Fuse svycoxph and cph objects for rms ecosystem compatibility

Takes a fitted cph object and a fitted svycoxph object with the same formula, and returns a modified cph-class object with survey-correct coefficients and variance-covariance matrix while preserving the rms $Design structure needed for anova.rms(), Predict(), summary.rms(), and related generics.

Usage

svycph_fuse(fit_cph, fit_svy)

Arguments

fit_cph: A cph object fitted with x=TRUE, y=TRUE, surv=TRUE.
fit_svy: A svycoxph object fitted with the same formula and data.

Value

A modified cph object with survey-correct inference. The $var slot contains the sandwich variance-covariance matrix from fit_svy; $coefficients contains the weighted partial likelihood estimates. The $Design and all structural slots are preserved from fit_cph.

Details

anova.rms() constructs Wald tests as $(L\beta)^\top (LVL^\top)^{-1} (L\beta)$ where $L$ is a contrast matrix derived from $Design and $V$ is $var. Substituting the survey-corrected $V$ from svycoxph yields design-correct Wald statistics while preserving the rms term-identification machinery.

Degrees of freedom in anova.rms() are based on contrast matrix rank, not survey PSU count. For fully correct F-test df, use survey::regTermTest() directly on fit_svy.

References

Binder, D.A. (1992). Fitting Cox's proportional hazards models from survey data. Biometrika, 79(1), 139–147.

Lin, D.Y. (2000). On fitting Cox's proportional hazards models to survey data. Biometrika, 87(1), 37–47.