This chapter contains examples that illustrate the use of PEANUTS for the analysis of ungrouped cohort survival data using partial likelihood methods for proportional hazards models. PEANUTS can be used to fit a broad class of semiparametric hazard function models with or without stratification. These models include the common exponential (often called multiplicative) and additive relative risk models. The general class of risk models available in PEANUTS was described in Fitting Models with Epicure. EPICURE transformations can be used to define time-dependent covariates, including time-dependent categorical variables. PEANUTS allows for late entry (left censored) data and can be used to analyze case-cohort studies (Prentice, 1986). Once a model has been fit, PEANUTS can be used to estimate the baseline hazard and can plot survival curves, hazard functions, and cumulative hazard functions. To facilitate the production of higher quality plots, lifetable and hazard function estimates produced by PEANUTS can be saved in various formats and text files, for use in other statistical or graphics packages. All of the EPICURE hypothesis testing and bound computation commands are available in PEANUTS.
AMFIT can be used to fit proportional and nonproportional hazards models to grouped cohort survival data using Poisson regression methods. These methods are particularly useful when working with large cohorts and time-dependent covariates or when you want to work with fully specified parametric hazard function models.
The script files for these examples should be located in My Documents/Epicure/examples/peanutsex directory. The data files are located in My Documents/Epicure/examples/exdata directory.Several data sets are used to illustrate the use of PEANUTS. The first few examples make use of data from a clinical trial involving leukemia patients in remission (data from Feigl as described by Kalbfleisch and Prentice, 1980) to illustrate basic principles of risk estimation in PEANUTS. The next few examples illustrate the use of more complex models involving stratification, hazard function estimation, and some of the plotting options. The data are from the Veteran's Administration Lung Cancer Trial (Kalbfleisch and Prentice, 1980, pp 60-62 and 89) in which patients with advanced, inoperable lung cancer were randomly assigned to either standard or test regimes. We will also use these data to demonstrate the use of alternatives to the exponential (multiplicative) relative risk model. The next series of examples use the Stanford heart transplant data (Clark et al, 1971; Crowley and Hu, 1977) to provide additional information on the use of time-dependent covariates in PEANUTS. In these examples we reproduce some of the analyses of these data originally carried out by Crowley and Hu (1977), Kalbfleisch and Prentice (1980), and Aitkin et al (1983). Case-cohort methods are illustrated using data from the study of breast cancer incidence among women exposed to radiation during treatment for tuberculosis (Boice et al, 1990).
Technical details regarding the likelihood function, handling of tied failure times, and hazard and survival probability estimates are given in Technical Details.