I'm trying to evaluate the toxicity of a certain treatment over time. I have 2 independent groups and I apply a treatment to one, and nothing to the second one. I thought about using logistic regression but since in the control group all of them are alive at the end, I would have a problem of separation, so I was trying to look for an alternative.
I've 6 time points (0h,2h,4h,6h,8h and 24h) and the data is binary (dead=0, alive=1). I was wondering if I could use the COX regression model to test if this treatment is toxic for this population over time, using R:
res.cox <- coxph(Surv( time, alive_dead) ~treatment, data=data_aliveD) summary(res.cox)
I get this:
Call: coxph(formula = Surv(time, alive_dead) ~ treatment, data = data_aliveD) n= 720, number of events= 654 coef exp(coef) se(coef) z Pr(>|z|) treatment T -0.30952 0.73380 0.07952 -3.893 9.92e-05 *** --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 exp(coef) exp(-coef) lower .95 upper .95 treatment T 0.7338 1.363 0.6279 0.8576 Concordance= 0.514 (se = 0.014 ) Rsquare= 0.021 (max possible= 1 ) Likelihood ratio test= 15.27 on 1 df, p=9e-05 Wald test = 15.15 on 1 df, p=1e-04 Score (logrank) test = 15.27 on 1 df, p=9e-05
Also I would like to know what is the difference between Cox Regression Model and Log-Rank test. If I use Long-rank test I have the following output:
log.rank <- survdiff(Surv( time, alive_dead) ~treatment, data=data_aliveD)
Call: survdiff(formula = Surv(time, alive_dead) ~ treatment, data = data_aliveD) N Observed Expected (O-E)^2/E (O-E)^2/V treatment=H2O 360 360 327 3.33 9.7 treatment=T 360 294 327 3.33 9.7 Chisq= 9.7 on 1 degrees of freedom, p= 0.002