Medical Research Statistical Analysis (analyze-stats)

What it does

• Profiles your data file first (shape, inferred types, missingness per column, first rows, categorical levels) and pins down the analysis unit (patient, exam, lesion, rater).
• Proposes an analysis plan and waits for approval before executing, routing each request to a method guide and template: Table 1, diagnostic accuracy (Se/Sp/PPV/NPV/AUC), agreement (kappa, ICC, Bland–Altman), logistic/linear regression, survival and competing risks, propensity scores (PSM/IPTW/SIPTW/overlap), repeated measures (LMM/GEE), survey-weighted analyses (NHANES/KNHANES), meta-analysis, mediation, MR, PRS.
• Enforces reproducibility: seed 42, every reported number must come from executed output, a lint gate (check_generated_code.py) that blocks missing seeds, hand-typed data literals, absolute paths and in-place overwrites of raw data, plus an _analysis_outputs.md manifest for downstream figure/paper skills.
• Pre-screens classic pitfalls: complete/quasi separation before fitting, no Shapiro-Wilk gating (skewness rule instead), Welch by default, strata disjointness before trend tests, structural zeros (never-smoker pack-years), and collider/mediator over-adjustment in cross-sectional models.
• Produces reporting-ready output: exact p-values, 95% CIs on every primary estimand, effect sizes with plain-language IQR translations, absolute risk difference and NNT, calibration and decision-curve analysis, and journal-specific table styling (Radiology/JAMA/NEJM/Lancet) with gtsummary code.
• Handles privacy: prompts for de-identification before reading raw files and never prints PHI values.

Who it's for

• Clinical, radiology and public-health researchers writing their own papers.
• Research teams that need every manuscript number reproducible and reviewer-proof.
• Analysts working with complex survey data or claims/ICD-10 cohorts.

Example uses

1. "Build Table 1 from cohort.csv" → variable/missingness report → mean (SD) vs median (IQR) chosen by skewness → CSV + markdown table + gtsummary code.
2. "Fit a logistic model for IDH mutation" → separation screen on all categorical predictors → if a pathognomonic sign separates perfectly, it proposes Firth penalization or a two-stage rule in the plan instead of silently reporting a broken OR.
3. "Compare my new imaging marker against the clinical model" → nested likelihood-ratio test on the new term, ΔC-statistic with CI, net-benefit/decision curve, and a drafted Results paragraph.

Read the instructions in this file and follow them to help me:
https://raw.githubusercontent.com/Aperivue/medsci-skills/HEAD/skills/analyze-stats/SKILL.md

What I want: (describe your task here)

Install the skill I found on Claude Skill Mart.
Copy the skills/analyze-stats folder from the GitHub repo Aperivue/medsci-skills into my ~/.claude/skills/analyze-stats/.
When it's done, tell me in one line what this skill can do.