Power and Sample Size (bioSkills)

A Claude Code skill for justifying trial size in a protocol or SAP — picking the right formula for the endpoint and design, setting a defensible non-inferiority margin, and writing the CONSORT item 16a justification.


Type	Claude Skill
Supplier	GPTomics bioSkills (community OSS, MIT)
Availability	GA — part of the bioSkills collection
Pricing	Free / OSS (MIT) — `statsmodels`, `scipy`, `numpy`, `pandas` and the R packages `pwr`, `gsDesign`, `gsDesign2`, `rpact`, `presize`, `npsurvSS`, `nph`, `simtrial`, `PowerTOST`, `clusterPower`, `ratesci`, `nlme` are separately installed OSS. The commercial alternatives (nQuery, PASS, East) are not installed.
Capabilities	Read/Write — Claude runs the skill’s workflow locally (Python and R), not as an MCP tool
Verified	works · 2026-08-13
Security	cleared · 2026-08-13 — GPTomics/bioSkills MIT, no external credentials

How to install

bioSkills is not an npm package — skills are plain markdown/code read directly by the agent. Clone the repo, then either run the installer for the whole category or copy the single skill directory.

Claude Code — clone and install via the bundled script:
```
git clone https://github.com/GPTomics/bioSkills
cd bioSkills
./install-claude.sh --categories "clinical-biostatistics"
```
The installer copies matching skills into ~/.claude/skills/ (default target). Use ./install-claude.sh --list to preview the skills first.
Claude Code / other agents — copy just this one skill:
```
cp -r bioSkills/clinical-biostatistics/power-and-sample-size ~/.claude/skills/
```
(run from inside your clone — the previous step left you in bioSkills/; otherwise replace bioSkills/ with the absolute path of your clone). Install the Python side on first use with pip install "statsmodels>=0.14" "scipy>=1.12" "numpy>=1.26" "pandas>=2.1"; survival and bioequivalence work additionally needs R with install.packages(c("gsDesign", "rpact", "npsurvSS", "PowerTOST")).

What it does

Covers continuous, binary and time-to-event endpoints across superiority, non-inferiority and equivalence framings, and drives statsmodels (primary) with R packages for the cases Python does not cover well:

Superiority sizing — effect size on the Cohen scale (d = 0.2 small, 0.5 medium, 0.8 large) at two-sided α = 0.05, with the skill’s rule that the design difference δ should be ≥ 1.5 × MCID, because setting δ equal to the minimum clinically important difference leaves no margin for sampling variation (Norman 2003).
Non-inferiority margins — the FDA 2016 M1/M2 framework at one-sided α = 0.025: M2 ≤ 0.5 × historical M1 (the “double discount” that guards against biocreep) and M2 ≤ 0.5 × MCID, since the loss you are willing to accept must be smaller than a difference that would be clinically meaningful.
Survival endpoints — Schoenfeld 1981 for the event-driven count under proportional hazards, and Lakatos 1988 or simulation when it does not hold. The skill flags that Schoenfeld under-estimates by 20–50% under non-proportional hazards (Lin 2020, NPH Working Group), which is the delayed-effect case typical of immuno-oncology.
Bioequivalence — Schuirmann’s two one-sided tests, targeting a 90% CI for the geometric mean ratio of C_max/AUC inside (0.80, 1.25) per the FDA 1992 guidance, via PowerTOST.
Inflation and design factors — dropout inflation n_final = n / (1 − q) (q = 0.20 → multiply by 1.25); the stratified-randomisation efficiency gain n / (1 − r²) (Senn 2013); the cluster design effect 1 + (m − 1)·ICC.
Adaptive interaction — if the size may be re-estimated at an interim, unblinded SSR needs Cui–Hung–Wang weighting to hold the Type-I error, and the Mehta–Pocock promising zone is calibrated for a conditional power range of roughly 30–80%.

Primary use cases: writing the sample-size section of a protocol or SAP, choosing a non-inferiority margin, sizing an event-driven survival trial.

Notes

MCID and δ are not the same number and the skill treats conflating them as the headline error. The MCID is a property of the outcome measure; δ is a design choice that must sit above it.

The skill notes that applying a continuity correction to binary comparisons wastes roughly 10% of the sample size (D’Agostino 1988, Am Stat 42:198) and that modern computing makes exact or score-based intervals (ratesci, Miettinen–Nurminen) the better default.

Regulatory anchors cited: the FDA 2016 non-inferiority guidance (M1/M2), the FDA 1992 bioequivalence guidance (80–125%), and CONSORT 2025 item 16a, which is the reporting requirement this skill’s output is written to satisfy.

A power calculation for a fixed design is a different problem from a group-sequential one — for interim looks, spending functions and sample-size re-estimation, use Adaptive Designs, which drives the same rpact/gsDesign stack.

Distributed as a SKILL.md (plus reference material) in the bioSkills collection — Claude executes the workflow locally rather than as an MCP server. The upstream skill front-matter name is bio-clinical-biostatistics-power-sample-size; if invoked as a namespaced plugin command it resolves under the bioSkills plugin, not as a bare /power-and-sample-size. Upstream directory: clinical-biostatistics/power-and-sample-size.

Related: Multiplicity and Graphical Procedures (α is an input to this calculation, and splitting it changes n), Subgroup Analysis (interaction tests need roughly 4× the main-effect sample size), Trial Reporting, statsmodels.

Sources

Installed this tool?

Share feedback — install path, OS, errors, workarounds. The form opens with this tool pre-selected and a link back to this page.