From 407600ab at student.eur.nl Wed Nov 29 18:14:57 2017 From: 407600ab at student.eur.nl (Lisa van der Burgh) Date: Wed, 29 Nov 2017 17:14:57 +0000 Subject: [R-sig-Epi] Question about AFT models Message-ID: Hello everyone, I am doing a survival analysis at the moment. I want to make an AFT model in R. I have fitted the following model: fit1 <- survreg(Surv(TimeVentilator, event) ~ (age + sex + mutation + BM1 + BM2)^2, data = DF, dist = "weibull") summary(fit1) I want to test if the weibull distribution is the good distribution. I tested this in the following way: fits <- fit1$linear.predictors resids <- (fit1$y[, 1] - fits) / fit1$scale resKM <- survfit(Surv(resids, event = ) ~ 1, data = DF) plot(resKM, mark.time = FALSE, xlab = "AFT Residuals", ylab = "Survival Probability", main = "Dataset: Pompe Disease") xx <- seq(min(resids), max(resids), length.out = 35) yy <- exp(- exp(xx)) lines(xx, yy, col = "red", lwd = 2) I saw that the weibull distribution is not a good distribution. I found that I can also test other distributions by changing the yy command. I found this: if you have fitted an AFT models assuming the log-normal distribution, you should use yy <- pnorm(xx, lower.tail = FALSE), and if you have fitted an AFT model assuming the log-logistic distribution, you should use yy <- plogis(xx, lower.tail = FALSE). So I tried that, but still, the distributions do not fit my data. Can you help me how to change the yy so that I can test other distributions for the AFT model? Like the students t distribution. I found the possible distributions, but I cannot find the R code for this. I really hope that someone can help me with this! Thankyou in advance, Best, Lisa [[alternative HTML version deleted]]