I am a bit confused by the hazard ratios from the flexsurvspline model (Royston-Parmar). For a categorical variable (say X_1 with two levels, (a) and (b)) with time-varying coefficients, I can specify the model as
flexsurvspline(Surv(time, event) ~ X_1+X_2+X_3,k=2, data = data, scale="hazard",anc=list(gamma1=~X_1, gamma2 = ~ X_1, gamma3 = ~ X_1))
where gamma1,2, and 3 are the ancillary parameters (see:https://cran.r-project.org/web/packages/flexsurv/flexsurv.pdf). Now, in the model summary, as expected, I will get the hazard ratios for the three ancillary parameters as gamma1(X_1(a) vs X_1(b)),gamma2(X_1(a) vs X_1(b)),gamma3(X_1(a) vs X_1(b)).
My confusion arises from the timepoints used by the model to get the three hazard ratios. As we have two knots (which by default are at 33% and 67% quantiles of log(time)), do they play any role in the timepoint selection for the hazard ratios?
