28sungmin/MEGIQ

bias-detection

Detect demographic bias in LYDUS data quality metrics by computing all 16 QUIQ quality metrics (completeness, range/date/format/sequence validity, preciseness, fidelity, class/instance diversity, sentence/vocabulary diversity, logical accuracy,…

cross-sectional-consistency

Calculate cross-sectional consistency for categorical string variables in a QUIQ-format table using a local exo LLM to group semantically equivalent values. Runs on an on-prem local exo LLM (closed-network safe; no external API key). Use for data quality…

Source text: Mixed languages

date-validity

Validate date values in a QUIQ-format table. Checks Event_date column and Mapping_info_1='date' rows using standard format parsing, Korean date formats, and optional LLM fallback. Use for data quality assessment of temporal data.

Source text: Mixed languages

format-validity

Validate format of medical codes (ICD-9/10/11, SNOMED-CT, RxNorm, LOINC, ATC) in a QUIQ-format table. Identifies code type from variable name/description, then validates each value against the corresponding regex. Use for data quality assessment of medical…

Source text: Mixed languages

logical-accuracy

Detect logical outliers in clinical variables using a multi-model ensemble (Quantile Regression + Gradient Boosting + Autoencoder for numeric; OneClassSVM + IsolationForest + Autoencoder for categorical). Uses an on-prem local exo LLM (OpenAI-compatible;…

note-accuracy

Evaluate accuracy of unstructured clinical notes and radiology reports in a QUIQ-format table using a local exo LLM. Detects diagnostic, procedural, drug, demographic, and date errors in clinical notes; identifies critical errors in radiology impressions.…

note-fidelity

Evaluate completeness (fidelity) of unstructured clinical notes and radiology reports in a QUIQ-format table. Uses a local exo LLM to check whether required template items are mentioned in each note. Runs on an on-prem local exo LLM (closed-network safe; no…

quiq

Convert MIMIC-IV clinical data into QUIQ (Quality Intelligence Unified Query) long-format table. Use when transforming MIMIC-IV tables to QUIQ format for data quality assessment, when applying QUIQ Mapping_info rules to MIMIC-IV variables, or when building a…

sequence-validity

Validate temporal ordering of date variables in a QUIQ-format table. Uses a local exo LLM to automatically identify start/end date variable pairs, then checks whether start_date <= end_date for each matched record. Runs on an on-prem local exo LLM…

Source text: Mixed languages

via

Generate VIA (Variable Information Archive) table from a user-provided documentation file (PDF or text). Requires a data dictionary or schema documentation to produce meaningful variable descriptions. If no document is provided, the skill will ask for one…

Source text: Mixed languages

structure-cohort-definition

Stage 2 study plan을 OHDSI ATLAS식 '구조화 코호트 정의'(entry event · inclusion rules · exit strategy · time-at-risk)로 변환한다. LATCH Planner Stage 3. 데이터 표준화(OMOP 표준 어휘 매핑)는 하지 않고 ATLAS의 구조만 반영한다. 이 SKILL.md 본문이 곧 LLM 시스템 프롬프트로 런타임 로드된다. 검증은 tools/cohort(C1~C8)가 담당.

decompose-study-plan

Stage 1의 6칸 명세를 파싱된 study plan JSON으로 변환한다. LATCH Planner Stage 2. 파생변수 재귀분해 + 조건 분리 + 영어 keyword 번역(clinical-translate 포함). 이 SKILL.md 본문이 곧 LLM 시스템 프롬프트로 런타임 로드된다.

structure-study-design

연구자의 자유 텍스트 가설을 LATCH 6칸 구조화 명세로 재배치한다. LATCH Planner Stage 1. 이 SKILL.md 본문이 곧 LLM 시스템 프롬프트로 런타임 로드된다 ({datasets}, {analysis_catalog}는 설정/스키마에서 주입). 검증은 tools(R1~R7, DS)가 담당.

Source text: Mixed languages

clinical-translate

한글 임상 용어를 표준 영어 임상 용어로 번역해 keyword를 DB 매칭 가능하게 만든다. LATCH Planner Stage 2의 한 부분. 원문은 label/original에 보존(규칙2 글자), keyword는 영어로 번역(규칙2 정신). 임의 의역·동의어 치환 금지, 표준 용어만.

Source text: Korean