tools-only M4 Skills Index
This directory contains skills for the M4 framework, covering clinical research concepts and system functionality.
Clinical Skills
Severity Scores
| Skill | Description |
|---|---|
| sofa-score | Sequential Organ Failure Assessment score calculation |
| apsiii-score | APACHE III (Acute Physiology Score III) with mortality prediction |
| sapsii-score | SAPS-II score with mortality prediction |
| oasis-score | Oxford Acute Severity of Illness Score (no labs required) |
| lods-score | Logistic Organ Dysfunction Score |
| sirs-criteria | Systemic Inflammatory Response Syndrome criteria |
| hfrs | Hospital Frailty Risk Score from ICD codes (Gilbert 2018) |
Sepsis and Infection
| Skill | Description |
|---|---|
| sepsis-3-cohort | Sepsis-3 cohort identification (SOFA >= 2 + infection) |
| suspicion-of-infection | Suspected infection events (antibiotic + culture) |
Organ Failure
| Skill | Description |
|---|---|
| kdigo-aki-staging | KDIGO AKI staging using creatinine and urine output |
Medications and Treatments
| Skill | Description |
|---|---|
| vasopressor-equivalents | Norepinephrine-equivalent dose calculation |
Laboratory and Measurements
| Skill | Description |
|---|---|
| baseline-creatinine | Baseline creatinine estimation for AKI staging |
| gcs-calculation | Glasgow Coma Scale extraction with intubation handling |
Cohort Definitions
| Skill | Description |
|---|---|
| first-icu-stay | First ICU stay selection and cohort construction |
Research Methodology
| Skill | Description |
|---|---|
| clinical-research-pitfalls | Common methodological mistakes and how to avoid them |
| equiflow | Equity-focused cohort flow diagrams with SMD bias detection (Ellen 2024) |
System Skills
Data Structure
| Skill | Description |
|---|---|
| mimic-table-relationships | MIMIC-IV table relationships and join patterns |
| mimic-eicu-mapping | Mapping between MIMIC-IV and eICU databases |
M4 Framework
| Skill | Description |
|---|---|
| m4-api | Python API for M4 clinical data queries |
| m4-research | Structured clinical research workflow and protocol drafting |
| create-m4-skill | Guide for creating new M4 skills |
Gaps and Future Work
Concepts Not Yet Extracted
The following valuable concepts exist in the source repositories but were not extracted:
- APACHE-II Score: Older scoring system, still used in some contexts
- Charlson Comorbidity Index: Important confounder adjustment
- Ventilation Duration: Time on mechanical ventilation
- Antibiotic Classification: Categorization by class/spectrum
- MELD Score: Model for End-Stage Liver Disease
- CRRT Concepts: Continuous renal replacement therapy details
- Code Status: DNR/DNI documentation
eICU-Specific Concepts Needed
- APACHE IV (pre-computed in eICU)
- eICU pivoted lab values
- eICU vasopressor concepts
- Hospital-level clustering
Additional Data Quality Skills
- Unit conversion guidelines
- Outlier detection thresholds
- Timestamp and time zone handling
Usage Notes
Dataset-Agnostic Design: Skills document concepts, not dataset-specific implementations. Dataset-specific SQL lives in each skill's
scripts/subdirectory.Pre-computed Tables: Most clinical skills reference pre-computed derived tables in
mimiciv_derivedschema. These are available on BigQuery and can be regenerated locally viam4 init-derived.Script Files: Full SQL implementations are in each skill's
scripts/subdirectory, with separate files per dataset where applicable.Format Reference: See SKILL_FORMAT.md for the canonical skill structure specification.
References
- MIMIC-IV: https://mimic.mit.edu/docs/iv/
- eICU: https://eicu-crd.mit.edu/
- mimic-code: https://github.com/MIT-LCP/mimic-code
- eicu-code: https://github.com/MIT-LCP/eicu-code
- Agent Skills Standard: https://agentskills.io