9 Checking Proportional Hazards

The Cox model assumes that each covariate multiplies the hazard by the same factor at every time. When this fails, the single hazard ratio the model reports is a kind of average over follow-up, which may describe no period well: a treatment that halves the risk in the first year and does nothing afterwards is reported as a modest constant effect. This section shows how to check the assumption, with plots and with tests based on the Schoenfeld residuals, and what to do when it fails, using the veteran model of the previous section, in which it does fail.

Questions on this section

Stuck on something here? Ask below and it stays attached to this topic.