A Clinical Team Needs an Evidence Brief by Tomorrow: A Practical Workflow

A request for an evidence brief "by tomorrow" is really a request for a defensible decision aid: a short document that states a focused question, shows how the evidence was found and selected, separates what the studies measured from what your team needs to know, and ends with a clear statement of what the evidence can and cannot support. The work is not to read everything. It is to build a traceable chain from a focused question to a bounded answer.

This article walks through that chain using one concrete example: a hospital-based clinical team asked whether a structured early-mobility program, compared with usual care, changes functional outcomes for adults hospitalized on general medical wards. The example is deliberately narrow. It lets us separate evidence level, applicability, indirectness, and limitations without drifting into individual treatment advice.

Most last-minute briefs begin in one of three states. Naming your state tells you which step to protect.

• A pile of PDFs and no question. Someone has collected papers, but no eligibility criteria exist. The risk is that the brief becomes a summary of whatever was downloaded.
• A question and no search record. The team agrees on the topic, but nobody can say which databases, terms, or filters produced the current reading list.
• A draft with claims that outrun the sources. The prose is written, but statements about benefit, harm, or generalizability are not tied to specific studies.

The early-mobility example usually starts in the second state. The team knows it cares about mobility on medical wards, but "early mobility" could mean a physical-therapy-led protocol, nurse-assisted walking, or a bed-exercise program. Those are different interventions with different comparators. Until that is settled, every search returns a mixture that cannot be synthesized cleanly.

The first artifact is a one-page question frame. Write the population, the intervention, the comparator, and the outcomes you will actually report, plus the settings you will accept. For the example:

• Population: adults hospitalized on general medical wards.
• Intervention: a structured early-mobility program delivered during the admission.
• Comparator: usual care or routine mobilization as described by the study.
• Outcomes: functional status or mobility measured during or shortly after admission, with adverse events reported separately.
• Settings: inpatient wards; note whether intensive-care or surgical populations are excluded.

This frame does two jobs. It defines eligibility before you search, which is the discipline the Cochrane Handbook recommends: plan searches from eligibility criteria rather than from whatever a database happens to return. And it gives you a place to record deviations later, when a promising paper turns out to study a slightly different population.

Keep the frame short enough to paste at the top of the brief. A reader should be able to see, in under a minute, what question the rest of the document answers.

The Cochrane Handbook's guidance on searching is a useful checklist for a one-day brief: derive terms from the eligibility criteria, use more than one source, combine free-text terms with controlled vocabulary, document the process, deduplicate, and screen in stages. Each of those maps to a concrete artifact.

Start with the concepts in your frame and list synonyms for each. For the mobility example, the intervention concept might include phrases around early mobilization, early ambulation, and structured mobility programs, while the population concept covers hospitalized adults and general medical wards. Controlled vocabulary helps here: NLM's Medical Subject Headings are a hierarchical vocabulary with preferred headings, entry terms, scope notes, and qualifiers, and they can be used to expand or focus a PubMed strategy. Checking the MeSH scope note for a candidate heading tells you whether the indexers used it the way your team means it.

PubMed supports field tags, phrase searching, filters, Advanced Search, saved searches, and a Search Details view that reveals how terms were translated and combined. That translation view is the fastest way to catch a phrase that was split into separate words or a filter that quietly removed the population you needed. Record the final string, the date run, and any filters applied.

If you want a scholarly search interface to run and refine the paper search itself, WisPaper search is the natural entry point; a public description of the same capability sits on the WisPaper Scholar Search page. Use whichever surface fits your team, but keep the same discipline: save the strategy, not just the results.

Trials often generate several papers: a primary outcome report, a secondary analysis, and a protocol. The Cochrane Handbook explicitly calls for linking multiple reports of one study. In practice, this means your screening sheet needs a study identifier column, not just a citation column, so that two papers from the same trial are counted once when you describe the body of evidence.

Staged screening is the difference between a brief and a book report. A workable sequence for one day:

1. Title and abstract screen against the question frame. Record include, exclude, or maybe, with a one-line reason.
2. Full-text screen for the maybes and for anything whose abstract is ambiguous about population or comparator.
3. Link reports of the same study under one identifier.
4. Chart the included studies into a fixed table: design, population and setting, intervention content and dose, comparator description, outcomes measured, and the limitations the authors report.

Two habits protect the timeline. First, write exclusion reasons as you go; reconstructing them later costs more time than recording them. Second, stop expanding the search when new results stop changing the included set, and note the date you stopped. A brief that says "search run on this date, screening completed on this date" is auditable in a way that a list of references is not.

This is where last-minute briefs most often go wrong. Four judgments are distinct, and collapsing them produces overconfident prose.

• Evidence level describes the design and rigor of the studies you found. A randomized trial and an observational cohort answer the same question with different strengths, and your chart should say which you have.
• Applicability asks whether the studied population, setting, and delivery model resemble your wards. A trial run in a dedicated unit with a full therapy team may not describe what a short-staffed medical ward can deliver.
• Indirectness asks whether the intervention or outcome in the studies matches your question. A trial that measured discharge destination rather than functional status is indirect for a functional-outcome question, even if everything else fits.
• Limitations are the constraints the authors and your charting identify: small samples, short follow-up, co-interventions, or outcome measurement that differs across studies.

For the mobility example, a reasonable brief might say: the included studies compare structured programs with usual care in hospitalized adults, but the programs differ in who delivers them and how often, and the outcomes range from walking distance to functional independence scores. That sentence is honest about heterogeneity without declaring a winner. It also tells the team what a local protocol would need to specify before any comparison to the literature is meaningful.

A one-day brief has three bottlenecks: finding papers, keeping them organized, and checking that the citations in the draft point to real work. Match the tool to the bottleneck rather than to the whole task.

Finding and refining. Run the search, then use the results to sharpen terms. If your team prefers a longer, agent-driven pass over the literature, Scholar Agent is designed for longer research tasks in the literature-review, inspiration, and reproduction space. It fits after the question frame exists, not before.

Organizing. Save the included and borderline papers in one place so the charting table and the reference list come from the same set. My Library is where papers can be saved or uploaded and organized. If your team runs several briefs, Research Projects gives each one its own workspace.

Interrogating the included set. Once the papers are in the library, questions like "which of these reported adverse events?" or "how did each study define usual care?" are faster to answer against the collected set than by reopening PDFs one at a time. Library QA supports asking questions based on papers uploaded to My Library or added from search results. Treat its answers as pointers back to the source text, and confirm each one in the paper before it enters the brief.

Checking the reference list. Before circulating the draft, screen the BibTeX. TrueCite checks BibTeX against real academic databases to flag hallucinated references. This is a concrete, bounded check: it catches references that do not correspond to real records, which is exactly the failure mode that wastes a reviewer's morning.

If you are new to the platform, the WisPaper homepage introduces it as an AI research assistant for finding, screening, organizing, and working with academic papers, and English readers can start from the English homepage.

A brief that a clinical team can use tomorrow has a predictable shape. Six short sections are enough.

1. Question and scope. The frame from earlier, plus what is explicitly out of scope.
2. Search summary. Sources used, the strategy in brief, dates run, and filters applied.
3. Included studies table. One row per study, with linked reports noted.
4. Findings by outcome. What the studies measured, kept close to how they measured it.
5. Applicability and indirectness. Where the evidence maps onto your setting and where it does not.
6. Limitations and open questions. What would change the picture, and what the team should specify locally.

Keep the findings section descriptive. "Two trials reported functional independence scores at discharge; the programs differed in session frequency" is supportable from a charting table. "The program improves function" is a causal claim that requires a synthesis you may not have time to do well, and it is the kind of sentence that invites a challenge you cannot answer by tomorrow.

Readiness is checkable. Run these tests before you send.

• Traceability. Every factual sentence in the findings and applicability sections points to a row in the included-studies table.
• Consistency. The population, intervention, and outcomes in the question frame match the columns in the table and the claims in the prose.
• Counts. The number of included studies equals the number of study identifiers, not the number of PDFs.
• Reference integrity. The BibTeX has been screened, and every entry resolves to a real record.
• Bounded conclusion. The final paragraph states what the evidence supports for this question and what remains unspecified, without recommending care for any individual.
• Reproducibility. A colleague could rerun the search from the summary and land in a similar place.

If any test fails, fix that section rather than adding caveats elsewhere. A short brief with a clean chain is more useful than a long one with a soft middle.

The specific topic will change; the artifacts will not. The question frame, the documented strategy, the staged screening log, the study-level chart, and the four-way separation of evidence level, applicability, indirectness, and limitations are reusable for any rapid evidence request. So is the habit of placing each tool at the bottleneck it addresses: search where you refine terms, a library where you organize, question-answering against the collected set where you need to interrogate it, and citation screening before the draft leaves your hands.

The next time the request arrives with a short deadline, the fastest start is not a search box. It is a blank question frame and a decision about what the brief will be allowed to claim.