How to Stop Writing a One-Study-at-a-Time Literature Review

A one-study-at-a-time review reads like a stack of paper summaries: Study A found this, Study B found that, Study C found something else. It is easy to write and hard to use, because the reader still cannot see what the body of evidence says, where it applies, or where it breaks down. The fix is not to write faster. It is to change the unit of work from the individual paper to the evidence table, and to build that table from a planned search rather than from whatever papers happened to arrive first.

This guide is for clinicians and medical researchers preparing a focused evidence brief, a research update, or a team discussion. It lays out a sequence with checkpoints, a worked example, a reusable checklist, and stop rules. It stays at the level of evidence appraisal and synthesis; it does not offer individual treatment advice.

The habit usually starts with how the search was done. If you begin from a handful of known papers and read outward, each new paper becomes a paragraph. Cochrane's guidance on searching and selecting studies points in the opposite direction: plan searches from explicit eligibility criteria, use multiple sources, combine free-text terms with subject headings, document the process, deduplicate records, and screen in stages. When those steps are skipped, there is no frame that tells you which studies belong together, so every study looks like its own story.

A second cause is tooling. Database interfaces reward finding one paper at a time. PubMed supports field tags, phrase searching, filters, Advanced Search, saved searches, Similar Articles, and citation lookup, and its search details show how terms were translated and combined. Those features are powerful, but they only help if you use them to build a reproducible set rather than to chase single references.

Before restructuring, decide which of these describes your draft. The diagnosis determines the next action.

• Paper list. You have citations but no shared eligibility criteria. Next action: write the question and criteria first.
• Annotated stack. You have summaries per paper but no comparison across them. Next action: convert summaries into an evidence table with common columns.
• Partial synthesis. You have grouped some studies but have not separated evidence level, applicability, indirectness, and limitations. Next action: add those four appraisal columns and re-read.
• Near-final brief. You have a table and a narrative but have not checked whether multiple reports belong to one study, or whether the search is current. Next action: link reports and rerun the search.

A quick test: can you state, in one sentence, the population, the outcome, and the comparison your brief addresses? If not, the review is still organized around papers rather than around a question.

Everything downstream depends on this step. Write the question so that a reader could apply your criteria without asking you. Cochrane's search guidance begins from eligibility criteria, so treat the criteria as the artifact that drives the search strings, the screening decisions, and the table columns.

A practical format for a focused brief:

• Population: who is included, and which groups are explicitly out.
• Outcome: the outcome you will synthesize, stated at a level you can actually compare across studies.
• Comparison or exposure: what is being compared, if applicable.
• Study designs: which designs you will include, and why.
• Setting and date limits: any restrictions, stated as decisions rather than defaults.

Checkpoint: circulate the criteria to one colleague before searching. If they cannot predict which of two sample papers you would include, tighten the wording.

A defensible search is documented, multi-source, and reproducible. Cochrane recommends planning from eligibility criteria, using multiple sources, combining free-text terms with subject headings, documenting the process, and deduplicating records. PubMed's field tags and search details help you see exactly how a query was translated, which is what makes rerunning possible later.

For a focused brief, a workable search record includes:

1. The databases and sources searched.
2. The full query strings, including subject headings and free-text terms.
3. Filters applied and the reason for each.
4. The date of the search.
5. The number of records retrieved and the number remaining after deduplication.

If you are building this set inside a scholarly search interface, WisPaper search is the place to run paper searches, and the public WisPaper Scholar Search page explains the entry point for finding academic papers. Save the query text alongside your results so the search can be rerun before the brief is finalized; Cochrane specifically recommends rerunning searches before publication.

Checkpoint: a second person should be able to reproduce your record count from your written query. If not, the search is not yet documented.

Screening is where one-study-at-a-time reviews quietly go wrong. Cochrane recommends screening in stages and linking multiple reports of one study. A trial can generate a methods paper, an interim report, and a final analysis; if you treat each as a separate study, your synthesis will double-count and your narrative will repeat itself.

A staged approach:

• Stage 1: title and abstract screening against the criteria, with reasons recorded for exclusions that are not obvious.
• Stage 2: full-text screening, again against the criteria.
• Stage 3: grouping. For each included record, identify the underlying study and link all reports to it.

Checkpoint: your included-records list and your included-studies list should be separate counts. If they are identical, you have probably not looked for linked reports.

The evidence table is the artifact that breaks the one-study habit. Each row is a study, not a report, and each column is a decision you will need to defend in the narrative.

Keep evidence level, applicability, indirectness, and limitations in separate columns. They answer different questions: how the study was built, whether it fits your population, whether it answers your question, and what constrains its interpretation. Collapsing them into a single "quality" score hides exactly the judgments your readers need to see.

Checkpoint: every included study has a row, and every row has an entry in all four appraisal columns. Blank cells are decisions you have not yet made.

Suppose your brief addresses a defined adult population and a single clinical outcome, and you want to know what the body of evidence shows about that outcome across included designs.

1. Question and criteria. You specify the adult population, the outcome, the comparisons you will accept, and the designs you will include. You exclude studies in populations that differ in a way that matters for the outcome.
2. Search. You write a query combining free-text terms with subject headings, run it across more than one source, record the date, and deduplicate. You note the record count at each step.
3. Screening. Title and abstract screening removes records that clearly fail the criteria. Full-text screening resolves the borderline cases, with reasons recorded.
4. Linking. Two reports turn out to describe the same study at different follow-up points. They become one row with two linked reports.
5. Table. Each study gets a row. One study is a stronger design but was conducted in a setting that differs from your population, so its applicability entry is cautious even though its evidence level is high. Another study matches your population closely but measures the outcome differently, so its indirectness entry notes the mismatch.
6. Synthesis. The narrative is organized by what the table shows: where studies agree, where the outcome definitions diverge, and which findings depend on populations that differ from yours. Individual studies appear as support for statements about the body of evidence, not as standalone paragraphs.

Notice what changed. The evidence level of a study and its applicability to your population are now separate statements. A high-level study with limited applicability is described that way, rather than being either over-credited or dismissed.

Tools help most when they are attached to a specific step in the sequence above.

• Building the search set. Run and refine queries in WisPaper search, and keep the query text with your search record.
• Organizing included papers. Save or upload the papers you retain into My Library so the included set lives in one place rather than across browser tabs.
• Asking questions of the included set. Once papers are in the library, Library QA supports asking questions based on papers uploaded to My Library or added from search results. Use it to interrogate the set you have already screened, not to replace screening.
• Longer synthesis tasks. For literature review work that extends beyond a single focused brief, Scholar Agent covers literature review, inspiration, and reproduction use cases.
• Checking citation entries. If your brief includes a BibTeX reference list, TrueCite checks BibTeX against real academic databases to flag hallucinated references. Run it at the reference-list stage, after the evidence table is settled.

Checkpoint: each tool you use maps to a step in your written process record. If you cannot say which step a tool served, it is not part of the workflow.

With the table complete, the narrative almost writes itself in a different shape. Organize by question, not by study:

• What the body of evidence shows for your population and outcome, stated at the level the table supports.
• Where studies diverge, and whether the divergence tracks population, outcome definition, or design.
• Where applicability is limited, naming the populations or settings that differ from yours.
• Where indirectness matters, distinguishing studies that answer a nearby question from studies that answer yours.
• What the limitations imply for how confidently the brief can be read.

A useful discipline: every sentence in the synthesis should be traceable to a column or a set of rows. If a sentence could only have come from reading one paper in isolation, it probably belongs in the table, not the narrative.

The brief is ready when these conditions hold:

• The question and eligibility criteria are written and were used to build the search.
• The search record includes sources, full query strings, filters, date, and counts, and the search has been rerun before finalizing.
• Screening was staged, and included reports are linked to included studies.
• Every included study has a complete evidence table row, with evidence level, applicability, indirectness, and limitations recorded separately.
• The synthesis is organized by question, and each claim is traceable to the table.
• The reference list has been checked, including any BibTeX entries.

If any condition fails, the fix is usually upstream. A thin synthesis often means an incomplete table; an unstable table often means the criteria were never sharp enough to screen against.

Use these rules to decide when to pause rather than push forward.

• Stop searching and revise criteria if screening keeps producing borderline cases that the criteria cannot resolve.
• Stop and link reports if you cannot tell whether two records describe the same study. Resolve the link before adding either to the table.
• Escalate to a second reviewer when an inclusion decision would change the direction of the synthesis, or when an applicability judgment is contested within the team.
• Escalate before finalizing if the search has not been rerun since the criteria were last changed.
• Pause the narrative if a paragraph cannot be traced to table rows; return to the table rather than writing around the gap.
• Escalate the reference list for a citation check whenever BibTeX entries were assembled or edited outside the database record.