fix: label by= strata from the fit, so subset= can drop a group

[ehrlinger]

Follow-up to #305, raised by Copilot in review on #308.

What was wrong

kaplan() and nelson() pass ... to survfit(). An option there that removes a whole group (subset, start.time) leaves the fit with fewer strata than the by column has values.

• On main (since fix: restart kaplan()/nelson() interval lags in every by= stratum #305) this stops: kaplan("time", "status", data = veteran, by = "trt", subset = veteran$trt == 1) gives "the 'by' column has 2 groups but the fit has 1 strata".
• On CRAN 3.5.3 it did not error, but labelled the surviving trt == 2 rows as group 1.

The change

• by is turned into a factor bound to the name grp (.strata_factor()), and both fits use survfit(srv ~ grp, ...). survfit() then names each stratum grp=<level>, and .label_strata() reads the labels back from those names, mapping them to the original values so groups keeps the column's type. The fit is the same as with strata(data[[by]]); I checked surv and the strata counts on veteran.
• When subset leaves a single group, survfit() returns no $strata, so there is nothing to read. That case takes the group from the rows the fit used: complete cases inside the subset subscript, applied to the row numbers so logical, positive and negative subscripts all work (.fit_rows()). A single stratum left by start.time keeps its name and is read like any other. If the kept rows do not come to one group, it is an error that says to subset data instead.
• nelson()'s weighted fit uses the same grp.

Verification

• New test, for both estimators: subset leaving one group (labelled 2 and equal to that arm fitted alone); a negative subset leaving the other; start.time = 200 dropping adeno, and start.time = 600 leaving only squamous; and subset dropping one of four celltype groups (labels and surv match each group fitted alone).
• devtools::document(): no changes. lintr::lint_package(): 0.
• NOT_CRAN=true VDIFFR_RUN_TESTS=true devtools::test(): 0 failures, 2317 passed, 6 skipped (existing skips). No snapshot files touched.
• R CMD check --as-cran with the manual, from a git archive export: 1 NOTE (CRAN incoming feasibility).

A backport to maint/v3 is ready locally and will follow once this merges.

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 85.71429% with 5 lines in your changes missing coverage. Please review.
✅ Project coverage is 91.14%. Comparing base (4c0b1ec) to head (3810943).

Files with missing lines	Patch %	Lines
R/utils.R	79.16%	5 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #310      +/-   ##
==========================================
- Coverage   91.15%   91.14%   -0.02%     
==========================================
  Files          60       60              
  Lines        5949     5964      +15     
==========================================
+ Hits         5423     5436      +13     
- Misses        526      528       +2     

Files with missing lines	Coverage Δ
R/kaplan.R	100.00% <100.00%> (ø)
R/nelson.R	96.96% <100.00%> (+0.04%)	⬆️
R/utils.R	96.09% <79.16%> (-1.33%)	⬇️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Copilot review overview

🟡 Changes recommended

Negative row-index subsets are mishandled, and the documented start.time behavior lacks regression coverage.

Review effort: Balanced
Findings: 1 · 1

Open (2)

• Negative subset indexing breaks strata label recovery · New
• Add start.time coverage for dropping early strata · New

What changed in this PR

Updates stratified Kaplan–Meier and Nelson–Aalen labeling when survfit() options remove groups.

Changes:

• Derives labels from fitted strata.
• Handles single-stratum fits using retained rows.
• Adds subset-based regression coverage.

File	Description
R/​utils.R	Adds stratification and fit-row helpers.
R/​kaplan.R	Uses fit-derived group labels.
R/​nelson.R	Applies matching labels to weighted and unweighted fits.
tests/​testthat/​test_kaplan_nelson.R	Tests groups removed by subset.
NEWS.md	Documents corrected labeling.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[ehrlinger]

Confirmed: with subset = -arm_2 the fit kept only trt == 1 and label recovery rejected it. Fixed in 3810943: the subscript is applied to the row numbers (seq_along(kept)[subset]), so logical, positive and negative subscripts mean what they mean to survfit(). The test adds the negative case for both estimators.

[ehrlinger]

Right, the test did not cover start.time, and checking it showed my PR description was also wrong. A start.time that leaves a single stratum does not lose its name: survfit() still returns grp=squamous, so it is labelled from the name and does not error. Only a subset down to one group returns no $strata. Added in 3810943: start.time = 200 drops adeno (last time 186) and labels the other three; start.time = 600 leaves only squamous, labelled correctly. Description corrected.