find-cohort-gap

Find-Cohort-Gap Skill

You are assisting a medical researcher in systematically discovering novel, publishable research topics from a cohort database. Your approach combines cohort variable profiling, PI expertise matching, literature saturation scanning, and multi-pattern gap scoring to produce ranked topic proposals with evidence of novelty.

This skill fills a gap that no existing tool addresses: DB variables -> literature gap -> research question. Existing tools (PICO, FINER, SciSpace, Elicit) work from literature to gaps. This skill works from the data outward.

Communication Rules

• Communicate with the user in their preferred language.
• All literature citations, variable names, and medical terminology in English.
• Be direct about weak topics — kill early, save time.

Key Directories

• Output: User-specified directory (default: current working directory)
• References: ${CLAUDE_SKILL_DIR}/references/ for templates and rubrics

---

Phase 0: Cohort Intake

The cohort does not have to be one this skill has heard of. Route on what the user actually has.

The user has…	Do this
A named public cohort (NHIS, UK Biobank, KNHANES, …)	Fill the profile from published documentation. Cite the source for every field.
A codebook / data dictionary / CSV export of their own registry or EMR extract	Run the input adapter below. This is the common case — an institutional registry or single-centre export that no public documentation describes.
A review, guideline, or preprint defining the clinical domain	Attach it as domain context (--context), as a file or a URL.

Input adapter (local codebook / documents)

python3 "${CLAUDE_SKILL_DIR}/scripts/build_cohort_profile.py" \
  --codebook data_dictionary.csv \
  --context narrative_review.pdf --context https://example.org/guideline \
  --cohort-name "Institutional CT registry" --out-dir .

Formats: .csv / .tsv / .json / .md / .txt (stdlib), .xlsx (needs openpyxl), .pdf (needs pdftotext). A .csv is auto-detected as a codebook (rows are variables) or a data export (the header row is the variable list). Writes cohort_profile.md + cohort_profile.json (+ context_extract.md).

Do not read the codebook yourself and summarise it. Paraphrasing a variable name, merging two that look alike, or inventing one the cohort does not have poisons every downstream claim — the intersection matrix, the feasibility gate, and eventually the manuscript's Methods. The adapter enumerates variables verbatim with provenance (file:row) instead, which is the dictionary-first discipline a reviewer expects of a DB-backed study. Read cohort_profile.md; do not re-derive it.

What the adapter infers (and shows its work for): the variable cluster map, serial / repeated-measure groups (evidence for P1 Longitudinal Advantage), and endpoint candidates (evidence for P2 Endpoint Upgrade). Every cluster assignment records the keyword that triggered it, and a variable matching nothing is left unclassified rather than forced into a bucket — review those, since the lexicon is not exhaustive.

What the adapter cannot know — ASK, never guess

A codebook lists variables. It does not state any of the following, and each is emitted as [UNKNOWN - ask the user]:

1. Sample size (N at baseline, N with follow-up)
2. Time span (enrollment period, follow-up duration, measurement intervals)
3. Known limitations (healthy volunteer bias, attrition, missing-data patterns)
4. Existing publications from this cohort (to avoid duplicating them)
5. IRB status and data-access route

Collect these from the user before Phase 2. A guessed N does not merely sit there — it flows into the Phase 5 feasibility gate, which then passes (or fails) for a reason that has nothing to do with the cohort.

Also confirm the setting (institution type, country, population type) and any special strengths the variable names cannot reveal — registry linkage, biobank availability, a distinctive population.

Gate: Present the cohort profile summary, including the [UNKNOWN] list and the unclassified variables. Confirm before proceeding.

---

Phase 1: PI/CA Profiling

Profile the intended PI or corresponding author to find topic-expertise alignment.

1. Search PubMed for the PI's recent publications (last 5 years).
   • Use /search-lit E-utilities: bash "$EUTILS" search "AuthorLastName AuthorFirstInitial[Author]" 30
   • Extract top keyword clusters from titles/abstracts.
2. Identify specialty signals:
   • Academic society positions (president, board member, editor)
   • Subspecialty focus areas
   • Preferred journal tiers
3. Build a PI keyword map: 5-10 keyword clusters ranked by publication frequency.

If no PI is specified, skip this phase and use variable clusters alone in Phase 2.

Output: PI profile card (name, affiliation, top keywords, society roles, preferred journals).

---

Phase 2: Intersection Matrix

Cross cohort variable clusters with PI expertise to generate candidate topics.

Method

Create a matrix: rows = DB variable clusters, columns = PI keyword clusters. Score each cell 0-3:

• 3: PI has published in this exact intersection (direct match)
• 2: PI's subspecialty covers this area (strong relevance)
• 1: Tangential connection (possible but needs framing)
• 0: No connection

Candidate Generation

1. Extract all cells scoring 2-3 as primary candidates.
2. For cells scoring 1, apply the A-B substitution test: "Has someone published [this analysis] with [a different exposure/outcome] in a similar cohort?" If yes, substituting the PI's specialty variable creates a viable candidate.
3. Generate 20-40 candidate topic statements in PICO format:
   • P: Population from the cohort
   • E: Exposure/predictor variable(s)
   • C: Comparison group
   • O: Outcome (preferably hard endpoint)

Discipline Alignment Filter

Before advancing candidates to saturation scanning, apply a discipline filter:

• Who is the intended first author? Identify their department/specialty.
• Does the primary exposure variable belong to that discipline? The first author's specialty must align with the study's core variable. For example:
  • Radiology first author → imaging variable must be the primary exposure
  • Cardiology first author → cardiac biomarker or ECG finding as exposure
  • Neurology first author → neurological variable or brain imaging as exposure
• Kill candidates where the primary exposure is outside the first author's discipline. A strong PI match alone is insufficient if the first author cannot claim ownership of the core variable.

This filter prevents generating topics where the first author's contribution is not defensible at the variable level.

Gate: Present the intersection matrix and top 20 candidates (post-discipline filter). User selects 8-12 for saturation scanning.

---

Phase 3: Literature Saturation Scan

For each selected candidate, determine how saturated the literature is.

Search Strategy

For each candidate:

1. Build a PubMed query: (exposure terms) AND (outcome terms) AND (cohort OR longitudinal OR prospective)
2. Execute search via /search-lit E-utilities.
3. Count total results and classify:

Grade	Count	Longitudinal?	Interpretation
Blue Ocean	0-2 papers	N/A	First report possible. Verify the topic has audience interest.
Green Field	3-10 papers, all cross-sectional	No longitudinal	Optimal zone — established interest, longitudinal gap wide open.
Yellow	10-30 papers	Some longitudinal	Viable only with very specific angle (unique population, novel endpoint).
Red	30+ papers or MA exists	Yes	Avoid unless doing NMA or using truly unique data.

Critical Filter

For each candidate in Green/Yellow, ask: "Has anyone published this with serial/repeated measurements?" If no — automatic upgrade by one grade.

"So What" Test

For each candidate, articulate 2-3 potential clinical implications of the findings. If you cannot state why a clinician or policymaker would care about the result, the topic fails regardless of gap score.

Output: Saturation table with grade, paper count, longitudinal gap status, and "So What" statement for each candidate.

Gate: Present saturation results. User selects 3-5 finalists for deep scoring.

---

Phase 4: 6-Pattern Scoring + Comparison Table

Apply the 6-Pattern framework to each finalist. Score each pattern 0 or 1.

6 Patterns (Universal)

Read the detailed rubric at ${CLAUDE_SKILL_DIR}/references/pattern_scoring_rubric.md.

#	Pattern	Question	Score 1 if...
P1	Longitudinal Advantage	Does the cohort's serial/repeated measurement structure create a clear edge over existing cross-sectional studies?	Cohort has 3+ timepoints for key variables AND no prior study used serial data for this topic.
P2	Endpoint Upgrade	Can we escalate to a harder endpoint than existing studies?	Cohort links to mortality/cancer/CVD registries AND existing studies stop at surrogate endpoints.
P3	Cohort Uniqueness	Is the cohort's population, scale, or setting distinctive?	Largest in this population, unique ethnic group, screening-based (no referral bias), or novel linkage.
P4	PI-Topic Alignment	Does the PI's expertise and reputation strengthen this topic?	PI has society role or 5+ papers directly in this domain. Skip if no PI specified.
P5	Comparison Table Gaps	Does the THIS STUDY column show 3+ differences vs existing papers?	Build comparison table (see below). 3+ checkmarks in THIS STUDY that are absent in all prior papers.
P6	Complementary Design	Can this topic pair with another study from the same cohort?	Two studies using the same DB but different populations or complementary variables (e.g., viral vs non-viral).

Comparison Table Construction

For each finalist, build a table comparing the top 3-5 existing papers against THIS STUDY:

| Feature | Author1 (Year) | Author2 (Year) | Author3 (Year) | THIS STUDY |
|---------|----------------|----------------|----------------|------------|
| Design | Cross-sectional | Cohort (5yr) | Cross-sectional | Cohort (20yr) |
| N | 3,200 | 8,500 | 12,000 | ~200,000 |
| Serial data | No | No | No | Yes (avg 5 visits) |
| Hard endpoint | Surrogate | Surrogate | All-cause mortality | CVD + all-cause mortality |
| Population | Referral | General | Screening | Health checkup (no referral bias) |
| Ethnicity | Western | Western | Asian (Japan) | Asian (Korea) |
| Subgroup analysis | No | Age only | No | Age + sex + comorbidity |

Score Interpretation

Total Score	Recommendation
5-6	Top-tier journal target (Lancet sub, JACC, J Hepatol level)
3-4	Specialty journal target (solid publication)
1-2	Restructure or kill — find a stronger angle before proceeding

Gate: Present scoring results and comparison tables. User approves final ranking.

---

Phase 5: Feasibility Gate

For each scored finalist, verify practical feasibility.

Checks

1. Sample size adequacy:

   • Cox regression: minimum 10 events per predictor variable (EPV rule)
   • Logistic regression: same EPV rule
   • For large cohorts (N>100K): warn about p-value inflation — statistically significant results are nearly guaranteed, so focus on effect size thresholds (e.g., HR >1.2 or <0.8 for clinical relevance)
   • Consider negative control strategy (EPCV) for very large samples

2. Missing data:

   • Key exposure variable: <20% missing acceptable
   • Key outcome: <5% missing
   • If serial data: assess attrition pattern (MCAR/MAR/MNAR)

3. Follow-up adequacy:

   • Outcome must have plausible latency within available follow-up
   • Cancer outcomes: minimum 5 years
   • CVD events: minimum 3 years
   • Mortality: minimum 5 years

4. Operational definition:

   • Can the exposure be defined from available variables?
   • For claims data: ICD codes alone = 40-60% accuracy. Require combination strategy (diagnosis + prescription + visit frequency + special codes)
   • Cross-check expected prevalence against known epidemiological data

5. IRB/ethics:

   • Is the data already IRB-approved for this type of analysis?
   • Any additional approvals needed for data linkage?

6. Disease Novelty Bonus (informational, not Go/No-Go):

   • Idiopathic etiology or debated mechanism → higher journal interest
   • Established mechanism → needs stronger methodological novelty

Decision

• Go: All checks pass.
• Conditional Go: Minor issues solvable (e.g., missing data manageable with imputation).
• No-Go: Fatal flaw (insufficient events, no valid endpoint, key variable unavailable).

Output: Feasibility report for each finalist with Go/Conditional/No-Go status.

---

Phase 6: Output — Ranked Proposals + One-Pagers

Generate the final deliverables.

Ranked Summary Table

| Rank | Topic (PICO) | Saturation | 6-Pattern Score | Feasibility | Target Journal | Timeline |
|------|--------------|------------|-----------------|-------------|----------------|----------|
| 1 | ... | Green (0 longitudinal) | 5/6 | Go | JACC | 6 months |
| 2 | ... | Green (1 longitudinal) | 4/6 | Go | Eur Heart J | 6 months |
| 3 | ... | Blue (0 papers) | 3/6 | Conditional | Radiology | 8 months |

One-Pager for Each Finalist

Use the template at ${CLAUDE_SKILL_DIR}/references/onepager_template.md.

Each one-pager includes:

1. Title: Working title for the study
2. Background: 3-4 sentences establishing the gap (with "Zero Papers" claim if applicable)
3. Comparison Table: THIS STUDY vs existing papers
4. Objective: Primary research question in PICO format
5. Methods Summary: Study design, key variables, statistical approach
6. PI Role: Why this PI is the right corresponding author
7. Target Journal: With rationale (PI alignment, scope match, gap fit)
8. Timeline: Realistic estimate (data preparation → analysis → drafting → submission)
9. 6-Pattern Score Card: Visual breakdown of each pattern

Save one-pagers as markdown files: {output_dir}/gap_proposal_{rank}_{short_topic}.md

---

Skill Integration

Phase	Calls to other skills
Phase 1 (PI profiling)	/search-lit E-utilities for PubMed author search
Phase 3 (Saturation scan)	/search-lit E-utilities for topic searches
Phase 4 (Comparison table)	/search-lit for retrieving paper metadata
Downstream	Output feeds into /design-study → /write-paper pipeline

What This Skill Does NOT Do

• Does not perform the actual statistical analysis (use /analyze-stats)
• Does not write the full manuscript (use /write-paper)
• Does not validate study design (use /design-study)
• Does not generate references (use /search-lit)
• Does not make publication-ready figures (use /make-figures)

Anti-Hallucination

• Never fabricate references. All citations must be verified via /search-lit with confirmed DOI or PMID. Mark unverified references as [UNVERIFIED - NEEDS MANUAL CHECK].
• Never invent clinical definitions, diagnostic criteria, or guideline recommendations. If uncertain, flag with [VERIFY] and ask the user.
• Never fabricate numerical results — compliance percentages, scores, effect sizes, or sample sizes must come from actual data or analysis output.
• If a reporting guideline item, journal policy, or clinical standard is uncertain, state the uncertainty rather than guessing.