chendun1995/tte-surgery
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Draft results: performance summary, draft report (Figs 1-3, Tables 2-5), D52 open questions

Browse Source 
Co-Authored-By: Claude Opus 5.5 <[email protected]>
This commit is contained in:
chendun1995
2026-10-28 06:59:49 -04:00
co-authored by Claude Opus 5.5
parent cfc9da2c87
commit fbc417662c
4 changed files with 19862 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/decisions.md
+6
View File
@@ -103,3 +103,9 @@ Each decision is listed with the date and the reason for it. Settled decisions a
| 49 | **V2 is the V1 fit with a physician-cluster-robust SE.** Under V4, physicians without an estimable fixed effect in a phase get probability 0, and the V4 population is restricted to physicians with variation in early PVI. | Same point estimate by construction (D15). |
| 50 | **Validation (W0, harm, 40 replicates of 50,000):** CCW A−B1 is +5.12 vs a truth of +5.11 on the disease composite (z = 0.0); sequential ITT is +5.83 vs +5.71 (z = 0.6); MALE and NCO-1 have |z| ≤ 0.7. The model SE is 0–20% below the empirical SE for CCW, so the bootstrap check in D28 still needs doing. | Plan §15 validation (a). |
| 51 | **Speed and scale.** The PVI model is fitted once per replicate and variant, and shared by CCW and the sequential trials. CCW clones and weights are built once for all outcomes. The deferred-period PVI model uses every PVI plus a fixed 10% sample of the other person-months, with offset log(10). On W0 validation the estimates are identical to 4 decimals. The main designs take about 39 s per replicate instead of about 130 s. **Scale:** n = 50,000 stays. Draft is 100 replicates; final is 200 replicates (replacing 500 in D29), and the final run extends the draft's replicates 1–100. Total about 12 hours on 11 cores. | User asked for a faster, smaller run. The sample size is not reduced, because CCW's strategy A already depends on about 250 patients treated in month 5 at n = 50,000. With 200 replicates, MCSE of bias ≈ empirical SE / 14 (CCW about 0.2 pp) and MCSE of coverage ≈ 1.5%. |
## Draft run (2026-10-28)
| # | Decision | Why |
|---|---|---|
| 52 | **Draft run done:** 8 worlds × 3 scenarios × 100 replicates of 50,000 patients; 9.4 h on 11 cores; 0 errors. 16 of 288,000 estimates are NA (sequential trials, V4, 4 replicates) and are dropped. Summary: `simulations/04_summarise.R` writes `results/summary/`. Report: `reports/draft-report.Rmd`. Two open questions: (Q1) the landmark propensity model adjusts for month-5 severity, a post-PVI mediator, which drives the landmark question gap (W1 own truth about 20 pp vs a reference of 7.2 pp); with baseline covariates only, the 20-replicate oracle mean is 8.1. (Q2) NCO-1 has no severity term in the DGP, contrary to D32, so it cannot detect W5/WR bias. The bootstrap check (D28) waits on Q1/Q2. | Surfaced by the draft results. |