Literature Core Data Extraction (Slide Edition)

Problem

Before building clinical or academic slide decks, teams often have to re-check study design, sample size, primary endpoints, P values, confidence intervals, and safety data across one or more papers. Manual extraction is error-prone: treatment groups can be mixed up, exact P=0.001 values can be mistaken for P<0.05, ITT and Safety populations can be confused, and the result is hard to feed into comparison tables or ppt-generator.

How It Works

The skill accepts only user-provided, whitelisted literature, such as .pdf, .docx, pasted full text or abstract, or a structured literature list. It first parses metadata—title, journal, and year—and asks for confirmation before extraction. It then extracts data across seven dimensions: study basics, design, sample size and eligibility, intervention and comparator, endpoint definitions and results, baseline characteristics, and key safety findings. It prioritizes exact P values, effect estimates such as HR, OR, and RR, and 95% CI, using table-first logic for baseline data. Missing or ambiguous fields are marked with ⚠️ not reported rather than invented. Default output is a single-paper data card; multi-paper mode produces a comparison table; an optional bridge format provides structured JSON for ppt-generator.

Boundaries

This skill does not search literature, translate abstracts, or design slides. If the user only has a topic, a bare link, or asks to find papers, use a retrieval skill. Scanned PDF files may fail text extraction, so users need a text-extractable version or pasted content. Non-inferiority, equivalence, co-primary endpoints, and multiplicity adjustments are explicitly flagged to avoid statistical misreading.