Batch Cohort Agent Skill

Batch Cohort Analysis Skill

You are assisting a medical researcher in generating multiple analysis scripts from a single

validated methodology template, each differing only in the exposure/outcome variable combination.

This replicates the "80-person research team" pattern: one PI designs the methodology, and

many researchers execute the same approach with different variable swaps.

When to Use

• Researcher has a validated analysis template (e.g., from /replicate-study or /cross-national)
• Wants to explore multiple exposure → outcome combinations on the same database
• Goal: systematic variable-swap code generation + batch execution + result matrix

Inputs

• Database path(s): CSV/SAS data files (KNHANES, NHANES, NHIS, or any cleaned cohort)
• Methodology template: One of:
• Path to a validated R/Python analysis script (from /replicate-study or /cross-national)
• A paper type template name: nhis_cohort, cross_national, survey_weighted
• A source paper to extract methodology from (falls back to /replicate-study Phase 1)
• Combination spec: A list of exposure/outcome pairs, provided as:
• Inline list: exposures: [depression, obesity, smoking]; outcomes: [diabetes, hypertension, CVD]
• CSV file with columns: exposure, outcome, (optional) subgroup_vars
• "all" keyword: generates all pairwise combinations from the lists

Optional Inputs

• Covariate set: Fixed covariate list for all analyses (default: use template's set)
• Subgroup variables: Variables to stratify by (default: sex, age group)
• Output format: code_only (just scripts) | execute (run + collect results) | full (code + results + summary)
• Cross-national mode: If TRUE, generates paired scripts for both countries per combination

Workflow

Phase 1: Template Validation

• Read the methodology template (R script or paper type reference).
• Identify the slot variables — parts that change per combination:
• EXPOSURE_VAR: raw variable name in the database
• EXPOSURE_LABEL: human-readable label for tables/figures
• EXPOSURE_CODING: how to derive binary/categorical exposure
• OUTCOME_VAR: raw variable name
• OUTCOME_LABEL: human-readable label
• OUTCOME_CODING: how to derive binary outcome
• Verify the template runs successfully on at least one combination before batch generation.
• Output: template summary with identified slots → user approval.

Phase 2: Variable Specification

For each exposure and outcome in the combination spec:

• Look up the variable in the database:
• KNHANES: check variable name exists in the CSV header
• NHANES: check which table contains the variable (use codebook.csv if available)
• NHIS: check claims code or variable name
• Define coding:
• Binary: threshold or category mapping (e.g., HE_glu >= 126 → diabetes = 1)
• Categorical: level definitions (e.g., smoking: current/former/never)
• Check covariate overlap: If the exposure IS one of the standard covariates, remove it from the adjustment set for that analysis (no self-adjustment).
• Output: combination matrix with all variable specifications.

| # | Exposure | Exposure Coding | Outcome | Outcome Coding | Covariates (adjusted) | Notes |
|---|----------|-----------------|---------|----------------|----------------------|-------|
| 1 | Depression (PHQ≥10) | BP_PHQ sum ≥10 | Diabetes | HE_glu≥126|HbA1c≥6.5|DE1_dg=1 | age,sex,edu,income,smoking,alcohol,obesity,CVD | — |
| 2 | Obesity (BMI≥25) | HE_obe ≥4 | Diabetes | same | age,sex,edu,income,smoking,alcohol,depression,CVD | obesity removed from covariates |
| ... | | | | | | |

Phase 3: Batch Code Generation

For each combination in the matrix:

• Clone the template script.
• Replace slot variables with the combination-specific values.
• Adjust covariates: Remove exposure variable from covariate list if present.
• Set output paths: Each combination gets its own results subdirectory.
• Generate a master runner script (run_all.R or run_all.sh) that:
• Executes all N scripts sequentially (or in parallel via future/parallel)
• Captures errors per script without stopping the batch
• Logs execution time per analysis

Phase 4: Batch Execution (if execute or full mode)

• Run the master script.
• Collect results from each combination's output directory.
• Handle failures gracefully:
• Log which combinations failed and why
• Common failures: convergence issues, too few events, empty subgroups
• Suggest fixes for failed combinations

Phase 5: Summary Matrix

Aggregate all results into a single summary:

Main Results Matrix (summary_matrix.csv):

| Exposure | Outcome | N | Events | Model 1 OR (95% CI) | Model 2 OR (95% CI) | Model 3 OR (95% CI) | p-value | Significant |

|----------|---------|---|--------|---------------------|---------------------|---------------------|---------|-------------|

| Depression | Diabetes | 5,811 | 487 | 2.14 (1.52–3.01) | 1.89 (1.33–2.69) | 1.36 (0.91–2.05) | 0.137 | No |

| Obesity | Diabetes | 5,811 | 487 | 3.45 (2.71–4.39) | 3.38 (2.65–4.32) | 3.12 (2.42–4.02) | <0.001 | Yes |

| ... | | | | | | | | |

Subgroup Summary (subgroup_matrix.csv): Same format, stratified by subgroup variables.

Heatmap (optional): Visual matrix of effect sizes × significance, exposure on Y-axis, outcome on X-axis.

Output Files

{working_dir}/batch_{timestamp}/
├── README.md                    — Batch run summary (N combinations, template used, date)
├── combination_matrix.csv       — All exposure/outcome specs with coding
├── template/
│   └── base_template.R          — The validated template (frozen copy)
├── scripts/
│   ├── 01_depression_diabetes.R
│   ├── 02_obesity_diabetes.R
│   ├── ...
│   └── run_all.R                — Master execution script
├── results/
│   ├── 01_depression_diabetes/
│   │   ├── table1.csv
│   │   ├── main_results.csv
│   │   └── subgroup_results.csv
│   ├── 02_obesity_diabetes/
│   │   └── ...
│   └── ...
├── summary/
│   ├── summary_matrix.csv       — Main results across all combinations
│   ├── subgroup_matrix.csv      — Subgroup results across all combinations
│   ├── failed_runs.csv          — Combinations that failed + error messages
│   └── heatmap.png              — Optional effect size × significance visual
└── logs/
    └── batch_execution.log      — Timing + error log

Critical Rules

• Never modify the core methodology across combinations — only swap exposure/outcome/covariates.
• Remove self-adjustment: If exposure = BMI, remove obesity from covariates. If exposure = education/income, remove the same variable from covariates. If outcome = MetS, consider removing obesity from covariates. Document all removals.
• Weighted analysis mandatory for KNHANES/NHANES/NHIS — inherited from template.
• Event count check: Before running, verify each outcome has ≥10 events per covariate (EPV rule). Flag underpowered combinations.
• Multiple comparisons: When generating >5 combinations, include a Bonferroni-corrected significance column in the summary matrix. Add a note about exploratory vs confirmatory framing.
• Reproducibility: Freeze the template version. Include a SHA256 hash of the data file in README.
• No p-hacking framing: The summary matrix is for hypothesis generation, not confirmation. State this explicitly in README and any manuscript output.
• Outcome definitions MUST include physician diagnosis: Diabetes = FPG≥126 OR HbA1c≥6.5 OR physician-diagnosed (KNHANES: DE1_dg=1, NHANES: DIQ010="Yes"). Hypertension = SBP≥140 OR DBP≥90 OR physician-diagnosed (KNHANES: DI1_dg=1, NHANES: BPQ020="Yes"). Lab-only definitions systematically overestimate exposure→outcome associations (validated: Joo 2026 replication showed US depression→DM wOR 1.92 without vs 1.54 with physician dx).
• Full covariate set is default: Always use 8 covariates (age, sex, education, income, smoking, alcohol, obesity, CVD) unless explicitly justified. Minimal models (age+sex+BMI only) overestimate effects due to residual confounding.

10. Generated-code quality gate: Because this skill emits N near-identical scripts, a single reproducibility slip (a missing seed, an absolute path, a hand-typed data literal) replicates across the whole batch. After Phase 3, lint the generated scripts with the /analyze-stats code-quality gate (check_generated_code.py --code-dir {batch_dir} --strict) and clear every Major (MISSING_SEED, HARDCODED_DATA_LITERAL, HARDCODED_ABS_PATH, INPLACE_SOURCE_OVERWRITE) before batch execution.

Cross-National Batch Mode

When cross_national: true:

• Generate paired scripts for each combination (Korea + US)
• Summary matrix includes both countries side-by-side
• Direction agreement column: ✓ if both countries show same direction of effect
• Uses /cross-national skill's dual-survey-design approach

Integration with Upstream Skills

| Need | Skill |

|------|-------|

| Variable coding lookup | analyze-stats survey_weighted guide |

| Template creation from paper | /replicate-study Phase 1–3 |

| Cross-national paired analysis | /cross-national |

| ICD-10 claims algorithms | analyze-stats nhis_icd10_mapping guide |

| Write manuscript from results | /write-paper (nhis_cohort or cross_national type) |

| Figure generation | /make-figures (forest plot of all combinations) |

Example Invocations

Basic: Single DB, Multiple Exposures × Single Outcome

/batch-cohort

DB: /path/to/knhanes/HN18.csv
Template: /path/to/validated_analysis.R
Exposures: [depression, obesity, smoking, heavy_drinking, low_income, low_education]
Outcome: diabetes
Mode: full

Cross-National: Full Matrix

/batch-cohort

DB Korea: /path/to/knhanes/HN18.csv
DB US: /path/to/nhanes/
Template: cross_national
Exposures: [depression, obesity, smoking]
Outcomes: [diabetes, hypertension, metabolic_syndrome]
cross_national: true
Mode: execute

NHIS Cohort: Claims-Based Batch

/batch-cohort

DB: /path/to/nhis_sample_cohort.csv
Template: nhis_cohort
Exposures: [atrial_fibrillation, heart_failure, COPD, CKD]
Outcomes: [all_cause_mortality, cardiovascular_death, stroke]
Mode: code_only

Anti-Hallucination

• Never fabricate variable names, dataset column names, or variable codings. If a variable mapping is uncertain, output [VERIFY: variable_name] and ask the user to confirm against the data dictionary.
• Never fabricate statistical results — no invented p-values, effect sizes, confidence intervals, or sample sizes. All numbers must come from executed code output.
• Never generate references from memory. Use /search-lit for all citations.
• If a function, package, or API does not exist or you are unsure, say so explicitly rather than guessing.