Stat 547: Survival Analysis

E. Neely Atkinson
Department of Biostatistics, M. D. Anderson Cancer Center

Xuelin Huang
Department of Biostatistics, M. D. Anderson Cancer Center

• Class description:
  • This class covers the basics of survival (time to event) analysis. We will cover basic estimation and testing as well as modeling.
  • The emphasis in this class is on the practical aspects of survival analysis. We will use both SAS and R.
  • The final course grade will be based on homework (approximately one per week), a midterm exam, and a final project. The weighting will be approximately homework 10%, midterm 45%, final 45%.
• Text: Survival Analysis: Techniques for Censored and Truncated Analysis, Klein and Moeschberger, 2nd ed.
• Time and place: 1:00 - 2:20 Wednesdays/Fridays, HB 22.
• Office hours: DH 2099, after class.
• Blog: To facilitate communication among class members, we have established a web log at stat547.wordpress.com. By posting comments you can ask questions, make suggestions, share useful information, etc.
• Homework:
  • Feel free to work in groups. Several people can hand in one assignment.
  • Assignment 1, due Wednesday 1/14/2009. Chap. 2, exercises 2.13, 2.15, 2.17, 2.19 *hint
  • Assignment 2, due Wednesday 1/21/2009. Chap. 3, exercises 3.1, 3.3, 3.5, 3.7
  • Assignment 3, due Wednesday 2/4/2009. Chap. 4, exercises 4.1, 4.3, 4.5, 4.7, 4.9
  • Assignment 4, due Wednesday 2/18/2009. Chap. 7, exercises 7.3 a & b, 7.5, 7.7, 7.9. Hint: all exercises can be done with procedures available in R. If R does not have exactly what the book wants, use the closest equivalent, e.g. Peto-Peto version of the Gehan test or using coxph to test for trend.
  • Assignment 5, due Friday 3/27/2009. Chap. 8, exercises 8.1, 8.2, 8.3
  • Assignment 6, due Wednesday 4/8/2009. Chap. 8, exercise 8.11, Chap. 9, exercises 9.4, 9.5. Note: The data for Exercise 8.11 does not have a variable indicating if infants were breast fed or not at birth. Please use the use the 13th variable, Month the child was weaned, i.e. let Z=0 if Month the child was weaned=0, and Z=1 otherwise. Another confusing point about the data is the time when a child was hospitalized for pneumonia. Please use the first variable (Age child had pneumonia, months), and ignore the last variable (Age child in the hospital for pneumonia, months). If you have done the analysis using the last one, you do not have to redo it. The two variables will give roughly the same conclusions. You can use any tie-handling method, not necessarily the discrete method as instructed by the authors.
• R code:
  • Computing mean residual life: mrl.R
  • Chapter 2 selected exercises: exercises.2.R
  • Chapter 4: section 2, section 3, section 3, section 4, section 5, median, section 6, section 7
  • Chapter 7: section 2, section 3 , log.rank.hypergeometric.R , section 4 , section 5 , section 6 , section 7 , section 7 (continued) , tab.surv.R
  • hazard estimation: est.haz.R , est.haz.2.R , est.haz.3.R , est.haz.4.R , est.haz.5.R , est.haz.6.R
  • Chapter 5: doubly censored data, interval censored data
• Documents for Using SAS and R:
  • Handling time-dependent covariates by R in the Cox model: R_time_dependent_covaraite.doc
  • Handling time-dependent covariates by SAS in the Cox model: Beginner_SAS.doc
  • Handling strata and covariate interactions by R and SAS: strata.doc
  • SAS code for HW6, Exercise 9.4 bmt_time_dep.sas and SAS output results
  • R code for HW6 contributed by Luis Leon Novelo and Roberto Bertolusso: ch08-09.R
  • 4/16/2009: Since the above R code has a problem when there are ties between the times of different types of events, I write a R code to take care of this problem when creating a data set with time-dependent covariates: BMT_time_dep.R. The data set I used: bmt.dat , and the data set created by this code: BMT_time_dep.txt .
  • Another example of handing time-dependent covariates by R TimeDep.Cox from Prof. Eric Slud's web site
  • R/Splus programs for the two approaches for fitting semi-parametric Accelerated Failure Time (AFT) models: least square regression and rank-based from Prof. Zhezhen Jin's web site
• Announcements:
  • I did not mean for the homework for chapter 4 to consume your lives. Do what you can. If a problem is taking too much time, just describe the approach you would take. The point is to learn the techniques. Here are my solutions. They are not pretty and they do not always agree with the book.
  • A helpful student pointed out that R contains a cumsum function, so a lot of my code is excessively verbose. Sorry.
  • The datasets from the text are available from CRAN as package KMsurv. These contain the same errors and inconsistencies as those from the authors's website.
  • I will hand out the midterm Wednesday, February 25. It will be due back Wednesday, March 11 at the beginning of class. It will be take home, open book, open notes, open Internet, etc., but no working together, please.
• Midterm:
  • Here's the midterm, if you want to go ahead and get started. I will have hard copies available in class Wednesday.
  • Here are the solutions to the midterm. We will go over these in class Wednesday, March 18, 2009.
• Final Exam:
  • An optional problem for final exam.