BIOSTAT / EPI 537
06 Jan 2009
OUTLINE

Introductions

Instructor

Teaching Assistant(s)

Students

Intro to Longitudinal Studies

Event times (Biostat 537)

Repeated measures (Biostat 540)

Biostat 537 – Course Overview

Introduction to Survival Data

Examples

Functions used to summarize distribution of times
My Introduction:

My statistical research does focus primarily on longitudinal studies – both repeated measures issues and survival.

Change is interesting!

Research and Teaching

Interested in B537/B540 in order to “bring together” and perhaps write a book.
Longitudinal Study – VPS

Prospective study with interest in both timetoevent and longitudinal health status measures.

Event = timeuntilAIDS

discrete time

some dropout

Longitudinal = knowledge
Longitudinal Study – Mayo PBC Data

outcome = timeuntildeath

Different research questions:

Compare two treatment groups

Predictive model?

Accuracy of predictive model/marker?
Longitudinal Study – MACS data

Prospective cohort study

First Event: Incident infection (time scale?)

Second Event: AIDS or death given infected = t0

Longitudinal analysis of changes in immune status as part of natural history of infection and/or in response to treatment.

Idea of timedependent covariates – can the current value of CD4 or viral load be used to predict risk of death?
Longitudinal Study – Breast Cancer Data

Prospective study subsequent to BC diagnosis

Event = timeuntildeath, but this is relatively rare (censoring!)

Some delay to enrollment (left truncation)

Compare the predictive ability of different cytometry measures.
Longitudinal Study – Cystic Fibrosis Registry

National registry of CF patients

Event = death

Goal = predict mortality, perhaps to guide selection of patients for lung transplantation.

Logistic regression with 5year status (0=alive; 1=dead)

Survival analysis using timedependent covariates to model the hazard of death.

Longitudinal analysis looks at performance status, FEV1, and changes over time – are some groups at great risk of decline in pulmonary function?
Summary

CHANGE

In a monotone discrete status such as vital status, or disease (first occurrence).

In continuous or discrete characteristics that are measured at regular time periods (e.g. yearly).

Survival methods are concerned with

Summarizing risk of death using hazard regression methods

Summarizing cumulative fraction that have died using survival curve methods (KaplanMeier)

Missing data from censoring!

Covariates may change with time (be careful!)

Longitudinal methods are concerned with

Summarizing changes in the mean over time and group.

Characterizing the sources of variation: between groups represented by covariate values (systematic); among individuals (random effects); and withinindividuals over time (error, or drift).
Overview of Biostat 537

Contact hours: (2) lectures; (1) discussion section starting 13 Jan 2008.

Evaluation: (2) Quizes (45 min); midterm (takehome) and final (inclass).

Weekly exercises

Discussion section = review key content, discuss exercise details and/or issues.

Web page
Introduction to Censored Survival Data – Overview

Survival time, observation time, and censoring indicator.

Causespecific and issues…

Hazard

Censoring and truncation

Recurrent events
Regression Methods

Review linear regression ideas

RANDOM = normal, equal variance

SYSTEMATIC = mean as fnx of X

Review logistic regression

RANDOM = binomial, Bernoulli

SYSTEMATIC = log odds as fnx of X

Typical survival regression

RANDOM = general distribution (sometimes parametric)

SYSTEMATIC = log hazard as function of time and X
Mathematical Summaries of a Population of Event Times
SUMMARY for LECTURE 1

Longitudinal studies generally offer two analyses that focus on time: survival; and repeated measures.

Survival data have a number of features that require some special methods:

censoring

causespecific events

models for time (or avoid it!)

Mathematically we can focus on any one of (5) key summaries – but for survival data we focus on the survival curve and the hazard function – much more about both of these
