How to Find the Next Research Question with a Research Gaps Matrix

A literature review that ends with “further research is needed” has not finished its job. That sentence is a shrug. A usable review should leave the reader with a next research question: who still needs to be studied, against what, on which endpoint, for how long, and with which design.

LitSynth’s Research Gaps Matrix is built for that last mile. It is not a novelty generator. It is a structured reading of limitations already sitting in the included evidence.

The method, in one paragraph

After screening and extraction, inspect the evidence table for six kinds of residual uncertainty: missing population, missing comparator, missing endpoint, short follow-up, weak design, and unresolved heterogeneity. For each gap that the included papers actually imply, write why it matters and what study would reduce it. Do not promote “this set did not contain X” into “the field has never studied X.”

You can paste this into a Methods section:

Why most “research gaps” sections are unusable

Three failure modes show up in student drafts, grant boilerplate, and AI-generated reviews:

1. Author-echo. The paper’s introduction said it would fill a gap. The review repeats that claim as if the field still has the same hole, without asking whether later included studies already filled it.
2. Local absence as global absence. Your 25 included cards do not mention a pediatric arm. That is a property of this corpus, not proof that no pediatric trial exists.
3. Design-free wishes. “Larger, longer, better studies are needed” cannot be turned into a protocol. A next question needs a population, an intervention or exposure, a comparator, an endpoint, a time horizon, and a design.

A gaps matrix exists to stop those three collapses.

Author-stated gaps are not residual gaps

Keep two objects separate.

Extraction can record the first as a field on a single paper. The Research Gaps Matrix is the second: a synthesis artefact that only becomes legitimate after screening and extraction. If you write residual gaps from titles alone, you are guessing.

That is why the order in LitSynth’s screening → extraction → appraisal tutorial matters. Appraisal without extraction is theatre. Gaps without extraction are marketing.

Six gap types that can become a next study

LitSynth’s paper analysis is instructed to derive gaps only from evidence you actually supplied, and to prefer clinically or scientifically usable next steps. The useful categories are boring on purpose:

1. Missing population. The included trials enrolled adults with preserved kidney function; the review question was about advanced disease.
2. Missing comparator. Every included study is single-arm or usual-care; the decision you need is versus an active standard.
3. Missing endpoint. Papers report biomarkers or imaging surrogates; the question was clinical events.
4. Short follow-up. Effects are measured on treatment; durability after withdrawal is the actual question.
5. Weak design. The only included evidence is observational, pre/post, or secondary analysis of the same parent trial counted twice.
6. Unresolved heterogeneity. Two or more included studies conflict, and the table shows why (dose, population, endpoint definition) without resolving it.

If a candidate gap does not fit one of these, it is usually a slogan. Drop it.

What a complete gap row contains

A matrix row is not a paragraph of vibes. In LitSynth it is a record with:

• category — which of the types above;
• description — what the included evidence cannot yet answer;
• why it matters — the decision, harm, or inference that stays blocked;
• suggested next step — a study you could actually write a protocol for;
• supporting papers — which included items implied this hole;
• evidence signal — what the current cards suggest, conservatively;
• study design needed — RCT, longer cohort, head-to-head, and so on.

The next-step sentence should name population, intervention or exposure, comparator, endpoint, and follow-up when the evidence allows it. “More RCTs are needed” is not a next step.

How to read the matrix without fooling yourself

Work the table like a methods reviewer:

1. Start from extracted limitations, not from the draft’s last heading. If the evidence table does not show short follow-up, you may not invent a durability gap.
2. Check supporting paper indices. A gap with no supporting cards is a hallucination of absence.
3. Stay inside the sampling frame. Phrase it as “the included evidence did not report…” unless a cited paper itself claims the field lacks the study.
4. Do not launder overlapping evidence. Two papers that are extensions of the same trial are not two independent reasons the gap is “well established.”
5. Keep abstract-only cards conservative. If extraction never saw full text, the next study suggestion is a hypothesis about this set, not a map of the literature.

This is the same honesty rule that belongs in data extraction methods: empty is better than invented.

Where the matrix sits in a PRISMA-lite workflow

The Research Gaps Matrix is a synthesis aid, not a search engine for unasked questions.

1. Freeze the review question and eligibility criteria.
2. Screen titles and abstracts with inspectable reasons.
3. Extract included papers into an evidence table.
4. Read limitations, missing PICO pieces, and conflicts from that table.
5. Fill the gaps matrix.
6. Write the review’s “Research gaps and future directions” section from the matrix, with citations.
7. If you then design a study or a grant, take the next-step row—not the narrative flourish.

The Evidence Matrix (themes, supporting versus conflicting items) answers “what does the current evidence say?” The Gaps Matrix answers “what can it not yet say, tightly enough to plan work?” Do not merge them. A conflict is not automatically a gap; sometimes it is a difference in endpoint definition that you should report as heterogeneity.

What LitSynth does not claim

• It does not search the entire field to prove a study has never been done.
• It does not replace a registered systematic review, living evidence map, or horizon scan.
• It does not assign GRADE certainty or Cochrane risk of bias. Residual uncertainty from weak design is not a RoB 2 judgement; see risk-of-bias methods.
• It does not turn an AI suggestion into a fundable question without a human who can name the endpoint and the ethics.

If your protocol requires a formal evidence-gap map, export the table and do that work against a stated search. Keep the matrix as the first structured pass, and say so in print.

A worked pattern (not a disease claim)

Suppose included cards on an intervention report 12–16 week surrogate outcomes, no active comparator, and no post-discontinuation follow-up. A defensible matrix row looks like:

• Category: short follow-up / missing comparator.
• Description: Included trials estimate on-treatment change versus baseline or usual care; they do not establish incremental benefit versus the current standard after treatment stops.
• Why it matters: Clinical and coverage decisions need durability and a head-to-head, not only a short surrogate.
• Suggested next step: A randomized comparison versus the relevant active standard, with the primary endpoint measured after a pre-specified off-treatment window.
• Design needed: Parallel-group RCT with withdrawal or maintenance follow-up.
• Wording to avoid: “No studies have evaluated long-term effects.”

That last sentence is the one AI tools love, and the one a methods committee will delete.

How to report this in a paper or a grant

Use the matrix as a methods object:

• State that gaps were taken from included evidence, not from model memory.
• Name the gap types you allowed.
• Keep citations on the gap claims that mention specific studies.
• If dual screening, full-text RoB 2, or GRADE were not done, do not hide that in the gaps section by sounding more definitive than the evidence table.

In a manuscript, the matrix becomes the “Research gaps and future directions” section: cited, bounded, and no stronger than the evidence table. In a grant, the same row becomes the specific aims seed—one primary question, one design, one endpoint—not a list of everything the field might someday study.

The PRISMA-lite product page and the systematic review workflow exist so you can point a committee at the same boundary in product language. This post is the method.

What to do next

If you are still choosing a workflow, start from screening, extraction, and appraisal. If you already have included papers, extract first, then ask the Gaps Matrix what the table cannot answer. The next research question is a property of the evidence you were willing to inspect—not a sentence the model was willing to invent.