Medical Meta-Analysis

Problem

Clinical meta-analysis workflows often fail not because of one R function, but because intent, effect size, column mapping, model selection, subgroup setup, bias checks, and report generation are scattered. This skill turns the R pipeline into a reproducible workflow: users provide data or an analysis intent, and the skill routes to a compute track or a topic track without free-form rewriting of statistics.

Core capabilities and steps

• Compute track: for clear tasks such as pooling OR, forest plots, funnel plots, I²/τ², and Egger/Begg checks; it runs the R workflow and returns an HTML report with a reproducible R script.
• Topic track: for undecided or feasibility-stage questions; it uses a literature probe to collect real hit counts, then produces a decision card or full report with four-dimension scoring.
• Task coverage: pairwise_meta, subgroup_analysis, metareg, nma, nma_rank, survival_meta, and diagnostic_meta; publication bias checks are embedded in relevant tasks.
• Output contract: numeric results must come from the R-side run; the HTML report is the main presentation surface, with forest/funnel SVGs inlined and last_run.json preserving request and result details for reproducibility.

Boundaries

This skill supports statistical analysis, not clinical judgment. Numerical computation is delegated to an external coze workflow, and users are notified before data leaves the local environment. If the cloud run fails, diagnostic retry requires user consent. It does not perform general literature database search; full-text retrieval is only relevant when the user supplies a DOI/PMID.