations qualify the reach of that finding without undermining it.
“This study has three principal limitations. First, overall survival data were immature at this analysis, with 41 and 22 patients remaining at risk at 24 months; the confidence interval around the hazard ratio does not exclude either a clinically relevant survival benefit or none, and longer follow-up is required before the effect on survival can be characterised. Second, the interaction observed by biomarker status was one of six prespecified subgroup analyses in a trial powered for the overall population rather than for subgroup comparisons; while the interaction was statistically significant and biologically plausible, it should be regarded as hypothesis-generating and requires prospective confirmation. Third, enrolment was restricted to patients with ECOG performance status 0 to 2, and 12% of randomised patients had no evaluable biomarker sample; the findings may therefore not generalise to patients with poorer performance status, and the biomarker analysis applies to a subset of the trial population.”
Check each against the three parts:
Limitation 1 — threat: OS immature, with the numbers at risk quoted. Consequence: interval excludes nothing in either direction. Response: longer follow-up.
Limitation 2 — threat: one of six subgroups, underpowered for interaction. Consequence: hypothesis-generating only. Response: prospective confirmation.
Limitation 3 — threat: restricted eligibility plus 12% missing biomarker data. Consequence: limited generalisability, and the biomarker analysis covers a subset. Response: implicit in the scope statement.
What this paragraph does not do: it does not apologise, it does not say the result may not be real, and it does not end on further research is needed. The primary endpoint was met in a well-conducted randomised trial, and the limitations qualify the reach of that finding without undermining it.
Reflection
Take your last paper's limitations paragraph. For each limitation, is the threat named, the consequence stated, and a response given?
Most limitations paragraphs contain only threats. Adding the consequence is what converts them from disclaimer to argument.
Next module
3.4 — The Discussion clinic
Week 3 · Module 4
The Discussion clinic
A full Discussion against a word budget, scored on structure, calibration and limitations at once
Your scenario
Same trial as the Results clinic. You have written the Results. Now write the argument — in 750 words, with a reviewer who will check every verb against the design.
How to do this
Timer at 45 minutes. Five moves in order. 750 words. Write the section as you would submit it, then score. Do not read the rubric first.
The brief
Write the full Discussion, including the conclusion.
Design: randomised, double-blind, placebo-controlled Phase III, second-line advanced disease, n = 402. Hierarchy PFS → OS → ORR → PRO.
Primary: median PFS 8.9 vs 5.2 months; HR 0.66 (95% CI 0.53–0.82); p < 0.001. PH assumption supported.
OS: not reached vs 21.7 months; HR 0.84 (0.62–1.14); p = 0.26. At risk at 24 months 41 vs 22.
ORR: 44% vs 29% (nominal p = 0.003).
PRO: −5.2 vs −1.8 (nominal p = 0.02); MCID = 8 points.
Subgroups: biomarker interaction p = 0.03 (HR 0.58 positive, 0.94 negative); one of six prespecified; 12% no evaluable sample.
Safety: grade ≥3 37.8% vs 31.2%; discontinuation for AE 8.0% vs 5.7%; no treatment-related deaths.
Prior literature: Trial A, same class, unselected population, no PFS benefit (HR 0.94). Trial B, biomarker-enriched, PFS HR 0.61 — consistent with the present biomarker-positive subgroup.
Eligibility: ECOG 0–2 only. Median follow-up 19.8 months.
Primary: median PFS 8.9 vs 5.2 months; HR 0.66 (95% CI 0.53–0.82); p < 0.001. PH assumption supported.
OS: not reached vs 21.7 months; HR 0.84 (0.62–1.14); p = 0.26. At risk at 24 months 41 vs 22.
ORR: 44% vs 29% (nominal p = 0.003).
PRO: −5.2 vs −1.8 (nominal p = 0.02); MCID = 8 points.
Subgroups: biomarker interaction p = 0.03 (HR 0.58 positive, 0.94 negative); one of six prespecified; 12% no evaluable sample.
Safety: grade ≥3 37.8% vs 31.2%; discontinuation for AE 8.0% vs 5.7%; no treatment-related deaths.
Prior literature: Trial A, same class, unselected population, no PFS benefit (HR 0.94). Trial B, biomarker-enriched, PFS HR 0.61 — consistent with the present biomarker-positive subgroup.
Eligibility: ECOG 0–2 only. Median follow-up 19.8 months.
Model Discussion
“In this randomised, placebo-controlled trial in previously treated advanced disease, the study regimen significantly prolonged progression-free survival, with a 34% reduction in the risk of progression or death. Overall survival data were immature at this analysis.
These findings extend earlier work in this drug class. Trial A found no progression-free survival benefit in an unselected population, whereas Trial B, which enrolled a biomarker-enriched population, reported a hazard ratio closely comparable to that observed here in biomarker-positive patients. Taken together, these results are consistent with the hypothesis that benefit in this class is concentrated in biomarker-positive disease, and they may help explain the discrepant findings of earlier unselected trials.
The magnitude of the progression-free survival benefit, and the accompanying difference in objective response rate, are consistent with the proposed mechanism of action. The between-group difference in patient-reported outcome score reached nominal significance but was below the published minimal clinically important difference of 8 points, and should not be interpreted as a demonstrated improvement in quality of life. The safety profile was consistent with the class, with a modest excess of grade 3 or higher events and of discontinuation for adverse events, and no treatment-related deaths.
This study has three principal limitations. First, overall survival data were immature, with 41 and 22 patients at risk at 24 months; the confidence interval excludes neither a clinically relevant survival benefit nor the absence of one, and the effect on survival cannot yet be characterised. Second, the interaction observed by biomarker status was one of six prespecified subgroup analyses in a trial powered for the overall population; although statistically significant and biologically plausible, and although it is consistent with Trial B, it remains hypothesis-generating and requires prospective confirmation. Twelve per cent of randomised patients had no evaluable biomarker sample, so the biomarker analysis applies to a subset of the trial population. Third, eligibility was restricted to patients with ECOG performance status 0 to 2, and the findings may not generalise to patients with poorer performance status.
In previously treated advanced disease, the study regimen significantly improved progression-free survival compared with placebo, with a manageable safety profile. Continued follow-up is required to determine its effect on overall survival, and prospective evaluation is needed before biomarker status can be used to direct treatment.”
Rubric — 16 points
Structure (5)
1. Opens with the principal finding, past tense, no citation
2. Prior work compared, not reviewed — two studies used to build an argument
3. Interpretation separated from comparison
4. Limitations fourth, not first and not last
5. Conclusion last, two sentences
Calibration (6)
6. Improved / prolonged used for the primary endpoint — the randomised design licenses it
7. OS described as immature, not as showing no benefit
8. ORR and PRO not called significant — hierarchy broke at OS
9. PRO explicitly compared against the MCID
10. Biomarker finding hedged: hypothesis-generating, needs prospective confirmation
11. No practice recommendation from the subgroup
Limitations (3)
12. Three limitations, each with threat, consequence and response
13. The 12% missing biomarker data named specifically
14. No apologising and no undermining of a met primary endpoint
Discipline (2)
15. No new results introduced
16. No full restatement of Results with intervals and p values
Scoring
13–16 — publishable. This is the standard a senior writer is held to.
9–12 — identify which block you lost points in and redo that module.
Under 9 — work through 3.1 to 3.3 again before retrying.
The three most people miss: item 8, because ORR has a smaller p value than the primary endpoint and the temptation to call it significant is strong; item 9, because a significant result below its MCID feels like good news; and item 11, because the biomarker story is the most interesting thing in the trial and it is the claim the design least supports.
“In this randomised, placebo-controlled trial in previously treated advanced disease, the study regimen significantly prolonged progression-free survival, with a 34% reduction in the risk of progression or death. Overall survival data were immature at this analysis.
These findings extend earlier work in this drug class. Trial A found no progression-free survival benefit in an unselected population, whereas Trial B, which enrolled a biomarker-enriched population, reported a hazard ratio closely comparable to that observed here in biomarker-positive patients. Taken together, these results are consistent with the hypothesis that benefit in this class is concentrated in biomarker-positive disease, and they may help explain the discrepant findings of earlier unselected trials.
The magnitude of the progression-free survival benefit, and the accompanying difference in objective response rate, are consistent with the proposed mechanism of action. The between-group difference in patient-reported outcome score reached nominal significance but was below the published minimal clinically important difference of 8 points, and should not be interpreted as a demonstrated improvement in quality of life. The safety profile was consistent with the class, with a modest excess of grade 3 or higher events and of discontinuation for adverse events, and no treatment-related deaths.
This study has three principal limitations. First, overall survival data were immature, with 41 and 22 patients at risk at 24 months; the confidence interval excludes neither a clinically relevant survival benefit nor the absence of one, and the effect on survival cannot yet be characterised. Second, the interaction observed by biomarker status was one of six prespecified subgroup analyses in a trial powered for the overall population; although statistically significant and biologically plausible, and although it is consistent with Trial B, it remains hypothesis-generating and requires prospective confirmation. Twelve per cent of randomised patients had no evaluable biomarker sample, so the biomarker analysis applies to a subset of the trial population. Third, eligibility was restricted to patients with ECOG performance status 0 to 2, and the findings may not generalise to patients with poorer performance status.
In previously treated advanced disease, the study regimen significantly improved progression-free survival compared with placebo, with a manageable safety profile. Continued follow-up is required to determine its effect on overall survival, and prospective evaluation is needed before biomarker status can be used to direct treatment.”
Rubric — 16 points
Structure (5)
1. Opens with the principal finding, past tense, no citation
2. Prior work compared, not reviewed — two studies used to build an argument
3. Interpretation separated from comparison
4. Limitations fourth, not first and not last
5. Conclusion last, two sentences
Calibration (6)
6. Improved / prolonged used for the primary endpoint — the randomised design licenses it
7. OS described as immature, not as showing no benefit
8. ORR and PRO not called significant — hierarchy broke at OS
9. PRO explicitly compared against the MCID
10. Biomarker finding hedged: hypothesis-generating, needs prospective confirmation
11. No practice recommendation from the subgroup
Limitations (3)
12. Three limitations, each with threat, consequence and response
13. The 12% missing biomarker data named specifically
14. No apologising and no undermining of a met primary endpoint
Discipline (2)
15. No new results introduced
16. No full restatement of Results with intervals and p values
Scoring
13–16 — publishable. This is the standard a senior writer is held to.
9–12 — identify which block you lost points in and redo that module.
Under 9 — work through 3.1 to 3.3 again before retrying.
The three most people miss: item 8, because ORR has a smaller p value than the primary endpoint and the temptation to call it significant is strong; item 9, because a significant result below its MCID feels like good news; and item 11, because the biomarker story is the most interesting thing in the trial and it is the claim the design least supports.
Reflection
Which rubric block cost you most — structure, calibration, limitations or discipline?
Next week
4.1 — The Introduction, written last
Week 4 · Module 1
The Introduction, written last
Three moves, four hundred words, and one job: make the reader believe the study needed to exist. It is taught last because it should be written last
Your scenario
A desk rejection arrives with one sentence: “The rationale for the study is not clearly established.” Your Introduction runs to nine hundred words and cites thirty-one papers. It reviews the field thoroughly and never states, in any single sentence, what was not known before you started.
Why last
An Introduction written first commits you to framing the study around what you hoped to find. Written after Results and Discussion, it can be framed around what you did find — which is the only version that will survive review.
The order that works
Methods → Results → Discussion → Introduction → Conclusion → Abstract → Title.
Methods first because it is largely mechanical and warms you up. Results next because everything downstream depends on it. Introduction late, so the gap you describe is the gap your findings actually address. Abstract and title last, because they compress a finished argument.
Methods first because it is largely mechanical and warms you up. Results next because everything downstream depends on it. Introduction late, so the gap you describe is the gap your findings actually address. Abstract and title last, because they compress a finished argument.
The three moves
| Move | Job | Length and form |
|---|---|---|
| 1. What is known | Establish that the problem matters, to this journal's readers | One paragraph. Epidemiology, current standard of care, why it is inadequate. Present tense. |
| 2. What is not known | Identify the specific gap | One paragraph, ending on a sentence a reader could quote as the gap. This is the paragraph most drafts lack. |
| 3. What this study did | State the objective, matching the primary endpoint | Two or three sentences. Past tense. No results. |
Four faults
No identifiable gap. The commonest cause of desk rejection on rationale. If no sentence says what was unknown, the reader never learns why the study was needed.
A literature review. Thirty citations establishing that the field exists. Cite what supports the gap; nothing else.
Results in the Introduction. The objective states what you set out to do, not what you found.
An objective that does not match the primary endpoint. If the Introduction promises survival and the primary endpoint is progression-free survival, the paper contradicts itself on page one.
A literature review. Thirty citations establishing that the field exists. Cite what supports the gap; nothing else.
Results in the Introduction. The objective states what you set out to do, not what you found.
An objective that does not match the primary endpoint. If the Introduction promises survival and the primary endpoint is progression-free survival, the paper contradicts itself on page one.
Drills
Drill 1 of 3
Which sentence is a properly stated gap?
A“Further research in this area is needed.”
B“Advanced disease remains a major cause of mortality.”
C“We therefore conducted a randomised trial.”
D“Agents in this class have been evaluated in unselected populations with inconsistent results, but their efficacy in biomarker-selected second-line disease has not been assessed in a randomised trial.”
Drill 2 of 3
When should the Introduction be written?
AAfter Results and Discussion
BFirst, to establish direction
CAt the same time as the abstract
DImmediately after Methods
Drill 3 of 3
Your Introduction cites 31 references across 900 words. What should you do?
ANothing — thorough citation demonstrates rigour
BCut to the references that establish the problem and the gap; target 400–600 words
CMove the extra references to a supplementary background section
DChoose a journal with a longer word limit
Production drill
Write the three-move Introduction for the trial you have been working with. Target 420 words.
Context: advanced disease, second line; standard therapy has changed little in a decade; median PFS with standard care around 5 months.
Prior work: Trial A, same class, unselected population, no PFS benefit (HR 0.94). Trial B, biomarker-enriched Phase II, PFS HR 0.61 — promising but uncontrolled for the biomarker question and not randomised.
The gap: no randomised evaluation of this class in second-line disease with prospective biomarker stratification.
This study: randomised, double-blind, placebo-controlled Phase III; primary endpoint PFS; stratified by region and prior lines; biomarker status prespecified as a subgroup.
Prior work: Trial A, same class, unselected population, no PFS benefit (HR 0.94). Trial B, biomarker-enriched Phase II, PFS HR 0.61 — promising but uncontrolled for the biomarker question and not randomised.
The gap: no randomised evaluation of this class in second-line disease with prospective biomarker stratification.
This study: randomised, double-blind, placebo-controlled Phase III; primary endpoint PFS; stratified by region and prior lines; biomarker status prespecified as a subgroup.
Model Introduction:
“Advanced disease remains a leading cause of cancer-related mortality, and outcomes after failure of first-line therapy are poor. Standard second-line treatment has changed little over the past decade, with median progression-free survival of approximately five months and limited improvement in patient-reported outcomes. New treatment options in this setting are needed.
Agents in this class have been investigated in second-line disease with inconsistent results. Trial A evaluated a comparable agent in an unselected population and found no improvement in progression-free survival. In contrast, Trial B enrolled a biomarker-enriched population and reported a substantially larger effect, raising the possibility that benefit is concentrated in biomarker-positive disease. However, Trial B was a single-arm study and did not permit a controlled comparison, and no randomised trial has evaluated an agent of this class in second-line disease with prospectively defined biomarker stratification.
We therefore conducted a randomised, double-blind, placebo-controlled Phase III trial of the study regimen in patients with previously treated advanced disease. The primary endpoint was progression-free survival. Randomisation was stratified by region and number of prior lines of therapy, and biomarker status was prespecified as a subgroup of interest.”
Check your version against these five:
1. Move 1 establishes the problem and ends by stating that new options are needed — the bridge into the gap.
2. Move 2 ends on an explicit gap sentence: no randomised trial has evaluated… A reader can quote it. This is the sentence most drafts do not contain.
3. Two prior studies, used to build the argument. Not a survey. Their disagreement is the rationale.
4. The objective matches the primary endpoint. PFS in both places.
5. No results appear. Biomarker status is named as a prespecified subgroup, not as a finding.
The test: delete the second paragraph. Does the study still have a reason to exist? If yes, your gap was in the wrong paragraph. If no, it is in the right one.
“Advanced disease remains a leading cause of cancer-related mortality, and outcomes after failure of first-line therapy are poor. Standard second-line treatment has changed little over the past decade, with median progression-free survival of approximately five months and limited improvement in patient-reported outcomes. New treatment options in this setting are needed.
Agents in this class have been investigated in second-line disease with inconsistent results. Trial A evaluated a comparable agent in an unselected population and found no improvement in progression-free survival. In contrast, Trial B enrolled a biomarker-enriched population and reported a substantially larger effect, raising the possibility that benefit is concentrated in biomarker-positive disease. However, Trial B was a single-arm study and did not permit a controlled comparison, and no randomised trial has evaluated an agent of this class in second-line disease with prospectively defined biomarker stratification.
We therefore conducted a randomised, double-blind, placebo-controlled Phase III trial of the study regimen in patients with previously treated advanced disease. The primary endpoint was progression-free survival. Randomisation was stratified by region and number of prior lines of therapy, and biomarker status was prespecified as a subgroup of interest.”
Check your version against these five:
1. Move 1 establishes the problem and ends by stating that new options are needed — the bridge into the gap.
2. Move 2 ends on an explicit gap sentence: no randomised trial has evaluated… A reader can quote it. This is the sentence most drafts do not contain.
3. Two prior studies, used to build the argument. Not a survey. Their disagreement is the rationale.
4. The objective matches the primary endpoint. PFS in both places.
5. No results appear. Biomarker status is named as a prespecified subgroup, not as a finding.
The test: delete the second paragraph. Does the study still have a reason to exist? If yes, your gap was in the wrong paragraph. If no, it is in the right one.
Reflection
In your last paper's Introduction, which single sentence states the gap? If you cannot point to one, that is why the rationale felt thin.
Next module
4.2 — Methods, abstract and title
Week 4 · Module 2
Methods, abstract and title
Reproducibility, compression and findability — three different jobs, and the abstract is the only part of your paper most people will ever read
Your scenario
Your paper is published. Over the next two years it is cited forty times. Fewer than five of those citing authors read past the abstract. The abstract is not a summary of the paper — for most of your readership it is the paper.
Methods: the reproducibility test
One standard: could another team repeat this study from your Methods section? Referring the reader to the protocol for essential detail does not satisfy it.
| Element | Must include |
|---|---|
| Design and setting | Design, number of sites, countries, dates of enrolment and of the data cut-off |
| Participants | Key eligibility criteria; the full list may go to supplementary |
| Randomisation | Sequence generation, stratification factors with their levels, allocation concealment mechanism |
| Blinding | Who was blinded, by role, and how — CONSORT item 11a |
| Interventions | Dose, schedule, route, duration, dose modification rules |
| Endpoints | Primary and secondary, each with its definition and assessment method |
| Sample size | Assumed effect size, power, alpha, resulting n |
| Statistical analysis | Analysis populations, tests, multiplicity strategy, handling of missing data |
| Ethics and registration | Committee approval, consent, registration number, protocol availability |
The sentence most drafts omit
The multiplicity strategy. If a testing hierarchy or adjustment was prespecified, it belongs in Methods, and its absence is a CONSORT gap. It also determines what the Results section is permitted to call significant — so leaving it out creates a problem two sections later.
The abstract
| Section | Words | Contains |
|---|---|---|
| Background | 30–50 | The standing problem and the gap, in two sentences. No citations. |
| Methods | 50–70 | Design, population, intervention, comparator, primary endpoint. Enough to judge the evidence level. |
| Results | 70–100 | Primary endpoint first, with denominators and intervals. Key safety. |
| Conclusions | 25–40 | What the data support. One implication. Hedged to the design. |
Abstract rules that get broken
It must stand alone. Every abbreviation redefined; no citations; no reference to figures or tables.
Every number must appear in the paper. Abstract-only numbers are a common late-stage error, usually caused by a data refresh after the abstract was drafted.
The conclusion must match the primary endpoint. A conclusion about survival when the primary endpoint was progression-free survival is the single most-flagged abstract fault.
Write it last. An abstract drafted first commits you to a message the data may not support.
Every number must appear in the paper. Abstract-only numbers are a common late-stage error, usually caused by a data refresh after the abstract was drafted.
The conclusion must match the primary endpoint. A conclusion about survival when the primary endpoint was progression-free survival is the single most-flagged abstract fault.
Write it last. An abstract drafted first commits you to a message the data may not support.
The title
| Type | Form | When |
|---|---|---|
| Descriptive | Names the intervention, population and design without stating the result | Default for clinical trials. Most journals require it. |
| Declarative | States the finding | Some journals permit it; risky, because it commits the title to a claim reviewers may soften |
| Question form | Poses the research question | Rare in clinical journals; more common in reviews |
What a title has to carry
Intervention · population · design · the trial acronym if there is one. It must be findable — the words a searcher would use should be in it. Avoid abbreviations except universally recognised ones, and never use an abbreviation in the title that is not expanded in the abstract.
“Study regimen versus placebo in previously treated advanced disease: a randomised, double-blind, placebo-controlled Phase III trial” — intervention, comparator, population, design. Nothing wasted.
“Study regimen versus placebo in previously treated advanced disease: a randomised, double-blind, placebo-controlled Phase III trial” — intervention, comparator, population, design. Nothing wasted.
Drills
Drill 1 of 3
Your Methods says “full eligibility criteria are provided in the protocol.” Is that adequate?
AYes — the protocol is the definitive source
BNo — every criterion must appear in the main text
CNo — key criteria must appear in the Methods; the full list may go to supplementary
DYes, if the trial registration number is given
Drill 2 of 3
The primary endpoint was progression-free survival. Which abstract conclusion is correct?
A“The study regimen improved survival in previously treated advanced disease.”
B“The study regimen should be considered standard of care.”
C“Biomarker-positive patients benefit from the study regimen.”
D“The study regimen significantly improved progression-free survival compared with placebo; overall survival data remain immature.”
Drill 3 of 3
Which title is best for a clinical trial report?
A“Study regimen versus placebo in previously treated advanced disease: a randomised, double-blind, placebo-controlled Phase III trial”
B“The study regimen improves outcomes in advanced disease”
C“A clinical trial in advanced disease”
D“TRIAL-2: PFS and OS outcomes in 2L AD”
Production drill
Write the structured abstract in under 250 words, plus the title.
Design: randomised, double-blind, placebo-controlled Phase III; 402 patients; second-line advanced disease; hierarchy PFS → OS → ORR → PRO.
PFS: 8.9 (95% CI 7.4–10.3) vs 5.2 (4.4–6.1) months; HR 0.66 (0.53–0.82); p < 0.001.
OS: not reached vs 21.7 months; HR 0.84 (0.62–1.14); p = 0.26.
ORR: 44% vs 29% (nominal p = 0.003), n = 371.
Safety (n = 389): grade ≥3 37.8% vs 31.2%; discontinuation for AE 8.0% vs 5.7%.
Follow-up: median 19.8 months.
PFS: 8.9 (95% CI 7.4–10.3) vs 5.2 (4.4–6.1) months; HR 0.66 (0.53–0.82); p < 0.001.
OS: not reached vs 21.7 months; HR 0.84 (0.62–1.14); p = 0.26.
ORR: 44% vs 29% (nominal p = 0.003), n = 371.
Safety (n = 389): grade ≥3 37.8% vs 31.2%; discontinuation for AE 8.0% vs 5.7%.
Follow-up: median 19.8 months.
Title: “Study regimen versus placebo in previously treated advanced disease: a randomised, double-blind, placebo-controlled Phase III trial”
Abstract — 218 words
Background: Outcomes after failure of first-line therapy for advanced disease remain poor, and standard second-line treatment has changed little in a decade. Agents of this class have shown inconsistent results in unselected populations, and none has been evaluated in a randomised second-line trial.
Methods: In this randomised, double-blind, placebo-controlled Phase III trial, 402 patients with previously treated advanced disease were randomly assigned 1:1 to the study regimen or placebo, stratified by region and number of prior lines of therapy. The primary endpoint was progression-free survival. Secondary endpoints were tested in a prespecified hierarchy.
Results: At a median follow-up of 19.8 months, median progression-free survival was 8.9 months (95% CI 7.4–10.3) with the study regimen and 5.2 months (95% CI 4.4–6.1) with placebo (hazard ratio 0.66, 95% CI 0.53–0.82; p < 0.001). Median overall survival was not reached versus 21.7 months (hazard ratio 0.84, 95% CI 0.62–1.14; p = 0.26); formal testing stopped at this endpoint. Objective response rate was 44% versus 29% (nominal p = 0.003). Grade 3 or higher adverse events occurred in 37.8% and 31.2% of treated patients.
Conclusions: The study regimen significantly prolonged progression-free survival compared with placebo in previously treated advanced disease. Overall survival data were immature, and continued follow-up is required.
Six things to check in yours:
1. Under 250 words with no facts lost.
2. Background ends on the gap, not on background.
3. Methods states the design clearly enough to judge the evidence level, and mentions the hierarchy.
4. Results give both arms with intervals, and state that testing stopped at OS. ORR is marked nominal.
5. Conclusion matches the primary endpoint — progression-free survival, not survival.
6. No abbreviations left undefined; no citations; no reference to tables or figures.
The one most people get wrong: point 4. Omitting the hierarchy break lets a reader treat ORR as a confirmed finding, and in an abstract that most readers will never look past, that omission does real work.
Abstract — 218 words
Background: Outcomes after failure of first-line therapy for advanced disease remain poor, and standard second-line treatment has changed little in a decade. Agents of this class have shown inconsistent results in unselected populations, and none has been evaluated in a randomised second-line trial.
Methods: In this randomised, double-blind, placebo-controlled Phase III trial, 402 patients with previously treated advanced disease were randomly assigned 1:1 to the study regimen or placebo, stratified by region and number of prior lines of therapy. The primary endpoint was progression-free survival. Secondary endpoints were tested in a prespecified hierarchy.
Results: At a median follow-up of 19.8 months, median progression-free survival was 8.9 months (95% CI 7.4–10.3) with the study regimen and 5.2 months (95% CI 4.4–6.1) with placebo (hazard ratio 0.66, 95% CI 0.53–0.82; p < 0.001). Median overall survival was not reached versus 21.7 months (hazard ratio 0.84, 95% CI 0.62–1.14; p = 0.26); formal testing stopped at this endpoint. Objective response rate was 44% versus 29% (nominal p = 0.003). Grade 3 or higher adverse events occurred in 37.8% and 31.2% of treated patients.
Conclusions: The study regimen significantly prolonged progression-free survival compared with placebo in previously treated advanced disease. Overall survival data were immature, and continued follow-up is required.
Six things to check in yours:
1. Under 250 words with no facts lost.
2. Background ends on the gap, not on background.
3. Methods states the design clearly enough to judge the evidence level, and mentions the hierarchy.
4. Results give both arms with intervals, and state that testing stopped at OS. ORR is marked nominal.
5. Conclusion matches the primary endpoint — progression-free survival, not survival.
6. No abbreviations left undefined; no citations; no reference to tables or figures.
The one most people get wrong: point 4. Omitting the hierarchy break lets a reader treat ORR as a confirmed finding, and in an abstract that most readers will never look past, that omission does real work.
Reflection
Read your last abstract alone, as a stranger would. Does its conclusion match its primary endpoint?
Next module
4.3 — Submission and peer review
Week 4 · Module 3
Submission and peer review
The cover letter, the submission package, and the response to reviewers — a document whose craft determines whether a revision is accepted
Your scenario
A major revision arrives with twenty-three comments across three reviewers. Two comments contradict each other. One is based on a misreading. One asks for an analysis you cannot do. How you write the response letter matters as much as what you change — editors accept or reject revisions substantially on the quality of that document.
The cover letter
The editor's first decision is fit, not quality. A cover letter that repeats the abstract wastes the only chance to answer that question directly.
| Element | One or two sentences each |
|---|---|
| What you are submitting | Title, article type, and the trial registration number |
| The question and the finding | Not the abstract — the single reason this paper exists |
| Why this journal | What it adds for their readership specifically |
| Declarations | Original, not under consideration elsewhere, all authors approved, conflicts disclosed |
| Compliance | Ethics approval, consent, reporting guideline followed, data sharing statement |
| Optional | Suggested or excluded reviewers, with brief reasons |
The response to reviewers
Figure 4.3 — Anatomy of a response. Editors decide on revisions under time pressure. A response letter that can be read on its own, without cross-referencing two other documents, is materially more likely to be accepted — and this is entirely within your control.
| Situation | How to handle it |
|---|---|
| You agree | Say so, make the change, quote it. No lengthy justification needed. |
| You disagree on evidence | Thank them, present the evidence, and change the text anyway so the next reader does not make the same inference. |
| The reviewer misread something | A misreading is usually a signal your text was unclear. Clarify the manuscript, not just the letter. |
| Two reviewers contradict | Name the conflict explicitly, explain your choice, and let the editor adjudicate. Do not silently follow one. |
| The request is impossible | Explain precisely why — data unavailable, not prespecified, would require unblinding — and offer the nearest feasible alternative. |
| The comment is hostile | Respond to the substance and ignore the tone entirely. Never match it. |
The three fatal moves
Leaving a comment unanswered. Editors check. An unanswered comment reads as evasion.
Claiming to have made a change you did not make. This is discovered and it ends the submission.
Arguing with tone rather than substance. However unfair the review, the response letter is read by the editor, and matching hostility loses papers that would otherwise have been accepted.
Claiming to have made a change you did not make. This is discovered and it ends the submission.
Arguing with tone rather than substance. However unfair the review, the response letter is read by the editor, and matching hostility loses papers that would otherwise have been accepted.
Drills
Drill 1 of 3
A reviewer criticises you for not adjusting for a variable you did in fact adjust for, and reported in Table 2. How do you respond?
A“The reviewer is mistaken; this adjustment is reported in Table 2.”
B“We apologise this was not clear. The adjustment is reported in Table 2; we have now stated it explicitly in the Methods (lines 148–150) and referenced it in the Results.”
CPerform the analysis again and present it as new
DLeave the comment unaddressed since the analysis was already done
Drill 2 of 3
Reviewer 1 wants the subgroup analysis expanded; Reviewer 2 wants it removed. What do you do?
AFollow Reviewer 1, since expansion adds content
BSplit the difference without mentioning the conflict
CName the conflict, explain your choice with reasoning, and invite the editor to direct
DRemove the analysis to end the dispute
Drill 3 of 3
What is the primary purpose of the cover letter?
ATo summarise the abstract
BTo list the authors and their affiliations
CTo request specific reviewers
DTo tell the editor why this paper fits this journal and this readership
The Rebuttal Matrix Professional Anatomy of a Point-by-Point Author Response
When drafting an author response letter, every reviewer critique must be addressed using the 3-Step Consensus Model: Acknowledge → Action → Evidence/Location.
| Reviewer Critique Type | Poor / Defensive Response (Rejection Risk) | Professional Industry Response (Acceptance Standard) |
|---|---|---|
| Statistical Confounding / Missing Data | "We used standard ITT and don't believe missing data alters the result." | "We thank Reviewer 2 for this critical observation. To evaluate the impact of missing data, we have performed a tipping-point sensitivity analysis under the pattern-mixture model. The treatment effect remained robust across all plausibility thresholds. We have added this analysis to Supplementary Table S4 and expanded the Methods (Page 9, Lines 182–191)." |
| Overclaimed Causal Interpretation | "Previous studies in mice show this drug cures fibrosis, so our conclusion is justified." | "We agree with Reviewer 1 that our original phrasing overstated causality in this observational cohort. We have revised the Discussion (Page 14, Lines 310–314) to replace 'demonstrates neuroprotection' with 'shows a significant cross-sectional association with preserved hippocampal volume' and highlighted this in the Study Limitations." |
| Contradictory Reviewer Requests (R1 vs R3) | "Reviewer 1 asked to add Table 2, but Reviewer 3 said it was too long, so we kept it." | "We appreciate the differing perspectives of Reviewers 1 and 3 regarding Table 2. To balance depth with readability, we have condensed Table 2 in the main text to highlight key primary outcomes, and moved the extensive subgroup stratification to Supplementary Appendix Table S2 (Page 11, Lines 240–248)." |
Production drill
Write the response to these three comments, using all four parts of the anatomy.
Reviewer 1, comment 3: “The authors present the biomarker subgroup as though it establishes a predictive effect. This is not supported by a single interaction test in a trial not powered for subgroups.”
Reviewer 2, comment 1: “Overall survival is reported as not reached, which suggests the data are missing. The authors should clarify.”
Reviewer 2, comment 5: “Please provide the analysis of PFS censored at the time of subsequent therapy.”
Context: this sensitivity analysis was not prespecified in the SAP and the data are not available at this cut.
Reviewer 2, comment 1: “Overall survival is reported as not reached, which suggests the data are missing. The authors should clarify.”
Reviewer 2, comment 5: “Please provide the analysis of PFS censored at the time of subsequent therapy.”
Context: this sensitivity analysis was not prespecified in the SAP and the data are not available at this cut.
Reviewer 1, comment 3
“The authors present the biomarker subgroup as though it establishes a predictive effect…”
We agree, and we thank the reviewer for this point. We have revised the Discussion to state explicitly that the interaction was one of six prespecified subgroup analyses in a trial powered for the overall population, and that the finding is hypothesis-generating (Discussion, paragraph 4, lines 298–304). The revised text reads: “Although the interaction by biomarker status was statistically significant and biologically plausible, the trial was not powered for subgroup comparisons and this finding should be regarded as hypothesis-generating, requiring prospective confirmation.” We have also removed the phrase identifies patients who should receive from the Conclusion.
Reviewer 2, comment 1
“Overall survival is reported as not reached, which suggests the data are missing…”
We thank the reviewer and apologise that this was unclear. Not reached indicates that fewer than half of the patients in that arm had died at the data cut-off, so the median could not be estimated; the data are not missing. We have clarified this in the Results (lines 214–217) and added the numbers at risk: “Median overall survival was not reached with the study regimen, indicating that fewer than half of patients had died at the data cut-off, and was 21.7 months with placebo; 41 and 22 patients respectively remained at risk at 24 months.”
Reviewer 2, comment 5
“Please provide the analysis of PFS censored at the time of subsequent therapy.”
We are unable to provide this analysis. It was not prespecified in the statistical analysis plan, and data on subsequent therapy are not captured at the present data cut-off; conducting it post hoc would risk a data-driven analysis of the kind the prespecified plan exists to prevent. As the nearest available alternative we have added the prespecified sensitivity analysis of PFS by investigator assessment, which was consistent with the primary analysis (Supplementary Table S4), and we have noted in the Limitations that the effect of subsequent therapy could not be assessed (lines 322–325).
What each response demonstrates:
1 — agreeing fully, and doing more than asked by also fixing the Conclusion.
2 — the reviewer misread, and the response fixes the manuscript rather than merely correcting the reviewer.
3 — declining, with a specific reason, an alternative offered, and a limitation added. Declining is acceptable; declining without explanation is not.
“The authors present the biomarker subgroup as though it establishes a predictive effect…”
We agree, and we thank the reviewer for this point. We have revised the Discussion to state explicitly that the interaction was one of six prespecified subgroup analyses in a trial powered for the overall population, and that the finding is hypothesis-generating (Discussion, paragraph 4, lines 298–304). The revised text reads: “Although the interaction by biomarker status was statistically significant and biologically plausible, the trial was not powered for subgroup comparisons and this finding should be regarded as hypothesis-generating, requiring prospective confirmation.” We have also removed the phrase identifies patients who should receive from the Conclusion.
Reviewer 2, comment 1
“Overall survival is reported as not reached, which suggests the data are missing…”
We thank the reviewer and apologise that this was unclear. Not reached indicates that fewer than half of the patients in that arm had died at the data cut-off, so the median could not be estimated; the data are not missing. We have clarified this in the Results (lines 214–217) and added the numbers at risk: “Median overall survival was not reached with the study regimen, indicating that fewer than half of patients had died at the data cut-off, and was 21.7 months with placebo; 41 and 22 patients respectively remained at risk at 24 months.”
Reviewer 2, comment 5
“Please provide the analysis of PFS censored at the time of subsequent therapy.”
We are unable to provide this analysis. It was not prespecified in the statistical analysis plan, and data on subsequent therapy are not captured at the present data cut-off; conducting it post hoc would risk a data-driven analysis of the kind the prespecified plan exists to prevent. As the nearest available alternative we have added the prespecified sensitivity analysis of PFS by investigator assessment, which was consistent with the primary analysis (Supplementary Table S4), and we have noted in the Limitations that the effect of subsequent therapy could not be assessed (lines 322–325).
What each response demonstrates:
1 — agreeing fully, and doing more than asked by also fixing the Conclusion.
2 — the reviewer misread, and the response fixes the manuscript rather than merely correcting the reviewer.
3 — declining, with a specific reason, an alternative offered, and a limitation added. Declining is acceptable; declining without explanation is not.
Reflection
Find a response letter you or a colleague has written. Does every comment have all four parts?
Part 4 — quoting the new text — is the one most often skipped, and it is the one that makes an editor's job easiest.
Next module
4.4 — Capstone and exit assessment
Week 4 · Module 4
Capstone and exit assessment
A finished draft with twenty seeded faults across every section — the review task a senior writer is actually paid to perform
Your scenario
You are handed a colleague's draft and asked whether it can be submitted. Not to write it — to judge it. This is what senior medical writing actually consists of, and it is the task an agency sets at final interview. Pass mark: 16 of 20.
Do this properly
Timer at 50 minutes. Read the draft, write your list of findings with the correction and the reason for each, and only then reveal the rubric.
The draft under review
List every fault with its correction and reason. Twenty are seeded.
Title: The study regimen improves outcomes in advanced disease
Introduction (extract): “Advanced disease is common and treatment options are limited. Many agents have been studied. Trial A, Trial B, Trial C, Trial D and Trial E have all investigated this class. We therefore conducted a trial and found a significant benefit in progression-free survival.”
Methods (extract): “Patients were randomised to the study regimen or placebo. Full eligibility criteria are in the protocol. The primary endpoint was progression-free survival. Statistical analyses were performed using standard methods.”
Results (extract): “Encouragingly, the primary endpoint was met, with median PFS of 8.9 vs 5 months (HR .66, CI 0.53-0.82, p=0.000), consistent with the mechanism of action. Baseline characteristics were well balanced (all p > 0.05). Surprisingly, OS did not differ (p = 0.26). ORR was significantly higher (44% vs 29%, p = 0.003). Grade 3 or higher events occurred in 36.6% of the 402 randomised patients. Quality of life improved significantly (p = 0.02). 402 patients were randomised.”
Discussion (extract): “Advanced disease affects millions worldwide. Trial A reported a hazard ratio of 0.94 and Trial B reported 0.61. Our finding of a median PFS of 8.9 versus 5.2 months (HR 0.66, 95% CI 0.53–0.82, p < 0.001) is notable. The biomarker subgroup identifies patients who should receive this regimen. This study has some limitations. In conclusion, the study regimen proves effective and should become standard of care.”
Introduction (extract): “Advanced disease is common and treatment options are limited. Many agents have been studied. Trial A, Trial B, Trial C, Trial D and Trial E have all investigated this class. We therefore conducted a trial and found a significant benefit in progression-free survival.”
Methods (extract): “Patients were randomised to the study regimen or placebo. Full eligibility criteria are in the protocol. The primary endpoint was progression-free survival. Statistical analyses were performed using standard methods.”
Results (extract): “Encouragingly, the primary endpoint was met, with median PFS of 8.9 vs 5 months (HR .66, CI 0.53-0.82, p=0.000), consistent with the mechanism of action. Baseline characteristics were well balanced (all p > 0.05). Surprisingly, OS did not differ (p = 0.26). ORR was significantly higher (44% vs 29%, p = 0.003). Grade 3 or higher events occurred in 36.6% of the 402 randomised patients. Quality of life improved significantly (p = 0.02). 402 patients were randomised.”
Discussion (extract): “Advanced disease affects millions worldwide. Trial A reported a hazard ratio of 0.94 and Trial B reported 0.61. Our finding of a median PFS of 8.9 versus 5.2 months (HR 0.66, 95% CI 0.53–0.82, p < 0.001) is notable. The biomarker subgroup identifies patients who should receive this regimen. This study has some limitations. In conclusion, the study regimen proves effective and should become standard of care.”
The twenty faults
Title and Introduction (5)
1. Declarative title stating a result, and improves outcomes is vaguer than the endpoint measured. Use a descriptive title naming intervention, comparator, population and design.
2. No gap stated. Five trials are listed and no sentence says what was unknown. This is the desk-rejection fault.
3. Literature listed, not used. Trials A–E are named without their findings or their relevance.
4. Results in the Introduction. …and found a significant benefit belongs in Results.
5. Objective missing. The Introduction never states what the study set out to do.
Methods (3)
6. Randomisation inadequately described — no ratio, no sequence generation, no stratification factors, no allocation concealment. CONSORT items 8–9.
7. Eligibility criteria deferred to the protocol. Key criteria belong in the Methods; reproducibility is the standard.
8. “Standard methods” is not a statistical analysis section. No populations, no tests named, no multiplicity strategy, no handling of missing data.
Results (8)
9. Sequence inverted — disposition appears last. Participant flow and baseline open the section.
10. Encouragingly / Surprisingly — editorialising adverbs.
11. “consistent with the mechanism of action” — interpretation; belongs in the Discussion.
12. Baseline p values — discouraged by CONSORT.
13. ORR called significant — the hierarchy broke at OS (p = 0.26), so ORR is nominal only.
14. Quality of life called significant with no magnitude — and the difference is below its MCID, which must be stated.
15. Safety uses the randomised denominator (402). It must use the treated population, changing 36.6% to 37.8%.
16. Reporting conventions: 5 should be 5.2; .66 needs a leading zero; CI needs its level; hyphen should be an en dash; p=0.000 is impossible; sentence opens on a numeral; median follow-up is absent entirely.
Discussion and Conclusion (4)
17. Move 1 missing — opens on epidemiology rather than the principal finding.
18. Results restated in full with medians, hazard ratio, interval and p value.
19. Prior trials cited without an argument, and the biomarker claim — identifies patients who should receive — is a practice recommendation from one interaction test.
20. Generic limitations (“some limitations”), and a conclusion that proves effectiveness and recommends standard of care from a single trial whose OS endpoint was not met.
Scoring
16–20 — pass. You can review a manuscript unsupervised, and you would have caught the three faults that reach patients: 13, 14 and 15.
11–15 — close. Note which section your misses cluster in and redo that week's clinic.
Under 11 — repeat the Results and Discussion clinics before retrying.
If you added faults that were not there: over-flagging is its own failure mode. A reviewer who marks correct text as wrong costs the team as much time as one who misses errors.
The three that matter beyond the paper: fault 13 lets a reader treat an untested endpoint as confirmed; fault 14 presents a statistically significant but clinically trivial result as a patient benefit; fault 15 understates an adverse event rate in a published article. The rest are craft. Those three are consequences.
Title and Introduction (5)
1. Declarative title stating a result, and improves outcomes is vaguer than the endpoint measured. Use a descriptive title naming intervention, comparator, population and design.
2. No gap stated. Five trials are listed and no sentence says what was unknown. This is the desk-rejection fault.
3. Literature listed, not used. Trials A–E are named without their findings or their relevance.
4. Results in the Introduction. …and found a significant benefit belongs in Results.
5. Objective missing. The Introduction never states what the study set out to do.
Methods (3)
6. Randomisation inadequately described — no ratio, no sequence generation, no stratification factors, no allocation concealment. CONSORT items 8–9.
7. Eligibility criteria deferred to the protocol. Key criteria belong in the Methods; reproducibility is the standard.
8. “Standard methods” is not a statistical analysis section. No populations, no tests named, no multiplicity strategy, no handling of missing data.
Results (8)
9. Sequence inverted — disposition appears last. Participant flow and baseline open the section.
10. Encouragingly / Surprisingly — editorialising adverbs.
11. “consistent with the mechanism of action” — interpretation; belongs in the Discussion.
12. Baseline p values — discouraged by CONSORT.
13. ORR called significant — the hierarchy broke at OS (p = 0.26), so ORR is nominal only.
14. Quality of life called significant with no magnitude — and the difference is below its MCID, which must be stated.
15. Safety uses the randomised denominator (402). It must use the treated population, changing 36.6% to 37.8%.
16. Reporting conventions: 5 should be 5.2; .66 needs a leading zero; CI needs its level; hyphen should be an en dash; p=0.000 is impossible; sentence opens on a numeral; median follow-up is absent entirely.
Discussion and Conclusion (4)
17. Move 1 missing — opens on epidemiology rather than the principal finding.
18. Results restated in full with medians, hazard ratio, interval and p value.
19. Prior trials cited without an argument, and the biomarker claim — identifies patients who should receive — is a practice recommendation from one interaction test.
20. Generic limitations (“some limitations”), and a conclusion that proves effectiveness and recommends standard of care from a single trial whose OS endpoint was not met.
Scoring
16–20 — pass. You can review a manuscript unsupervised, and you would have caught the three faults that reach patients: 13, 14 and 15.
11–15 — close. Note which section your misses cluster in and redo that week's clinic.
Under 11 — repeat the Results and Discussion clinics before retrying.
If you added faults that were not there: over-flagging is its own failure mode. A reviewer who marks correct text as wrong costs the team as much time as one who misses errors.
The three that matter beyond the paper: fault 13 lets a reader treat an untested endpoint as confirmed; fault 14 presents a statistically significant but clinically trivial result as a patient benefit; fault 15 understates an adverse event rate in a published article. The rest are craft. Those three are consequences.
Retake the diagnostic
Return to Module 1.1 and take the 24-item diagnostic again. Your first score was stored as the baseline; the difference is the only honest measure of what these four weeks changed. Expect movement of one band.
What to do next
Build the habits, not the knowledge. On your next manuscript: agree the author list in week one; write the core message before drafting; build the number sheet before writing Results; and run the eleven interpretation words over the Results section before it leaves your desk. Four habits, and they catch most of what this course covered.
Then pair it with the foundations. The Statistics course teaches what the numbers mean; the Scientific English course teaches how to write the sentences. This course teaches what each section is allowed to say. The three overlap deliberately at the point that matters most — matching the strength of a claim to the strength of the evidence.
Then pair it with the foundations. The Statistics course teaches what the numbers mean; the Scientific English course teaches how to write the sentences. This course teaches what each section is allowed to say. The three overlap deliberately at the point that matters most — matching the strength of a claim to the strength of the evidence.
Final reflection
Your diagnostic score at the start, and now. Which module changed how you write most? What still feels uncertain?
You have completed the course
Writing the Manuscript
IMRAD Academy — Read like a scientist. Write like a professional.
Writing the Manuscript · v1.0