STK4080 - Autumn 2012

PRACTICAL EXERCISE 12

The Diabetic Retinopathy Study was conducted in the US by the National Eye Institute to assess the effect of laser photocoagulation in delaying onset of severe visual loss ("blindness") in patients with diabetic retinopathy. One eye of each patient was randomly selected for photocoagulation and the other was observed without treatment. The patients were followed over several years for the occurrence of blindness in their left and right eyes. Censoring is caused by death, dropout, or end of the study. We consider only a subset of the original data set containing 197 high risk patients. See Huster et al, 1989, Biometrics, for a discussion and further references.

There are two lines for each patient in the data set: one line for each eye. The variables are: 

  • id: patient number 
  • trteye:   treated eye (1=right 2=left)
  • ageonset:   age at diagnosis of diabetes
  • typediab:  type of diabetes (1= juvenile age at onset < 20 years , 2=adult)
  • status:   status for eye (0=censored, 1=blindness)
  • time:   follow-up time in months
  • trt:   treatment of eye (1=control eye, 2=treated eye)

Before you start, you will have to read the data. Use the R-command:

 

eye=read.table("http://folk.uio.no/borgan/BGC1-2012/data/retinopathy.txt", header=T)

 

You also need to install and load the parfm library, which fits parametric frailty models for clustered and recurrent event data.

 

We start out by only using treatment as a covariate.

 

a) We first fit a model with Weibull baseline and no frailty by the command:

 

parfm(Surv(time,status)~factor(trt),data=eye)

 

Perform the command, and make sure that you understand what the output tells you.

 

b) We then fit a model with Weibull baseline and gamma frailty:

 

parfm(Surv(time,status)~factor(trt),cluster="id", frailty="gamma", data=eye)

 

Perform the command, and compare with the result for the model without frailty. Is there a significant frailty effect?

 

c) Fit a model with gamma frailty and treatment, age at onset and type of diabetes as covariates. Is there a significant effect of type of diabetes and age at onset?