Readme file

SERIES C  
Applied Statistics

A unified model for covariate measurement adjustment in an occupational health study while accounting for non-detectable exposures, by K. A. Wannemuehler and R. H. Lyles et al
Journal of the Royal Statistical Society, Series C, Applied Statistics, Volume 54 (2005) part 1, 259-271

Purpose: Use of PROC NLMIXED (SAS v. 8.01) to fit likelihood to address both associated measurement error and non-detectable exposures. The trust region method is utilized in this example.

data:
data1 is a simulated dataset (beta=0.2) from a single group of subjects.
Each subject has 1 response measurement (resp_ind=1)
and between 2 & 6 exposure measurements (resp_ind=0).
Approximately 15% of the exposure measurements fall below a limit of detection (in this example LOD = -1.07 on the natural log scale).

Data dictionary:
subj = Subject ID
n_obs = Record number for a given subject
resp_ind = indicates the record that contains the response measurement (ycen)
cens_ind = indicates whether the exposure measurement (resp_ind=0) is observed
      (cens_ind=0) or fell below the LOD (cens_ind=1) ycen = response measurement when resp_ind=1
      = exposure measurement on the natural log scale when resp_ind=0
mnxcen = the mean of the exposure measurements on the original scale
age = covariate age in years
smoke = covariate indicating whether subject is a smoker

Syntax: nlmixed_syntax.txt

To use the syntax, read or import the data file, data1, into SAS. To obtain starting values for the exposure model (sigsqb, sigsqw, mu), fit a random intercept model to the exposure measurements. To obtain the starting values for the response model (alpha, beta, gam1, gam2, sigsq) fit a regression model with the surrogate expsosure (mean of the exposure measurements on the original scale) and the covariates age and smoke. These starting values are then given to NLMIXED via the parms statement. These values have been included already.

Likelihood function is based on a simplified version of equation 5 (drop 'g')
with f* from equation 6 replacing f(R_i|delta_i,C_i,Theta) in eq 5.

K. A. Wannemuehler
Biostatistics Department
Rollins School of Public Health
Grace Crum Rollins Building
Emory University
1518 Clifton Road
Atlanta
GA 30322
USA

E-mail: kwannem@sph.emory.edu

• Datasets (data1.txt, size - 8KB)
• Datasets (nlmixed_syntax.txt, size - 3KB)

SERIES A
Statistics in Society

SERIES B
Statistical Methodology

SERIES C
Applied Statistics

SERIES D
The Statistician