Week 1 · Module 1
Source document diagnostic
Twenty-four questions on the documents you will be handed and expected to work from — and a personal study order for the next four weeks
Your scenario
First week on a global Phase III cardiovascular and metabolic trial across 45 international sites. You are handed Protocol v4.0, two historical amendment summaries, the SAP, and an unblinded interim safety pack alongside the 52-week primary database lock output. Your lead requests a draft manuscript in four days. Nobody explains which protocol amendment altered the primary 3-point MACE definition, which data cut generated each table, or why baseline demographics in Table 14.1.1 show 402 randomised vs 389 treated patients. What you verify in the first two hours decides whether your draft survives medical review.
Why this course exists
Every medical writing course teaches you to write. Almost none teaches you to read — and reading is where the working day actually goes. A writer who cannot navigate a protocol, locate an estimand in a SAP, or spot that a safety table has the wrong denominator will produce fluent text that is wrong.
This course is about the documents themselves: what each one is for, how it is structured, what it can and cannot tell you, and which discrepancies to raise rather than resolve.
Take the diagnostic honestly
Do not look anything up. A wrong answer here is worth more than a right one, because it is the only way the diagnostic can find the gap.
—
—
Your twelve areas — study the red ones first
Before you move on
Write down your three weakest areas. Return here after the capstone.
Next module
1.2 — The document ecosystem
Week 1 · Module 2
The document ecosystem
Twelve documents, who owns each one, what it can tell you, and — most importantly — what it cannot
Your scenario
You quote a median from the clinical study report. Two weeks later the statistician points out that the CSR was finalised at an earlier data cut, and the figure you published in the abstract is superseded. The CSR was not wrong. You used it for something it was never meant to provide.
The map
Figure 1.2 — The ecosystem. Documents produced before the study define what was intended; documents produced after database lock describe what happened. The single most useful habit in this course is following the gold path: every figure you publish traces to an output, not to prose about an output.
What each document can and cannot give you
| Document | Use it for | Never use it for |
|---|---|---|
| Protocol | Design, eligibility, interventions, endpoint definitions, ethics, registration | Results — it describes intent, not outcome |
| SAP | Analysis populations, estimand, testing hierarchy, missing data handling, prespecified subgroups | Numbers — it is the rulebook, not the scoreboard |
| Investigator's brochure | Background pharmacology, prior clinical experience, reference safety information | Anything about the current study |
| CRF / data specification | Establishing whether a variable was captured at all | Values — it is a blank instrument |
| Output pack (TFLs) | Every number you quote, anywhere | Interpretation — outputs do not say what matters |
| CSR | Narrative context, protocol deviations, disposition detail, safety narratives | Definitive figures — derived text, possibly at an earlier cut |
| DSUR / PBRER | Cumulative safety picture across studies | Study-specific efficacy |
| SmPC / USPI / CCDS | Approved indication, dosing, contraindications, warnings | Investigational or off-label uses |
The version and date problem
Every one of these documents exists in versions. A protocol has amendments. A SAP has amendments. An IB is reissued annually. An output pack has a data cut-off.
Before you use any document, check two things: its version number and its date. Working from a superseded protocol or an earlier data cut is the single most common source of errors that survive every review, because the text reads perfectly.
Before you use any document, check two things: its version number and its date. Working from a superseded protocol or an earlier data cut is the single most common source of errors that survive every review, because the text reads perfectly.
Clinical Trial Registries (ClinicalTrials.gov & EU CTIS) — The Public Source Anchor
Fresh medical writers frequently overlook trial registry records. Under FDAAA 801 and the EU Clinical Trials Regulation (CTR 536/2014), clinical trial protocols and summary results must be posted to ClinicalTrials.gov (PRS) and the EU Clinical Trials Information System (CTIS).
- Protocol Registration: Establishes the official National Clinical Trial identifier (NCT number), primary completion date (PCD), and prespecified primary/secondary outcome measures before the first participant is enrolled.
- Summary Results Module: Requires structured tabular reporting of Participant Flow, Baseline Characteristics, Outcome Measures (with statistical test results), and All-Cause Mortality / Serious Adverse Events within 12 months of PCD.
- Writing Rule: When preparing manuscripts or briefing docs, verify that the primary endpoint timing, outcome definitions, and trial registration ID perfectly mirror the public registry record to avoid journal submission rejection.
Drills
Drill 1 of 3
You need to state the definition of the primary endpoint. Which source?
AThe protocol, cross-checked against the SAP
BThe output pack
CThe CSR
DThe investigator’s brochure
Drill 2 of 3
You want to know whether frailty was recorded, so you can address it in a limitations paragraph. Fastest route?
ASearch the CSR
BCheck the CRF or the data specification
CLook through the output pack
DEmail the sponsor
Drill 3 of 3
A CSR states median PFS as 8.1 months; the current output pack states 8.9 months. What is the most likely explanation?
AA calculation error in the output pack
BThe CSR is authoritative and the output is wrong
CThe CSR was finalised at an earlier data cut-off
DAverage the two
Production drill
For each question, name the document you would open and the section within it.
1. Was this trial registered, and where?
2. Which patients were included in the primary efficacy analysis?
3. How many patients discontinued because of an adverse event?
4. Is grade 3 hepatotoxicity an expected reaction for this product?
5. Why were 14 patients excluded from the per-protocol population?
6. Was smoking history collected?
7. What is the approved indication in the European Union?
8. What happened to patient 1042, who died on study?
2. Which patients were included in the primary efficacy analysis?
3. How many patients discontinued because of an adverse event?
4. Is grade 3 hepatotoxicity an expected reaction for this product?
5. Why were 14 patients excluded from the per-protocol population?
6. Was smoking history collected?
7. What is the approved indication in the European Union?
8. What happened to patient 1042, who died on study?
1. Protocol — administrative or ethics section; cross-check the registry record itself for the registration date, since prospective registration is what matters.
2. SAP for the population definition; output pack table 14.1.1 for the actual n. Two documents, one answer — and this is the pairing most people get wrong by using only one.
3. Output pack, 14.3.x safety series. Note the denominator will be the treated population, not the randomised one.
4. Investigator’s brochure — reference safety information. This is the only section that determines expectedness, and getting it wrong has regulatory consequences.
5. CSR section 10, protocol deviations and analysis populations. The output pack gives the count; the CSR explains the reasons.
6. CRF or data specification. If the field is not there, it was not collected.
7. SmPC. Not the IB, which covers an investigational product, and not the CCDS, which is the internal reference from which national labels are derived and may differ.
8. Patient narrative in the CSR, supported by patient-level listings in section 16.2.
What this drill is really testing: whether you reach for the right document first. A writer who searches the CSR for all eight will find plausible-looking answers to most of them and be wrong on 2, 4, 6 and 7.
2. SAP for the population definition; output pack table 14.1.1 for the actual n. Two documents, one answer — and this is the pairing most people get wrong by using only one.
3. Output pack, 14.3.x safety series. Note the denominator will be the treated population, not the randomised one.
4. Investigator’s brochure — reference safety information. This is the only section that determines expectedness, and getting it wrong has regulatory consequences.
5. CSR section 10, protocol deviations and analysis populations. The output pack gives the count; the CSR explains the reasons.
6. CRF or data specification. If the field is not there, it was not collected.
7. SmPC. Not the IB, which covers an investigational product, and not the CCDS, which is the internal reference from which national labels are derived and may differ.
8. Patient narrative in the CSR, supported by patient-level listings in section 16.2.
What this drill is really testing: whether you reach for the right document first. A writer who searches the CSR for all eight will find plausible-looking answers to most of them and be wrong on 2, 4, 6 and 7.
Reflection
Of the eight documents in the table, how many have you actually opened and read a section of? Name the ones you have not.
Most writers with several years of experience have never opened a CRF or an IB. Both take an hour and change how you read everything else.
Next module
1.3 — The protocol
Week 1 · Module 3
The protocol
A hundred and fifty pages with a fixed structure. Learn the structure and you can answer any design question in ninety seconds
Your scenario
A reviewer asks whether outcome assessors were blinded. You search the protocol PDF for “blind” and get forty-one hits across ninety pages. You spend an hour and are still not certain. The answer was in one paragraph, in a section you did not know existed.
The standard structure
ICH E6 specifies what a protocol must contain. Sponsors vary the numbering, but the content blocks are consistent and appear in roughly this order.
| Block | Answers | Writer uses it for |
|---|---|---|
| Synopsis | Everything, briefly | Orientation only. Never quote from it — go to the body text. |
| Background and rationale | Why the study exists | Introduction background; the gap statement |
| Objectives and endpoints | What is being measured | Endpoint definitions, primary versus secondary |
| Study design | Structure, duration, phases, randomisation ratio | The design sentence in Methods |
| Selection and withdrawal of subjects | Inclusion, exclusion, withdrawal criteria | Eligibility paragraph; generalisability limits |
| Treatment of subjects | Dose, schedule, route, modifications, concomitant medication | Interventions paragraph |
| Assessment of efficacy and safety | How and when each endpoint is measured | Assessment methods; visit schedule |
| Statistics | Sample size, planned analyses, interim analyses | Outline only — the SAP is definitive |
| Ethics, consent, data handling | Approvals, consent, confidentiality, monitoring | Ethics statement; committee names |
| Schedule of assessments | The visit grid | Fastest way to see what was measured and when |
Start with the schedule of assessments
It is usually a single large table near the back, and it is the most information-dense page in the protocol. It shows every assessment, at every visit, for the whole study. Two minutes there tells you more about the study than twenty minutes of reading prose — and it immediately answers whether a variable was measured, and how often.
Amendments
| Question to ask | Why it matters |
|---|---|
| Which version governs the data I am reporting? | Using a superseded version means describing a study that was not conducted |
| What changed, and when? | Substantial amendments must be described in the CSR and usually in the manuscript |
| Did any amendment occur after unblinding? | A change to endpoints or analysis after unblinding is a serious concern and must be disclosed |
| Did eligibility change mid-study? | The population may differ between enrolment periods, which affects generalisability |
Blinding is not one section
Protocols scatter blinding information. Allocation concealment sits in the randomisation section; who is blinded sits in the design or blinding section; unblinding procedures sit in a separate emergency section; and outcome assessor blinding may only appear in the efficacy assessment section, often as a single sentence about central review.
To answer “was this properly blinded” you need all four. Searching for the word will not assemble them.
To answer “was this properly blinded” you need all four. Searching for the word will not assemble them.
Drills
Drill 1 of 3
Fastest way to establish whether quality of life was measured, and at which visits?
ASearch the protocol for “quality of life”
BRead the synopsis
CCheck the output pack
DOpen the schedule of assessments
Drill 2 of 3
The protocol was amended after the interim analysis to change the primary endpoint. What should you do?
AEstablish whether the amendment preceded unblinding, disclose it and its rationale, and raise it with the team
BReport the final endpoint and say nothing about the change
CReport the original endpoint instead
DRefuse to write the paper
Drill 3 of 3
Where does the protocol tell you whether outcome assessors were blinded?
AThe randomisation section
BOften only in the efficacy assessment section, as a note on central review
CThe synopsis
DThe statistics section
Production drill
Write the Methods design paragraph from these protocol extracts in a Severe Eosinophilic Asthma trial, and list what is still missing.
Section 6, Study design: Multicentre, randomised, double-blind, placebo-controlled Phase III trial in severe asthma. 1:1 allocation. 48 sites in 8 countries. Enrolment Jan–Sep 2024.
Section 6.3, Randomisation: Centralised via interactive web response system; stratified by region and baseline blood eosinophil count (<300 vs ≥300 cells/μL).
Section 7, Eligibility: Adults 18–75 years with severe asthma inadequately controlled despite high-dose inhaled corticosteroids plus long-acting β2-agonists; ≥2 documented asthma exacerbations in prior 12 months; pre-bronchodilator FEV1 <80% predicted.
Section 8, Interventions: Investigational biologic mAb 100 mg subcutaneous injection every 4 weeks vs matching placebo for 52 weeks.
Section 9, Endpoints: Primary endpoint is the annualized rate of moderate-to-severe asthma exacerbations at Week 52. Key secondary: Change from baseline in pre-bronchodilator trough FEV1 and Asthma Control Questionnaire (ACQ-5) score at Week 52.
Section 6.3, Randomisation: Centralised via interactive web response system; stratified by region and baseline blood eosinophil count (<300 vs ≥300 cells/μL).
Section 7, Eligibility: Adults 18–75 years with severe asthma inadequately controlled despite high-dose inhaled corticosteroids plus long-acting β2-agonists; ≥2 documented asthma exacerbations in prior 12 months; pre-bronchodilator FEV1 <80% predicted.
Section 8, Interventions: Investigational biologic mAb 100 mg subcutaneous injection every 4 weeks vs matching placebo for 52 weeks.
Section 9, Endpoints: Primary endpoint is the annualized rate of moderate-to-severe asthma exacerbations at Week 52. Key secondary: Change from baseline in pre-bronchodilator trough FEV1 and Asthma Control Questionnaire (ACQ-5) score at Week 52.
Draft paragraph:
“This was a multicentre, randomised, double-blind, placebo-controlled Phase III trial conducted across 48 sites in 8 countries between January and September 2024. Eligible participants were adults aged 18 to 75 years with severe asthma inadequately controlled on high-dose inhaled corticosteroids plus long-acting β2-agonists who had experienced two or more documented exacerbations in the previous 12 months and had a baseline pre-bronchodilator FEV1 <80% predicted. Participants were randomised (1:1) via a centralised interactive web response system, stratified by region and baseline blood eosinophils (<300 vs ≥300 cells/μL), to receive either the investigational monoclonal antibody (100 mg subcutaneously every 4 weeks) or matching placebo for 52 weeks. The primary endpoint was the annualized rate of moderate-to-severe asthma exacerbations over 52 weeks; key secondary endpoints included change from baseline in pre-dose trough FEV1 and ACQ-5 score.”
Still missing — must confirm from protocol before drafting:
1. Exact clinical definition of an “exacerbation” (e.g. systemic corticosteroid burst ≥3 days, emergency visit, or hospitalisation).
2. Washout and permitted background controller medications during the 4-week run-in period.
3. The definition of the primary analysis population (mFAS vs per-protocol completers).
4. Schedule of spirometry assessments (pre-dose at Weeks 0, 4, 12, 24, 36, and 52).
5. Institutional ethics approval, Helsinki declaration adherence, and ClinicalTrials.gov NCT identifier.
“This was a multicentre, randomised, double-blind, placebo-controlled Phase III trial conducted across 48 sites in 8 countries between January and September 2024. Eligible participants were adults aged 18 to 75 years with severe asthma inadequately controlled on high-dose inhaled corticosteroids plus long-acting β2-agonists who had experienced two or more documented exacerbations in the previous 12 months and had a baseline pre-bronchodilator FEV1 <80% predicted. Participants were randomised (1:1) via a centralised interactive web response system, stratified by region and baseline blood eosinophils (<300 vs ≥300 cells/μL), to receive either the investigational monoclonal antibody (100 mg subcutaneously every 4 weeks) or matching placebo for 52 weeks. The primary endpoint was the annualized rate of moderate-to-severe asthma exacerbations over 52 weeks; key secondary endpoints included change from baseline in pre-dose trough FEV1 and ACQ-5 score.”
Still missing — must confirm from protocol before drafting:
1. Exact clinical definition of an “exacerbation” (e.g. systemic corticosteroid burst ≥3 days, emergency visit, or hospitalisation).
2. Washout and permitted background controller medications during the 4-week run-in period.
3. The definition of the primary analysis population (mFAS vs per-protocol completers).
4. Schedule of spirometry assessments (pre-dose at Weeks 0, 4, 12, 24, 36, and 52).
5. Institutional ethics approval, Helsinki declaration adherence, and ClinicalTrials.gov NCT identifier.
Reflection
Open any protocol you have access to and find its schedule of assessments. How long did it take, and what did it tell you that you did not know?
Next module
1.4 — The investigator's brochure
Week 1 · Module 4
The investigator's brochure
The document that tells you what is already known about the product — and contains the one section with direct regulatory consequences
Your scenario
A safety physician asks whether grade 3 hepatotoxicity is expected for the product. You check the CSR, find it listed among adverse events, and say yes. That is the wrong document and the wrong answer. Expectedness is determined solely by the reference safety information in the current investigator's brochure — and if the event is not listed there, it is unexpected, which triggers expedited reporting.
Structure
| Section | Contains | Writer uses it for |
|---|---|---|
| Summary | Two or three pages covering everything | Fast orientation on an unfamiliar product |
| Physical, chemical and pharmaceutical properties | Formulation, stability, storage | Rarely; occasionally for methods detail |
| Non-clinical studies | Pharmacology, toxicology, animal PK | Mechanism paragraphs in an Introduction |
| Effects in humans | Clinical PK, metabolism, safety and efficacy from prior studies | Background on prior clinical experience; prior trial results |
| Reference safety information | The list of expected serious adverse reactions | Expectedness assessment. The highest-consequence section. |
| Guidance for the investigator | Risk management, monitoring, dose modification advice | Understanding why the protocol handles safety as it does |
Reference safety information
The RSI is the definitive list of expected serious adverse reactions for an investigational product. Its function is regulatory, not descriptive.
An event not in the RSI is unexpected. A serious, unexpected, suspected adverse reaction is a SUSAR, and SUSARs must be reported to regulators on an expedited timeline.
Three rules follow. Use the current version of the IB, since the RSI changes. Use the RSI and nothing else — not the CSR, not the SmPC of a related product, not your memory. And if you are uncertain, ask; the consequences of guessing here are not editorial.
An event not in the RSI is unexpected. A serious, unexpected, suspected adverse reaction is a SUSAR, and SUSARs must be reported to regulators on an expedited timeline.
Three rules follow. Use the current version of the IB, since the RSI changes. Use the RSI and nothing else — not the CSR, not the SmPC of a related product, not your memory. And if you are uncertain, ask; the consequences of guessing here are not editorial.
The IB is not the label
An IB describes an investigational product. An SmPC or USPI describes an approved one.
They serve different purposes, they are governed by different processes, and their safety sections do not match. Using the IB to answer a question about an approved indication, or the SmPC to assess expectedness in a trial, is a category error — and both directions occur regularly.
They serve different purposes, they are governed by different processes, and their safety sections do not match. Using the IB to answer a question about an approved indication, or the SmPC to assess expectedness in a trial, is a category error — and both directions occur regularly.
Drills
Drill 1 of 3
Which document determines whether a serious adverse reaction is expected?
AThe clinical study report
BThe SmPC
CThe reference safety information in the current IB
DThe protocol safety section
Drill 2 of 3
You are writing an Introduction and need prior clinical experience with the product. Where do you look?
AThe reference safety information
BThe non-clinical studies section
CThe current CSR
DThe IB, effects in humans section
Drill 3 of 3
The IB you have is dated eighteen months ago. What is the concern?
AThe RSI may have been updated, changing which reactions are expected
BNone — IB structure does not change
CIt has expired and is invalid
DThe non-clinical data will be wrong
Production drill
For each request, name the document, the section, and one thing that could go wrong if you used the wrong source.
1. A safety physician asks whether grade 4 neutropenia is expected for the investigational product.
2. A medical science liaison asks what the approved dose is in Germany.
3. An author asks for the mechanism of action for the Introduction.
4. A regulator asks what non-clinical toxicology supported the starting dose.
5. A colleague asks whether an event seen in this trial has been reported before.
2. A medical science liaison asks what the approved dose is in Germany.
3. An author asks for the mechanism of action for the Introduction.
4. A regulator asks what non-clinical toxicology supported the starting dose.
5. A colleague asks whether an event seen in this trial has been reported before.
1. IB — reference safety information. Using the CSR instead would treat an observed event as an expected one; if it is genuinely unexpected, an expedited SUSAR report is missed. This is the highest-consequence error in the whole set.
2. SmPC for Germany (national label). Using the IB would describe an investigational dose that is not approved; using the CCDS would give the company reference, which may differ from what is authorised in that market.
3. IB — non-clinical studies, and effects in humans for clinical pharmacology. Using the protocol background instead risks repeating a summary written years earlier and since superseded.
4. IB — non-clinical studies, toxicology. This is exactly what the section exists for; the CSR contains nothing relevant.
5. IB — effects in humans and RSI, plus the DSUR for the cumulative cross-study picture. The current CSR alone tells you only about this trial.
The pattern: four of the five are answered by the IB, and each by a different section. Writers who have never opened an IB tend to route all five to the CSR, which is wrong for every one of them.
2. SmPC for Germany (national label). Using the IB would describe an investigational dose that is not approved; using the CCDS would give the company reference, which may differ from what is authorised in that market.
3. IB — non-clinical studies, and effects in humans for clinical pharmacology. Using the protocol background instead risks repeating a summary written years earlier and since superseded.
4. IB — non-clinical studies, toxicology. This is exactly what the section exists for; the CSR contains nothing relevant.
5. IB — effects in humans and RSI, plus the DSUR for the cumulative cross-study picture. The current CSR alone tells you only about this trial.
The pattern: four of the five are answered by the IB, and each by a different section. Writers who have never opened an IB tend to route all five to the CSR, which is wrong for every one of them.
Reflection
Have you ever read a reference safety information section? If not, that is this week's task — ask for one and read it.
It is usually short, often a single table. Reading one once permanently changes how you think about the word "expected".
Next week
2.1 — The statistical analysis plan
Week 2 · Module 1
The statistical analysis plan
One document, usually forty pages, that governs what every later claim is allowed to say. If you read only one page of it, read the multiplicity section
Your scenario
In a 52-week severe asthma biologic trial, you report the primary efficacy outcome (annualized exacerbation rate reduction) based on patients who completed all 52 weeks without protocol deviations. During review, the biostatistics director flags a critical error: the SAP specified a Modified Full Analysis Set (mFAS) using a Negative Binomial regression model with Jump-to-Reference multiple imputation for patients who discontinued due to COVID-19. Your draft excluded 34 discontinued patients, inadvertently inflating the treatment effect by 14%. A ten-minute check of SAP Section 5 would have prevented this major audit finding.
The five things to find, in order
| # | What | Why it governs your writing |
|---|---|---|
| 1 | Analysis population definitions | Determines every denominator. ITT, mITT, per-protocol, evaluable, safety — and mITT has no standard definition, so read it. |
| 2 | The estimand | Under ICH E9(R1): population, variable, intercurrent events and their handling, summary measure, treatment condition. Defines what question the trial answers. |
| 3 | Multiplicity strategy | Testing hierarchy or adjustment. Decides what may be called significant. The single highest-value page. |
| 4 | Handling of missing data | Imputation method and sensitivity analyses. Feeds directly into limitations. |
| 5 | Prespecified subgroups | Distinguishes reportable subgroups from exploratory ones. An output pack often contains far more than the SAP prespecified. |
The hierarchy rule
In a fixed-sequence procedure, endpoints are tested in a prespecified order and testing stops at the first failure. Everything below the break was not formally tested, whatever its p value.
Such endpoints are reported with a nominal p value and an explicit statement that formal testing stopped. Describing them as significant is one of the few writing errors with regulatory consequences, and it is entirely avoidable by reading one page.
Such endpoints are reported with a nominal p value and an explicit statement that formal testing stopped. Describing them as significant is one of the few writing errors with regulatory consequences, and it is entirely avoidable by reading one page.
SAP versus protocol
Where the two disagree, the SAP governs the analysis — it is normally finalised later and in more detail. But a discrepancy is itself worth raising, because it may indicate that one document was not updated.
Check the SAP version and date, and check whether it was finalised before database lock and unblinding. A SAP amended after unblinding is a serious concern and must be disclosed.
Check the SAP version and date, and check whether it was finalised before database lock and unblinding. A SAP amended after unblinding is a serious concern and must be disclosed.
Drills
Drill 1 of 3
The SAP hierarchy is PFS → OS → ORR. PFS is significant, OS is not, ORR has p = 0.001. How is ORR reported?
AAs significant, since p = 0.001
BWith a nominal p value, stating that formal testing stopped at OS
COmitted
DReordered above OS in the hierarchy
Drill 2 of 3
The SAP refers to the “mITT population” without further detail. What do you do?
AAssume it means all randomised patients who received one dose
BUse the ITT population instead
CFind the explicit definition; if absent, raise it as a query
DReport it as mITT without defining it
Drill 3 of 3
The output pack contains 14 subgroup analyses; the SAP prespecifies 6. What governs?
AThe output pack — report all 14
BReport whichever eight are most interesting
CReport none — subgroups are always exploratory
DThe SAP — report the six, labelling any others exploratory
Production drill
From this SAP extract in a Cardiovascular Outcome Trial (CVOT), write the statistical methods paragraph and list every constraint it places on your Results section.
SAP v3.0, final, dated 12 Dec 2024 (database lock 20 Dec 2024).
Populations: Full Analysis Set (FAS/ITT) = all randomised patients, by assigned arm. Safety Population = all patients receiving ≥1 dose of study drug, by actual treatment received. Modified FAS (mFAS) = randomised patients excluding sites closed prior to first visit.
Primary endpoint: Time to first 3-point Major Adverse Cardiovascular Event (3-point MACE: cardiovascular death, non-fatal MI, or non-fatal ischemic stroke), centrally adjudicated by a blinded Clinical Events Committee (CEC).
Hypothesis testing: Stratified Cox proportional hazards model at two-sided α = 0.05 (stratified by baseline eGFR <60 vs ≥60 mL/min/1.73m² and prior MI). Hierarchy: Primary 3-point MACE → All-cause mortality → Heart failure hospitalisation → KCCQ health status score. Testing stops at first failure.
Handling of missing data: Primary analysis censors patients without event at date of last contact; sensitivity analysis applies Jump-to-Reference multiple imputation for treatment discontinuations.
Populations: Full Analysis Set (FAS/ITT) = all randomised patients, by assigned arm. Safety Population = all patients receiving ≥1 dose of study drug, by actual treatment received. Modified FAS (mFAS) = randomised patients excluding sites closed prior to first visit.
Primary endpoint: Time to first 3-point Major Adverse Cardiovascular Event (3-point MACE: cardiovascular death, non-fatal MI, or non-fatal ischemic stroke), centrally adjudicated by a blinded Clinical Events Committee (CEC).
Hypothesis testing: Stratified Cox proportional hazards model at two-sided α = 0.05 (stratified by baseline eGFR <60 vs ≥60 mL/min/1.73m² and prior MI). Hierarchy: Primary 3-point MACE → All-cause mortality → Heart failure hospitalisation → KCCQ health status score. Testing stops at first failure.
Handling of missing data: Primary analysis censors patients without event at date of last contact; sensitivity analysis applies Jump-to-Reference multiple imputation for treatment discontinuations.
Statistical methods draft:
“Primary efficacy analyses were conducted in the intention-to-treat (Full Analysis Set) population, comprising all randomised patients analysed according to assigned arm. The primary composite endpoint of time to first 3-point Major Adverse Cardiovascular Event (CV death, non-fatal myocardial infarction, or non-fatal stroke) was evaluated using a two-sided stratified Cox proportional hazards model at an alpha level of 0.05, stratified by baseline renal function (eGFR <60 vs ≥60 mL/min/1.73m²) and prior myocardial infarction history. Multiplicity across secondary endpoints was controlled via a pre-specified fixed-sequence testing hierarchy (3-point MACE → all-cause mortality → heart failure hospitalisation → KCCQ health status score); formal hypothesis testing terminated at the first non-significant outcome. Sensitivity analyses utilized Jump-to-Reference multiple imputation for intercurrent treatment cessations. Safety was evaluated in all participants who received at least one dose of study medication.”
Constraints this SAP places on your Results section:
1. Hierarchy rule: if all-cause mortality fails to reach significance (p ≥ 0.05), heart failure hospitalisation cannot be claimed as statistically significant however small its p value — it must be described as nominal/exploratory.
2. Adjudication rule: only CEC-adjudicated events may be counted in primary efficacy tables; investigator-reported events belong exclusively in safety listings.
3. Stratification rule: the hazard ratio quoted must derive from the stratified model pre-specified in the SAP.
4. Censoring rule: the primary time-to-event finding must report the total patient-years of follow-up and number of adjudicated events in each arm.
5. Sensitivity rule: the Jump-to-Reference multiple imputation result must be reported alongside the primary analysis to prove robustness against dropout.
“Primary efficacy analyses were conducted in the intention-to-treat (Full Analysis Set) population, comprising all randomised patients analysed according to assigned arm. The primary composite endpoint of time to first 3-point Major Adverse Cardiovascular Event (CV death, non-fatal myocardial infarction, or non-fatal stroke) was evaluated using a two-sided stratified Cox proportional hazards model at an alpha level of 0.05, stratified by baseline renal function (eGFR <60 vs ≥60 mL/min/1.73m²) and prior myocardial infarction history. Multiplicity across secondary endpoints was controlled via a pre-specified fixed-sequence testing hierarchy (3-point MACE → all-cause mortality → heart failure hospitalisation → KCCQ health status score); formal hypothesis testing terminated at the first non-significant outcome. Sensitivity analyses utilized Jump-to-Reference multiple imputation for intercurrent treatment cessations. Safety was evaluated in all participants who received at least one dose of study medication.”
Constraints this SAP places on your Results section:
1. Hierarchy rule: if all-cause mortality fails to reach significance (p ≥ 0.05), heart failure hospitalisation cannot be claimed as statistically significant however small its p value — it must be described as nominal/exploratory.
2. Adjudication rule: only CEC-adjudicated events may be counted in primary efficacy tables; investigator-reported events belong exclusively in safety listings.
3. Stratification rule: the hazard ratio quoted must derive from the stratified model pre-specified in the SAP.
4. Censoring rule: the primary time-to-event finding must report the total patient-years of follow-up and number of adjudicated events in each arm.
5. Sensitivity rule: the Jump-to-Reference multiple imputation result must be reported alongside the primary analysis to prove robustness against dropout.
Reflection
For a study you have worked on, what was the multiplicity strategy? If you cannot answer, find the SAP and locate it — it is usually one paragraph.
Next module
2.2 — Navigating the output pack
Week 2 · Module 2
Navigating the output pack
Nine hundred pages with a numbering system inherited from ICH E3. Learn the map and you can find anything in under a minute
Your scenario
Four days to draft. You spend the first day scrolling. On day three you are still re-finding the same three tables because you never wrote down where they were. The pack was not the problem.
The numbering map
Figure 2.2 — The output pack map. Four series, one logic. 14.1 tells you who; 14.2 what happened to them; 14.3 at what cost; 16.2 what happened to any individual. Note the flag on 14.3 — safety outputs use the treated population, so their denominators differ from 14.2 by design.
Your first two hours
| Order | Do this | Because |
|---|---|---|
| 1 | Open 14.1.1 and write down every population n | Everything downstream uses these numbers |
| 2 | Open the SAP and find the estimand, hierarchy and population definitions | One page that governs every claim you may make |
| 3 | Locate the primary endpoint table and its figure; read them together | Confirms they agree and share a data cut |
| 4 | Build the number sheet | Makes your draft checkable and stops you re-finding figures |
| 5 | Write your queries before drafting | Answers arrive while you write rather than after |
Read every footer
A table footer carries the data cut-off, the population, the program name and the generation date. Two seconds each, and it catches the commonest cause of two outputs disagreeing.
If a figure and a table give different values, check their footers before doing anything else. Nine times out of ten the answer is there.
If a figure and a table give different values, check their footers before doing anything else. Nine times out of ten the answer is there.
The CDISC Clinical Data Pipeline — From Electronic CRF to Final CSR
As a medical writer, you are not writing SAS code, but understanding how raw patient data transforms into the Tables, Figures, and Listings (TFLs) on your desk eliminates confusion when reading table titles and footers.
| Pipeline Stage | CDISC Standard / Layer | What It Contains | Medical Writer Significance |
|---|---|---|---|
| 1. Data Collection | eCRF / EDC (e.g., Medidata Rave) | Raw patient records entered by clinical site coordinators at patient visits. | Defines whether a clinical variable was captured at all. If it is not on the eCRF, it cannot be analysed. |
| 2. Tabulation | SDTM (Study Data Tabulation Model) | Standardised raw domains: DM (Demographics), AE (Adverse Events), LB (Laboratory), VS (Vital Signs), CM (Concomitant Meds). |
Direct representation of raw site entries without derivations. Used by regulators for independent data audits. |
| 3. Analysis Ready | ADaM (Analysis Data Model) | Derived analysis datasets: ADSL (Subject-Level), ADAE (Adverse Events Analysis), ADTTE (Time-to-Event), ADLB (Lab Analysis). |
The definitive source datasets behind every TFL. When a table footer reads Source: ADSL, ADAE, it references these datasets. |
| 4. Output Display | TFL Output Pack (ICH E3) | Post-processed summary tables (14.1, 14.2, 14.3) and patient-level listings (16.2). | The primary quantitative source from which you draft clinical study reports, manuscripts, and briefing packages. |
Drills
Drill 1 of 3
You need the number of patients who discontinued because of an adverse event. Which series?
A14.3.x
B14.1.x
C14.2.x
D16.2.x
Drill 2 of 3
A table and a figure give different medians for the same endpoint. What is the first thing you check?
AUse the table, since tables are more precise
BThe footers of both, for data cut-off and population
CEmail the statistician immediately
DUse the figure, since readers will see the curve
Drill 3 of 3
What should the number sheet record for each figure you plan to quote?
AThe value only
BValue and table reference
CValue, table reference, population and n, data cut-off, and where it will appear
DValue and the name of whoever gave it to you
Production drill
Build the number sheet for an abstract from this pack, and note which outputs you still need.
You must write a 250-word abstract covering: design, population, primary endpoint result, key secondary, response rate, and headline safety.
Available in the pack:
14.1.1 Disposition — 402 randomised, 389 treated. Cut 15 Jan.
14.1.2 Baseline characteristics (ITT). Cut 15 Jan.
14.2.1 PFS primary analysis (ITT). Cut 15 Jan.
14.2.2 PFS Kaplan–Meier figure. Cut 30 Sep.
14.2.5 OS analysis (ITT). Cut 15 Jan.
14.2.8 ORR — header states “response-evaluable”, no n given.
14.2.11 Subgroup forest plot — 11 subgroups.
14.3.1 TEAE summary — header states n = 402.
14.3.6 Discontinuation due to AE.
Available in the pack:
14.1.1 Disposition — 402 randomised, 389 treated. Cut 15 Jan.
14.1.2 Baseline characteristics (ITT). Cut 15 Jan.
14.2.1 PFS primary analysis (ITT). Cut 15 Jan.
14.2.2 PFS Kaplan–Meier figure. Cut 30 Sep.
14.2.5 OS analysis (ITT). Cut 15 Jan.
14.2.8 ORR — header states “response-evaluable”, no n given.
14.2.11 Subgroup forest plot — 11 subgroups.
14.3.1 TEAE summary — header states n = 402.
14.3.6 Discontinuation due to AE.
Number sheet (abstract only):
· Randomised / treated n — 14.1.1 — 402 / 389 — cut 15 Jan — Methods, Results
· Median PFS both arms — 14.2.1 — ITT n=402 — cut 15 Jan — Results
· PFS HR + 95% CI + p — 14.2.1 — ITT — cut 15 Jan — Results
· Median OS both arms + HR — 14.2.5 — ITT — cut 15 Jan — Results
· ORR both arms — 14.2.8 — population n TBC — cut TBC — Results
· Grade ≥3 TEAE — 14.3.1 — denominator wrong — cut 15 Jan — Results
Four queries before drafting:
1. Table 14.3.1 uses n = 402. Safety must use the treated population, 389. Every percentage in that table is understated. Most serious of the four.
2. Figure 14.2.2 is at the 30 September cut while table 14.2.1 is at 15 January. Do not quote from the figure; request a regenerated version.
3. Table 14.2.8 does not state its population n. Response-evaluable is normally a smaller denominator than ITT. Confirm before quoting percentages.
4. Forest plot 14.2.11 shows 11 subgroups where the SAP prespecified six. Identify which six.
Not needed for the abstract: baseline table (Results text only), full subgroup detail, and discontinuation figures unless the word count allows.
The habit this builds: writing down what you do not yet have is as valuable as writing down what you do. Those four queries go out before you draft a sentence, and the answers arrive while you write.
· Randomised / treated n — 14.1.1 — 402 / 389 — cut 15 Jan — Methods, Results
· Median PFS both arms — 14.2.1 — ITT n=402 — cut 15 Jan — Results
· PFS HR + 95% CI + p — 14.2.1 — ITT — cut 15 Jan — Results
· Median OS both arms + HR — 14.2.5 — ITT — cut 15 Jan — Results
· ORR both arms — 14.2.8 — population n TBC — cut TBC — Results
· Grade ≥3 TEAE — 14.3.1 — denominator wrong — cut 15 Jan — Results
Four queries before drafting:
1. Table 14.3.1 uses n = 402. Safety must use the treated population, 389. Every percentage in that table is understated. Most serious of the four.
2. Figure 14.2.2 is at the 30 September cut while table 14.2.1 is at 15 January. Do not quote from the figure; request a regenerated version.
3. Table 14.2.8 does not state its population n. Response-evaluable is normally a smaller denominator than ITT. Confirm before quoting percentages.
4. Forest plot 14.2.11 shows 11 subgroups where the SAP prespecified six. Identify which six.
Not needed for the abstract: baseline table (Results text only), full subgroup detail, and discontinuation figures unless the word count allows.
The habit this builds: writing down what you do not yet have is as valuable as writing down what you do. Those four queries go out before you draft a sentence, and the answers arrive while you write.
Reflection
Next time you get an output pack, time your first two hours against the five-step order. How much re-finding did it save?
Next module
2.3 — Reading tables
Week 2 · Module 3
Reading tables
Denominators, footers, and seven cross-checks that catch most of what quality control would otherwise send back
Your scenario
Every percentage in your safety section is slightly wrong, in the sponsor's favour. Not because anyone intended it — because the table header said n = 402 and you used it, when the treated population was 389. The error is invisible in the prose and survives every review that does not go back to the source.
Anatomy of an output table
| Part | What to check |
|---|---|
| Title | The endpoint and the population, usually both named |
| Column headers | The n per arm. Compare against the disposition table. |
| Row labels | Whether categories are mutually exclusive, and whether a total row is meaningful |
| Cells | Whether percentages recompute from counts using the stated denominator |
| Footer | Data cut-off, population definition, program name, generation date, abbreviation key |
| Footnotes | Where the exceptions live — missing data, alternative denominators, censoring rules |
The seven cross-checks
Run these on every pack. They take about twenty minutes and catch most quality control findings, and occasionally a real error in the outputs.
1. Do arm totals in each table match the disposition table?
2. Does the safety denominator equal the treated population?
3. Do table medians match where the curve crosses 50%?
4. Do subgroup n values sum to the overall n?
5. Does the event count match between the primary table and its figure?
6. Do percentages recompute from their counts?
7. Do all outputs share the same data cut-off?
1. Do arm totals in each table match the disposition table?
2. Does the safety denominator equal the treated population?
3. Do table medians match where the curve crosses 50%?
4. Do subgroup n values sum to the overall n?
5. Does the event count match between the primary table and its figure?
6. Do percentages recompute from their counts?
7. Do all outputs share the same data cut-off?
Never resolve a discrepancy yourself
Do not pick the number that looks right. Do not recalculate and substitute. Do not quietly use the one that suits the narrative.
Raise it in writing, with both table references and both values. Usually there is an explanation. Occasionally you have found a real error — and you will have found it before a regulator did.
Raise it in writing, with both table references and both values. Usually there is an explanation. Occasionally you have found a real error — and you will have found it before a regulator did.
Drills
Drill 1 of 3
The efficacy table shows n = 402; the safety table shows n = 389. What do you do?
AQuery it as an inconsistency
BUse 402 throughout for consistency
CUse 389 throughout for consistency
DNothing — different populations, both correct
Drill 2 of 3
Subgroup n values by biomarker status sum to 371, but the ITT population is 402. What does this mean?
A31 patients had no evaluable result; report the biomarker-evaluable denominator explicitly
BThe table is wrong; query it as an error
CAssign the 31 to the larger subgroup
DUse 402 as the denominator for both subgroups
Drill 3 of 3
A percentage in a table does not recompute from its stated count and denominator. Most likely cause?
ARounding
BA footnote specifies a different denominator for that row
CA programming error
DUse your own recalculated figure
Production drill
Run the seven cross-checks on this Type 2 Diabetes / Obesity trial output pack and list every finding.
14.1.1 Disposition: 402 randomised (201 Tirzepatide/Semaglutide analogue vs 201 Placebo); 389 treated (196 active / 193 placebo); 371 completed 52-week treatment.
14.2.1 HbA1c & Weight Loss (mITT, n = 395): LS Mean HbA1c change −1.42% vs −0.11% (p < 0.001); Mean body weight change −12.4% vs −2.1% (p < 0.001). Data cut: 15 Jan 2025.
14.2.2 HbA1c (Per-protocol completers, n = 348): LS Mean difference −1.55% (95% CI −1.76 to −1.34). Data cut: 15 Jan 2025.
14.3.1 TEAE Overview (Header reads: n = 402): Any TEAE 88.4% vs 79.2%; Grade ≥3 TEAEs 14.2% vs 11.4%; Serious AEs 6.1% vs 5.7%. Data cut: 15 Jan 2025.
14.3.2 Most Frequent Gastrointestinal TEAEs (Safety Population, n = 196 active / 193 placebo): Nausea 31.6% vs 8.3%; Diarrhoea 22.4% vs 7.8%; Vomiting 14.3% vs 3.1%. Data cut: 30 Sep 2024 (Interim Run).
14.2.1 HbA1c & Weight Loss (mITT, n = 395): LS Mean HbA1c change −1.42% vs −0.11% (p < 0.001); Mean body weight change −12.4% vs −2.1% (p < 0.001). Data cut: 15 Jan 2025.
14.2.2 HbA1c (Per-protocol completers, n = 348): LS Mean difference −1.55% (95% CI −1.76 to −1.34). Data cut: 15 Jan 2025.
14.3.1 TEAE Overview (Header reads: n = 402): Any TEAE 88.4% vs 79.2%; Grade ≥3 TEAEs 14.2% vs 11.4%; Serious AEs 6.1% vs 5.7%. Data cut: 15 Jan 2025.
14.3.2 Most Frequent Gastrointestinal TEAEs (Safety Population, n = 196 active / 193 placebo): Nausea 31.6% vs 8.3%; Diarrhoea 22.4% vs 7.8%; Vomiting 14.3% vs 3.1%. Data cut: 30 Sep 2024 (Interim Run).
Findings to raise before writing:
1. Table 14.3.1 denominator error: The safety overview header uses n = 402 (the randomised cohort) rather than the safety treated population (n = 389: 196 active / 193 placebo). Every incidence percentage in this table is calculated against the wrong base. Stop drafting safety until this table is re-run with correct safety denominators.
2. Data cut mismatch: Table 14.3.2 (Gastrointestinal TEAEs) was generated from an obsolete interim cut (30 Sep 2024); all other outputs are from the final database lock (15 Jan 2025). Quoting Table 14.3.2 will create direct discrepancies with Table 14.3.1. Request the 15 Jan 2025 refresh of Table 14.3.2.
3. Per-protocol exclusions: 54 patients (402 − 348) were excluded from per-protocol completers analysis. Check Section 14.1.2 protocol deviations for specific exclusion reasons (e.g. non-compliance <80% dosing, forbidden rescue medications) before drafting Methods.
4. Efficacy population denominator: Table 14.2.1 reports mITT n = 395 (7 randomised patients excluded due to no post-baseline HbA1c). Confirm the pre-specified mITT definition in the SAP before writing the primary Results sentence.
Notice what you did: You caught two serious data mismatches, a denominator flaw, and verified population exclusions in under ten minutes by comparing table titles, footers, and N counts across the output pack.
1. Table 14.3.1 denominator error: The safety overview header uses n = 402 (the randomised cohort) rather than the safety treated population (n = 389: 196 active / 193 placebo). Every incidence percentage in this table is calculated against the wrong base. Stop drafting safety until this table is re-run with correct safety denominators.
2. Data cut mismatch: Table 14.3.2 (Gastrointestinal TEAEs) was generated from an obsolete interim cut (30 Sep 2024); all other outputs are from the final database lock (15 Jan 2025). Quoting Table 14.3.2 will create direct discrepancies with Table 14.3.1. Request the 15 Jan 2025 refresh of Table 14.3.2.
3. Per-protocol exclusions: 54 patients (402 − 348) were excluded from per-protocol completers analysis. Check Section 14.1.2 protocol deviations for specific exclusion reasons (e.g. non-compliance <80% dosing, forbidden rescue medications) before drafting Methods.
4. Efficacy population denominator: Table 14.2.1 reports mITT n = 395 (7 randomised patients excluded due to no post-baseline HbA1c). Confirm the pre-specified mITT definition in the SAP before writing the primary Results sentence.
Notice what you did: You caught two serious data mismatches, a denominator flaw, and verified population exclusions in under ten minutes by comparing table titles, footers, and N counts across the output pack.
Reflection
Run the seven checks on the last output pack you worked from. How many findings?
Next module
2.4 — Reading figures and listings
Week 2 · Module 4
Reading figures and listings
Survival curves, forest plots and patient-level data — what each one will tell you, and what it will let you claim if you read it carelessly
Your scenario
A medical affairs colleague wants a slide built around a survival curve that appears to plateau at 28%. You check the numbers-at-risk row: five patients and three patients at that timepoint. The plateau is four people. Your answer to that request is a test of whether you can read the figure rather than look at it.
Kaplan–Meier curves
| Feature | What it tells you |
|---|---|
| Each step down | An event. Steps grow taller as fewer patients remain at risk. |
| Tick marks | Censored patients — event-free when last observed. Heavy early ticks mean early dropout; find out why. |
| Where the curve crosses 0.5 | The median. If it never crosses, the median is not reached — which means fewer than half had the event, not that data are missing. |
| Numbers at risk | The most important row on the figure. Single figures mean the tail is unreliable. |
| Curves that cross or separate late | Proportional hazards is likely violated; a single hazard ratio may mislead. |
Forest plots
| Element | What it tells you, and what it does not |
|---|---|
| Square | The point estimate. Its size usually reflects subgroup size — a small square is a warning before you read its position. |
| Horizontal line | The confidence interval. Read width before position. |
| Diamond | The overall estimate. The reliable number on the figure. |
| Interaction p value | Whether the effect genuinely differs between subgroups. The only test that matters, usually printed small at the right. |
| Number of rows | Your multiplicity denominator. Fifteen subgroups means several will look extreme by chance. |
The subgroup trap
Comparing whether one subgroup is significant and another is not is not a valid way to claim they differ. A subgroup of 34 patients will be non-significant almost regardless of the true effect, simply because it is small.
Read the right-hand interaction column before the middle of the figure. The most extreme row on a forest plot is, by construction, the one most likely to be a chance finding.
Read the right-hand interaction column before the middle of the figure. The most extreme row on a forest plot is, by construction, the one most likely to be a chance finding.
Patient-level listings
| Use listings for | Never use them for |
|---|---|
| Writing patient narratives for deaths, serious events and discontinuations | Any figure in a publication |
| Resolving a discrepancy about a specific case | Deriving your own aggregate counts |
| Understanding the sequence of events for one patient | Anything identifiable, in any external document |
Confidentiality
Patient-level listings contain identifiable or re-identifiable information — subject numbers, dates, sites, sometimes ages and rare conditions in combination.
They stay inside the sponsor environment. Narratives written from them are de-identified: no dates of birth, no exact dates where a study day will serve, no site identifiers. If you would not want it in a published article, it does not leave the listing.
They stay inside the sponsor environment. Narratives written from them are de-identified: no dates of birth, no exact dates where a study day will serve, no site identifiers. If you would not want it in a published article, it does not leave the listing.
Drills
Drill 1 of 3
A curve appears to plateau at 28%. Numbers at risk at that timepoint read 5 and 3. What can you write?
A“A plateau was observed, suggesting a cured fraction of 28%.”
BOmit the long-term data
C“The 36-month estimate was 28% (95% CI …), though few patients remained at risk beyond this point.”
D“28% achieved long-term response.”
Drill 2 of 3
A forest plot has 14 rows. One subgroup (n = 31) shows HR 0.34 (95% CI 0.11–1.05); interaction p = 0.28. What can you write?
A“The regimen is particularly effective in this subgroup.”
B“66% of patients in this subgroup benefited.”
COmit the forest plot
D“Treatment effect was consistent across subgroups; no significant interaction was observed (p = 0.28).”
Drill 3 of 3
You are writing a narrative for a patient who died on study. What may appear in it?
AStudy day, age band, relevant history, event sequence, treatment and causality assessment
BDate of birth, exact event dates and the site number
CThe listing row copied verbatim
DNothing — narratives are optional
Production drill
Write what you can defensibly say about each clinical figure, and what you would refuse to say.
Figure A — Time-to-First 3-Point MACE Kaplan–Meier Curve (Cardiovascular Trial). Curves separate from Month 6. Estimated 3-year event-free survival 84.2% vs 76.8%; HR 0.74 (95% CI 0.58–0.94); Log-rank p = 0.014. Numbers at risk at Month 36: Active arm 14 patients, Placebo arm 9 patients.
Figure B — Subgroup Forest Plot for ACR50 Response (Rheumatoid Arthritis Trial). Twelve prespecified subgroups. Subgroup 11 (Prior Biologic Exposure: Experienced, n = 34): OR 1.12 (95% CI 0.45–2.78, p = 0.81); Subgroup 12 (Biologic-Naïve, n = 368): OR 2.45 (95% CI 1.68–3.57, p < 0.001). Interaction p value for prior biologic exposure = 0.22.
Figure C — Time-to-Treatment Discontinuation with Crossing Survival Curves. Curves overlap and cross at Month 4 before separating. Schoenfeld residual test for proportional hazards: p = 0.02.
Figure B — Subgroup Forest Plot for ACR50 Response (Rheumatoid Arthritis Trial). Twelve prespecified subgroups. Subgroup 11 (Prior Biologic Exposure: Experienced, n = 34): OR 1.12 (95% CI 0.45–2.78, p = 0.81); Subgroup 12 (Biologic-Naïve, n = 368): OR 2.45 (95% CI 1.68–3.57, p < 0.001). Interaction p value for prior biologic exposure = 0.22.
Figure C — Time-to-Treatment Discontinuation with Crossing Survival Curves. Curves overlap and cross at Month 4 before separating. Schoenfeld residual test for proportional hazards: p = 0.02.
Figure A (Cardiovascular MACE KM Curve)
Can say: Significant reduction in time to first 3-point MACE over 36 months (HR 0.74, 95% CI 0.58–0.94; p = 0.014). Report 12-, 24-, and 36-month event-free survival estimates alongside the numbers-at-risk table.
Would refuse: Claiming an “extended survival plateau” after Month 30 — only 14 and 9 patients remain at risk. Quote the tail numbers responsibly.
Figure B (Rheumatoid Arthritis Forest Plot)
Can say: Treatment effect for ACR50 was statistically significant in the overall trial population. Subgroup findings are exploratory.
Would refuse: Claiming that the drug “is only effective in biologic-naïve patients and fails in biologic-experienced patients.” The interaction p value is 0.22 (non-significant), meaning there is no statistical evidence that treatment efficacy differs by prior biologic history. The wide CI in experienced patients reflects small sample size (n = 34), not lack of efficacy.
Figure C (Crossing Survival Curves / PH Violation)
Can say: Early curves overlap during the first 4 months before sustained divergence favouring active treatment; report milestone landmark rates and restricted mean survival time (RMST).
Would refuse: Quoting a single global Cox hazard ratio without acknowledging the violation. Schoenfeld p = 0.02 confirms non-proportional hazards, meaning the single HR is an averaged artefact of two distinct phases.
The pattern across all three: The number that governs what you may write is never just the headline p value — it is the numbers-at-risk row, the subgroup interaction p value, and the proportional hazards diagnostic.
Can say: Significant reduction in time to first 3-point MACE over 36 months (HR 0.74, 95% CI 0.58–0.94; p = 0.014). Report 12-, 24-, and 36-month event-free survival estimates alongside the numbers-at-risk table.
Would refuse: Claiming an “extended survival plateau” after Month 30 — only 14 and 9 patients remain at risk. Quote the tail numbers responsibly.
Figure B (Rheumatoid Arthritis Forest Plot)
Can say: Treatment effect for ACR50 was statistically significant in the overall trial population. Subgroup findings are exploratory.
Would refuse: Claiming that the drug “is only effective in biologic-naïve patients and fails in biologic-experienced patients.” The interaction p value is 0.22 (non-significant), meaning there is no statistical evidence that treatment efficacy differs by prior biologic history. The wide CI in experienced patients reflects small sample size (n = 34), not lack of efficacy.
Figure C (Crossing Survival Curves / PH Violation)
Can say: Early curves overlap during the first 4 months before sustained divergence favouring active treatment; report milestone landmark rates and restricted mean survival time (RMST).
Would refuse: Quoting a single global Cox hazard ratio without acknowledging the violation. Schoenfeld p = 0.02 confirms non-proportional hazards, meaning the single HR is an averaged artefact of two distinct phases.
The pattern across all three: The number that governs what you may write is never just the headline p value — it is the numbers-at-risk row, the subgroup interaction p value, and the proportional hazards diagnostic.
Reflection
Find a Kaplan–Meier figure in a recent paper. What are the numbers at risk at the last timepoint shown? Does the text describe the tail more confidently than those numbers support?
Next week
3.1 — The clinical study report
Week 3 · Module 1
The clinical study report
ICH E3 gives every CSR the same skeleton. Learn sixteen section numbers once and you can navigate any sponsor's report
Your scenario
You are asked why fourteen patients were excluded from the per-protocol population. The output pack gives you the count and nothing else. The reasons are in the CSR, in a section most writers never open — and they are exactly what your Methods paragraph needs.
The ICH E3 skeleton
Figure 3.1 — CSR structure. The numbering is fixed by ICH E3, so it is identical across sponsors. Section 10 is the natural entry point because every denominator in the report depends on it, and section 14 is where the output pack lives — which is why output tables are numbered 14.x.
The CSR is not a source for figures
CSR narrative text is derived. It was written at a particular data cut and may have been superseded. Quoting a median from CSR prose rather than from a table in section 14 is the most common way a stale number reaches publication.
Use the CSR for reasons, context and narrative — why patients were excluded, what deviations occurred, what happened to an individual. Use section 14 for values.
Use the CSR for reasons, context and narrative — why patients were excluded, what deviations occurred, what happened to an individual. Use section 14 for values.
Drills
Drill 1 of 3
You need to know why 14 patients were excluded from the per-protocol population. Where do you look?
ASection 9
BSection 10
CSection 12
DSection 14
Drill 2 of 3
The CSR discussion concludes the regimen has a favourable benefit-risk profile. How should you treat this?
ACopy it into the manuscript Discussion
BIgnore it entirely
CRead it as the sponsor’s interpretation, then verify the claim against the design
DTreat it as a regulatory finding
Drill 3 of 3
Where in the CSR would you find the protocol and SAP as they were actually filed?
ASection 9
BSection 14
CSection 16.2
DSection 16.1
Production drill
For each question, give the CSR section and say whether the CSR is the right source at all.
1. What was the median PFS?
2. How many patients had a major protocol deviation, and of what kind?
3. Was the trial registered?
4. What happened to the patient who died on day 84?
5. What were the stratification factors?
6. Which version of the SAP governed the analysis?
7. Did the sponsor consider the benefit-risk favourable?
2. How many patients had a major protocol deviation, and of what kind?
3. Was the trial registered?
4. What happened to the patient who died on day 84?
5. What were the stratification factors?
6. Which version of the SAP governed the analysis?
7. Did the sponsor consider the benefit-risk favourable?
1. Section 14 tables. The CSR narrative is not the right source — go to the output table and check its data cut-off.
2. Section 10. Right source. Counts and reasons together, which no output table gives you.
3. Sections 1–8 give the registration number, but verify against the registry itself — what matters is whether registration was prospective, which the CSR will not tell you.
4. Section 12 narrative, supported by listings in 16.2. Right source, and the only place the sequence of events is written out.
5. Section 9 for the narrative description; cross-check against the SAP in 16.1, since strata are occasionally described loosely in the CSR.
6. Section 16.1. Right source — and check the SAP date against the database lock date.
7. Section 13. Right source for the sponsor’s view, but it is an interpretation, not a finding. Read it, verify it, do not inherit it.
The pattern: the CSR is the right source for five of seven. For question 1 it is actively the wrong source, and for question 3 it is incomplete. Knowing which is which is what stops stale numbers reaching print.
2. Section 10. Right source. Counts and reasons together, which no output table gives you.
3. Sections 1–8 give the registration number, but verify against the registry itself — what matters is whether registration was prospective, which the CSR will not tell you.
4. Section 12 narrative, supported by listings in 16.2. Right source, and the only place the sequence of events is written out.
5. Section 9 for the narrative description; cross-check against the SAP in 16.1, since strata are occasionally described loosely in the CSR.
6. Section 16.1. Right source — and check the SAP date against the database lock date.
7. Section 13. Right source for the sponsor’s view, but it is an interpretation, not a finding. Read it, verify it, do not inherit it.
The pattern: the CSR is the right source for five of seven. For question 1 it is actively the wrong source, and for question 3 it is incomplete. Knowing which is which is what stops stale numbers reaching print.
Reflection
Have you ever read section 10 of a CSR? If you have access to one, read it now — it is usually eight to twelve pages.
Next module
3.2 — Safety documents
Week 3 · Module 2
Safety documents
DSURs, PBRERs, narratives and line listings — a document family with its own vocabulary, its own timelines, and consequences for getting it wrong
Your scenario
You are asked for a cumulative picture of hepatic events across the whole development programme. You have the CSR for the pivotal trial. It is the wrong document — one study cannot give a cumulative picture, and the document that can is one most writers have never opened.
The family
| Document | Scope and timing | Writer uses it for |
|---|---|---|
| DSUR Development Safety Update Report | Annual, investigational products, across all studies | Cumulative safety picture during development; background for an Introduction |
| PBRER Periodic Benefit-Risk Evaluation Report (formerly PSUR) | Periodic, authorised products | Post-marketing safety and benefit-risk; input to labelling changes |
| SUSAR report | Expedited, per event | Rarely written by medical writers, but the concept governs expectedness work |
| Patient narrative | Per patient, in the CSR | Required for deaths, other serious events, and discontinuations for adverse events |
| Line listing | Per event, tabular | Source for narratives; never a publication source |
| Risk management plan | Product-level, regulatory | Identified and potential risks, missing information, pharmacovigilance activities |
Three terms that are not interchangeable
Adverse event — any untoward occurrence in a participant, with no causal judgement.
Adverse reaction — an event where a causal relationship to the product is at least a reasonable possibility.
Serious — a regulatory category: death, life-threatening, hospitalisation or prolongation of it, persistent disability, congenital anomaly, or an important medical event. Serious is not the same as severe. A severe headache may be non-serious; an asymptomatic finding that causes hospitalisation is serious.
Writers mix serious and severe constantly. In a safety document that is a factual error, not a word choice.
Adverse reaction — an event where a causal relationship to the product is at least a reasonable possibility.
Serious — a regulatory category: death, life-threatening, hospitalisation or prolongation of it, persistent disability, congenital anomaly, or an important medical event. Serious is not the same as severe. A severe headache may be non-serious; an asymptomatic finding that causes hospitalisation is serious.
Writers mix serious and severe constantly. In a safety document that is a factual error, not a word choice.
Anatomy of a narrative
| Element | Written as |
|---|---|
| Patient identifier | Study subject number only. No initials, no date of birth. |
| Demographics | Age band and sex. Exact age only where clinically relevant. |
| Relevant history | Conditions and concomitant medications bearing on the event |
| Study treatment | Dose, start day, and any modification before the event |
| Event sequence | By study day, not calendar date. Onset, course, interventions, outcome. |
| Causality | The investigator's assessment, reported as such — not the writer's opinion |
| Outcome | Resolved, resolving, ongoing, fatal, unknown |
Two rules for narratives
Study days, not dates. Calendar dates combined with a site identifier and a rare condition can re-identify a patient. Study day 84 carries the clinical meaning without the risk.
Report the investigator’s causality assessment, do not form your own. The narrative states what the investigator concluded. A writer who upgrades or downgrades causality has altered a regulatory record.
Report the investigator’s causality assessment, do not form your own. The narrative states what the investigator concluded. A writer who upgrades or downgrades causality has altered a regulatory record.
Drills
Drill 1 of 3
You need the cumulative incidence of hepatic events across all studies of an investigational product. Which document?
AThe DSUR
BThe CSR for the pivotal trial
CThe PBRER
DThe SmPC
Drill 2 of 3
A patient has a grade 3 headache requiring no intervention and no hospitalisation. How is it classified?
ASerious, because it is grade 3
BSevere but not serious
CNeither severe nor serious
DSerious and severe are the same thing
Drill 3 of 3
The listing records the investigator's causality as “possibly related”. You think the event was clearly unrelated. What do you write?
A“Unrelated to study treatment.”
BOmit the causality assessment
C“The investigator assessed the event as possibly related to study treatment.”
DGive both assessments side by side
Production drill
Write the safety narrative from this line listing extract for a Psoriatic Arthritis patient receiving an oral JAK inhibitor. De-identify appropriately.
Listing 16.2.7, subject 1042-0087
DOB 14 Mar 1961 · Female · Site 1042, Chennai · Psoriatic Arthritis Phase III Trial
Randomised 08 Feb 2024, oral investigational JAK1 inhibitor 15 mg once daily from 09 Feb 2024
Medical history: Psoriatic arthritis, hypertension, BMI 31.2 kg/m². Concomitant: Methotrexate 15 mg/week, Amlodipine 5 mg/day
02 May 2024 (Study Day 84) — Severe left calf pain, swelling, and erythema. Ultrasound confirms acute left popliteal Deep Vein Thrombosis (DVT). CTCAE Grade 3.
02 May 2024 — Study drug permanently discontinued.
04 May 2024 — Admitted to hospital for therapeutic anticoagulation initiation (Apixaban).
19 May 2024 — Discharged with resolved acute symptoms on maintenance oral anticoagulation.
Investigator causality: Related. Sponsor causality: Related (Adverse Event of Special Interest / Class Warning).
Seriousness criterion: Inpatient hospitalisation.
DOB 14 Mar 1961 · Female · Site 1042, Chennai · Psoriatic Arthritis Phase III Trial
Randomised 08 Feb 2024, oral investigational JAK1 inhibitor 15 mg once daily from 09 Feb 2024
Medical history: Psoriatic arthritis, hypertension, BMI 31.2 kg/m². Concomitant: Methotrexate 15 mg/week, Amlodipine 5 mg/day
02 May 2024 (Study Day 84) — Severe left calf pain, swelling, and erythema. Ultrasound confirms acute left popliteal Deep Vein Thrombosis (DVT). CTCAE Grade 3.
02 May 2024 — Study drug permanently discontinued.
04 May 2024 — Admitted to hospital for therapeutic anticoagulation initiation (Apixaban).
19 May 2024 — Discharged with resolved acute symptoms on maintenance oral anticoagulation.
Investigator causality: Related. Sponsor causality: Related (Adverse Event of Special Interest / Class Warning).
Seriousness criterion: Inpatient hospitalisation.
Model narrative:
“Subject 1042-0087, a woman in her early sixties with a history of psoriatic arthritis, hypertension, and obesity (BMI 31.2 kg/m²) receiving background methotrexate and amlodipine, was randomised to the investigational JAK1 inhibitor 15 mg once daily, commencing therapy on study day 1.
On study day 84, the subject developed acute pain and erythema in the left lower extremity; compression Doppler ultrasonography confirmed a Grade 3 acute left popliteal deep vein thrombosis (DVT). The investigational product was permanently discontinued on the same day. On study day 86, the subject was admitted to hospital for therapeutic anticoagulation management with apixaban, thereby meeting the seriousness criterion of inpatient hospitalisation.
The acute symptoms resolved, and the subject was discharged on study day 101 on maintenance oral anticoagulation. Both the investigator and the sponsor assessed the deep vein thrombosis as related to study treatment (an adverse event of special interest for the JAK inhibitor class). The clinical outcome was recorded as resolved with ongoing secondary thromboprophylaxis.”
Six de-identification and safety writing decisions:
1. Date of birth removed — replaced with an age band (“woman in her early sixties”).
2. Site name and geographic city removed. Site number and city combined with date of birth pose re-identification risks under HIPAA / GDPR.
3. Calendar dates converted to relative study days (e.g. Study Day 1, Study Day 84, Study Day 86).
4. Pre-specified seriousness criterion explicitly articulated (“meeting the seriousness criterion of inpatient hospitalisation”).
5. Adverse Event of Special Interest (AESI) framing — documenting both investigator and sponsor causality views.
6. Objective diagnostic confirmation stated (Doppler ultrasonography) before reporting drug discontinuation.
“Subject 1042-0087, a woman in her early sixties with a history of psoriatic arthritis, hypertension, and obesity (BMI 31.2 kg/m²) receiving background methotrexate and amlodipine, was randomised to the investigational JAK1 inhibitor 15 mg once daily, commencing therapy on study day 1.
On study day 84, the subject developed acute pain and erythema in the left lower extremity; compression Doppler ultrasonography confirmed a Grade 3 acute left popliteal deep vein thrombosis (DVT). The investigational product was permanently discontinued on the same day. On study day 86, the subject was admitted to hospital for therapeutic anticoagulation management with apixaban, thereby meeting the seriousness criterion of inpatient hospitalisation.
The acute symptoms resolved, and the subject was discharged on study day 101 on maintenance oral anticoagulation. Both the investigator and the sponsor assessed the deep vein thrombosis as related to study treatment (an adverse event of special interest for the JAK inhibitor class). The clinical outcome was recorded as resolved with ongoing secondary thromboprophylaxis.”
Six de-identification and safety writing decisions:
1. Date of birth removed — replaced with an age band (“woman in her early sixties”).
2. Site name and geographic city removed. Site number and city combined with date of birth pose re-identification risks under HIPAA / GDPR.
3. Calendar dates converted to relative study days (e.g. Study Day 1, Study Day 84, Study Day 86).
4. Pre-specified seriousness criterion explicitly articulated (“meeting the seriousness criterion of inpatient hospitalisation”).
5. Adverse Event of Special Interest (AESI) framing — documenting both investigator and sponsor causality views.
6. Objective diagnostic confirmation stated (Doppler ultrasonography) before reporting drug discontinuation.
Reflection
In your own writing, have you used "serious" and "severe" interchangeably? Search your last safety text.
Next module
3.3 — CRFs and patient-level data
Week 3 · Module 3
CRFs and patient-level data
The fastest way to answer whether a variable exists at all — and the boundary you must not cross when handling individual records
Your scenario
A reviewer asks you to adjust for smoking history. You spend two days searching the CSR and the output pack for it. Ten minutes with the case report form would have told you it was never collected — which is not a failure, but it is a limitation you should have written in the first draft.
The CRF as an index of what exists
The case report form is the instrument used to collect protocol-required data on each participant. Its blank version is a complete index of what the study captured. If a field is not there, the variable does not exist in the database and no analysis can recover it.
| Question | How the CRF answers it in minutes |
|---|---|
| Was this variable collected? | Scan the module list. Absent field means absent variable. |
| At which visits? | CRF modules map to visits; cross-check the schedule of assessments. |
| In what form? | Free text, coded list, or a scale — determines what analysis was possible. |
| Which coding dictionary? | MedDRA for events, WHO Drug for medications. Note the version. |
| Could this confounder have been adjusted for? | Only if the field exists. This is your limitations paragraph. |
The limitations shortcut
When you need to name unmeasured confounders for a limitations paragraph, do not guess and do not write a generic sentence. Open the CRF and list what is missing.
Performance status trajectory, frailty, socioeconomic status, smoking, adherence and patient preference are commonly absent from trial and database CRFs alike. Naming the two or three that actually matter for your question is far more persuasive than a generic residual confounding sentence — and it takes ten minutes.
Performance status trajectory, frailty, socioeconomic status, smoking, adherence and patient preference are commonly absent from trial and database CRFs alike. Naming the two or three that actually matter for your question is far more persuasive than a generic residual confounding sentence — and it takes ten minutes.
Handling patient-level data
| Permitted | Not permitted |
|---|---|
| Reading listings inside the sponsor environment to write narratives | Copying listing rows into any document that will circulate |
| Resolving a discrepancy about a specific case | Deriving your own aggregate counts from listings |
| Quoting a de-identified narrative in a CSR | Dates of birth, exact event dates, site names, initials |
| Describing a case by study day and age band | Any combination that could re-identify a rare case |
Never derive your own aggregates
It is tempting, when a count is missing from the output pack, to tally it from the listing yourself. Do not.
You will not apply the same population definition, the same censoring rules or the same coding hierarchy that the programmed output uses. Your number will differ from the one in the pack, and nobody will be able to reconcile them. Request the output instead.
You will not apply the same population definition, the same censoring rules or the same coding hierarchy that the programmed output uses. Your number will differ from the one in the pack, and nobody will be able to reconcile them. Request the output instead.
Drills
Drill 1 of 3
A reviewer asks you to adjust for a variable. Fastest way to establish whether that is possible?
ASearch the CSR
BScan the output pack
CCount it yourself from the listings
DCheck the blank CRF or data specification
Drill 2 of 3
A count you need is not in the output pack, but you could tally it from listing 16.2.4. What do you do?
ARequest the output from the statistician
BTally it and use your figure
CTally it and ask someone to check
DOmit the figure
Drill 3 of 3
Which detail must not appear in a narrative that will be included in a CSR?
AThe study subject number
BThe date of birth and the site name
CThe study day of onset
DThe investigator’s causality assessment
Production drill
From this CRF module list, write the limitations sentence on unmeasured confounding, and answer the three requests.
CRF modules present: demographics (age, sex, race, ethnicity) · medical history (coded) · concomitant medications (coded) · ECOG at baseline only · disease characteristics (stage, histology, biomarker) · prior therapy (lines, agents) · study drug administration · adverse events (MedDRA) · laboratory · vital signs · tumour assessment (RECIST) · survival follow-up
Requests:
1. A reviewer asks you to adjust for smoking status.
2. An author wants to describe change in performance status over time.
3. A colleague wants to report adherence to study drug.
Requests:
1. A reviewer asks you to adjust for smoking status.
2. An author wants to describe change in performance status over time.
3. A colleague wants to report adherence to study drug.
Limitations sentence:
“Several potentially prognostic variables were not captured, including smoking history, socioeconomic status, and performance status beyond baseline; residual confounding by these factors cannot be excluded, and their absence particularly limits interpretation of the observed differences in tolerability.”
Request 1 — smoking status. Not collected. Respond: “Smoking status was not captured on the case report form and is therefore not available for adjustment. We have added this to the limitations.” Naming the CRF as the reason is what makes the response credible rather than evasive.
Request 2 — ECOG over time. ECOG was collected at baseline only. A change-over-time analysis is impossible. This is the request most likely to be made in good faith by an author who assumes longitudinal collection, and the CRF answers it in seconds.
Request 3 — adherence. There is a study drug administration module, so some exposure data exist — but whether it supports an adherence metric depends on whether it records planned versus actual doses. This one is a query, not a yes or no. Ask what the module captures before promising anything.
What this drill demonstrates: ten minutes with a CRF answers three requests that would otherwise take days, and produces a specific limitations sentence rather than a generic one. The third answer — recognising that the CRF gets you to a precise question rather than a final answer — is the mark of someone who has actually used one.
“Several potentially prognostic variables were not captured, including smoking history, socioeconomic status, and performance status beyond baseline; residual confounding by these factors cannot be excluded, and their absence particularly limits interpretation of the observed differences in tolerability.”
Request 1 — smoking status. Not collected. Respond: “Smoking status was not captured on the case report form and is therefore not available for adjustment. We have added this to the limitations.” Naming the CRF as the reason is what makes the response credible rather than evasive.
Request 2 — ECOG over time. ECOG was collected at baseline only. A change-over-time analysis is impossible. This is the request most likely to be made in good faith by an author who assumes longitudinal collection, and the CRF answers it in seconds.
Request 3 — adherence. There is a study drug administration module, so some exposure data exist — but whether it supports an adherence metric depends on whether it records planned versus actual doses. This one is a query, not a yes or no. Ask what the module captures before promising anything.
What this drill demonstrates: ten minutes with a CRF answers three requests that would otherwise take days, and produces a specific limitations sentence rather than a generic one. The third answer — recognising that the CRF gets you to a precise question rather than a final answer — is the mark of someone who has actually used one.
Reflection
Have you ever seen a blank CRF? If not, request one for a study you work on. It is the fastest hour of learning available in this course.
Next module
3.4 — Labelling and regulatory documents
Week 3 · Module 4
Labelling and regulatory documents
SmPC, USPI, CCDS and the eCTD — the documents that define what may be said about an approved product, and which one governs where
Your scenario
A medical information enquiry asks about a dose used in a published trial. You answer from the trial publication. That dose is not in the approved label. You have just provided off-label information in response to what may have been a routine request, without recognising it as such.
The label family
| Document | Scope | What it governs |
|---|---|---|
| CCDS Company Core Data Sheet | Company-wide, internal | The reference from which national labels are derived. Not itself an approved label anywhere. |
| SmPC Summary of Product Characteristics | European Union | The approved product information. Definitive for EU indication, dosing, contraindications, warnings. |
| USPI US Prescribing Information | United States | The approved US label, with its own structure including boxed warnings. |
| PIL / patient leaflet | Patient-facing | Plain-language version derived from the SmPC. Different register, same content. |
| eCTD Module 2 | Regulatory submission | Clinical Overview (2.5) and Clinical Summary (2.7) — the sponsor's argument to the regulator |
Labels differ by market
The same product may have a different approved indication, dose or contraindication in the EU, the US, Japan and India. Always establish which market the question concerns before answering it.
The CCDS is the company's internal reference and may contain safety information not yet reflected in every national label. It is not a substitute for the label of the market in question, and quoting it as though it were is a common and avoidable error.
The CCDS is the company's internal reference and may contain safety information not yet reflected in every national label. It is not a substitute for the label of the market in question, and quoting it as though it were is a common and avoidable error.
On-label and off-label
Anything outside the approved indication, population, dose or route is off-label in that market — regardless of how strong the published evidence is.
Trial data are not label data. A dose studied in a Phase III trial is off-label until it appears in the approved label. Responding to an unsolicited request about off-label use is permitted in most jurisdictions under defined conditions; proactively promoting it is not. Recognising which you are doing is the whole skill, and Module 4.3 returns to it.
Trial data are not label data. A dose studied in a Phase III trial is off-label until it appears in the approved label. Responding to an unsolicited request about off-label use is permitted in most jurisdictions under defined conditions; proactively promoting it is not. Recognising which you are doing is the whole skill, and Module 4.3 returns to it.
Drills
Drill 1 of 3
A clinician in Germany asks about the approved dose. Which document?
AThe CCDS
BThe USPI
CThe SmPC
DThe investigator’s brochure
Drill 2 of 3
A published Phase III trial used a dose higher than the approved one. A clinician asks about it. How is that request classified?
AOn-label, since the dose is published
BOn-label, since the trial was sponsor-run
CIt cannot be answered at all
DOff-label, and it must be handled under the off-label response process
Drill 3 of 3
What is eCTD Module 2.5?
AThe Clinical Overview — the sponsor’s critical benefit-risk argument
BThe detailed summary of clinical efficacy and safety
CThe clinical study report
DThe quality and manufacturing section
Production drill
For each medical enquiry regarding an oral anticoagulant, name the definitive source document and flag whether it is on- or off-label.
1. A US cardiologist asks whether the product carries a Boxed Warning regarding premature discontinuation and spinal/epidural hematoma risk.
2. An EU hospital pharmacist asks about renal dosage adjustment (Section 4.2 Posology) for an atrial fibrillation patient with CrCl 24 mL/min.
3. An electrophysiologist asks whether the drug can be used in patients with mechanical prosthetic heart valves (where it is contra-indicated).
4. A health authority reviewer requests the sponsor's integrated benefit-risk synthesis for the stroke prevention indication.
5. A pharmacovigilance safety officer asks if a newly identified interaction with P-gp inhibitors has been added to the global company core safety document.
6. A physician in Mumbai asks whether the twice-daily 15 mg regimen is approved by the CDSCO in India.
2. An EU hospital pharmacist asks about renal dosage adjustment (Section 4.2 Posology) for an atrial fibrillation patient with CrCl 24 mL/min.
3. An electrophysiologist asks whether the drug can be used in patients with mechanical prosthetic heart valves (where it is contra-indicated).
4. A health authority reviewer requests the sponsor's integrated benefit-risk synthesis for the stroke prevention indication.
5. A pharmacovigilance safety officer asks if a newly identified interaction with P-gp inhibitors has been added to the global company core safety document.
6. A physician in Mumbai asks whether the twice-daily 15 mg regimen is approved by the CDSCO in India.
1. USPI (US Package Insert). Boxed Warnings are a US FDA statutory labelling instrument; the European SmPC has no direct equivalent (placing major risks under Section 4.4 Warnings and Precautions). On-label.
2. SmPC (Summary of Product Characteristics), Section 4.2 (Posology and Method of Administration) and Section 5.2 (Pharmacokinetics). On-label if explicit dosing recommendations for severe renal impairment appear; off-label if unapproved.
3. SmPC Section 4.3 (Contraindications) / USPI Section 4. Off-label / Contra-indicated. Must never be answered casually from clinical trial publications — route strictly through medical information off-label SOPs.
4. eCTD Module 2.5 (Clinical Overview). Contains the sponsor’s formal benefit-risk justification and critical analysis of clinical efficacy and safety.
5. Company Core Data Sheet (CCDS). The sponsor’s internal global standard document. A safety update in the CCDS triggers mandatory local variation filings across all regional markets.
6. CDSCO-approved Indian Package Insert (Prescribing Information) — not the USPI or SmPC. Dosing indications differ across national jurisdictions, and local approvals govern local medical practice.
The pattern: Six regulatory questions, five distinct source documents. The essential rule is to establish the exact jurisdiction and regulatory status before citing any source.
2. SmPC (Summary of Product Characteristics), Section 4.2 (Posology and Method of Administration) and Section 5.2 (Pharmacokinetics). On-label if explicit dosing recommendations for severe renal impairment appear; off-label if unapproved.
3. SmPC Section 4.3 (Contraindications) / USPI Section 4. Off-label / Contra-indicated. Must never be answered casually from clinical trial publications — route strictly through medical information off-label SOPs.
4. eCTD Module 2.5 (Clinical Overview). Contains the sponsor’s formal benefit-risk justification and critical analysis of clinical efficacy and safety.
5. Company Core Data Sheet (CCDS). The sponsor’s internal global standard document. A safety update in the CCDS triggers mandatory local variation filings across all regional markets.
6. CDSCO-approved Indian Package Insert (Prescribing Information) — not the USPI or SmPC. Dosing indications differ across national jurisdictions, and local approvals govern local medical practice.
The pattern: Six regulatory questions, five distinct source documents. The essential rule is to establish the exact jurisdiction and regulatory status before citing any source.
Reflection
For a product you work on, do you know whether the EU and US indications are identical? If not, compare the two labels — it takes twenty minutes and is frequently surprising.
Next week
4.1 — Traceability and the number sheet
Week 4 · Module 1
Traceability and the number sheet
Every number in every deliverable must trace back to one output. The discipline that makes your work checkable — by you, by quality control, and by the statistician
Your scenario
A number in your abstract does not match the same number in your Results table. Nobody can remember where either came from. You spend a day reconstructing the chain that a single page, built in ninety minutes at the start, would have preserved.
The chain
Figure 4.1 — The traceability chain. The number sheet is the link that makes the last step reversible. Without it, a figure in a deliverable is an assertion; with it, anyone can verify it in one step and a data refresh can be applied systematically rather than by memory.
The number sheet
| Column | Why it is there |
|---|---|
| The number as it will appear | Decimals decided once, so every mention matches |
| Table or figure reference | Lets you, quality control and the statistician verify in seconds |
| Population and n | Catches the randomised-versus-treated error before it reaches the page |
| Data cut-off | Catches mixed-cut outputs |
| Where it appears | Abstract, text, table, legend, slide — so a change propagates everywhere |
| Status | Confirmed, or query raised and pending |
When the data refresh arrives
A new data cut is not a disaster if you have a number sheet. Walk down the sheet, update each value from the new output, and use the where it appears column to find every location that must change.
Without it, updating a manuscript for a new data cut means re-reading the whole document and hoping. This is the single situation where the ninety minutes pays for itself several times over.
Without it, updating a manuscript for a new data cut means re-reading the whole document and hoping. This is the single situation where the ninety minutes pays for itself several times over.
Drills
Drill 1 of 3
You need a percentage that is not in the output pack, but you can compute it from two numbers that are. What do you do?
ACompute it and use it
BRequest the output, or ask the statistician to confirm the derivation in writing
CCompute it and flag it in review
DOmit it
Drill 2 of 3
A new data cut arrives midway through drafting. What is your first move?
ARe-read the draft and change what you notice
BWait until the draft is finished
CUpdate the number sheet, then propagate using the location column
DKeep the old numbers and note the cut in a footnote
Drill 3 of 3
What does the "status" column on a number sheet track?
AWhether the section has been reviewed
BWhether the author has approved it
CWhether it has been submitted
DWhether the number is confirmed or has a query outstanding
Production drill
Build the number sheet for these six figures, marking status and any query.
You are writing: a 250-word abstract and a Results section.
From the pack:
14.1.1 — 402 randomised, 389 treated, 371 with post-baseline assessment. Cut 15 Jan.
14.2.1 — median PFS 8.9 vs 5.2 months; HR 0.66 (95% CI 0.53–0.82); p < 0.001. ITT. Cut 15 Jan.
14.2.2 — KM figure. Cut 30 Sep.
14.2.5 — median OS not reached vs 21.7 months; HR 0.84 (0.62–1.14). ITT. Cut 15 Jan.
14.2.8 — ORR 44% vs 29%. Header states n = 402.
14.3.1 — grade ≥3 TEAE 147 and 121. Header states n = 402.
From the pack:
14.1.1 — 402 randomised, 389 treated, 371 with post-baseline assessment. Cut 15 Jan.
14.2.1 — median PFS 8.9 vs 5.2 months; HR 0.66 (95% CI 0.53–0.82); p < 0.001. ITT. Cut 15 Jan.
14.2.2 — KM figure. Cut 30 Sep.
14.2.5 — median OS not reached vs 21.7 months; HR 0.84 (0.62–1.14). ITT. Cut 15 Jan.
14.2.8 — ORR 44% vs 29%. Header states n = 402.
14.3.1 — grade ≥3 TEAE 147 and 121. Header states n = 402.
Number sheet
1. 402 randomised / 389 treated · 14.1.1 · disposition · 15 Jan · abstract Methods, Results, flow diagram · confirmed
2. Median PFS 8.9 vs 5.2 months · 14.2.1 · ITT n=402 · 15 Jan · abstract Results, Results text, Fig 2 legend · confirmed
3. HR 0.66 (95% CI 0.53–0.82), p < 0.001 · 14.2.1 · ITT · 15 Jan · abstract, Results, Fig 2 legend · confirmed
4. Median OS NR vs 21.7 months, HR 0.84 (0.62–1.14) · 14.2.5 · ITT · 15 Jan · abstract, Results · confirmed
5. ORR 44% vs 29% · 14.2.8 · population n queried · cut not stated · abstract, Results · QUERY 1 — header says 402, but response is normally assessed in the 371 with a post-baseline scan. Confirm population and cut-off.
6. Grade ≥3 TEAE 147 / 121 · 14.3.1 · denominator wrong · 15 Jan · abstract, Results, safety table · QUERY 2 — header states 402; safety must use the treated population, 196/193. All percentages need recomputing.
Not on the sheet, deliberately: figure 14.2.2. It is at the 30 September cut, so no number from it enters the draft. QUERY 3 — request a regenerated figure at the 15 January cut.
What the sheet has already bought you: three queries identified before a sentence was written, two of which would otherwise have produced published errors. The two provisional numbers are visibly marked, so you will not forget they are unconfirmed when the draft goes for review.
1. 402 randomised / 389 treated · 14.1.1 · disposition · 15 Jan · abstract Methods, Results, flow diagram · confirmed
2. Median PFS 8.9 vs 5.2 months · 14.2.1 · ITT n=402 · 15 Jan · abstract Results, Results text, Fig 2 legend · confirmed
3. HR 0.66 (95% CI 0.53–0.82), p < 0.001 · 14.2.1 · ITT · 15 Jan · abstract, Results, Fig 2 legend · confirmed
4. Median OS NR vs 21.7 months, HR 0.84 (0.62–1.14) · 14.2.5 · ITT · 15 Jan · abstract, Results · confirmed
5. ORR 44% vs 29% · 14.2.8 · population n queried · cut not stated · abstract, Results · QUERY 1 — header says 402, but response is normally assessed in the 371 with a post-baseline scan. Confirm population and cut-off.
6. Grade ≥3 TEAE 147 / 121 · 14.3.1 · denominator wrong · 15 Jan · abstract, Results, safety table · QUERY 2 — header states 402; safety must use the treated population, 196/193. All percentages need recomputing.
Not on the sheet, deliberately: figure 14.2.2. It is at the 30 September cut, so no number from it enters the draft. QUERY 3 — request a regenerated figure at the 15 January cut.
What the sheet has already bought you: three queries identified before a sentence was written, two of which would otherwise have produced published errors. The two provisional numbers are visibly marked, so you will not forget they are unconfirmed when the draft goes for review.
Reflection
On your next project, build the sheet before drafting. How many queries did it surface, and how many would you have caught later?
Next module
4.2 — Writing the query
Week 4 · Module 2
Writing the query
What to raise, who to ask, and how to write it so it gets answered in one round. The skill that most quickly changes how senior colleagues regard you
Your scenario
You email the statistician: “The safety numbers look wrong, can you check?” Four days later she replies asking which table. Two more days pass. The same query, written properly, would have been answered in an hour — and the difference is entirely in how you wrote it.
What to raise, and what not to
| Raise it | Handle it yourself |
|---|---|
| Two outputs disagree | A footnote explains the apparent discrepancy — read it first |
| A denominator does not match the disposition table | A different population is stated in the footer and is correct |
| An output is missing that the SAP prespecified | An output exists but you had not found it |
| A population definition is absent or ambiguous | The definition is in the SAP and you have not looked |
| Outputs come from different data cuts | All footers show the same cut |
| Percentages do not recompute and no footnote explains it | Rounding accounts for the difference |
Read the footers and footnotes first
Roughly half of what looks like a discrepancy is explained in a footnote you have not read. Querying it anyway costs a colleague time and costs you credibility.
The other half is real, and finding it is exactly what you are for. The discipline is: check the footer, check the footnote, check the SAP — then query.
The other half is real, and finding it is exactly what you are for. The discipline is: check the footer, check the footnote, check the SAP — then query.
Anatomy of a good query
| # | Element | Example |
|---|---|---|
| 1 | The specific output | Table 14.3.1, page 412 |
| 2 | What you observe | Header states n = 402 for both arms |
| 3 | Why it looks wrong | Disposition table 14.1.1 gives 196 and 193 treated; the SAP defines safety as the treated population |
| 4 | What you think it should be, tentatively | Should the denominators be 196 and 193, with percentages recomputed? |
| 5 | What it blocks | This affects every safety percentage in the abstract and Results |
| 6 | When you need it | By Thursday, to keep the draft on schedule |
Who to ask
Statistician — anything about outputs, populations, analyses, derivations.
Clinical lead or medical monitor — clinical interpretation, event attribution, whether a finding is expected.
Data management — whether a variable was collected, coding dictionary versions.
Safety or pharmacovigilance — expectedness, causality, narrative content.
Regulatory — label wording, market-specific indications.
Sending every query to the project manager to route is slower and marks you as unfamiliar with the team.
Clinical lead or medical monitor — clinical interpretation, event attribution, whether a finding is expected.
Data management — whether a variable was collected, coding dictionary versions.
Safety or pharmacovigilance — expectedness, causality, narrative content.
Regulatory — label wording, market-specific indications.
Sending every query to the project manager to route is slower and marks you as unfamiliar with the team.
Drills
Drill 1 of 3
Which query will be answered fastest?
A“Table 14.3.1 states n = 402 for both arms; 14.1.1 gives 196 and 193 treated, and the SAP defines safety as the treated population. Should the denominators be 196 and 193? This affects every safety percentage in the abstract — I need it by Thursday.”
B“The safety numbers look wrong, can you check?”
C“Table 14.3.1 is wrong.”
D“Please review the attached output pack.”
Drill 2 of 3
A percentage does not recompute from its counts. What do you do first?
AQuery it immediately
BRead the footnotes for an alternative denominator
CRecalculate and use your own figure
DIgnore it as rounding
Drill 3 of 3
You need to know whether an observed hepatic event is expected for the product. Who do you ask?
AThe statistician
BData management
CSafety or pharmacovigilance, referencing the current IB
DThe project manager, to route
Production drill
Write all four queries from the Week 2 pack, addressed to the right person, using the six-part anatomy.
Findings to raise:
1. Table 14.3.1 uses n = 402; disposition gives 196 and 193 treated.
2. Figure 14.2.2 is at the 30 Sep cut; Table 14.2.1 is at 15 Jan.
3. Table 14.2.8 (ORR) does not state its population.
4. Forest plot shows 11 subgroups; the SAP prespecified 6.
Your deadline: first draft due Friday.
1. Table 14.3.1 uses n = 402; disposition gives 196 and 193 treated.
2. Figure 14.2.2 is at the 30 Sep cut; Table 14.2.1 is at 15 Jan.
3. Table 14.2.8 (ORR) does not state its population.
4. Forest plot shows 11 subgroups; the SAP prespecified 6.
Your deadline: first draft due Friday.
To the statistician — four queries on the output pack, draft due Friday
1. Table 14.3.1 — safety denominator. The header states n = 402 for both arms, but Table 14.1.1 gives 196 and 193 patients treated, and the SAP defines the safety population as all patients receiving at least one dose. Should the denominators be 196 and 193, with percentages recomputed? This affects every safety percentage in the abstract, the Results text and the safety table. Needed by Wednesday.
2. Figure 14.2.2 — data cut-off. The figure footer shows a cut-off of 30 September, while Table 14.2.1 shows 15 January, and the medians differ accordingly (8.1 versus 8.9 months). Could the figure be regenerated at the 15 January cut? I am not quoting anything from the figure until then. Needed by Thursday.
3. Table 14.2.8 — ORR population. The table does not state its analysis population or n. Response is usually assessed in patients with at least one post-baseline assessment, which Table 14.1.1 gives as 371. Can you confirm the population and the denominator, and the data cut-off? This changes the percentages in the abstract. Needed by Wednesday.
4. Forest plot 14.2.11 — subgroup count. The plot shows 11 subgroups; the SAP prespecifies six (age, sex, region, ECOG, prior lines, biomarker status). Could you confirm which six are prespecified, so the remainder can be labelled exploratory or excluded? Needed by Thursday.
Separately, to safety: none arising from this pack.
Four things this letter does:
1. One recipient, one message, numbered items. Four separate emails would be four separate delays.
2. Each query shows the reasoning, so the recipient can confirm or correct without opening three documents.
3. Each states the impact, so priority is obvious without you asserting it.
4. Staggered deadlines reflecting which blocks what. Query 1 blocks the most, so it is due first.
Note what is absent: no apology, no hedging about whether it might be your mistake, and no assertion that the table is wrong. You describe what you observe and propose a resolution. That tone is what gets queries answered quickly and gets you invited to more projects.
1. Table 14.3.1 — safety denominator. The header states n = 402 for both arms, but Table 14.1.1 gives 196 and 193 patients treated, and the SAP defines the safety population as all patients receiving at least one dose. Should the denominators be 196 and 193, with percentages recomputed? This affects every safety percentage in the abstract, the Results text and the safety table. Needed by Wednesday.
2. Figure 14.2.2 — data cut-off. The figure footer shows a cut-off of 30 September, while Table 14.2.1 shows 15 January, and the medians differ accordingly (8.1 versus 8.9 months). Could the figure be regenerated at the 15 January cut? I am not quoting anything from the figure until then. Needed by Thursday.
3. Table 14.2.8 — ORR population. The table does not state its analysis population or n. Response is usually assessed in patients with at least one post-baseline assessment, which Table 14.1.1 gives as 371. Can you confirm the population and the denominator, and the data cut-off? This changes the percentages in the abstract. Needed by Wednesday.
4. Forest plot 14.2.11 — subgroup count. The plot shows 11 subgroups; the SAP prespecifies six (age, sex, region, ECOG, prior lines, biomarker status). Could you confirm which six are prespecified, so the remainder can be labelled exploratory or excluded? Needed by Thursday.
Separately, to safety: none arising from this pack.
Four things this letter does:
1. One recipient, one message, numbered items. Four separate emails would be four separate delays.
2. Each query shows the reasoning, so the recipient can confirm or correct without opening three documents.
3. Each states the impact, so priority is obvious without you asserting it.
4. Staggered deadlines reflecting which blocks what. Query 1 blocks the most, so it is due first.
Note what is absent: no apology, no hedging about whether it might be your mistake, and no assertion that the table is wrong. You describe what you observe and propose a resolution. That tone is what gets queries answered quickly and gets you invited to more projects.
Reflection
Find the last query you sent. Did it contain all six elements? How many rounds did it take to resolve?
Queries missing element 3 — the reasoning — are the ones that take three rounds.
Next module
4.3 — From source to deliverable
Week 4 · Module 3
From source to deliverable
The same output pack feeds a manuscript, a congress abstract, a slide deck and a medical information response — and each one may use it differently
Your scenario
You reuse a Results paragraph from the manuscript in an MSL slide. Medical review rejects it. The paragraph was accurate — but a promotional-adjacent deck cannot carry a claim about an endpoint that sits below a hierarchy break, and cannot present off-label data at all. The same true sentence has different standing in different documents.
What changes, deliverable by deliverable
| Deliverable | Uses the source for | Constraint that does not apply elsewhere |
|---|---|---|
| Manuscript | Complete, balanced reporting of all prespecified endpoints | Must report negative and non-significant results; CONSORT applies |
| Congress abstract | The primary endpoint and headline safety, under a hard word limit | Embargo rules; encore restrictions; data must be final at submission |
| Slide deck for field use | Data supporting approved use, referenced to source | On-label only; medical-legal-regulatory review; every claim referenced |
| Medical information response | The specific question asked, and nothing beyond it | Reactive only; off-label permitted only in response to an unsolicited request |
| Lay summary | The primary result and what it means for patients | Plain language; no promotional tone; often a regulatory requirement |
The rule that catches people
A statement being true and published does not make it usable in every document.
A hierarchy-break endpoint is reportable in a manuscript, with a nominal p value, because a manuscript reports everything prespecified. In a field-facing deck it may not be presentable as a benefit at all. An off-label dose can be discussed in a reactive medical information response and cannot appear in proactive material.
The source does not change. What you may build from it does.
A hierarchy-break endpoint is reportable in a manuscript, with a nominal p value, because a manuscript reports everything prespecified. In a field-facing deck it may not be presentable as a benefit at all. An off-label dose can be discussed in a reactive medical information response and cannot appear in proactive material.
The source does not change. What you may build from it does.
Referencing
| Deliverable | Reference back to |
|---|---|
| Manuscript | Published literature; the study's own data need no citation |
| Congress abstract | Usually no references permitted |
| Field slide deck | Every data point referenced to a publication or the label — not to internal outputs |
| Medical information response | Publications, label, and data on file where nothing published exists |
Data on file
When a figure exists only in the output pack and nowhere published, a field-facing document may cite it as data on file — but only under a controlled process, with the underlying output retained and retrievable.
What you cannot do is quote an internal output as though it were published, or use one to support a claim the published data do not carry.
What you cannot do is quote an internal output as though it were published, or use one to support a claim the published data do not carry.
Drills
Drill 1 of 3
An endpoint below the hierarchy break was nominally significant. Where may it appear?
ANowhere — it was not formally tested
BAnywhere, once published
COnly in field material
DIn the manuscript as nominal, but not as a benefit claim in field material
Drill 2 of 3
A field deck needs a figure that exists only in the output pack. What is required?
AReference it as data on file, with the output retained and retrievable
BCite the internal table number
CUse it without a reference
DCite the manuscript, even though it does not contain that figure
Drill 3 of 3
A clinician sends an unsolicited question about an off-label dose used in a published trial. What is permitted?
ANothing may be provided
BA reactive, balanced response to the specific question, under the off-label process
CA full response plus additional off-label data of likely interest
DThe standard promotional deck
Production drill
For each medical deliverable in an Immunology / Severe Plaque Psoriasis biologic trial, state what you would take from this pack and what you would leave out — and why.
The pack contains: Primary PASI-90 response rate at Week 16 met (68.4% vs 14.2%, OR 12.8, p < 0.001); Key secondary static Physician Global Assessment (sPGA 0/1) met (p < 0.001); Dermatology Life Quality Index (DLQI 0/1) score met statistically (p = 0.03) but below the pre-specified Minimal Clinically Important Difference (MCID); Exploratory HLA-Cw6 positive biomarker subgroup showed enhanced response (interaction p = 0.02); Safety showed mild-to-moderate nasopharyngitis and injection-site erythema.
Regulatory product status: Approved for moderate-to-severe plaque psoriasis. Pharmacogenomic HLA-Cw6 biomarker-directed selection is NOT approved in the official label.
Deliverables: (1) Peer-reviewed clinical manuscript, (2) 250-word congress abstract (AAD), (3) Field Medical MSL slide deck, (4) Medical Information standard response letter to an unsolicited HCP enquiry about HLA-Cw6 biomarker efficacy.
Regulatory product status: Approved for moderate-to-severe plaque psoriasis. Pharmacogenomic HLA-Cw6 biomarker-directed selection is NOT approved in the official label.
Deliverables: (1) Peer-reviewed clinical manuscript, (2) 250-word congress abstract (AAD), (3) Field Medical MSL slide deck, (4) Medical Information standard response letter to an unsolicited HCP enquiry about HLA-Cw6 biomarker efficacy.
1. Manuscript — include everything. All prespecified primary (PASI-90) and secondary endpoints (sPGA, DLQI), the DLQI MCID contextualisation, the exploratory HLA-Cw6 subgroup interaction analysis, and complete safety profiles. Omitting any prespecified protocol endpoint constitutes selective outcome reporting under ICMJE / CONSORT.
2. Congress abstract (AAD) — primary endpoint, key secondary, headline safety. Word count constraints demand strict prioritisation: report PASI-90 and sPGA response rates with 95% CIs and p values, plus overall adverse event rates. Leave out the DLQI score and HLA-Cw6 subgroup — both require extensive qualifying methodology that the 250-word limit cannot support.
3. MSL field deck — on-label PASI-90, sPGA, and safety only. Every slide must reference approved label data. Leave out the HLA-Cw6 biomarker subgroup entirely — promoting biomarker-directed enrichment when unapproved violates FDA / IFPMA commercial compliance rules. Present DLQI with caution: describe numerical change without claiming a validated patient-reported quality-of-life benefit since it failed MCID thresholds.
4. Medical information response — the HLA-Cw6 subgroup only. The enquiry was unsolicited, specific, and from a healthcare professional; therefore, a scientifically balanced reactive response is legally permissible. Detail the subgroup trial methodology, state that it was an exploratory analysis in an unpowered subgroup, explicitly state that biomarker-directed therapy is not indicated in the approved label, and stop. Do not volunteer unsolicited efficacy claims or general marketing summaries.
The pattern: One source pack, four distinct deliverables. The HLA-Cw6 biomarker finding is fully reported in the manuscript, omitted from the abstract, prohibited in proactive field decks, and permitted strictly reactively in medical information. The underlying data never changed — only the regulatory and audience context.
2. Congress abstract (AAD) — primary endpoint, key secondary, headline safety. Word count constraints demand strict prioritisation: report PASI-90 and sPGA response rates with 95% CIs and p values, plus overall adverse event rates. Leave out the DLQI score and HLA-Cw6 subgroup — both require extensive qualifying methodology that the 250-word limit cannot support.
3. MSL field deck — on-label PASI-90, sPGA, and safety only. Every slide must reference approved label data. Leave out the HLA-Cw6 biomarker subgroup entirely — promoting biomarker-directed enrichment when unapproved violates FDA / IFPMA commercial compliance rules. Present DLQI with caution: describe numerical change without claiming a validated patient-reported quality-of-life benefit since it failed MCID thresholds.
4. Medical information response — the HLA-Cw6 subgroup only. The enquiry was unsolicited, specific, and from a healthcare professional; therefore, a scientifically balanced reactive response is legally permissible. Detail the subgroup trial methodology, state that it was an exploratory analysis in an unpowered subgroup, explicitly state that biomarker-directed therapy is not indicated in the approved label, and stop. Do not volunteer unsolicited efficacy claims or general marketing summaries.
The pattern: One source pack, four distinct deliverables. The HLA-Cw6 biomarker finding is fully reported in the manuscript, omitted from the abstract, prohibited in proactive field decks, and permitted strictly reactively in medical information. The underlying data never changed — only the regulatory and audience context.
Reflection
Think of a claim you have written into two different document types. Did the constraints differ? Did you notice at the time?
Next module
4.4 — Capstone and exit assessment
Week 4 · Module 4
Capstone and exit assessment
A full document set with eighteen planted problems. The task is not to write anything — it is to find what is wrong before anyone writes a word
Your scenario
This is the real first day. A document set, a deadline, and nobody to explain it. Pass mark: 14 of 18.
How to do this
Timer at 50 minutes. Work through the set and produce two outputs: a number sheet for the abstract, and a query letter. Do not read the rubric first.
The document set
Find every problem. Produce the number sheet and the query letter.
PROTOCOL v2.0, dated 04 Mar 2024. Randomised, double-blind, placebo-controlled Phase III. Primary endpoint: progression-free survival. Stratified by region and prior lines. Amendment 3 dated 18 Nov 2024 changed the primary endpoint from overall survival to progression-free survival.
SAP v2.0, dated 02 Jan 2025. Database lock 20 Dec 2024. Populations: ITT = all randomised; safety = all treated; response-evaluable = ITT with ≥1 post-baseline scan. Hierarchy: PFS → OS → ORR. Prespecified subgroups: 6.
IB edition 7, dated Aug 2023. RSI lists neutropenia, fatigue, nausea, rash.
14.1.1 Disposition. 402 randomised (201/201); 389 treated (196/193); 371 with post-baseline scan. Cut 15 Jan 2025.
14.2.1 PFS (ITT n = 402). Median 8.9 vs 5.2 months; HR 0.66 (95% CI 0.53–0.82); p < 0.001. Cut 15 Jan 2025.
14.2.2 PFS KM figure. Median appears ~8.1 months. Cut 30 Sep 2024.
14.2.5 OS (ITT). NR vs 21.7 months; HR 0.84 (0.62–1.14); p = 0.26. Cut 15 Jan 2025.
14.2.8 ORR. Header "n = 402". 176 (44%) vs 117 (29%); p = 0.003.
14.2.11 Subgroups. 11 subgroups shown. Biomarker-positive 252 + negative 119 = 371.
14.3.1 TEAE. Header "n = 402". Grade ≥3: 147 and 121. Cut 15 Jan 2025.
16.2.7 Listing. Subject 1042-0087: grade 3 hepatic transaminase elevation, hospitalised, investigator causality "probably related".
Draft abstract sentence supplied by an author: "The regimen significantly improved response rate (44% vs 29%, p = 0.003) and quality of life, and was particularly effective in biomarker-positive patients."
SAP v2.0, dated 02 Jan 2025. Database lock 20 Dec 2024. Populations: ITT = all randomised; safety = all treated; response-evaluable = ITT with ≥1 post-baseline scan. Hierarchy: PFS → OS → ORR. Prespecified subgroups: 6.
IB edition 7, dated Aug 2023. RSI lists neutropenia, fatigue, nausea, rash.
14.1.1 Disposition. 402 randomised (201/201); 389 treated (196/193); 371 with post-baseline scan. Cut 15 Jan 2025.
14.2.1 PFS (ITT n = 402). Median 8.9 vs 5.2 months; HR 0.66 (95% CI 0.53–0.82); p < 0.001. Cut 15 Jan 2025.
14.2.2 PFS KM figure. Median appears ~8.1 months. Cut 30 Sep 2024.
14.2.5 OS (ITT). NR vs 21.7 months; HR 0.84 (0.62–1.14); p = 0.26. Cut 15 Jan 2025.
14.2.8 ORR. Header "n = 402". 176 (44%) vs 117 (29%); p = 0.003.
14.2.11 Subgroups. 11 subgroups shown. Biomarker-positive 252 + negative 119 = 371.
14.3.1 TEAE. Header "n = 402". Grade ≥3: 147 and 121. Cut 15 Jan 2025.
16.2.7 Listing. Subject 1042-0087: grade 3 hepatic transaminase elevation, hospitalised, investigator causality "probably related".
Draft abstract sentence supplied by an author: "The regimen significantly improved response rate (44% vs 29%, p = 0.003) and quality of life, and was particularly effective in biomarker-positive patients."
The eighteen problems
Governing documents (5)
1. SAP v2.0 dated 02 Jan 2025; database lock was 20 Dec 2024. The SAP was finalised after lock. This is the most serious finding in the set and must be raised and disclosed.
2. Protocol amendment 3 changed the primary endpoint from OS to PFS. Establish whether it preceded unblinding; disclose the amendment and its rationale regardless.
3. IB is edition 7 from August 2023 — over a year old. Request the current edition before any expectedness assessment.
4. Hepatic events are not in the RSI (neutropenia, fatigue, nausea, rash). Subject 1042-0087’s event is therefore unexpected, which has expedited reporting implications. Raise with safety immediately — and note this depends on point 3, since a newer IB may have added it.
5. Protocol is v2.0 but amendment 3 exists. Version numbering is inconsistent; obtain the correct current version.
Output pack (7)
6. 14.3.1 uses n = 402. Safety must use the treated population, 196 and 193. Every safety percentage is wrong.
7. 14.2.2 is at the 30 Sep 2024 cut while all other outputs are at 15 Jan 2025 — explaining 8.1 versus 8.9 months. Request regeneration; quote nothing from it.
8. 14.2.8 header says n = 402 but ORR should use the response-evaluable population, 371. 176/371 = 47.4%, not 44%. Confirm.
9. 14.2.11 shows 11 subgroups; the SAP prespecifies 6. Identify which six.
10. Subgroup n values sum to 371, not 402 — 31 patients had no evaluable biomarker result. Report the biomarker-evaluable denominator.
11. No data cut-off stated on 14.2.8. Request it.
12. No PRO output exists in the pack at all, yet the draft abstract claims a quality-of-life benefit. There is nothing to support it.
The draft abstract sentence (4)
13. “Significantly improved response rate” — ORR sits below the hierarchy break at OS (p = 0.26), so it was not formally tested. Nominal only.
14. “and quality of life” — no supporting output exists. Unsupported claim.
15. “particularly effective in biomarker-positive patients” — no interaction p value is available, the subgroup is one of eleven presented against six prespecified, and 31 patients are unaccounted for.
16. The sentence omits the primary endpoint entirely. An abstract sentence leading on a hierarchy-break endpoint while ignoring PFS inverts the study’s own hierarchy.
Process (2)
17. No number in the draft sentence traces to a checked output. Build the number sheet before any further drafting.
18. The queries split across three recipients — statistician (6–11), safety (4), clinical or regulatory lead (1, 2, 3, 5). Sending them all to one person delays everything.
Scoring
14–18 — pass. You can be handed a document set unsupervised. You caught the SAP timing and the unexpected hepatic event, which are the two with consequences beyond the paper.
9–13 — close. Note which week your misses cluster in and redo that module.
Under 9 — work through Weeks 2 and 3 again before retrying.
The two that matter most: problem 1, because a SAP finalised after database lock undermines the credibility of every prespecification claim in the paper; and problem 4, because an unexpected serious adverse reaction carries expedited reporting obligations and a missed one is a regulatory failure, not an editorial one.
If you found only the output-pack problems (6–11): that is the commonest result, and it is the difference between a writer who checks tables and one who reads a document set. The governing documents were where the serious findings were.
Governing documents (5)
1. SAP v2.0 dated 02 Jan 2025; database lock was 20 Dec 2024. The SAP was finalised after lock. This is the most serious finding in the set and must be raised and disclosed.
2. Protocol amendment 3 changed the primary endpoint from OS to PFS. Establish whether it preceded unblinding; disclose the amendment and its rationale regardless.
3. IB is edition 7 from August 2023 — over a year old. Request the current edition before any expectedness assessment.
4. Hepatic events are not in the RSI (neutropenia, fatigue, nausea, rash). Subject 1042-0087’s event is therefore unexpected, which has expedited reporting implications. Raise with safety immediately — and note this depends on point 3, since a newer IB may have added it.
5. Protocol is v2.0 but amendment 3 exists. Version numbering is inconsistent; obtain the correct current version.
Output pack (7)
6. 14.3.1 uses n = 402. Safety must use the treated population, 196 and 193. Every safety percentage is wrong.
7. 14.2.2 is at the 30 Sep 2024 cut while all other outputs are at 15 Jan 2025 — explaining 8.1 versus 8.9 months. Request regeneration; quote nothing from it.
8. 14.2.8 header says n = 402 but ORR should use the response-evaluable population, 371. 176/371 = 47.4%, not 44%. Confirm.
9. 14.2.11 shows 11 subgroups; the SAP prespecifies 6. Identify which six.
10. Subgroup n values sum to 371, not 402 — 31 patients had no evaluable biomarker result. Report the biomarker-evaluable denominator.
11. No data cut-off stated on 14.2.8. Request it.
12. No PRO output exists in the pack at all, yet the draft abstract claims a quality-of-life benefit. There is nothing to support it.
The draft abstract sentence (4)
13. “Significantly improved response rate” — ORR sits below the hierarchy break at OS (p = 0.26), so it was not formally tested. Nominal only.
14. “and quality of life” — no supporting output exists. Unsupported claim.
15. “particularly effective in biomarker-positive patients” — no interaction p value is available, the subgroup is one of eleven presented against six prespecified, and 31 patients are unaccounted for.
16. The sentence omits the primary endpoint entirely. An abstract sentence leading on a hierarchy-break endpoint while ignoring PFS inverts the study’s own hierarchy.
Process (2)
17. No number in the draft sentence traces to a checked output. Build the number sheet before any further drafting.
18. The queries split across three recipients — statistician (6–11), safety (4), clinical or regulatory lead (1, 2, 3, 5). Sending them all to one person delays everything.
Scoring
14–18 — pass. You can be handed a document set unsupervised. You caught the SAP timing and the unexpected hepatic event, which are the two with consequences beyond the paper.
9–13 — close. Note which week your misses cluster in and redo that module.
Under 9 — work through Weeks 2 and 3 again before retrying.
The two that matter most: problem 1, because a SAP finalised after database lock undermines the credibility of every prespecification claim in the paper; and problem 4, because an unexpected serious adverse reaction carries expedited reporting obligations and a missed one is a regulatory failure, not an editorial one.
If you found only the output-pack problems (6–11): that is the commonest result, and it is the difference between a writer who checks tables and one who reads a document set. The governing documents were where the serious findings were.
Retake the diagnostic
Return to Module 1.1 and take the 24-item diagnostic again. Your first score is the baseline; the difference is the only honest measure of what changed.
What to do next
Four habits, starting on your next project. Read the footer of every output before quoting it. Build the number sheet before drafting. Check the SAP date against the database lock date. Open the CRF before writing a limitations paragraph.
Then go on to the document courses. This course taught you to read the sources; Writing the Manuscript teaches what each section may say, Statistics teaches what the numbers mean, and Scientific English teaches the sentences. This one comes first because the others assume it.
Then go on to the document courses. This course taught you to read the sources; Writing the Manuscript teaches what each section may say, Statistics teaches what the numbers mean, and Scientific English teaches the sentences. This one comes first because the others assume it.
Final reflection
Your diagnostic score at the start, and now. Which document had you never opened before this course?
You have completed the course
Reading Source Documents
IMRAD Academy — Read like a scientist. Write like a professional.
Reading Source Documents · v1.0