Medical Writers' Handbook.

Chapter 01· 8 min read

Medical Writing Is Evidence Work

The easiest way to misunderstand medical writing is to treat it as a language service. Language is central, but it is not the whole job. A document can be grammatical and elegant while still being misleading, incomplete, structurally wrong, unsuitable for its audience, or inconsistent with the underlying data. Successful medical communication joins a defined outcome and audience to sound science, applicable guidance, suitable language, and disciplined tools.^1,2 Remove any one of those elements and the document weakens.

You are transforming evidence, not decorating it

Every serious medical document starts with something that exists outside the prose: a research question, protocol, dataset, statistical analysis, study report, body of literature, clinical experience, authority requirement, patient need, or program decision. The writer's task is to understand that material, decide what it can support, and transform it into a document whose form serves its purpose.^1,3

That transformation involves selection. A study report cannot reproduce every row of every listing in its narrative. A manuscript cannot contain the entire clinical study report. A plain-language summary cannot retain every technical detail. Selection is unavoidable, which is why it must be principled. The writer has to decide what is material, what is background, what belongs in a visual, what needs a limitation, and what cannot safely be omitted.⁴

Selection also involves sequence. Readers need a path through information: why the question matters, what was done, what was found, what the findings mean, and where uncertainty remains. The sequence differs by document, but the need for a deliberate path does not.

Fidelity has two directions

Fidelity has two directions. Internal fidelity means precision and consistency inside the document: if the Methods define an objective, endpoint, or analysis, the Results must account for it, and the Results must not introduce analyses that the Methods never established.⁴ The abstract agrees with the body, endpoint names do not drift, numbers match across text and tables, and the conclusion reflects the results. External fidelity means accuracy against the evidence: data agree with validated outputs, claims agree with cited literature, methods agree with the protocol, and descriptions agree with the governing record.^4,5

Writers often notice external errors because they look like fact-checking problems. Internal errors can be subtler. A value may be copied correctly from a source but paired with the wrong denominator elsewhere. A conclusion may be individually plausible but stronger than the Results section permits. A method may be described accurately in one section and shortened into an incompatible form in the abstract. Fidelity therefore requires both source comparison and whole-document comparison.

This is why traceability is not clerical work left for the end. If you do not know where a claim came from while drafting, you create an expensive verification problem later. If you cannot explain why a sentence exists, the sentence is not under control.

Fair balance is the working ethic

Fair balance is a defining quality of medical communication. It is not achieved by inserting one negative sentence into otherwise selective advocacy. It requires a representative account of the relevant evidence, meaningful limitations, countervailing findings, and conclusions whose strength matches the study design, results, and evidence hierarchy.^4,6

The practical temptations are familiar. A preferred message arrives in the brief. One study produces an attractive result while another does not. Efficacy receives narrative space while safety is compressed. An exploratory subgroup is written as if it had been a prespecified primary question. A statistically significant result is presented without its magnitude or precision. A review cites the papers that support its direction and quietly leaves out the awkward ones.⁴

Fair balance resists those temptations. It asks whether the introduction represents the clinical or scientific context rather than one product alone. It asks whether methods and results stay aligned, whether safety and tolerability receive appropriate attention, whether the Discussion tests alternative explanations, and whether limitations are specific enough to change how a reader interprets the result.

Fair balance also depends on the audience. Physicians, patients, payers, policymakers, regulators, journal reviewers, investigators, and internal teams may need different detail and may judge relevance differently. Tailoring is not distortion. The same evidence can be framed differently for different readers, provided the meaning, uncertainty, and material context remain intact.

A clear sentence can still be wrong

Brevity and clarity matter, but neither is absolute. A shorter sentence is useful only if it preserves the information necessary for correct interpretation. Removing a qualifier may improve rhythm and destroy accuracy. Replacing a technical term with a familiar word may help a lay reader and introduce a false equivalence. Compressing a risk into one percentage may hide the denominator, time horizon, or comparator that gives the number meaning.^4,7

Clarity begins with clear thinking. Lead with the news in short, direct, declarative sentences instead of mistaking complexity for authority.⁴ The Methods section is central to manuscript logic because it establishes what was done and what the Results can legitimately report. Topic sentences, parallel structure, descriptive table titles, and explicit transitions help readers follow an argument, but these devices work only when the argument itself is coherent.^8,9

Clear medical prose avoids unnecessary jargon, careless word choice, ambiguous abbreviations, misplaced modifiers, and avoidable complexity.^8,10 These are not cosmetic defects. A dangling modifier can attach an action to the wrong subject; a pronoun can have two plausible antecedents; and one abbreviation can carry several, even contradictory, clinical meanings.¹¹ Editing for clarity is therefore part of accuracy.

The writer holds the whole picture

Medical writers work among specialists. Statisticians understand analysis. Clinicians understand disease and practice. Regulatory colleagues understand authority expectations. Authors and investigators understand the study. Designers understand visual communication. Patients understand the lived experience. The writer's distinctive responsibility is often integrative: to understand enough of each perspective to make the document coherent, expose missing decisions, and prevent one specialist's language from becoming misleading to another reader.³

That responsibility requires curiosity and humility. A writer should know when a term is being used inconsistently, when an endpoint does not map cleanly from protocol to results, when an author contribution is too thin for authorship, and when a statistical statement seems stronger than the design. Think like a peer reviewer during outlining and ask the difficult questions while major changes are still cheap.⁸ The writer should also know when to stop and ask the statistician, clinician, regulatory expert, patient representative, translator, or quality specialist.¹²

The senior move is not to conceal uncertainty with polished prose. It is to make the uncertainty legible and route it to the person who can resolve it.

Integrity is visible in the process

Scientific and medical integrity includes the obvious prohibitions against fabrication, falsification, and plagiarism, but it also reaches incomplete information, selective and duplicate publication, inappropriate authorship, undisclosed conflicts, unacknowledged writing support, mishandled patient information, and conclusions that outrun the data.^1,4

Authorship deserves special attention because writing teams can blur contribution. The ICMJE's four criteria require substantial contribution to the work, drafting or critical revision, final approval, and — since the 2013 revision — agreement to be accountable for the work as a whole; approval alone is not a sufficient intellectual contribution.⁸ Contributors who do not qualify should be acknowledged appropriately. Professional medical-writing support should be transparent, including the writer's identity, employer, funding, and intellectual contribution where applicable.¹³ A comparative journal study found that acknowledgment of medical writers rose from 5.1% of sampled articles in 2000 to 11.3% in 2007 as disclosure practice developed.¹⁴

Privacy is equally structural. Clinical documents and patient narratives can contain information that is scientifically interesting but unnecessary for the communication purpose. Before submission, remove personal identifiers — names, addresses, initials, hospital numbers, photographs — and obtain written informed consent when specific patient details must remain.¹⁰ Include what the document needs, protect sensitive information, and remember that a data point may describe a person.¹⁵

The document is an argument under constraints

An argument does not mean advocacy regardless of evidence. It means a controlled line of reasoning. A research manuscript argues that a question mattered, that a method addressed it, that the results answer it to a stated degree, and that the interpretation follows. A regulatory overview argues that a body of evidence supports a development or approval position while accounting for limitations. A patient document argues through explanation: this is what happened, this is what it means, and this is what you may need to do.

Every argument operates under constraints. The target may prescribe sections, length, terminology, disclosures, or submission format. The study design limits causal language. The source package limits what can be claimed. The review chain limits who can approve what. The audience limits assumed knowledge. The timeline limits iteration. A professional document makes these constraints explicit enough to manage.

Chapter 02· 8 min read

Define the Job Before You Touch the Draft

Many weak documents are faithful responses to weak assignments. The request says, “Write a review,” “Turn this study into a paper,” “Prepare the protocol,” or “Make this patient friendly.” The writer begins immediately, only to discover that the target is unknown, the audience is contested, the data are incomplete, authors disagree about the message, or the template arrived after the draft was built.

Start by determining the nature of the document, desired outcome, applicable guidance, target audience, key messages, sources of information, and template.¹ These are not preliminary niceties. Together, they define the writing problem.

Name the document precisely

“Manuscript” is not precise enough. Is it an original research report, narrative review, systematic review, meta-analysis, case report, commentary, or protocol paper? “Summary” is not precise enough. Is it a clinical overview, lay summary, executive summary, study synopsis, advisory-board report, or summary of a body of evidence? “Presentation” is not precise enough. Is it a congress talk, internal decision deck, field-training deck, patient education presentation, or continuing-education activity?

Document identity matters because it changes the expected evidence, structure, governing material, and review path.¹ A narrative review can organize literature around a reasoned thematic argument; a systematic review treats the search and selection process as part of the research method. A protocol describes intended conduct; a study report explains actual conduct and results. A plain-language summary must preserve scientific meaning while changing the assumed vocabulary and explanatory burden.

If the output has no settled label, describe it operationally: who will receive it, what they will do with it, what information it must contain, what source package it draws from, and what format it must become. That description can route the work even before the final name is chosen.

Define the desired outcome

The desired outcome is not simply “a completed document.” It is the change in understanding, decision, or action the document should enable.¹ A journal paper should make a contribution clear enough for editors and readers to judge. A protocol should make intended conduct reproducible and reviewable. A study report should provide an accurate account of methods, deviations, results, and interpretation. A patient communication should help a non-specialist understand information relevant to participation, treatment, safety, or results.

Outcome definition sharpens inclusion. If a section does not help the reader reach the intended understanding or satisfy a governing requirement, it may not belong. Outcome definition also exposes conflict. A sponsor may want a persuasive message; a journal requires balanced reporting; an author wants methodological detail; a reader wants the clinical implication. The brief should record these pressures rather than allowing them to surface as late-stage edits.

Put yourself in the reader's place

Imagine the target reader's response before choosing the level of detail or the argument's order.¹ This is more demanding than naming an audience category. “Clinicians” may include generalists and specialists with different knowledge. “Patients” is not a single literacy level, language, culture, experience, or information need. “Regulators” read within a formal review task. “Authors” may focus on scientific reputation, while journal editors focus on fit, novelty, methods, and reader value.

Ask what the reader already knows, what they need explained, what decision they face, what might confuse them, what might make them distrust the document, and what evidence they will expect to inspect. Audience affects terminology, detail, sentence structure, visual design, ordering, and the amount of context required. It should not change the underlying facts.

For public-facing material, involve intended readers where possible. User testing is performance based: it shows how a document behaves in the hands of its audience and reveals weaknesses that expert review or a readability formula may miss.⁷ Consent-document teams can combine plain language, validated glossaries, and literacy review.¹⁶ The reader is not an abstract persona when the work can include real reader input.

Treat key messages as hypotheses

Key messages help organize a document. They can also become a source of bias if accepted before the evidence is examined. The safest form is provisional: “The team expects the evidence to show X; confirm the scope, strength, and qualifications.” This wording lets the message guide investigation without ordering the writer to prove it.

Separate three categories in the brief:

• Required facts: Items that must be present because of the document's purpose or governing material.
• Proposed messages: Interpretations or emphases the team wants the evidence to support.
• Open questions: Issues the evidence, authors, statisticians, reviewers, or authorities still need to resolve.

Then connect each proposed message to its anticipated evidence. If the evidence is missing, weak, contradictory, or exploratory, record that state. A message without a plausible source path is not ready to become a drafting instruction.

Map the people around the document

Medical writing is collaborative, but collaboration without roles becomes comment accumulation. Identify the people who supply evidence, interpret science, decide scope, qualify for authorship, review language, check statistics, approve compliance, perform QC, and authorize release. The same person may hold several roles; the important point is that the roles are visible.

For publications, discuss authorship early and revisit it as contributions develop.⁸ For regulatory documents, distinguish document contributors from formal signatories or accountable functions.¹⁷ For patient material, include the people who can evaluate comprehension and relevance. For statistics-heavy work, identify the statistician who can answer analysis questions rather than relying on secondhand interpretation.

Name the decision owner for predictable conflicts: target journal, key message, analysis wording, author order, inclusion of a limitation, acceptance of a review comment, or release with an open item. If nobody owns the decision, the writer becomes the accidental adjudicator.

Define scope and boundaries

Scope is the agreement about what the project includes and excludes. It should cover the output, source package, therapeutic or scientific boundaries, time range for literature, number and type of visuals, reference volume, review rounds, expected rewrite depth, final format, and any associated files such as response letters or claim matrices.

Scope also protects quality. Fact-checking and source verification are not equivalent to copyediting. A “light edit” cannot safely include reconstruction of an incoherent analysis narrative unless the timeline and authority permit it. Plain-language adaptation is not merely shortening sentences. A systematic review cannot be produced from an undocumented convenience set of papers and still be represented as systematic.

Record what the writer is not being asked to decide. A writer may explain an analysis but not select it. A writer may flag a regulatory interpretation but not act as legal counsel. A writer may identify an authorship problem but not unilaterally resolve the author list. Boundaries make escalation professional rather than evasive.

Build the project timeline around decisions

A timeline made only of draft dates is too weak. Good milestones correspond to changes in project state: brief accepted, sources received, evidence gaps reviewed, target confirmed, Structure accepted, first section completed, data checks complete, author review reconciled, QC scope confirmed, final export validated.

A writing project moves through initiation, planning, execution, monitoring and control, and closure; risk identification and close-out documentation are core work, not ceremony.¹⁷ Clear workflows, roles, responsibilities, review stages, and collaboration rules reduce avoidable delay and ambiguity.¹⁸ The vocabulary matters less than the discipline: the project needs a plan, a way to detect drift, and a clear closing state.

Plan review time according to document risk and length. A long study report cannot receive a reliable data and consistency check in an implausibly short window. A multi-author paper needs time to reconcile comments rather than merely collect them. A patient document needs time for user input if that input is part of the method.

Capture the working brief

A useful one-page project frame contains:

The brief should remain editable while the project is being framed. Once accepted, changes should be visible. A shifting target is manageable when it is named; it is destructive when it silently rewrites the work.

Chapter 03· 8 min read

Build the Project Dossier

Medical writers often receive a pile: protocol, amendments, statistical analysis plan, tables, figures, listings, study report, slide deck, selected papers, author notes, a journal link, a prior manuscript, and an email containing the actual instruction. The pile may be rich in information and poor in control.^5,18

The dossier is the organized body of material from which the document will be built and defended. Its purpose is not aesthetic file management. It is to let the writer answer, quickly and accurately, “What source should govern this decision?” Version status, finality, and completeness determine whether an apparently relevant file can safely support the draft.⁵

Classify by role, not extension

A PDF tells you how a file is stored, not what it means. The same format may contain a protocol, article, guidance document, template, prior deliverable, or scanned comments. Classify sources by their role in the project:

• Evidence: Material that supports scientific, medical, statistical, operational, or factual claims.
• Directive: Material that instructs the work, such as a brief, authority guidance, journal instructions, SOP, approved terminology, or review decision.
• Template or exemplar: Material that shapes form, house style, sequence, or expected level of detail.
• Working record: Decisions, queries, correspondence, review comments, search logs, and version notes.
• Deliverable: The document being produced and its controlled derivatives.
• Excluded material: Content intentionally kept outside the evidence path because it is superseded, irrelevant, unverified, or prohibited.

Add domain-specific roles beneath those broad functions. A clinical-study project may contain a protocol, amendments, SAP, CSR, TFLs, patient narratives, case-report forms, registry information, and results slides. A narrative review may depend on a curated literature set, clinical-practice guidance, seminal papers, and expert input. A device evidence project may draw on evaluation plans, prior reports, state-of-the-art literature, post-market data, risk files, and device descriptions. The deep-dive pages expand these patterns.

One source can have several jobs

A clinical study report can supply methods, results, safety context, deviations, and appendices. It is a pivotal resource for industry-supported manuscripts and may change as analyses and reporting mature, so updates must be tracked.⁴ It may also reflect amendments and analysis decisions that require comparison with the final protocol and SAP. A published paper can provide background evidence, a model for journal structure, and a source of references, but those roles should not be confused. An old deliverable can be a style exemplar without becoming evidence for a new claim.

Allowing several roles prevents false simplification. It also creates a duty to state which role is active at the point of use. “Use the prior manuscript” might mean copy its tone, inherit its references, preserve its terminology, or update its conclusions. Those are very different instructions.

Record identity and provenance

For each material source, capture enough identity to distinguish it from similar or superseded versions:

• Title or descriptive name
• Source type and project role
• Author, owner, or originating function
• Version, date, and status
• Relationship to other versions or amendments
• Access level and confidentiality
• Whether full text or only metadata/abstract is available
• Whether it can be cited publicly, used only as project context, or used only as an exemplar
• Any known quality, conversion, OCR, or completeness problem

This can be lightweight for a small project. The point is not to create a database for its own sake. The point is to prevent “final,” “final 2,” and “final approved” from silently governing different sections. Shared repositories, explicit naming conventions, and one controlled master reduce the risk of comments being lost or applied to the wrong version.^18,19

Establish the source-of-truth hierarchy

Different sources may disagree because they describe different project states. The protocol may define a planned endpoint; an amendment may change it; the SAP may operationalize it; the CSR may describe what happened; a table may present the analyzed result; a slide may simplify it for discussion. The writer needs to know which source governs which question. CSR quality depends on coordinated review of the statistical analysis plan, final data, and validated outputs, with data issues resolved before results enter the narrative.¹²

Create a hierarchy by question, not one universal ranking. For intended study conduct, the final protocol and applicable amendments may govern. For prespecified analysis, the final SAP matters. For numerical reporting, validated outputs may govern. For interpretation, the document may require integration of results, study design, literature, and expert review. For wording and format, a current target instruction or controlled template may govern.

When two authoritative sources conflict, do not resolve the difference by choosing the more convenient one. Log the conflict, identify the owner, and keep the affected drafting area open until the decision is made.

Separate public citation from project traceability

Published literature normally enters a public reference list. Client materials, protocols, tables, analysis plans, prior drafts, and internal decisions often support the work without belonging in that reference list. Both kinds of sources matter.

The dossier should therefore distinguish:

• Citable literature: Sources intended to appear in formal references when used.
• Context evidence: Project materials that support claims, numbers, methods, and decisions but may be represented through internal traceability rather than public citation.
• Directive sources: Items that explain how to write but do not support a scientific claim.
• Exemplars: Items that influence form or style without serving as evidence.

This separation prevents a common mistake: treating anything not destined for the bibliography as if it does not need traceability.

Make gaps first-class project information

A missing file is not only an inconvenience. It changes what can be written. Record gaps with a useful status:

• Missing: Expected and not supplied.
• Partial: Some of the role is covered, but material pieces are absent.
• Inaccessible: Known to exist but not available to the writer.
• Conflicting: Available sources disagree.
• Unverified: Present but not yet authenticated or quality checked.
• Deferred: Intentionally postponed to a later gate.
• Not applicable: Considered and deliberately excluded.

State the consequence. “Final SAP missing” is less useful than “prespecification status of subgroup analyses cannot be confirmed until the final SAP is available.” A gap becomes manageable when it names the blocked decision.

Thin dossiers do not always stop work. The writer can build a Structure, draft stable background, or mark exact placeholders. But the document should not conceal the gap by converting an assumption into a fact. If a source is missing, under revision, or not final, raise the gap early enough to avoid a late trail of “unable to verify” comments.⁵

Inspect conversion and readability

Medical-writing sources arrive in Word documents, spreadsheets, slide decks, PDFs, reference exports, images, scanned pages, and systems. Conversion can introduce errors: broken symbols, merged columns, lost footnotes, missing superscripts, malformed tables, or OCR mistakes. A source can look present while being unusable for precise verification.

Spot-check extracted content against the original where stakes are high. Pay particular attention to minus signs, inequality symbols, decimal points, superscripts, Greek letters, units, table alignment, footnotes, and patient-level identifiers. If a source cannot be reliably searched, note the limitation and plan a manual inspection path. Source verification requires access to everything used to create the claim, not only the most convenient summary.^5,20

Produce the dossier readback

Before building the document Structure, summarize the dossier in plain language:

“The project contains the final protocol and two amendments, the final SAP, validated efficacy and safety tables, a draft CSR, 46 citable references of which 39 have full text, the target journal's author instructions, and a prior manuscript used only for style. The CSR and one results slide disagree on the analysis population for a secondary endpoint. Patient narratives are not yet available. The subgroup analysis is present in a table but its prespecification status still needs confirmation.”

This readback does three things. It demonstrates understanding, invites correction before drafting, and exposes the decisions the Structure must carry. Clear roles, source access, and planned decision points make collaborative work faster to review and easier to recover when the project changes.¹⁸

Chapter 04· 8 min read

Find, Appraise & Organize the Evidence

Literature is often described as a set of papers. For the writer, it is better understood as a chain of decisions: what question is being answered, where evidence is sought, what is retrieved, what is included, how quality and relevance are judged, what each source supports, and what remains uncertain.^6,21

The chain can be brief or highly formal. A manuscript introduction may need a targeted search for disease burden, current management, and the precise evidence gap. A narrative review may require a broad and balanced exploration of a field. A systematic review treats the search, screening, and synthesis process as a scientific method that must be reproducible. The label must match the process.^21,22

Begin with the evidence need

Translate the document brief into evidence questions. A useful question identifies the topic, population or setting, intervention or exposure where relevant, comparator where relevant, outcomes or concepts of interest, time horizon, and the role the evidence will play.

Not every question needs a formal PICO framework. The discipline is to move from “find references on this topic” to “find evidence that lets us explain this specific claim, context, comparison, mechanism, burden, practice gap, or uncertainty.”

List the likely evidence functions:

• Establish the importance or burden of the problem.
• Describe current practice or standard approaches.
• Explain a mechanism, method, or conceptual model.
• Support the rationale for a study or intervention.
• Compare the new findings with prior work.
• Identify limitations, contradictions, or safety concerns.
• Support recommendations or implications at the strength permitted by the evidence.

This list becomes a search map and later a source-to-section map.

Choose the rigor before the search expands

Use a simple ladder:

1. Targeted support: A bounded search for discrete claims or background needs. Record enough detail to repeat the search reasonably, but do not represent it as comprehensive.
2. Structured narrative review: A planned search across defined concepts and sources, with documented selection principles and balanced coverage, but without claiming the full reproducibility and exhaustive methods of a systematic review.
3. Systematic review: An explicit research method with predefined eligibility, reproducible search strategies, documented screening and exclusions, appraisal, and transparent synthesis. A meta-analysis adds statistical combination when appropriate.

Convenience samples must not masquerade as systematic evidence: even in a narrative review, the literature search is a critical source of selection bias.²¹ A systematic review, by contrast, begins with focused questions, a specified search, defined selection criteria, critical appraisal, and an explicit synthesis.⁶ A writer can search carefully and still produce a narrative review. The problem is not the absence of the “systematic” label; it is claiming a rigor the process did not have.

Build search concepts, not one giant sentence

Break the question into concept blocks, synonyms, spelling variants, abbreviations, indexing terms, drug or device names, disease terms, population terms, and study-design filters where justified. Search strings should be broad enough to capture relevant language and narrow enough to remain reviewable.

Test the strategy against known relevant papers. If a seminal or clearly eligible source does not appear, inspect which concept excluded it. Avoid making the strategy so dependent on exact terminology that it misses older language, regional terms, or indexing differences.

Select databases and current search guidance according to the project. Whatever the search level, document where you looked, what you asked, when you searched, which limits you applied, and how the results were handled.^6,23

Separate discovery from inclusion

A search result is a candidate, not evidence already accepted. Maintain distinct states:

• Retrieved metadata
• Abstract available
• Full text available
• Duplicate resolved
• Screened for eligibility
• Included or excluded with reason
• Appraised for validity and relevance
• Assigned to one or more evidence functions
• Cited or used in the document

This prevents a common collapse in which a title is collected, an abstract is skimmed, and the paper enters a reference list without full-text verification. Across medical fields, a systematic review of quotation accuracy estimated a mean cited-claim error rate of 14.7%; most content errors were major mismatches in which the reference failed to substantiate, was unrelated to, or contradicted the claim.²⁴

Metadata quality matters. Incorrect authors, title, year, journal, DOI, or pagination make references hard to retrieve and can propagate through citation tools. Deduplication should compare more than one field because the same work can appear through variant records.

Appraise before you summarize

Evidence appraisal asks two broad questions: can this source be trusted for the claim, and does it directly answer the question being asked?

Consider:

• Study design and whether it fits the inference.
• Population, setting, intervention, comparator, and outcome relevance.
• Sample size and precision.
• Risk of bias, confounding, missing data, and selective reporting.
• Prespecified versus exploratory status.
• Directness to the current document's population and use.
• Recency and whether later evidence changes interpretation.
• Independence, funding, and conflicts where relevant.
• Whether the full method and result are available for verification.

Do not reduce appraisal to a prestige shortcut. A highly cited paper can be indirect. A guideline can be authoritative within scope and irrelevant outside it. An observational study can answer a real-world question that a trial did not address, while remaining unable to support the same causal language. Match the inference to design, conduct, precision, and applicability—not to reputation.^6,25

Search for disconfirmation

Fair balance improves when the search explicitly looks for evidence that might change the preferred message.⁴ Search alternative terminology, contradictory findings, negative studies, adverse outcomes, limitations, and competing explanations. Citation chasing can reveal influential work missed by the first strategy, but record the path so that snowballing does not become invisible cherry picking.

The writer should be suspicious of an evidence set that is uniformly convenient. Uniformity may be real, but it deserves testing. A balanced review does not give every opinion equal weight; it represents material evidence in proportion to its quality and relevance.

Treat systematic reviews as research projects

Systematic reviews are investigations whose “subjects” are prior studies.²² Their credibility depends on explicit methods. Define the question and eligibility criteria before screening. Preserve the exact search strategies, dates, sources searched, deduplication, screening decisions, reasons for exclusion, included-study characteristics, appraisal, and synthesis method.^6,26

A meta-analysis requires statistical expertise. Pooling does not automatically solve disagreement among studies; heterogeneity, incompatible definitions, bias, and differing designs can make a single summary misleading.²⁶ The medical writer should understand the analysis well enough to report it and collaborate closely with a qualified statistician or evidence-synthesis specialist.

The systematic-review burden also changes drafting. Methods must let another reader understand and, in principle, reproduce the process. Results must account for the flow of studies and the characteristics and findings of those included — the PRISMA statement's 27-item checklist and study-flow diagram exist precisely to make that reporting transparent.²⁶ Discussion must separate what the synthesis shows from what the limitations of the evidence base allow.

Organize evidence for writing

A reference manager stores citations. A writing evidence system stores decisions about use. Add structured notes:

Do not write interpretive notes that are stronger than the paper. Whenever possible, retain the exact location supporting the note: page, table, figure, section, or paragraph. This reduces later source hunting.

Know when to stop

Searches expand easily because every paper contains more references and more terminology. Stop criteria depend on method. A systematic review follows its protocol and update rules. A targeted search can stop when the defined evidence needs are adequately covered, important contradictions have been investigated, and additional searching yields diminishing relevance rather than new conceptual information.

Stopping is not the same as declaring certainty. Record the search limit, coverage, and unresolved gaps. A sentence such as “No evidence was found” is much stronger than “The defined search did not identify evidence.” The latter describes what the process can actually support.

Open the systematic-review and review-article deep dive ↗

Chapter 05· 6 min read

Know What Governs the Document

Medical documents may be shaped by ICH families for drug development, reporting recommendations for study designs, good publication practice, journal instructions, regulatory requirements, templates, style guides, plain-language principles, statistical guidance, device frameworks, and organizational procedures.^1,27,28 These sources do not all have the same authority, purpose, or scope.

The writer's first guideline skill is therefore classification.

Several different things are called guidance

Use a practical taxonomy:

• Law and regulation: Binding within a defined jurisdiction and scope.
• Regulatory guidance: An authority's expectations or recommended approach, with a stated status.
• Technical or consensus standard: A harmonized framework such as the ICH material used across drug-development documents.^1,28
• Reporting guideline: A checklist or explanation for complete and transparent reporting of a study design or evidence method.^6,27
• Ethical and publication-practice policy: Authorship, disclosure, trial registration, data transparency, editorial conduct, and professional-writing support.
• Target instructions: Requirements from a journal, congress, authority, payer, platform, or other destination.
• SOP or house rule: Sponsor, client, institution, agency, or team procedure.
• Template: A reusable structural artifact, form, or content shell.
• Clinical-practice or society guideline: Domain evidence about care, usually cited as content rather than used as the reporting structure.

The distinctions prevent category errors. A clinical-practice guideline can support a statement about current management but does not tell the writer how to report a randomized trial. A reporting checklist can improve completeness but may not determine the file format or word limit. A sponsor template can dictate headings without overriding the underlying regulatory requirement.

Determine applicability before extracting rules

For every candidate governing source, ask:

1. What document or process does it govern?
2. Which study design, product, population, or lifecycle stage is in scope?
3. Which jurisdiction or destination applies?
4. What is its status: final, draft, superseded, optional, mandatory within an organization, or informative?
5. Which version and effective date apply to this project?
6. Does a target-specific instruction narrow or extend it?
7. Does an extension apply because of the study or audience?

Do not rely on memory for high-stakes requirements. Standards, policies, questions-and-answers documents, and implementation practice evolve; good publication practice alone has been revised repeatedly since 2003, including the GPP 2022 update.²⁹ Check the authoritative text, version, status, scope, and effective date instead of assuming that a remembered summary remains current.²⁸

Build the governing stack

The governing stack is the set of applicable sources arranged by role and authority. It might contain a regulatory requirement, technical guideline, sponsor SOP, controlled template, statistical reporting guidance, target-journal instructions, authorship policy, and house style. Each contributes something different.

Represent the stack as a register:

The register turns a reading task into a production tool. Instead of remembering that a source “applies,” the writer records what it changes.

Extract requirements into the Structure

Guidance becomes useful when translated into operations:

• Required sections and sequence
• Mandatory content items
• Definitions and controlled terminology
• Data presentation and statistical reporting expectations
• Safety and balance requirements
• Authorship, acknowledgment, disclosure, registration, and transparency statements
• Word, table, figure, reference, and supplement limits
• Required appendices, forms, checklists, or schemas
• File types, naming, submission components, and destination behavior
• Review, approval, and recordkeeping requirements

Map each requirement to a section, document-level check, or workflow gate. If it has no destination, it is likely to be forgotten.

Templates are scaffolds, not substitutes for thought

Templates can save time, improve consistency, and encode required structure. They can also encourage mechanical completion. A heading may remain after its purpose has changed. Boilerplate may survive into an incompatible study. A prior document may carry obsolete language, a superseded requirement, or an assumption that does not fit the new evidence.

Before drafting into a template, inspect every section. Ask what requirement it serves, what source belongs there, whether it applies, and whether the current project needs additional structure. Keep required sections even when content is unavailable, but mark the gap clearly rather than filling space with generic prose.

Reusable components can reduce repeated work, but template-based authoring remains labor-intensive and component reuse pays off only when it is managed deliberately.³⁰ Common formats, controlled text, checklists, and collaborative authoring can improve consistency and shorten development when ownership and review are explicit.^1,18 Reuse is safest when content identity, approved context, source support, and update responsibility are controlled. Copy-and-paste reuse without provenance is not a system; it is version drift.

Resolve conflicts explicitly

Conflicts may arise between a target instruction and a house template, between two versions of guidance, between a sponsor preference and a reporting requirement, or between a word limit and the content needed for complete interpretation.

Use a disciplined response:

1. Confirm that both sources actually apply.
2. Check version, status, and scope.
3. Identify whether the conflict is real or whether one source is more specific.
4. Define the consequence for the document.
5. Route the decision to the accountable owner.
6. Record the rationale and affected sections.
7. Recheck the final document against the resolved rule.

Do not hide conflicts inside stylistic editing. A choice that affects compliance, scientific interpretation, or disclosure deserves a visible decision.

Guidelines improve completeness, not judgment by themselves

Reporting guidelines help writers remember essential items, but they are aids rather than substitutes for judgment and often require application to the specific manuscript.^6,27 They do not decide whether the study was well designed, whether a claim is fair, whether a limitation is material, or whether the reader will understand the document. Checklist completion can coexist with poor reasoning.

Use a checklist twice: once while building the Structure, so required content has a home, and again during final review, so the completed text can be tested. Record “not applicable” deliberately. A blank checkbox is ambiguity; a reasoned exclusion is a decision.

Open the full guideline and standards shelf ↗

Chapter 06· 6 min read

Build the Structure Before the Prose

Writers who skip structure often spend their time solving structural problems sentence by sentence. Background expands because its boundary is unknown. Results repeat tables because the narrative purpose was never defined. Discussion becomes a second literature review. Reviewers request major reorganization after weeks of drafting because they never saw the intended shape.

The Structure makes the document arguable before it becomes polished.

A heading list is not enough

A useful Structure has two layers:

1. Skeleton: The ordered headings and subheadings required by the document type, target, and argument.
2. Section instructions: A concise statement of what each section must accomplish, which sources it uses, which claims or questions it handles, what limits apply, and how completion will be judged.

Compare:

Discussion

with:

Discussion — opening synthesis: Summarize the primary, secondary, and safety findings in Results order without repeating all values. State whether each prespecified objective was met. Use the final tables for values and the protocol/SAP for endpoint status. Do not introduce new results. Flag the unresolved analysis-population discrepancy before drafting.

The second version can guide a writer, reviewer, or tool. It also exposes what the team has not yet decided. A detailed outline can contain nearly all information intended for the final document and can be circulated while structural changes are still cheap.^4,31

Start from three maps

Build the Structure by overlaying:

• Requirement map: Sections and items required by guidance, target instructions, template, ethics, and submission format.
• Evidence map: Sources, tables, analyses, references, and gaps available to support each section.
• Reader map: The sequence of questions the audience needs answered.

Where the maps align, drafting is straightforward. Where they diverge, the Structure should show the problem. A required section without evidence becomes a gap. Important evidence without a section may signal missing structure or irrelevant material. A logical reader question without either requirement or source may require new research or a scope decision.

Give each section one primary job

Sections can contain several elements, but each needs a dominant purpose. In a research paper, the Introduction establishes the problem, current knowledge, gap, and objective. Methods establish what was done and analyzed. Results report findings without turning into interpretation. Discussion explains what the findings mean in context, including limitations and implications.

A one- or two-sentence problem statement that distinguishes the broader problem from the specific problem addressed is useful beyond manuscripts.³¹ Define what is insufficient or unresolved, what the current situation is, what the document contributes, and what the reader should understand as a result. That logic can shape an introduction, executive summary, briefing document, review, or educational activity.

Write section instructions as verbs: establish, define, describe, report, compare, interpret, qualify, recommend, disclose. A heading names a place; a verb names the work.

Route sources before drafting

For every section, list the sources that govern it and the sources that support it. Keep these categories distinct. A journal instruction may govern abstract length; the results tables support its numerical content. A sponsor template governs headings; the protocol explains planned conduct; the CSR and outputs explain actual results.

Source routing prevents opportunistic searching during drafting, where the first convenient source becomes the basis for the paragraph. It also enables parallel work because a reviewer can see which materials a section is supposed to use.

When a source is large, route to the exact location: protocol section, SAP analysis, CSR table, figure, patient narrative, or article passage. This is the beginning of traceability.

Plan claims and counterclaims

List the main propositions the document expects to communicate and map each to evidence. Then list the material qualifications, alternative explanations, negative findings, and limitations that the reader needs for fair interpretation.

For a study Discussion, this might include:

• Primary outcome and precision
• Secondary and exploratory findings with status clear
• Safety and tolerability
• Agreement or disagreement with prior literature
• Methodological strengths
• Specific limitations and likely direction of their effect
• Generalizability
• Clinical or scientific implication stated proportionately

This pre-writing exercise catches unsupported enthusiasm before it becomes polished prose.

Design tables and figures early

Design tables and figures early.^1,4 A visual is not decoration appended after the text; it is a communication unit that may carry the document's most important evidence. The form should fit the intended message, data, audience, and use while preserving scientific accuracy.³²

Decide which reader question each visual answers, which source supplies its data, and what the text should say that the visual does not. Avoid presenting the same data fully in text, table, and figure. Plan titles, legends, footnotes, abbreviations, populations, denominators, units, and statistical notation as part of the Structure.

Visual planning also tests the argument. If no clear display can be designed for a proposed message, the message may be poorly defined or unsupported.

Set budgets and constraints

Assign approximate word, table, figure, and reference budgets before prose expands. Budgets force prioritization. Give more space to the document's central reasoning and less to background the audience already knows.

Budgets are not rigid quotas. They are signals. If one section requires twice its allocation, inspect whether the Structure is wrong, the scope has expanded, or the prose is carrying information that belongs in a visual, appendix, supplement, or deep link.

Carry gaps honestly

Use precise placeholders:

• [Confirm final analysis population against SAP and Table 14.2.1]
• [Insert patient-disposition values after validated output is received]
• [Author decision required: include exploratory subgroup in abstract?]
• [Target journal not selected; abstract structure and reference limit remain provisional]

Avoid placeholders such as [TBD] that preserve no reasoning. A good placeholder states what is missing, why it matters, and what will resolve it.

Review the Structure with the right questions

Ask reviewers to evaluate shape, not sentence style:

• Does the sequence answer the reader's questions?
• Are all governing requirements represented?
• Does every key message have a credible source path?
• Are limitations and safety visible enough?
• Are planned visuals necessary and nonredundant?
• Are any sections overlapping or missing?
• Are gaps accurately stated?
• Is the scope realistic within the target and timeline?

Structure approval is valuable because it makes disagreement cheap. Moving a heading or changing a claim plan takes minutes; major changes in content or organization are far easier at the outline stage, before time has gone into crafting sentences.³¹ Rewriting a polished section can take days and generate version noise.

Let reality revise the plan

An accepted Structure is a contract, not a prison. Drafting may reveal that evidence is weaker, a section needs to split, a visual works better than prose, or an important contradiction deserves more space. Reopen the Structure explicitly, record the change, and check downstream effects.

Silent structural drift is the real danger. If the working document changes shape without the plan changing, reviewers no longer know what was agreed.

Open Structure patterns by document type ↗

Chapter 07· 7 min read

Draft from Evidence, Section by Section

Drafting feels like the central act of writing because it produces visible pages. In a controlled workflow, drafting executes decisions already made about purpose, audience, sources, governing requirements, Structure, and claims. This does not make drafting mechanical. It creates room for better judgment because the writer is not solving every project problem inside every sentence.^9,31

Start with the section contract

Before writing, read the section instruction and answer:

• What must the reader understand after this section?
• Which sources support it?
• Which source governs terminology or form?
• Which claims are planned?
• Which gaps or conflicts remain?
• What must not be introduced here?
• How long should it be?
• Which table or figure carries part of the message?

If these questions cannot be answered, the section is not ready. Return to the dossier or Structure instead of filling space.

Build a claim-evidence mini-plan

For each subsection, list the main claim, supporting source, material qualifier, and intended paragraph role. A simple plan might look like:

The plan makes it harder to write beyond the evidence. It also reveals when several sources support different parts of a sentence and should not be collapsed into one citation cluster.

Lead with the point

Readers often scan. Topic sentences let them follow the argument by reading the first line of each paragraph. The first sentence should normally state the paragraph's primary point; the rest supplies evidence, explanation, qualification, or transition.^4,9

Avoid openings that delay meaning with long scene-setting clauses. Put the important subject and action early. Keep related words close enough to prevent ambiguity. Use parallel structure when presenting comparable items. Repeat key terms when consistency helps more than stylistic variety.

Brevity comes from removing material that does not serve the reader, not merely from shortening words. A compact paragraph can still be dense. A longer paragraph can be clear if its logic is explicit. Readability is a property of information design as much as sentence length.

One paragraph, one main job

A paragraph can establish context, explain a method, report a result, compare evidence, interpret a finding, qualify a claim, or transition. When it tries to do all of these, the reader cannot tell which sentence controls.

Use a durable pattern:

1. Point
2. Evidence or explanation
3. Qualification or limitation where material
4. Connection to the next point

This pattern is flexible. Results paragraphs may lead with a finding and then provide magnitude and precision. Discussion paragraphs may lead with an interpretation, compare literature, consider alternatives, and end with the implication. The important discipline is that the paragraph's logic is visible.⁹

Separate the levels of statement

Medical prose often blurs four different acts:

• Observation: What the study, source, or analysis reported.
• Interpretation: What the observation may mean.
• Implication: Why the meaning matters in a broader context.
• Recommendation: What somebody should do.

Each step requires additional support. A difference observed in a study does not automatically establish a causal explanation. A causal explanation does not automatically justify a practice recommendation. The farther the sentence moves from observation, the more carefully the writer must examine evidence, scope, and authority.^6,25

Signal the level with verbs. “Was associated with” does different work from “caused.” “Suggests” differs from “demonstrates.” “May inform” differs from “should change.” Do not use cautious verbs to disguise a fundamentally unsupported claim; qualification must be substantive.

Match strength to design and precision

Lack of statistical significance is not proof of similarity, a p-value alone is inadequate, and estimates and confidence intervals carry essential meaning.^33,34 The prose should reflect study design, effect size, precision, and limitations.

Randomization, blinding, comparator choice, sample size, follow-up, missing data, analysis populations, multiplicity, confounding, and exploratory analyses all affect what can be said. The writer does not need to become the statistician, but must understand enough to avoid changing the inference in translation.

Keep “significant” for contexts where its meaning is explicit. If you mean clinically important, say so and support the threshold. If you mean statistically significant, provide the result in the form expected by the project. If you mean notable, choose a less ambiguous word.

Preserve expected and unexpected findings

Documents often become distorted when they narrate only the anticipated result. Report findings in the order established by the objectives or Structure, including results that are null, negative, inconsistent, or difficult to explain. Give safety and tolerability appropriate space rather than treating them as an appendix to efficacy.^4,6

Unexpected findings do not need speculative explanation. State what occurred, confirm the analysis, inspect related evidence, and distinguish plausible explanations from demonstrated ones. A Discussion can acknowledge uncertainty without becoming weak. It becomes stronger when readers can see the boundary between result and interpretation.

Use quotations and paraphrases sparingly

Medical writing usually depends on synthesis rather than quotation. Paraphrase after understanding the source, not by changing a few words. Preserve the original meaning and scope. Cite the source supporting the idea. Direct quotations are appropriate when exact wording is itself material, but they should not substitute for analysis.

Plagiarism risk rises when writers work too closely from source sentences or reuse old documents without tracking origin. Copying previously published material, including one's own work, without appropriate attribution or permission is unethical even when it is unintentional.⁸ Draft from notes and claim plans, then verify against the source. Reused language should follow the project's permissions, disclosure, and originality requirements.

Draft with visible uncertainty

Use explicit query markers rather than silent guesses:

• [QUERY—The table labels this population FAS; the CSR text says ITT. Confirm preferred term.]
• [SOURCE NEEDED—Current practice statement lacks a full-text citation.]
• [AUTHOR INPUT—Alternative explanation for the unexpected safety pattern.]
• [STATISTICAL REVIEW—Confirm wording of noninferiority conclusion.]

Queries are not signs of failure. They are evidence that the document is under active control. Remove or resolve every marker before finalization; do not let uncertainty vanish merely because the prose around it sounds complete.

Review the section before moving on

At the end of a section, perform a short review:

• Does it fulfill the section instruction?
• Does every material claim have the right support?
• Are numbers and populations consistent with sources?
• Are observation and interpretation separated?
• Are limitations and uncertainty proportionate?
• Is any information duplicated elsewhere?
• Does the last sentence hand the reader to the next section?

Early section review catches conceptual drift before it spreads into the abstract, summary, visuals, and conclusion.

Chapter 08· 8 min read

Turn Data into Accurate Narrative and Visuals

Numbers create an illusion of solidity. A value with two decimal places looks authoritative even when its denominator is wrong, its population is unclear, or its analysis was exploratory. The medical writer's job is not to make numerical output sound confident. It is to preserve what the analysis actually means.

Context comes before calculation. Who was studied? What was measured? How was it measured? What comparison was made? How large was the difference? How precise was it? What would count as clinically meaningful? What can be generalized?^33,34

Find the source of truth

Before writing from a table, identify its status. Is it a validated output, draft output, exploratory analysis, slide transcription, or manually created summary? Check whether the same result appears elsewhere and whether those versions agree.^5,12

For every material number, confirm:

• Analysis population
• Treatment or comparison groups
• Endpoint definition
• Time point and analysis window
• Unit and transformation
• Denominator and missing-data handling
• Estimate and measure of variability or precision
• Statistical method
• Prespecified, sensitivity, supportive, subgroup, or exploratory status
• Data cutoff or version

Do not infer these details from layout. A column header can be abbreviated or inherited from a template. Use the protocol, SAP, output notes, and statistical review as appropriate.

Build number provenance

Keep a compact provenance record for high-risk values:

The record does not need to accompany every final document. Its value is that a reviewer can reconstruct the number without searching hundreds of pages.

Number provenance is especially important when the same value is transformed. Percentages may be calculated from counts, rates may use person-time, changes may be derived from baseline and follow-up, and summaries may round. Record the transformation and avoid repeated manual recalculation in several sections.

Do not report a p-value alone

Never report a p-value alone: estimates and confidence intervals carry meaning that a p-value cannot.³³ A p-value addresses a narrow question under assumptions; it does not state the size, clinical importance, or precision of an effect. Report the estimate and an appropriate measure of uncertainty in the form required by the document.

Likewise, a nonsignificant result does not prove equivalence or absence of effect. It may reflect imprecision, limited sample size, variability, or an effect smaller than the study could reliably detect. Use wording such as “the study did not detect a statistically significant difference” when that is what the analysis supports, rather than “the treatments were the same.”³⁴

Statistical significance and clinical importance are separate. If a minimum clinically important difference or other threshold is invoked, identify its source and applicability. Do not label an effect meaningful merely because the p-value crossed a convention.

Respect analysis status

Prespecification affects credibility. Primary and secondary endpoints, multiplicity control, analysis populations, and sensitivity analyses belong to an agreed statistical plan. Post-hoc and exploratory analyses can be valuable, but they should be labeled as such and interpreted as hypothesis-generating unless stronger justification exists.

Subgroup findings require particular care. Small numbers, multiple comparisons, interaction testing, and lack of prespecification can make an apparently striking pattern unstable. Do not write a subgroup result as a general treatment claim. State the subgroup, analysis status, estimate, precision, and relevant limitation.

The same principle applies to real-world evidence and observational analyses. Confounding, selection, misclassification, missingness, and data provenance affect inference. Adjustment can reduce some problems but cannot transform an observational design into randomization. Let the design govern the verbs.³⁵

Report the population, not just the percentage

Percentages without counts can hide small denominators. Counts without denominators can hide differing group size. Rates without time can hide exposure differences. Always ask what a reader needs to interpret the value.

For adverse events, distinguish patients from events and incidence from exposure-adjusted rates. For diagnostic measures, keep sensitivity, specificity, predictive values, and prevalence context distinct. For survival analyses, report the relevant estimate, time horizon, censoring context, and uncertainty. For patient-reported outcomes, explain the instrument, direction, scale, timing, and interpretive threshold when used.

Units and decimal places carry meaning. Use precision appropriate to the measurement and analysis. Excess decimal places imply information the study may not possess; inconsistent rounding creates apparent disagreement.

Write Results in the order of the questions

Results should follow the Structure established by objectives and methods. A common sequence is participant flow and analysis populations, baseline characteristics, primary outcome, secondary outcomes, other analyses, and safety. The exact order depends on the document, but it should not be rearranged merely to place the most favorable finding first.

Use parallel Methods and Results headings where possible. This makes it easier to see whether every promised analysis was reported and whether any result appears without a described method.

Narrative text should highlight the pattern, magnitude, and interpretation needed to navigate the outputs. It should not transcribe every cell. If a table provides the full values, the text can state the main result and refer to the table. If the exact value is central to the claim, include it.

Make safety visible

Fair balance requires appropriate attention to safety and tolerability.⁴ Avoid placing detailed efficacy language beside a single generic sentence that treatment was “well tolerated.” Report the defined safety population, exposure, adverse-event categories, serious events, discontinuations, deaths where applicable, and events of special interest according to the document and source data.

Do not infer causality from temporal occurrence unless the analysis and assessment support it. Preserve the distinction among adverse events, adverse reactions, and other safety classifications used by the project. Use consistent terminology across narrative, tables, and conclusions.

Choose the right visual

Visuals are analytical communication, and the form should be chosen according to the reader's question, the structure of the data, and the risk of distortion.^1,32 Choose the form according to the question:

• Use a table when readers need exact values or comparison across several dimensions.
• Use a line graph for change or trend over an ordered continuum when the scale supports it.
• Use a bar or dot display for category comparisons when position and magnitude matter.
• Use a flow diagram for progression through study or review stages.
• Use a forest plot when effect estimates and confidence intervals across studies or subgroups are the point.
• Use a diagram when relationships, processes, or mechanisms matter more than exact values.

Avoid decorative three-dimensional effects, truncated axes that exaggerate differences, inconsistent scales, and colors that imply categories or importance without explanation. Elaborate designs can confuse and distract rather than inform.^10,32 Visual choices can mislead even when every plotted value is correct.¹

Make the visual stand alone

A reader may encounter a table or figure without reading the surrounding paragraph. Give it an informative title, define populations and time points, explain symbols and abbreviations, identify units, describe statistical summaries, and include material footnotes.⁴ A title should say what is shown, not merely “Results.”

Keep the visual concise enough to read. Large tables may need grouping, hierarchy, or movement to supplementary material. Do not solve crowding by shrinking text until the information is technically present and practically inaccessible.

For public-facing material, visuals may reduce cognitive burden, but they also require testing. Icons, risk graphics, timelines, and diagrams can be misunderstood. Health literacy, cultural relevance, accessibility, patient involvement, and human review all affect whether a visual communicates as intended.^7,32,36 A familiar-looking symbol is not automatically universal.

Run the cross-document number check

Compare repeated values across:

• Abstract or executive summary
• Main narrative
• Tables and figures
• Conclusions
• Plain-language derivative
• Poster or slides
• Response letter or briefing points

Check endpoint names, populations, denominators, direction, units, rounding, significance statements, and time points. If two values intentionally differ, explain why.

Open the statistics and visual-reporting deep dive ↗

Chapter 09· 6 min read

Make Every Claim Traceable

Reference management and claim support are related but different tasks. Reference management asks whether the citation is formatted and listed correctly. Claim verification asks whether the source supports the sentence the document makes.

A sentence can have a perfect Vancouver citation and still misrepresent the paper.

Define the claim before checking it

Claims may be factual, numerical, methodological, causal, comparative, interpretive, or prescriptive. Break compound sentences apart during verification. One sentence may contain a disease-burden estimate, a statement about current practice, and a conclusion about unmet need; no single source may support all three.

For each claim, identify:

• Subject and population
• Intervention, exposure, condition, or method
• Comparator where relevant
• Outcome or fact asserted
• Direction and magnitude
• Time and setting
• Degree of certainty
• Whether the statement is observation, interpretation, implication, or recommendation

Then inspect whether the source matches those elements.

Cite what the source says, not what you hoped it would say

Common mismatches include:

• A source concerns adults while the claim says patients generally.
• An observational association is written causally.
• A subgroup result becomes a population-wide conclusion.
• A surrogate endpoint becomes a clinical outcome claim.
• A narrative review is cited as proof of a precise estimate available only in the primary study.
• An abstract is used for detail not present without full text.
• A paper reports statistical significance while the claim adds clinical importance.
• A guideline recommendation is used outside its population or jurisdiction.

Source relevance is not enough. The source must support the exact proposition at the strength stated. Verifying cited claims is a core medical-writer responsibility, not an editorial nicety.²⁴

Use the most direct support available

Prefer the primary source for a study-specific claim when it is available and appropriate. Reviews and guidelines can support synthesis, context, and recommendations, but secondary citation can blur the original population, method, or result. If a secondary source is used because the primary source is unavailable, know that the verification boundary is weaker and state the limitation where material.^5,24

For project-specific claims, use the direct internal source: validated table for the number, protocol for intended design, SAP for planned analysis, report for actual conduct and integrated interpretation, decision record for an approved deviation. A slide deck may be convenient but should not silently outrank the source from which it was derived.

Separate two evidence trails

Formal publication citations serve readers and feed the reference list. Project traceability links serve writers, reviewers, clients, quality specialists, and regulators who need to inspect nonbibliographic sources. A clinical manuscript may need both: journal citations for background and context, plus internal links from results to tables or report sections during development.^4,5

Do not pollute the public reference list with internal file references that do not belong there. Do not omit internal support merely because it will not appear in the bibliography. Maintain the two trails until handoff requirements determine what remains visible.

Verify at the right granularity

The more consequential the claim, the more precise the anchor should be. A general background statement may be supported at article level. A numerical claim should point to the exact table, figure, page, or result. A statement about protocol-defined analysis should point to the relevant section. A safety narrative may require patient-level source comparison.^20,24

Granularity makes review efficient. “Supported by CSR” is weak when the CSR contains hundreds of pages. “CSR Section 12.2, Table 14.3.1, row X” is inspectable.

Treat reference accuracy as a safety net

Check references for correctness and completeness.^4,5,10 Confirm authors, title, journal or publisher, year, volume, pages, identifier, and access to the actual source. Confirm that the in-text citation maps to the intended reference and that every listed reference is cited appropriately.

Citation tools reduce formatting work but can import incorrect metadata, duplicate records, or attach the wrong full text. Automated output requires review. When a citation changes position in a numbered system, check that numbering and cross-references remain intact after revisions.

Detect the four dangerous states

1. Unsupported: No source currently supports the material claim.
2. Source mismatch: A source is present but does not support the exact claim.
3. Overclaimed: The source supports a narrower or less certain statement.
4. Conflicting evidence: Credible sources disagree in a way material to the reader.

These states need different responses. Unsupported claims require evidence, removal, or reframing. Mismatches require a better source or corrected sentence. Overclaims require reduced strength or additional support. Conflicts require balanced synthesis rather than choosing the preferred paper.

Build a claim-support matrix when risk justifies it

Not every document needs a full matrix. High-risk claims, regulated documents, key-message platforms, and derivative assets benefit most. The matrix also reveals when one approved claim is being reused across outputs with different context.

Check the sentence after every material revision

A reviewer may improve style while changing meaning. Removing “in this study,” “was associated with,” or “exploratory” can convert a bounded observation into a general causal claim. Adding “therefore” can create a logical step the evidence does not support. After substantive changes, repeat the claim-source check.

Traceability must survive review. If support is attached only to a drafting note that disappears when text moves, the evidence trail is brittle. Use stable anchors or a maintained matrix.

Practice ethical synthesis

Citation choices shape the scientific record. Avoid citation padding, selective citation, and citing a source merely because it is prominent. Acknowledge materially contradictory evidence. Represent the contribution of prior work accurately. Avoid text recycling and patchwriting.⁴ Make professional writing assistance and authorship contributions transparent according to the publication context.¹³

The goal is not the largest reference list. It is a reference set that lets the reader inspect the intellectual path of the document.

Open the publication ethics, authorship, and disclosure deep dive ↗

Chapter 10· 6 min read

Use AI Without Handing Over Judgment

Generative AI can support structured authoring, automated narratives, literature review, content reuse, and quality checks, while introducing risks involving confidentiality, fabricated content, bias, inconsistency, and misplaced trust.^37,38,39 AI can accelerate bounded parts of the workflow, but the writer remains responsible for whether the document is true, fit for purpose, and safe to release.

Start with the document task and workflow

Define what work you want to improve. Suitable candidates may include formatting references or abbreviation lists, converting images to editable text, preparing slide scripts, creating search strings, writing macros, basic data exploration, content mapping, structured reuse, drafting from defined inputs, and assisting quality checks.^37,39

Classify tasks by consequence:

• Low-risk transformation: Format, classify, deduplicate, compare, or convert material where errors are easy to detect.
• Assisted production: Propose an outline, summary, paragraph, script, query, or visual from specified sources, with mandatory review.
• High-risk judgment: Decide evidence quality, interpret ambiguous statistics, resolve conflicting guidance, assign authorship, determine regulatory acceptability, or approve final claims.

AI is most defensible when the task is bounded, the input is controlled, and the output can be checked. High-risk judgment should remain with qualified people, even when AI organizes the material they inspect.

Understand probabilistic output

Deterministic expert systems are predictable and template-like, whereas language models generate probable continuations from learned statistical patterns.^37,39 A language model does not understand a study as a human investigator does, and it may not be able to identify the source behind a statement. It can produce fluent falsehoods, invented citations, outdated practices, biased framing, or inconsistent answers to the same instruction.

Fluency is therefore a risk factor. Awkward output invites scrutiny; polished output can pass unnoticed. Treat generated content as a proposal whose confidence is unknown until verified.

Retrieval-augmented workflows can provide relevant source snippets to a model and may allow citations, but retrieval does not guarantee that the right passage was selected or that the generation represents it correctly. The evidence still needs human inspection.

Protect data before adding project material

Permission, employer and client policies, nonpublic content, proprietary information, personal data, and protected health information must be resolved before project material enters an AI-enabled workflow.^15,37,39 Determine:

• Whether the content is public, confidential, personal, proprietary, or regulated.
• Whether the client, employer, publisher, or institution permits the use.
• Whether inputs are retained, used for training, exposed to vendors, or capable of leaking.
• Where data are processed and stored.
• Who can access project instructions and outputs.
• Whether the workflow and contract meet the project's security requirements.

Removing a name does not automatically anonymize clinical information. Patient narratives and small subgroups can remain identifiable through combinations of facts. If the permission or system boundary is unclear, do not upload the content.

Set the production contract before drafting

Useful project instructions resemble useful section instructions. Specify:

• Role and task
• Approved source set
• Audience
• Output type and length
• Required content
• Prohibited inference
• Terminology and style constraints
• How to show citations or source locations
• How to mark uncertainty, missing information, and conflicting evidence
• Expected format

The PLANTS reminder captures six useful constraints: persona, length, audience, nuance, type, and style guide.³⁷ The deeper principle is constraint. An unconstrained request asks the system to invent both the task and the answer. A production contract reduces ambiguity and makes review possible.

Require visible provenance

Ask the system to identify which source supports each material statement and to quote or point to the supporting location for reviewer inspection. Do not allow it to fill gaps with material outside the defined dossier when the task is meant to use a controlled evidence set.

If no source supports a requested claim, the correct output is a gap, not a plausible sentence. Design the workflow to reward abstention. “Not found in the supplied sources” is more valuable than a fluent invention.

Verify by risk, not by convenience

Check generated output for:

• Factual and numerical accuracy
• Source match and citation existence
• Population, design, time point, and analysis status
• Missing qualifiers and limitations
• Unsupported causal or comparative language
• Omitted contradictory or safety information
• Hallucinated structure, requirement, or terminology
• Confidentiality and privacy leakage
• Bias and inappropriate generalization
• Style consistency and accessibility

Do not ask the same model to be the sole verifier of its own work. Automated checks can help identify candidates, but material claims need independent comparison with the source.

Keep gates human

Human decision points should include:

• Accepting the project frame and evidence boundary
• Choosing the governing stack
• Accepting the Structure
• Approving key messages and limitations
• Resolving source conflicts
• Accepting or rejecting substantive generated changes
• Confirming statistical and specialist interpretation
• Approving final release

Removing writers from the workflow would also remove document leadership: the work of breaking down tasks, setting timelines, critically evaluating sources, coordinating expertise, and gaining consensus.^37,39 That integrative work is central to medical writing.

Evaluate the workflow against the use case

Assess the underlying model or system, modifications, data access, retrieval, determinism, security, output consistency, auditability, and ability to meet the task's risk profile. Confirm that the workflow preserves authorized sources, visible provenance, controlled access, reviewability, and the handoffs required to produce the document safely.

Pilot one or two narrow use cases. Record what works, what fails, what instructions and examples improve consistency, how much review time remains, and whether the workflow actually reduces total effort. Include the time needed for training, validation, governance, and correction.

Make use transparent

Follow applicable client, employer, journal, and institutional expectations for disclosure. Maintain an internal record where the quality system requires it: system, task, source boundary, output, reviewer, decision, and version. Transparency does not mean publishing every project instruction in every context; it means that material assistance is not hidden from the people accountable for the document.

Open the AI governance deep dive ↗

Chapter 11· 6 min read

Review, Quality Control & Handoff

Review can improve a document or dissolve it. Multiple reviewers add expertise, but without roles and reconciliation they also add repetition, contradiction, scope drift, and late rewrites. The writer needs to manage review as a controlled decision process.^18,19

Design review rounds

Assign a dominant purpose to each round:

• Concept and Structure review: Purpose, audience, scope, logic, evidence plan, and required content.
• Scientific and medical review: Accuracy, interpretation, context, balance, and clinical relevance.
• Statistical review: Analysis identity, estimates, precision, wording, tables, and limitations.
• Author or contributor review: Intellectual input, accountability, and consensus.
• Editorial review: Organization, clarity, coherence, language, target fit, and usability.
• Regulatory, legal, or compliance review: Applicable requirements, risk, claims, privacy, and approval boundaries.
• Quality control: Source data, consistency, correctness, formatting, cross-references, and completion.
• Final proof and release: Confirmation that approved changes and export behavior are correct.

The roles may overlap, but the scope should not. A QC specialist cannot perform a reliable data check if the writer expects an undeclared developmental edit. A statistical reviewer cannot resolve author strategy by implication. Tell reviewers what to inspect and what state the document is in.

Prepare the review package

Provide the current document, relevant source access, a short status note, known gaps, decisions needed, and review deadline. Identify whether tracked changes should be accepted, whether prior comments are resolved, and which parts are stable.

For complex work, include the accepted Structure, claim-support matrix, key output tables, or decision log. This helps reviewers evaluate reasoning rather than starting from a blank understanding.

Do not send an unproofread draft for high-value QC. Small inconsistencies distract attention from key messages, while dense source tables make incorrect transcription easy to miss.²⁰ Clean what the writer can clean so specialists can use their attention on risk.

Triage comments

Classify feedback:

• Correction: Fixes an error or inconsistency.
• Required change: Follows applicable authority, target, or approved procedure.
• Scientific or strategic decision: Changes interpretation, scope, or message.
• Clarification: Improves reader understanding without changing meaning.
• Preference: A stylistic choice among acceptable forms.
• Out of scope: Belongs to another document, phase, or decision.
• Conflict: Contradicts another reviewer or governing source.

This prevents all comments from receiving equal weight. A senior title does not turn preference into requirement. A minor copyedit can still be essential if it corrects ambiguity. Evaluate the comment, its source, and its effect.

Reconcile, do not stack

When feedback conflicts, state the options and consequence. Route the decision to the agreed owner. Avoid hybrid sentences that attempt to satisfy incompatible requests and satisfy neither.

Maintain a response or decision record for substantive comments. Record what changed, why, who decided, and whether downstream elements need updates. For journal peer review, answer every comment respectfully and specifically, show where the manuscript changed — including revised text with page and line numbers — and explain reasoned disagreement without hostility.⁴⁰

Review is complete when decisions are resolved, not when comments have been collected.

Control versions

Use one source of truth. Record document version, status, owner, and review round. Avoid parallel files that each contain accepted changes. If review must occur in several copies, reconcile into a controlled master and document the process.^5,18

Version control also applies to sources. A revised table, protocol amendment, or updated guidance can invalidate completed text. Record the cutoff and assess change impact rather than silently swapping the file.

Separate editing from QC

Editing improves structure, clarity, coherence, and language. QC checks the near-final document against sources and defined standards; QC specialists provide an independent safety net for source-data agreement, consistency, correctness, and clarity.²⁰ Proofreading catches surface errors and final-production defects. These activities support one another but are not interchangeable.

A risk-based QC plan may include:

• Data and numerical verification
• Claim and source verification
• Internal consistency
• Terminology, abbreviation, and style consistency
• Reference and citation checking
• Guideline and required-content conformance
• Tables, figures, legends, and footnotes
• Cross-references, numbering, links, and navigation
• Privacy, redaction, and disclosure checks
• Formatting and destination requirements
• Final-file behavior

Define whether QC checks samples or every instance. High-risk elements may require complete verification and independent review.

Use multiple passes

Focused passes reduce the errors that readers miss when trying to check everything at once.⁴¹ A useful sequence is:

1. Structure and completeness
2. Scientific logic and claims
3. Numbers and source comparison
4. Terminology and internal consistency
5. References and cross-references
6. Language and formatting
7. Visuals and appendices
8. Final export and navigation

Change the viewing method between passes where helpful: print, PDF, read aloud, different zoom, table-only review, or first-sentence scan. Familiarity makes writers see intended text rather than actual text; a fresh reviewer and a changed view interrupt that effect.⁴¹

Recheck change impact

Every substantive change can affect other places. If an endpoint name changes, search the whole document and derivatives. If a number changes, check abstract, text, table, figure, conclusion, PLS, poster, and response letter. If authorship changes, revisit contribution, acknowledgment, disclosure, and title-page files. If a limitation changes, revisit conclusion strength.

Do not assume tracked changes identify all effects. Some effects are logical rather than textual.

Validate the final artifact

The exported PDF, Word file, slide deck, or submission package is the deliverable. Inspect it after conversion:

• Fonts, symbols, equations, and special characters
• Page and slide breaks
• Table splitting and figure resolution
• Hyperlinks, bookmarks, contents, and cross-references
• Citation rendering and reference order
• Headers, footers, page numbers, and file metadata
• Accessibility elements and reading order where applicable
• Hidden comments, tracked changes, notes, or personal information
• Required filenames and component completeness

A clean source file can produce a broken export. Final validation belongs to the writing workflow.

Package the handoff

Deliver what the recipient needs to use and defend the document. Depending on context, that may include:

• Final artifact
• Editable source
• Clean and tracked versions
• Reference library or bibliography
• Claim-support matrix
• Guideline checklist
• QC record
• Comment reconciliation or response letter
• Source list and data cutoff
• Open-item or assumption log
• Version and approval record

Do not include internal working material by default. Include it deliberately according to purpose, privacy, contract, and quality requirements.

Chapter 12· 7 min read

Publications & Medical Communications

Publications are the clearest place to see the common engine. A study paper, review, abstract, poster, case report, and letter look different, but each begins with a purpose and audience, draws on an evidence set, follows target requirements, needs a deliberate Structure, and must survive source verification and review.

Original research manuscripts

An expanded IMRAD model comprises title, authors, abstract, registration where applicable, keywords, Introduction, Methods, Results, Discussion, references, acknowledgments, disclosures, and visuals.^4,22 Target instructions determine the exact form.

The Introduction should establish the problem, current knowledge, gap, and objective. It is not a complete history of the field. The Methods should let readers understand what was done and how the analysis addressed the question. The Results should report the planned sequence without interpretation or selective emphasis. The Discussion should synthesize the principal findings, compare them with prior work, explain plausible meaning, acknowledge strengths and limitations, and end with conclusions proportionate to the evidence.

For primary trial manuscripts, the protocol, amendments, SAP, validated outputs, CSR, registry record, and reporting guidance form the central dossier. The manuscript is not a shortened CSR. It is a reader-oriented report of the research question and findings, with enough methodological transparency for evaluation.

For secondary or post-hoc manuscripts, analysis status is central. The Structure should state whether the question and analysis were prespecified, and the prose should not make exploratory findings sound confirmatory. Multiplicity, subgroup size, and relation to the primary publication require explicit handling.

For observational and real-world studies, data origin, cohort construction, confounding, missingness, and design-specific bias need more space. Causal language should match the method.³⁵

Review articles

A narrative review organizes a field for a defined reader. It should have a clear concept, balanced source strategy, and explicit organizing logic. The writer must avoid collecting papers without synthesis or treating personal familiarity as comprehensive evidence.²¹

A systematic review is a research report. Its search, screening, appraisal, and synthesis are methods, not backstage activity.^6,22,26 A meta-analysis adds statistical analysis and requires suitable expertise. Titles and abstracts should accurately identify the method.

Review conclusions are particularly vulnerable to scope drift. A review of a limited set of procedures cannot automatically support recommendations about all procedures. Keep the conclusion inside the eligibility, included evidence, and quality limits.

Case reports, commentaries, and letters

Case reports communicate unusual presentations, diagnostic or treatment lessons, adverse events, or observations that may generate hypotheses. They require careful patient privacy, consent as applicable to the context, and avoidance of generalizing from one case.

Editorials and commentaries are argumentative forms, but they still require transparent evidence and fair representation of opposing views. Letters to the editor may critique, clarify, or extend published work. Their brevity increases the need for a narrow point and precise support.

Abstracts

An abstract must stand alone while remaining faithful to the full work. Target headings and length control the form. Include enough method to interpret the result, report the most important estimate and uncertainty, and avoid conclusions that become stronger merely because qualifications were cut for space.

Congress abstracts may precede the final manuscript and sometimes precede complete analysis. State the data cutoff and avoid presenting provisional findings as final. When the full paper later differs, preserve the explanation.

Posters and oral presentations

Posters and talks are visual and time-bound. They need a single clear communication objective, readable hierarchy, restrained text, legible figures, and a narrative that can be understood at different depths. A poster reader may spend seconds scanning before deciding whether to engage.^10,32

Do not fill the canvas with the manuscript. Select the essential context, methods, results, and conclusion. Use visuals for comparison and flow. Keep citations sufficient for inspection without overwhelming the layout. For oral presentations, align the spoken script with the visual rather than reading dense slides aloud.

Plain-language publication summaries

A plain-language summary transforms a publication for non-specialists. It should explain why the work was done, what happened, what was found, what the findings do and do not mean, and where more information is available. Replace or explain jargon, make denominators and risk concrete, and preserve uncertainty.

Plain language is not childish language. It requires respect, health literacy, involvement of intended readers, concise structure, cultural relevance, visual support, and user testing.^36,42 Readability formulas can signal complexity but cannot prove understanding.⁷

Medical information and response documents

Medical-information responses answer unsolicited or defined questions using approved, balanced evidence. The document should restate the question, define scope, synthesize relevant evidence, include material limitations and safety context, and remain consistent with current approved content and references.

Because these documents are reused, source currency and version control are critical. A response that was accurate when approved can become outdated as evidence, labeling, or policy changes.

Scientific platforms and core narratives

A scientific platform organizes the core evidence, terminology, claims, and supporting references around a product, disease, or program. Its value lies in consistency and traceability across materials. It should distinguish evidence from interpretation, approved claims from hypotheses, and current content from superseded versions.

The platform does not remove the need to tailor. A field slide, publication, patient resource, and advisory-board brief use different language and detail. Reuse the evidence and approved meaning, not the exact sentence regardless of audience.

Advisory boards

Medical writers may support advisory boards through preparation, questionnaires, pre-reading, presentations, meeting capture, executive summaries, and reports.³ Begin with the meeting objective and decisions sought. Distinguish participant statements from validated evidence and from sponsor interpretation. Attribute views appropriately and protect confidentiality.

A report should synthesize themes without manufacturing consensus. Minority or divergent views may be material. Review the report according to the agreed participant and sponsor process.

Continuing medical education and training

Educational content begins with a need or practice gap, learning objectives, audience, and instructional format. Needs assessment, desired outcomes, format choice, and evaluation should be connected through an explicit instructional-design framework.⁴³ Content should be evidence-based, balanced, appropriate to professional independence requirements, and designed to help the learner do something—not merely receive information.

Training decks may have more explicit performance objectives but still need claim substantiation, audience fit, clear visuals, and controlled review. Promotional boundaries and applicable policies must be established in the brief.

Derivatives inherit claims, not automatic approval

A single evidence base may produce a manuscript, abstract, poster, slide deck, summary, and response document. Reuse improves consistency, but each transformation changes context. Shorter forms may drop qualifications. Visuals may imply stronger comparisons. Plain language may simplify uncertainty. Every derivative needs its own check against the source and its audience.

Use a parent-child register:

Open the complete publications and medical-communications playbooks ↗

Chapter 13· 8 min read

Regulatory, Safety, Devices, Market Access & Public-Facing Work

Regulatory writing applies the same fundamentals under a denser network of dependencies. A protocol affects conduct, analysis, consent, registration, and later reporting. A study report must reconcile planned and actual work. An investigator's brochure integrates changing knowledge across studies. Submission summaries interpret large bodies of evidence. Safety documents recur as the product lifecycle evolves.^1,28

The writer needs document craft and lifecycle awareness.

Regulatory documents are a connected system

Written communication runs throughout drug development, including nonclinical records, clinical programs, authority interactions, protocols, reports, investigator information, and Common Technical Document components.^1,28 Each document consumes and produces information for others.

Build a relationship map:

• Development strategy shapes studies.
• Protocols define intended conduct.
• SAPs define planned analyses.
• Study conduct generates data, deviations, and narratives.
• CSRs integrate methods, conduct, analyses, results, and interpretation.
• Investigator's brochures integrate available knowledge for investigators.
• Submission overviews and summaries synthesize across programs.
• Labeling, risk, and safety documents evolve with evidence and authority decisions.

Consistency across the system is not word-for-word identity. It is agreement on facts, terminology, study identity, data cutoff, conclusions, and the reason for any difference.

Protocols

A protocol should make the study rationale, objectives, design, population, interventions, outcomes, safety considerations, data handling, analysis approach, ethics, and operational plan clear enough for conduct and review. Its future-facing language must be consistent: what will be done, under what conditions, by whom, and how deviations or amendments are controlled.

The writer should connect objectives, endpoints, assessments, timing, and analysis. An objective without an endpoint or an endpoint without an analysis path creates downstream ambiguity. Eligibility criteria, withdrawal rules, safety monitoring, and data-management descriptions need operational clarity.

Templates help, but protocol quality depends on cross-functional input. Clinical, statistical, operational, safety, regulatory, data-management, and patient perspectives may all matter. Amendments should preserve a clear history and trigger impact assessment across consent, registration, SAP, and related materials.^19,28

Clinical study reports

A CSR explains what happened in relation to what was planned. The final protocol, amendments, SAP, validated outputs, deviations, exposure, disposition, efficacy, safety, and patient narratives form a large source system.

The writer must reconcile planned methods with actual conduct, map every objective and endpoint to results, distinguish analysis populations, and maintain consistency from synopsis through body and appendices. Safety narratives require source comparison, clinical coherence, and privacy protection. Modern CSR practice also requires structure, readability, completeness, and transparency, while the live project remains governed by its current controlling materials.^12,44

Planning, teamwork, and prototyping matter because a CSR can be too large to repair late. Agree tables, shells, key messages, and section responsibilities before all outputs arrive. Draft stable methods from controlled sources while leaving results placeholders precise.

Investigator's brochures and integrated summaries

An investigator's brochure integrates nonclinical and clinical information relevant to the safe and informed conduct of trials. It is a living document. The writer must manage new evidence, product identity, safety information, risk communication, and consistency across sections and versions.

CTD overviews and summaries are more than compression. They integrate evidence, explain development logic, and support evaluative conclusions. Start from strategy, select suitable expertise, and avoid summaries that merely repeat lower-level documents.^1,28 Integration means comparing studies, explaining patterns and differences, and relating evidence to the decision the document supports.

Safety and pharmacovigilance writing

Safety writing sits at the intersection of recurring documents, case information, aggregate analysis, risk management, and authority expectations. The writer needs controlled terminology, source-system knowledge, privacy awareness, and close collaboration with safety scientists and physicians.⁴⁵

At case level, narratives should tell a clinically coherent story from verified data while including only necessary personal detail. At aggregate level, documents must reconcile cases, exposure, signals, known risks, new information, and conclusions over a defined reporting period. Recurrent cycles make versioning and change detection essential. Health-authority feedback on aggregate safety reports repeatedly identifies missing information, weak assessment, internal inconsistency, and insufficient explanation as material quality problems.⁴⁵

Do not soften uncertainty into reassurance or list events without interpretation. Do not infer causality beyond the assessment framework. Escalate discrepancies in case data, coding, dates, seriousness, outcome, or listedness to the accountable safety function.

Medical devices and IVDs

Medical devices and IVDs vary widely in intended purpose, risk, evidence needs, and lifecycle documentation. Their written evidence often forms linked suites rather than one report: plans, clinical or performance evaluations, investigation reports, post-market follow-up, periodic reporting, and public summaries.²³

Evidence may include literature, equivalence or similarity considerations, clinical investigations, analytical and clinical performance, usability, risk management, post-market surveillance, complaints, vigilance, and state-of-the-art context. The writer needs to understand the device, intended purpose, users, patient population, claims, lifecycle stage, and applicable framework.

Living evaluations require a documented literature protocol, reproducible searches, critical appraisal, and update logic. Record what evidence is new, what conclusions change, and what linked documents are affected.²³ Device terminology must remain consistent with intended-use and labeling language. Public summaries add plain-language and accessibility demands without removing safety or performance qualifications.

HEOR and market access

HEOR and HTA writing supports decisions about value, resource use, outcomes, and access. The audience may include payers, assessment bodies, clinicians, and policymakers. Evidence can combine clinical studies, real-world data, epidemiology, utilities, costs, models, and comparative effectiveness.^35,46

Define the decision problem, perspective, population, comparator, time horizon, outcomes, and jurisdictional context. Model reports need traceability from narrative claims to inputs, assumptions, equations, scenarios, and sensitivity analyses. A result should not be presented without the perspective and assumptions that produced it.

Avoid making a model appear empirical merely because it produces precise numbers. Explain parameter and structural uncertainty, scenario dependence, assumptions, and limitations.⁴⁶ Market-access submissions are target-specific; current local methods and templates must govern live work.

Public disclosure and transparency

Clinical-trial transparency includes registration, results posting, public release of regulatory documents, and protection of personal and commercially confidential information.^15,47 Transparency work requires consistency across records and careful redaction or anonymization.

Disclosure is not an afterthought to final publication. Registry information, protocol information, results, CSR content, publications, and public summaries can be compared. Differences in endpoints, populations, dates, or conclusions require explanation.

The writer should include only necessary patient information, preserve scientific utility where possible, and use the applicable disclosure process. Privacy review is not simply deleting obvious identifiers.

Patient information and informed consent

Patient-facing material must be understandable, accurate, relevant, respectful, and sufficient for its purpose. Keep information simple and concise where possible, involve intended readers, preserve necessary content, and test whether real users can understand and use it.^7,16,36

For informed consent, simplicity must not remove information necessary for an informed decision. Explain purpose, procedures, risks, potential benefits, alternatives, privacy, voluntariness, and contacts according to the live governing context. Organize around the questions a participant is likely to ask.

For trial-result summaries, explain why the study was done, who took part, what happened, what was found, adverse reactions or safety findings, limitations, and where further information can be found. Patient involvement can improve relevance, tone, and comprehensibility, while AI-assisted drafting still requires human verification and accountability.^36,48 Avoid implying individual medical advice.

Plain language, translation, and accessibility

Plain language depends on audience, context, and testing. Use familiar words, define necessary technical terms, shorten and organize information, make actions explicit, and use headings and lists to reduce memory load. Explain absolute numbers and denominators. Avoid ambiguous abbreviations.

Translation is not literal substitution. It is a transcultural activity in which word-for-word rendering can be inadequate or dangerous.⁴⁹ Work with qualified translators or medical linguists, provide context and terminology, preserve source meaning, and perform independent linguistic and subject-matter review. Cultural relevance, examples, images, and risk communication may need adaptation.

Accessibility includes visual, cognitive, language, and technical access. Use legible type, contrast, descriptive links, alt text, meaningful reading order, and alternatives to color-only meaning. Visual aids should clarify rather than decorate and should be tested with intended users where feasible.^32,36

Know the expertise gate

Specialist work needs specialist review. Escalate when the document depends on:

• Statistical method or interpretation beyond the writer's competence
• Regulatory or legal interpretation
• Safety causality, signal, or case assessment
• Health-economic model design
• Device engineering or performance interpretation
• Patient privacy or anonymization
• Translation and cultural adaptation
• Accessibility testing
• Clinical recommendations

The writer's role is not diminished by escalation. It is fulfilled by knowing where accountable expertise belongs.

Open the regulated and specialist document playbooks ↗

Chapter 14· 4 min read

The Whole Workflow in One Piece

The craft of medical writing can feel fragmented because the profession produces so many documents and works among so many specialists. Beneath those differences is a stable pattern: every credible document creates a chain from purpose to evidence to form to wording to review.^1,2,18

The fourteen principles

1. Write for a use and reader. A document succeeds when the intended reader can use accurate information for the intended purpose.¹
2. Let evidence set the boundary. Desired messages can guide questions; they cannot manufacture support.⁴
3. Build a dossier, not a pile. Know what every source is, which version it represents, and what role it may play.^5,18
4. Match search rigor to the claim. Do not call a convenience set systematic, and do not make a targeted search carry more authority than it has.^6,21
5. Know what governs the output. Classify and apply current requirements, reporting guidance, target instructions, procedures, and templates.²⁸
6. Approve the shape before the prose. Give every section a purpose, evidence route, constraint set, and completion test.³¹
7. Draft one controlled section at a time. Build paragraphs from claim and evidence, not from momentum.⁹
8. Preserve numerical identity. Population, endpoint, time point, denominator, unit, estimate, precision, and analysis status travel with the number.^12,33
9. Make material claims inspectable. Citation formatting is not claim verification; support the sentence actually written.²⁴
10. Use tools without outsourcing accountability. Bound AI tasks, protect data, require provenance, and keep human gates.^37,39
11. Design review as decision-making. Give every round a question, scope, owner, and reconciliation path.^18,19
12. Separate creation from quality control. Fresh, focused checks catch what familiarity hides.^20,41
13. Treat derivatives as new communications. Reuse the evidence trail and recheck the transformed claim.⁴
14. Deliver the artifact and its appropriate receipts. The work is complete when the final form is verified and defensible.^18,20

The production loop

Brief
  ↓
Document identity + purpose + audience
  ↓
Dossier roles + source versions + gaps
  ↓
Evidence questions + search + appraisal
  ↓
Governing stack + template + constraints
  ↓
Structure + source routing + planned claims
  ↓
Section drafting + data + visuals
  ↓
Claim traceability + internal consistency
  ↓
Human review + reconciliation
  ↓
QC + export validation
  ↓
Deliverable + appropriate evidence trail

The arrows are not one-way. New evidence can revise the Structure. Statistical review can change a claim. A target can change format. A reviewer can expose a missing source. The discipline is to move the affected decision back to its owner rather than patching the final sentence alone.

The failure chain

A hidden assumption in the brief becomes an unsupported key message. The message shapes a selective source search. The selected sources shape an outline that omits contradictory evidence. The outline produces fluent prose. The prose is copied into the abstract, poster, and plain-language summary. Review focuses on style because the evidence trail is invisible. The final claim now appears in four polished places.

The chain is broken by visible contracts: brief, dossier readback, search method, governing register, accepted Structure, claim support, review decisions, and QC. These are not bureaucracy. They are the places where a team can still correct itself.

Your durable advantage is judgment with receipts

Tools can retrieve, format, classify, compare, draft, and check. Templates can encode structure. Guidelines can list required items. None of them decides, by itself, which evidence is material, which uncertainty changes interpretation, which conflict deserves escalation, or which sentence is both useful and fair.

The medical writer's advantage is disciplined judgment made inspectable. You understand the reader, interrogate the evidence, shape the argument, coordinate expertise, and leave a trail strong enough for somebody else to see why the document says what it says.

Monday morning · the first hour of the next assignment

1. Write one sentence naming the document, reader, and intended outcome.
2. List the governing sources you need to confirm.
3. Inventory the files and references you already have by role.
4. Mark missing, conflicting, and unverified material.
5. Write the proposed messages as questions the evidence must answer.
6. Identify authors, reviewers, specialists, and decision owners.
7. Draft the top-level Structure with one instruction under every heading.
8. Send the project frame and dossier readback for correction.

Do not begin by polishing the introduction. Begin by making the job true.