Solved – Calculating constant hazards in exponential survival distributions in R using survreg()

hazardrsurvival

In an ecological seed removal experiment, we have seed removal data from 720 seed plates with 25 seeds each. For each plate, we know the number of "surviving" seeds at several times ti until the end of the experiment. This results in right-censored data and a survival object for each plate that looks like this: 

> s1<-with(Site_80D1, Surv(Gone,Status))
> s1
 [1]  66  128  128  128  183  183  183  183  183  183  183  183  183  183  247  247 
[17] 247  247  367  367  367  367  367  367+ 367+

using survreg(), we estimate a parametric model assuming exponential survival distribution (and thus a constant hazard $h(t)=λ$).

> model1<-survreg(s1 ~ 1, dist="exponential")
> summary(model1)

Call:
survreg(formula = s1 ~ 1, dist = "exponential")
            Value Std. Error    z         p
(Intercept)  5.54      0.209 26.6 2.43e-155

Scale fixed at 1 

Exponential distribution
Loglik(model)= -150.3   Loglik(intercept only)= -150.3
Number of Newton-Raphson Iterations: 3 
n= 25

assuming that $h(t) = λ = \exp( -\text{Intercept})$, we calculate $λ$ by

> exp(-(summary(model1)$table[,1]))
[1] 0.00394038

The goal is to eventually be able to gather constant hazards for all 720 plates and conduct a LME (since the plates are grouped on 2 different spatial scales) including predictors that might act on the hazard as dependent variable.

My question now is: Am I correct in assuming that $h(t) = λ = \exp( -\text{Intercept})$ returns the constant hazard rate?

Thanks a lot in advance, and bear with me since this is my first time to go to CrossValidated for help on statistics.

Best Answer

Yes that should be the estimate of the constant hazard rate.

To my understanding, the model is of the form $\log T = \alpha + W$, so $\alpha$ should represent the log of the (population) mean survival time. For an exponential model at least, 1/mean.survival will be the hazard rate, so I believe you're correct. As a result, $\exp(-\hat{\alpha})$ should be the MLE of the constant hazard rate.

Presumably those times are days, in which case that estimate would be the instantaneous hazard rate (on the per-day scale). [edit: no turns out they're minutes so substitute minute for day everywhere above]

Related Question