Statistics Glossary → Proportional hazards assumption
Proportional hazards assumptionconcept
The assumption in the Cox model that the risk ratio between two subjects or groups stays constant over time.
The Cox model gives a single hazard ratio for each factor, assuming the ratio is the same at the start and at the end of follow-up. If a treatment halves the risk at first and then loses its effect over time, a single HR does not describe this correctly.
The assumption is checked with Schoenfeld residuals: if the residuals are significantly related to time (p < α), the assumption is violated for that factor. Remedies are stratifying on that factor (a separate baseline hazard per stratum), an interaction of the factor with time, or time-dependent covariates.
An everyday example
If an operation raises the risk in the first months because of complications but lowers it in the long run, the risk ratio between operated and non-operated patients changes over time; a single 'operation HR' hides this changing relationship. The proportional hazards test detects it.
In MerQur
In Cox regression, the 'Proportional hazards test (Schoenfeld)' option with its time transformation (Kaplan-Meier, rank, time, log t); the 'Strata variable (STRATA)' allows stratifying on a violating factor.
Analyses where this concept appears
Cox Regression (Proportional Hazards Model)