UKBAnalytica — Subgroup & Sensitivity Analysis

What it does

This skill helps agents produce reproducible R scripts and workflow guidance to run subgroup interaction tests and sensitivity analyses on UK Biobank Research Analysis Platform (RAP) cohorts using the UKBAnalytica package. It covers stratum-specific effect estimation across multiple model families (cox, logistic, linear, glm, negative-binomial), interaction p-values, lag (exclude early events) analyses, and complete-case sensitivity flows. The skill intentionally enforces privacy guardrails: the agent must not read or process participant-level RAP data and should only generate scripts for the user to execute inside RAP.

When to use it

Load this skill when the user requests subgroup analyses, interaction p-values, effect modification checks, sensitivity-exclude-early-events, or complete-case re-analysis. Use it when you need code templates for run_subgroup_analysis, run_multi_subgroup, or the sensitivity helper functions for downstream aggregate reporting.

What's included

• Scripts: none embedded, but the skill documents key functions (run_subgroup_analysis, run_multi_subgroup, sensitivity_exclude_early_events, sensitivity_exclude_missing_covariates) and example R snippets to copy into RAP.
• References: repository references and examples demonstrating function signatures and common pitfalls.
• Instructions: detailed pipeline phases, prerequisites, example calls, expected output tables, and interpretation notes (p_interaction meaning, multiple comparisons cautions, NA handling).

Compatible agents

Agents constrained to produce code (R) and workflow instructions are compatible. This skill is intended for environments where the agent cannot access private RAP data; it is for script generation, explanation, and interpretation of aggregate results only.