Skip to main content

Table 2 Association between Homocysteine and Serum Cotinine in Non-Smokers: Results by Different Analytical Methods for Analyzing Left-Censored Biomarker Data

From: Toward improved statistical methods for analyzing Cotinine-Biomarker health association data

Total n = 9,488

 

Regression Models

 

#subjects included

Univariate*

Multiple*

Method

 

Estimate (SE)

Estimate/SE

p-value

Estimate (SE)

Estimate/SE

p-value

Complete Case

5,865

0.027 (0.011)

2.455

0.0137

0.053 (0.009)

5.889

< 0.0001

Single Imputation with

   0

9,488

0.055 (0.011)

5.000

< 0.0001

0.083 (0.008)

10.375

< 0.0001

   LOD

9,488

0.053 (0.011)

4.818

< 0.0001

0.079 (0.009)

8.778

< 0.0001

   LOD/sqrt(2)

9,488

0.054 (0.011)

4.909

< 0.0001

0.081 (0.009)

9.000

< 0.0001

Multiple Imputation

9,488

0.001 (0.025)

0.040

0.9787

0.020 (0.026)

0.769

0.4367

Logistic Regression***

9,488

0.524 (0.055)

9.527

< 0.0001

1.093 (0.076)

14.382

< 0.0001

"Reverse" Kaplan-Meier ***

9,488

0.012 (0.015)

0.800

< 0.0001

0.222 (0.017)

13.059

< 0.0001

  1. LOD: limit of detection; sqrt: squared root
  2. Estimate (SE): regression coefficient and its standard error between Serum Cotinine and Homocysteine
  3. *Univariate regression models for inflammatory marker Homocysteine include serum cotinine only as a covariate; Type-I error rate = 5%.
  4. **Multiple regression models for inflammatory marker Homocysteine include serum cotinine, age in years, gender (female/male), race/ethnicity (non-Hispanic White/Others), and second hand smoking status (yes/no) as covariates; Type-I error rate = 5%.
  5. *** Outcome of these methods is left-censored serum cotinine.
Back to article page

Tobacco Induced Diseases

ISSN: 1617-9625

Contact us