Transcript Powerpoint Slides - University of Pittsburgh

Methodological Research
and Collaboration
Lan Kong
Assistant professor
Department of Biostatistics
April 6, 2016
Faculty Research Interest Seminar Series
Graduate School of Public Health, UNIVERSITY of PITTSBURGH
Outline
Introduction
Methodological Research
Survival analysis
Clinical trials
Collaboration
Projects
Motivated problems
Dept of Biostatistics, GSPH, UNIVERSITY of PITTSBRGH
Introduction
Joined the Biostat dept in August 2003
PhD in Biostat from UNC-Chapel Hill
In collaboration with critical care
medicine dept.
On doctoral exam committee
Courses intend to teach: survival analysis,
advanced categorical data analysis, or
estimating equation method.
Dept of Biostatistics, GSPH, UNIVERSITY of PITTSBRGH
Methodological research
Survival analysis
Study design: case-cohort
Model: semiparametric transformation
models and accelerated failure time model.
Approach: estimating equation method,
inverse of probability weighting technique
Statistical theory: U-statistics, finite
population sampling, martingales,
empirical/stochastic process
Dept of Biostatistics, GSPH, UNIVERSITY of PITTSBRGH
Case-cohort design (Prentice, 1986)
Subcohort
Full cohort
Full cohort
Failures/cases
Complete data available for Subcohort + Additional Cases outside the
subcohort
Dept of Biostatistics, GSPH, UNIVERSITY of PITTSBRGH
Semiparametric Survival Models
Cox model
λ(t)=λ0(t)exp(β’Z), λ0(t) unspecified
Transformation models
h(T)= -β’Z+ε OR g{Sz(t)}=h(t)+β’Z
h unknown, ε has known CDF F(.), g-1=1-F.
Accelerated failure time model
log(T)= -β’Z+ε, ε has unspecified CDF.
Dept of Biostatistics, GSPH, UNIVERSITY of PITTSBRGH
Methodological Research (cont.)
Clinical trials
Problem: Multiplicity issues
Example:
Multi-dose clinical trials--correlations
among multiple comparisons
Statistical concerns: familywise Type I
error and power
Dept of Biostatistics, GSPH, UNIVERSITY of PITTSBRGH
Family-Wise Error (FWE)
Probability that at least one hypothesis is
incorrectly declared significant.
Strongly control of FWE:
FWE is protected for any composite null
hypothesis
Closed testing procedures strongly control
the FWE
Dept of Biostatistics, GSPH, UNIVERSITY of PITTSBRGH
Multiple testing procedures
Analysis strategies:
Perform a sequence of tests in a pre-specified
order through the closed testing principle :
1. Global assessment of any dose effect
2. Comparisons of doses to Placebo
3. Other comparisons among doses
Manage multiplicity within the respective
steps by Hochberg method (BKA, 1988), closed
testing procedure.
Dept of Biostatistics, GSPH, UNIVERSITY of PITTSBRGH
Vaccine trials
Problem:
Inferiority/Equivalence assessment
Multiplicity due to multiple endpoints
(immune responses)
Statistical concerns:
How correlations among endpoints affect
study design of inferiority/EQ trials (power,
sample size)?
Dept of Biostatistics, GSPH, UNIVERSITY of PITTSBRGH
Hypotheses
H0 (inf)
H1 (noninf)
T-C
-K
H0 (noneq)
0
H1 (eq)
H0 (noneq)
T-C
-K
0
H0 (inf)
K
H1 (sup)
T-C
0
T: treatment group, C: control group
Dept of Biostatistics, GSPH, UNIVERSITY of PITTSBRGH
Further questions of interest
How to handle multiple endpoints in
multi-dose clinical trials?
How to simultaneously assess
inferiority/EQ on some endpoints and
superiority on the others?
Dept of Biostatistics, GSPH, UNIVERSITY of PITTSBRGH
Collaboration
Projects mainly involved:
Genetic and inflammatory markers of
Sepsis (GenIMS)
Economic Analysis of the Pulmonary
Artery Catheter Use (EA-PAC)
Prolonged Outcomes of Nitric Oxide for
Ventilated Premature Babies (PRONOX)
Dept of Biostatistics, GSPH, UNIVERSITY of PITTSBRGH
Statistical problems motivated
from collaborative projects
Dept of Biostatistics, GSPH, UNIVERSITY of PITTSBRGH
Genetic data
Explore how the candidate markers
interactively affect the outcome (Selection
of subset from a list of candidate genetic
markers)
Statistical learning method
Pattern recognition approach
Haplotype analysis for survival data to
accommodate missing genotypes, casecohort design
Dept of Biostatistics, GSPH, UNIVERSITY of PITTSBRGH
Missing and/or truncated data
Examples: inflammatory marker data are below
detectable limit, the measurement on the severity
of sepsis are heavily missing
Longitudinal analysis of truncated inflammatory
marker (extension of Tobit model)
Jointly modeling several truncated inflammatory
markers
Testing whether the missing is informative in the
non-monotone missing pattern
Jointly modeling the longitudinal outcome and
longitudinal covariates
Dept of Biostatistics, GSPH, UNIVERSITY of PITTSBRGH
Quality of Life and Cost data
Various types of outcomes: survival data, cost
data, quality of life data.
Informative censoring
Modeling quality adjusted survival
Jointly modeling QOL with survival data
Cost-effective analysis in presence of repeated
measures and missing data
Dept of Biostatistics, GSPH, UNIVERSITY of PITTSBRGH
Related papers
Kong L, Cai J, Sen PK. Weighted estimating equations
for semiparametric transformation models with
censored data from a case-cohort design.
Biometrika 2004; 91(2):305-319.
Kong L, Cai J, Sen PK. Asymptotic results for fitting
semiparametric transformation models to failure
time data from a case-cohort design. Statistica
Sinica 2004, in press.
Kong L. Analysis of case-cohort data with accelerated
failure time model, in preparation
Dept of Biostatistics, GSPH, UNIVERSITY of PITTSBRGH
Related papers (cont.)
Kong L, Kohberger R, Koch G. Type I error and power
in non-inferiority/equivalence trials with
correlated multiple endpoints: an example from
vaccine development trials. Journal of
Biopharmaceutical Statistics 2004; 14(4):893-907.
Kong L, Koch G, Liu T, Wang H. Performance of some
multiple testing procedures to compare three doses
of a test drug and placebo. Pharmaceutical
Statistics 2004, in press.
Kong L, Kohberger R, Koch G. Equivalence/noninferiority assessment on multiple proportion
outcomes. In preparation.
Dept of Biostatistics, GSPH, UNIVERSITY of PITTSBRGH