Proportional hazard model is a generic term for models (particularly survival models in medicine) that have the form
| L(t | x1, x2, ¼, xn) = h(t) exp(b1 x1 + ¼+ bn xn), |
where L is the hazard function or hazard rate, {xi} are covariates, {bi} are coefficients of the model – effects of the corresponding covariates, and h(t) gives the effect of duration on the hazard rate.
In a proportional hazard model, the effect of an independent variable on the hazard rate is assumed to be multiplicative. For example, the variable “smoking” in a model might have the effect of increasing the hazard rate 30%.
Examples of proportional hazard model are exponential, Weibull, and Gompertz models given respectively by
| h(t) = 1; |
| h(t) = exp(g ln(t)); |
| h(t) = exp(gt). |
Cox proposed an ingenious principle for estimating all proportional hazard models without knowing the function h(t) or even the base hazard rate h0(t). Using this principle one estimates the effects {bi} of the covariates {xi}, but not the effect of duration h(t).
This is known as the Cox Proportional Hazard Model.
If the symbols do not display properly, try
the graphic version of this page