Conclusion
R provides a credible, transparent, and reproducible alternative to commercial software for many standard sample size calculations in oncology.
For most practical two‑arm designs, alignment with East and nQuery is excellent.
Some niche designs,especially one‑arm survival or exact binary models still require careful interpretation and sometimes additional validation.
Overview of Methods Compared
| Endpoint | Design Type | Test | Software | R Packages |
|---|---|---|---|---|
| Survival | Two‑arm Fixed | Log-rank | East, nQuery | rpact, rashnu, gsDesign2 |
| Survival | Two‑arm Group‑Sequential | Log-rank | East, nQuery | rpact, gsDesign2 |
| Survival | One‑arm Fixed | Log-rank | East, nQuery | OneArm2stage, SampleSizeSingleArmSurvival, rashnu |
| Binary | Two‑arm Fixed | Z‑Pooled | East, nQuery | rpact |
| Binary | Two‑arm Fixed | Exact | East, nQuery | bbssr |
| Binary | Two‑arm Group‑Sequential | Z‑Pooled | East | rpact |
| Binary | One‑arm Fixed | Exact | East | A’Hern (no package) |
| Binary | One‑arm Fixed | 1‑arm Z‑test | East, nQuery | rpact |
Summary of Agreement Between R and Commercial Software
| Endpoint | Design type | Test | R/Software matching |
|---|---|---|---|
| survival | 2‑Arm fixed design | Log‑rank | 🟢 Perfect |
| survival | 2‑Arm group‑sequential design | Log‑rank | 🟡 Good |
| survival | 1‑Arm fixed design | Log‑rank | 🔵 Poor |
| Binary | 2‑Arm fixed design | Exact | 🟡 Good |
| Binary | 2‑Arm fixed design | Z‑Pooled | 🟢 Perfect |
| Binary | 2‑Arm group‑sequential design | Z‑Pooled | 🟢 Perfect |
| Binary | 1‑Arm fixed design | Exact | 🟡 Good |
| Binary | 1‑Arm fixed design | 1‑Arm Z test | 🟢 Perfect |
Legend:
🟢 Perfect = identical or extremely close values
🟡 Good = acceptable differences
🔵 Poor = substantial discrepancies requiring caution
Practical Guidance for Package Selection
If you need…
→ Two‑arm survival sample size (fixed or GS)
Use rpact or gsDesign2.
They are stable, well-documented, and closely match commercial software.
→ Two‑arm binary designs
→ One‑arm binary designs
- Exact (A’Hern) → implemented manually (no package), results consistent.
- One‑arm Z‑test → rpact again performs well.
→ One‑arm survival
Proceed with caution. Existing R packages use methods that differ from nQuery, leading to inconsistent results. Without consensus, caution is needed and further work may be required.