Accuracy and stance
This leaf answers how a medical writer keeps what a document says true to its evidence and says it with the stance the writer means: checking details, claiming as firmly as the evidence allows and no more, and choosing words whose attitude towards a contested act, a severe outcome or a hostile later reader is the one intended. Saying something specific covers the individual words that soften or overstate a claim, references and attribution checking that a cited work exists and supports the claim, describing people accurately the terms for participants and professionals, and communicating risk and uncertainty how a risk is framed for the reader who has to weigh it.
Separate facts from opinions, values and assumptions
Robert B. Taylor's Medical Writing: A Guide for Clinicians, Educators, and Researchers sets the terms: medical writing records what is known and makes evidence available for examination. Distinguish observations, interpretations, and opinions. The strength of a conclusion should follow the evidence, and the wording should let the reader understand what the evidence actually supports.
Communication is affected by the facts, opinions, values and assumptions that people bring to an interaction, and Cynthia Baur's 2013 McGovern Award lecture to the American Medical Writers Association sorts those four in a way that assigns the communicator a job. Facts are verifiable, testable information; opinions are aggregates of beliefs and attitudes; and people's values and assumptions influence what they consider to be a fact and what they consider an opinion. Two corollaries follow, and writers get them wrong in opposite directions. Facts are not the exclusive domain of scientists — lay people can also generate and marshal facts. And scientists' own values, opinions and assumptions can affect how they construct and present information. The role Baur assigns to professional communicators, including medical writers, is to help distinguish among the four, and then to choose how to present health, medical and scientific information or to help others make that choice.
Check details throughout production
An inaccurate number or ambiguous instruction can defeat an otherwise useful document. Taylor uses published correction notices to illustrate how drug concentrations and similar-looking scientific terms can be confused. Check the original evidence and the intended meaning during drafting, editing, and proof review. These historical examples demonstrate error mechanisms, not treatment instructions.
Precision includes supplying the detail needed to interpret a claim or act on an instruction. A vague category, broad range, or missing duration can leave the actual meaning unresolved. Verify both the value and its context, rather than relying on grammatical polish as a sign of accuracy.
Some of these checks belong to the editor rather than to the author, and Krithika Muthukumaran's September 2018 article for EMWA's journal states three of them together. Editors are responsible for checking the validity of any data presented, for confirming that the data are represented accurately and neither hyped nor understated, and for seeing that the facts are adequately referenced. Her exercises show the two arithmetic failures that survive a content read: a results sentence reporting the highest percentage in one group and the next in another and then naming a third group with no figure at all, and a percentage that does not follow from its own numerator and denominator — ten of 200 participants reported as 10 per cent. The issue was admitted from an existing Markdown conversion without content repair, and its known limitations are broken ligatures, mangled multi-column tables, interleaved page headers and footers, and figure text represented as marked text.
One further check is about stance rather than accuracy. Any point of view or critique in the text must be well researched and written responsibly rather than turned into a personal attack, which is a judgement the editor is better placed to make than the author.
Match confidence to support
Overstatement turns limited observations into sweeping recommendations or labels a modest finding as a breakthrough. Claims about all clinicians, all patients, or future certainty need especially careful scrutiny. Equally, unnecessary hedging can obscure a well-supported result. Review emphatic and tentative wording in context; uncertainty words are useful when uncertainty is real.
Reduce stacked hedges one qualification at a time
Stacked hedges are a specific and recognisable form of the second failure. Phillip Leventhal and Alistair Reeves, writing on abstracts for EMWA's journal in 2012, name multiple hedges — several qualifications piled on one claim — and Leventhal returns to them in a 2017 checklist for the same journal. Both accounts trace them to the same two pressures, which are worth separating because only one of them is about evidence: the scientific training that tells writers to say nothing definite, leaving the door open to later evidence that contradicts a conclusion, and a legal department or company compliance officers asking for nothing definite to be said for legal reasons. Their example carries four in a row: these preliminary results suggest the possibility that the drug might be effective at reducing the incidence of the disease in some populations, although further studies are needed to confirm this. The 2017 checklist counts the four as preliminary, possibility, might and in some populations, and observes that a sentence hedged four times ends up concluding nothing.
Their suggested repair — these results suggest that the drug will reduce the incidence of the disease — shows why removing hedges is a claim decision and not an editing one. It drops the preliminary status of the results, converts a possibility into a prediction, removes the restriction to some populations, and deletes the acknowledgement that confirmation is outstanding. Where the evidence supports the shorter statement, the original was padding; where it does not, the repair has changed what the paper asserts. Reduce the number of qualifications by deciding which of them the evidence actually requires, one at a time, rather than by treating the pile as a stylistic tic.
Leventhal's rule is that hedges are fine but you only need one, and Christine Møller, an assistant editor of the pathology journal APMIS writing in the same 2017 issue, gives the same rule — one hedge or indefinite word is enough — with two examples of her own: a rather somewhat unclear status is simply an unclear status, and one possible explanation … could be keeps the could be and drops the one possible. Leventhal adds a second target under the same heading: avoid ending a text by saying that further studies are needed, since further studies are always needed — if they were not, science would stop — so the sentence is a very weak way to close. Where it has to be said, say which study, as in but this needs to be validated in a large randomised controlled trial.
Choose between near-synonyms by what the evidence supports
A hedging choice that individual writers hold strong views about may not be a shared convention at all, and an exchange printed in EMWA's journal Medical Writing in December 2012 is a rare direct measurement of that. Alistair Kidd, a medical editor based in Umeå in Sweden, argued that "seems" and "appears" differ — "seems" describing a perceived state and so suggesting subjectivity, "appears" describing an observable state and so suggesting objectivity — and that "seems" should be avoided altogether in a scientific context: it can occasionally creep into the discussion of a paper, but it has no place in the introduction, methods or results, and certainly not in the abstract. His own prompt was noticing that a colleague with a background in English and history saw no difference and had never thought about it.
The journal then polled experienced writer, editor and translator colleagues in Europe and the USA and printed the answers grouped by position: no difference; there is a difference but it is too subtle for most; there is a difference and it should be preserved; there is a difference but it is a matter of usage; and not sure, never thought about it. The range of reasoning recorded is what makes the poll useful. Both words are hedging devices with the same meaning in the example given. It is worth cautioning professional writers against expecting much awareness of linguistic subtlety in other scientists, most of whom will not intentionally use one over the other to indicate a different degree of certainty. Whichever verb is used, the main thing clear to the reader is that some element of uncertainty or speculation is involved. "Seems" at best suggests the author is unsure and at worst that they are hedging their bets. "Seems" is in a lower register. And, from two respondents, the pair is one of English's many doublets with a Latin- or French-derived term beside an Anglo-Saxon or Germanic one, where the difference is register rather than semantics — "appear" from the French, "seem" from a Germanic root — so the choice is a matter of usage and collocation. One respondent with more than twenty years of teaching research writing reported that the distinction had never once surfaced, probably because of infrequent usage, while agreeing with it and adding that "seems" in the present tense could be excluded from most sections of a journal article except for an inference in the discussion.
Kidd's reply keeps the right amount of weight on it: several of those who agreed had scientific backgrounds and several who did not had linguistic ones, but the numbers are small and no serious conclusions can be drawn — and, he adds, many languages have only one word covering both, Swedish att verka being his example, so any subtlety is more likely to be lost when a writer is working from such a language. Two cautions on using this material. The response table is badly damaged in conversion: the group headings are interleaved into the name and background cells, so which respondent sits under which position cannot be reconstructed reliably, and the reply's own tally of five respondents with a linguistic background disagreeing does not reconcile with the backgrounds as printed. Take the range of positions as the finding and treat the individual attributions as unreliable. And take the finding itself for what it is — a small informal poll of colleagues at one association, reported as such by the people who ran it. What it establishes is not which verb is right but that a house rule on this pair is a house rule rather than a convention the reader can be assumed to share.
A hedge that hides who is doing the hedging fails the same test from the other side. An editor wrote to the AMWA Journal's usage column in 2008 about a paragraph reviewing N-acetylcysteine for contrast-induced nephropathy which closed: use of NAC, although not recommended, is suggested for patients at very high risk of the condition. She read the authors as using recommend for approval and suggest for possibility, having had to read the sentence three times before she saw it, and asked whether so fine a distinction could do anything but create confusion. Edie Schwager, a freelance writer and medical editor who answered usage questions in the Journal's Dear Edie column, gave an answer blunter than the question: the authors are weaseling, and they have failed in the aim of every medical article, which is to enlighten. The specific defect she names is that the sentence never says who recommended and who suggested — the authors, the review they cite, any of the authors cited within that review, or other anonymous sources — so the reader cannot locate the owner of either position. Where two near-synonyms are being made to carry a difference in confidence, name the party holding each position rather than relying on the pair to carry it.
A hedge can be right and still be the wrong word for what is known, and a 2010 usage question about consumer health brochures turns on exactly that. The sentence in question was "Eating healthy may lower your risk of heart disease", and the editor asked whether something can possibly lower risk or whether the sentence should simply assert that it does. Edie Schwager's answer separates the two issues. May properly conveys possibility, so the sentence is correct as it stands; but where the effect is established, can is the word that says so — since we know this to be true, eating healthily can lower your risk of heart disease. Choosing between them is a claim about the state of the evidence rather than a matter of tone. She also corrects the adverb: it is eating healthily, not eating healthy.
The same column flags a phrase that has since won. Diagnosed with, applied to a person, was becoming more and more commonplace in 2010 and she records it as an example of usage change while making her own preference clear — she would consign the phrase to oblivion and prefer a construction naming the diagnosis itself. A writer who dislikes a construction on those grounds should be able to say whether the objection is to accuracy or to idiom, because only the first survives the change.
Cut the qualifiers and intensifiers that weaken a claim
Three families of word are worth separating when a draft is being checked for stance, because they fail in different directions. Møller's advice on hedging is the blunt form: beware of using too many hedging words — may, might, can, could, possibly — and instead decide what you want to say and say it. Imprecise words such as quite, rather, fairly, relatively and somewhat are usually redundant and best eliminated, her example being a trend described as fairly similar to trends in other countries. And intensifiers such as very, highly and extremely paradoxically weaken the message, because absolutely essential says no more than essential, very true no more than true and highly central no more than central.
Barbara Every, a biomedical editor writing in the same issue, adds the reason qualifiers are a stance problem rather than a wordiness one. Qualifiers are imprecise and ambiguous in scientific writing, add little meaning, and lose their effectiveness if used too often, so they should be used sparingly if at all, or replaced by a stronger adjective — very hard by difficult, very important by crucial. The case she treats as a rule rather than a preference is the incomparable word: absolute, final, infinite, scant, total and unique represent an extreme state, so they cannot be quantified — very unique, somewhat final — or compared — more absolute, less scant, most total, least infinite — because the result is illogical. Removing those uses shortens the sentence and strengthens it without losing anything. The columns of this conversion are interleaved, so the qualifier argument and the redundancy argument are spliced into one another in the admitted text and are separated here by their subject matter.
Calibrate a generalisation for a lay reader
Generalising for a lay reader is where that match is hardest to hold, because leaving out the detail can change what the reader concludes rather than only how much they know. Joselita T. Salita's 2015 survey in EMWA's journal of why scientists find lay writing difficult supplies examples running in both directions: a discovery in medicine may be misinterpreted to suggest that a cure will soon be available, or a new virus may be hyped into unnecessary panic — and on the other side, some hold that scientists are partly to blame for public confusion and indifference about climate change, on the argument that using too many levels of likelihood terms such as likely, unlikely and most likely makes them sound unsure of themselves. Either failure makes the dissemination of scientific information counterproductive, which is the reason the calibration matters at all in a public-facing document.
Notice the sentence adverb that inserts an opinion
An adverb at the front of a sentence can insert an opinion without announcing one, and Laurie Endicott Thomas's 2013 column in the AMWA Journal is a compact account of the mechanism. The grammar underneath it is the distinction between an adjunct and a disjunct. An adjunct is an adverb or adverbial phrase attached to a verb, typically expressing time, place, frequency, degree or manner; a disjunct is an adverb or adverbial phrase loosely connected to a clause or sentence that conveys the writer's comment on its content, truth or manner. Suddenly, it began to rain modifies the verb; Fortunately, it began to rain states the writer's view of the weather. Editors and readers who can recognise the second kind are better equipped to sort fact from opinion, which is the reason the distinction is worth teaching at all. The issue was admitted from an existing Markdown conversion without content repair, in which figure labels can be inlined into prose, cover and contents columns can interleave, and reference URLs can be malformed or truncated.
The device earns its place by compression. A clumsy expletive construction collapses into a single sentence adverb — I feel that it is unfortunate that potato chips are fattening becomes Unfortunately, potato chips are fattening — and the same substitution works across a range of opinions: it is obvious to me that becomes obviously, the evidence suggests to me that becomes evidently, it would be ideal for me if becomes ideally, it is surprising to me that becomes surprisingly, I am certain that becomes certainly, I am thankful that becomes thankfully, it seems curious to me that becomes curiously, and I am hopeful that becomes hopefully. What is lost in the exchange is the personal pronoun, and with it the visible signal that an opinion is being expressed — which is exactly what makes the construction worth watching for in someone else's draft.
Her treatment of the contested case is a piece of reasoning rather than a ruling. A dangling modifier is one intended to modify something left out of the sentence, as in While walking to school today, my book fell into the mud, repaired by supplying the participle's subject and an auxiliary verb. Hopefully used as a sentence adverb does not dangle, because it is not modifying anything missing; it expresses the writer's feeling about what the sentence says. Her observation is that those who object to it do not seem to object to obviously, curiously, evidently or thankfully, and her answer to the suggested substitute one hopes that is that the sentence adverb really means I hope that — so a writer expressing an opinion may as well take ownership of it. Some adverbs work either way depending on context: in it is better to travel hopefully than to arrive the word is an adjunct describing the traveller, while in hopefully, they will arrive before nightfall it is a disjunct describing the speaker.
Punctuation and word order follow the ordinary rules rather than special ones: a sentence adverb and an adjunctive adverb are both set off with a comma when they open a clause or a sentence and neither is set off in the middle of one, and either can sit between the auxiliary and the stem of the verb it relates to. The closing caution is the one to apply on review. Writing evidently, certainly or obviously asserts that the evidence impresses you, that you are certain, or that something is obvious to you; writing fortunately or ideally asserts your own view of what is fortunate or would be ideal. What seems obvious to you may not be obvious to other people and may not even be true, and what would be good fortune for you may be misfortune for someone else.
Choose the wording that carries the severity you mean
A qualifier on a severe outcome is a stance decision disguised as a style question. Asked in the AMWA Journal's usage column in 2009 whether conventions exist for the common formula in which a nutrient deficiency or excess leads to a list of conditions "and even death" — whether the "even" is needed, and whether "and" or "or" matters — Edie Schwager's answer starts from the reader. She has often seen "or, rarely [or sometimes, or frequently], death", which takes the threat of imminent doom off the table by supplying a frequency; for a professional audience where the other conditions need no softening she would simply write "or even death". The conjunction is not neutral: "and even death" implies that death might, or almost certainly would, follow the conditions listed before it, where "or" presents it as an alternative outcome. Adding a frequency word is the repair that adds information rather than removing it.
The motto the column ran under states the disposition behind that answer: in medical writing there is no danger in being too precise, only in being imprecise. Read it as a corrective to the instinct to soften rather than as a licence for detail nobody needs. The answers printed beneath it all buy their precision by adding a qualifier the reader can use, not by adding words.
Treat a word that overstates as an accuracy problem, not a marketing one
The words that mislead most reliably are not technical, and the discipline of noticing them transfers from advertising to medical text. Arnold Melnick's 2008 column in the AMWA Journal takes the dictionary definition of a buzzword — an important-sounding word of little meaning, used chiefly to impress the layman — and works through five that a reader meets weekly. "Free" rarely means without charge or obligation: his worked example is a book club mailing carrying the word three times in half-inch letters alongside "no obligation", whose fourth book costs $11.99 and whose obligation is to buy it, and he contrasts it with a genuinely free thing, a team of athletes giving filled backpacks to children in a poor neighbourhood. "Buy one, get one free" is the same structure with the condition stated: the item is not free, it is attached to a condition.
The other four fail in more interesting ways. "Guaranteed" is unfalsifiable without an answer to by whom and under what circumstances — a retirement home carrying a national hotel chain's name guaranteeing return of an entrance fee backed only by the single-site operator rather than the chain, another returning the money only once a new applicant pays for the same apartment, a one-man repair business of a few years' standing offering a lifetime guarantee. "Independent" invites the question independent of what, since almost everything is dependent on government rules, state laws or organisational bureaucracy, and whether it makes any difference. "Nonprofit" carries an assumption of better operation at lower cost that nothing guarantees; what it means in law is only that no individual shares the bottom line, so a positive bottom line is booked as an excess of income over expenses while managers and officers may still draw large salaries. And "no obligation" is the claim to examine hardest, because the obligation is usually elsewhere in the same document.
Melnick is careful about what he is alleging and the care is part of the method: advertisers are attempting to create an attractive market with suggestive wording rather than acting fraudulently, and not every use of these terms is misleading. His conclusion is where it becomes a medical writing question. These are not words medical communicators typically use, and they illustrate the standard the whole vocabulary has to meet — never use words casually or carelessly, so that nothing you write can be suspected of being deliberately misleading. The test to carry across is his, applied to a word rather than to a claim: ask what the word would have to mean to be true, and whether the document makes that meaning available to the reader.
Test a contested term against the act it names
Where two names compete for the same act, the way to choose between them is to describe the act and then ask which name is true of it. Nancy Buermeyer, a public policy lobbyist for an organisation that supports and advocates for choice and care at the end of life, argued at the same 2008 session for death with dignity in preference to physician-assisted suicide, which she holds to be biased, inaccurate and inappropriate. The Oregon statute she cites allows a terminally ill, mentally competent patient to self-administer a lethal dose of medication; self-administration on those terms represents empowerment, independence and self-control, and not the despair, futility, self-destructiveness and hopelessness that suicide names. She reports that patients and their families appreciate the less value-laden phrase, and reads surveys of the level of advocacy and support for death with dignity as showing that society is more accepting of the behaviour when it is described that way. Asked from the floor how much use the provision gets, she put it at approximately 36 people a year in Oregon, adding that simply having the medicine on hand is a great comfort to many more terminally ill patients.
Two things are worth separating out of that. The method transfers: set out what the term denotes, then test each of its connotations against that description, which is the same move Knatterud makes on failure and on the elderly and can be made on a term neither of them considered. The conclusion does not. Buermeyer was speaking as an advocate for one side of a contested question, and what the record holds is a short summary written up by an attendee for the association journal's conference coverage, so her account of the statute and her figures are second-hand and her framing is a party to the argument rather than a resolution of it.
Who resists a change of wording, and why, is worth establishing before proposing one. A study of the rhetoric of race-based medicine used by scientists and clinicians found the geneticists amenable to geographic rather than race-based language, while physicians were adamantly against the intervention, arguing that a colour-blind approach amounts to whitewashing medicine and places the blame for health inequities on people and places rather than on unalterable genetic differences. Those are not the same kind of objection, and only one of them is about accuracy. The writer's stake in the question is stated in the account that reports the study: writers and editors have both the ability and the responsibility to select the language used in the biomedical literature, and that language shapes conversations and influences the wider culture. That account is an AMWA member's 2013 description of her doctoral reading in rhetoric, published in the same issue of the association's journal as Thomas's column on sentence adverbs and subject to the same conversion limitations.
Choose a tone the reader can trust
Tone is hard to define, but the working definition generally accepted is that it reflects the author's attitude to the subject — reflects, not reveals, since an author can read as fully engaged, even enthusiastic, while being bored by the subject and in need of a holiday. That leaves the question of which attitude medical writing should reflect, and Alistair Kidd's answer in Medical Writing in 2013 is a scientific one: objective and impartial. A neutral, balanced tone makes the reader confident that the conclusions are not overly biased, which is a separate matter from — and no substitute for — bias in which data were selected for presentation at all.
The reader also has to feel treated as an equal rather than patronised, which is why a word like "obviously" has no place in scientific writing: it tells a reader who did not find the point obvious that they should have. Kidd's other levers on tone are the avoidance of strong language, humour and contractions, grammar good enough for the authors to be taken seriously, and control over how much abstraction and how many strings of nouns the prose carries.
Two stances break the effect completely, and a book review in the same issue catches both. Stephen Gilliver's review of a bioethics collection objects that its author revels in his own ignorance — confessing that he cannot read the human genome papers in Nature and Science, that he is clueless about what nanotechnology means, and that "we don't really know how to turn genomics on and off", which the reviewer reads as showing that he had not bothered trying to understand the subject. The second is partisanship: a tone the reviewer describes as bordering on the offensive, in a book whose national framing colours its account of research done elsewhere. Neither failure is about accuracy in the narrow sense, and both cost the author the reader's willingness to accept their judgement on anything else.
Write a sensitive document expecting an opponent to read it
Most writers assume that having the data puts them on the right side of the law, and a lawyer's account of the exposure is a useful corrective. Joanne Flitcroft, a lawyer who had worked in the pharmaceutical industry for seventeen years specialising in defending product safety issues and maintaining corporate reputations, set the position out in EMWA's journal in December 2017. The industry is watched closely by regulatory and governmental authorities, competitors, patients, the media and lawyers acting for potential plaintiffs, and rival company lawyers monitor all information coming from competitors — so a medical writer's documents are written into an adversarial reading environment whether or not anyone intends one. The issue was admitted from an existing Markdown conversion without content repair, and its known limitations are broken ligatures, mangled multi-column tables, interleaved page headers and footers, and figure text represented as marked text.
Two principal legal risks arise from inappropriately written documentation: damage to corporate reputation related to the handling of pharmacovigilance and regulatory issues, and product liability claims and litigation. It is not only email and social media that carry the risk; all clinical, regulatory and patient safety documentation written without regard to the legal implications can affect patients, corporate reputation and an organisation's ability to work effectively, and a failure to identify and mitigate those risks can force a company to divert resources from drug development into defending safety or clinical issues, alongside protracted discussions with regulators. Where a pharmacovigilance issue is litigated or investigated, the related documentation is likely to be thoroughly examined and could become public: authorities, plaintiffs and their lawyers may be granted permission to obtain any documentation or communication relevant to the investigation. An investigation necessarily focuses on documents dealing with sensitive information — knowledge, data, opinions, hypotheses, analyses and ideas — which may implicate an organisation's legal position or its reputation if disclosed, and pharmacovigilance and regulatory documentation is by its nature sensitive. Reputational damage in the life sciences can affect a company's success and its ability to attract investors, and the costs of investigation and litigation are diverted from the core business.
Discovery is the mechanism that makes this a writing problem rather than a filing one. It is the legal process of obtaining evidence by searching documentation or interviewing the authors of it, and the rules for disclosing documents as evidence are wide-ranging and liberal and vary by jurisdiction; discoverable documentation is not limited to pharmacovigilance or regulatory documents but may include emails, calendars and even text messages that pertain to the medical issues discussed in them. Mitigation is therefore a communication skill: clarity and method are critical, coupled with appropriate classification of documents — whether they are confidential or subject to legal privilege — and, equally important, with avoiding the concealment or censoring of sensitive pharmacovigilance information or limiting how much information is communicated. Flitcroft's framing of the underlying problem is that sharing medical and scientific information is not the same as communicating it: what an author says is not necessarily what others hear. Her four preliminary questions are the ones to answer before beginning any medical communication — what is it that you want to say; how do you want to say it; what does your audience need to know; and what exactly do you want the audience to do with the information — followed by reflection on the audience's different needs, potential cultural differences and perceptions, and on how the document may look from an internal corporate perspective and from an external regulatory or public one.
Six practical rules follow, and each one is a stance decision rather than a procedural one.
Choose the method of communication deliberately. Consider whether the medium you have chosen is the right way to document the issue at all, and write documents on sensitive subjects with the expectation that they may become public at some point.
State facts, not opinions, unless you have specifically been asked for an opinion, are qualified to give one, and it is the purpose of the document; avoid commenting on matters outside your area of expertise. Where opinions or conclusions have to be documented, identify the source of the opinion, the conclusion or the information received — because otherwise a third party can later argue that the opinions were yours and that you had verified the information.
Be accurate, clear and concise. Inadvertent errors of fact may be interpreted as incompetence. Do not speculate or embellish with adjectives and adverbs, which are by nature open to more than one interpretation and easily create ambiguity, and avoid sarcasm, irony, exaggeration and gratuitous or flippant language. Recipients may forward the document to others without your knowledge, and an audience may misunderstand and draw the wrong conclusion. Beware of abbreviations and technical vocabulary, and assume the reader has less specialised knowledge and a more limited grasp of the terminology than you do.
Keep the tone neutral. Avoid expressing strong feelings, which can overwhelm clear thinking; emotionally charged expressions may carry unintended weight or meaning and are open to misinterpretation. Avoid defensive and critical comments for the same reason.
Record only what the role requires. Do not record information that is unnecessary or merely nice to have, and understand how to store information so that it complies with document retention requirements.
Mark what a document does not yet know. Many documents go through more than one iteration; where one on a sensitive subject is incomplete, say that it is in draft form and subject to change, and where there is more than one draft, number them in the order written. Writing from limited information is perfectly acceptable — what is required is to indicate clearly that the document rests on incomplete information and that further work is needed.
The test she ends on is a single question to put to any sensitive document before it leaves your hands: how would I feel if this became public? Poorly drafted documents are open to misinterpretation, and the language used in any document can create the wrong impression when taken out of context.