I every body!

Before to present my problem I apologize that I creates the same topic on the forum stackoverflow, but I think the best place for my question was here.

I'm a PhD student and I'm working on survival data currently, but I don't feel comfortable with survival analysis. I have to correlated problem I think I try to explain it in details

I my case, I'm working with samples in 6 cities which are in 2 treatments conditions (control and chemical stress). I haven't censored data.

The biological question is how genetic diversity (illustrated by the city) can interact with environment modifications (the treatment)?

To answer I created a survival parametric model with weibull distribution (AIC method to evaluate best models).

And I tested first if in control condition only, survival was different.

    data <- subset(survival.data, survival.data$treatment=="treated")
    surv <- with(data, Surv(time,event))
    weibull.model <- survreg(surv~city,dist="weibull", data=data)

The ouput model confirm significant differences (p.val<0.05) and this is my first problem. If I want to compare differences between cities, how can I be sure this is a treatment effect or just because a lineage have a longer longevity in mean? My idea was to correct it by the maximum lifespan mean for every city before to use the model, but I didn't see any information for that. I discovered the relative survival but I think it's not my case. I tried it and I transformed my data which are represented as % of total life per city [treated]/[max(untreated)]*100

If I resolved this point next analysis should be differents

If not, I will work withclassical survival data but I want to test the interaction term which represent the genotype x environnement interaction With ANOVA I can have this term and use a post-hoc test as TukeyHSD function in R, but here the output of the model use a parameter of the factors as reference and I can't have the interaction.

coxph(formula = surv~treatment*city,dist="weibull", data=data)

                               coef exp(coef) se(coef)     z       p
treatmentpq                   5.178   177.335    0.568  9.11 < 2e-16
cityparis                    -0.674     0.510    0.866 -0.78 0.43671
citysapporo                   2.354    10.524    0.539  4.37 1.2e-05
citysokol                     0.277     1.320    0.671  0.41 0.67931
citytokyo                     1.893     6.641    0.553  3.42 0.00062
citywatsonville               1.389     4.009    0.572  2.43 0.01522
treatmentpq:cityparis         0.644     1.904    0.904  0.71 0.47630
treatmentpq:citysapporo      -1.707     0.181    0.601 -2.84 0.00450
treatmentpq:citysokol        -1.085     0.338    0.720 -1.51 0.13191
treatmentpq:citytokyo        -2.098     0.123    0.613 -3.42 0.00062
treatmentpq:citywatsonville  -0.542     0.581    0.631 -0.86 0.39009

Likelihood ratio test=488.6  on 11 df, p=<2e-16
n= 360, number of events= 249 

In forum, publications or google I didn't see this type of analysis... Is it possible with survival data ? Maybe I can modelize it with glm approach but i'm not sure.

Thank you! ps: Sorry for English native speaker if I'm unclear...


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.