Efficient parameter estimation in longitudinal data analysis using a hybrid GEE method

School of Economics, Singapore Management University, 90 Stamford Road, Singapore.
Biostatistics (Impact Factor: 2.65). 05/2009; 10(3):436-45. DOI: 10.1093/biostatistics/kxp002
Source: PubMed


The method of generalized estimating equations (GEEs) provides consistent estimates of the regression parameters in a marginal regression model for longitudinal data, even when the working correlation model is misspecified (Liang and Zeger, 1986). However, the efficiency of a GEE estimate can be seriously affected by the choice of the working correlation model. This study addresses this problem by proposing a hybrid method that combines multiple GEEs based on different working correlation models, using the empirical likelihood method (Qin and Lawless, 1994). Analyses show that this hybrid method is more efficient than a GEE using a misspecified working correlation model. Furthermore, if one of the working correlation structures correctly models the within-subject correlations, then this hybrid method provides the most efficient parameter estimates. In simulations, the hybrid method's finite-sample performance is superior to a GEE under any of the commonly used working correlation models and is almost fully efficient in all scenarios studied. The hybrid method is illustrated using data from a longitudinal study of the respiratory infection rates in 275 Indonesian children.

Download full-text


Available from: You-Gan Wang, Feb 23, 2015
29 Reads
  • Source
    • "The findings in our Theorem 1 show that it is meaningful to investigate how the within-subject correlation or the within-block dependence affect the efficiency of the EL inference. More recently, by using the EL method, Leung et al. (2009) proposed a hybrid method that combined multiple GEEs based on different working correlation structures. They showed that is more robust and efficient for the choice of the working correlation structures. "
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper, we introduce the empirical likelihood (EL) method to longitudinal studies. By considering the dependence within subjects in the auxiliary random vectors, we propose a new weighted empirical likelihood (WEL) inference for generalized linear models with longitudinal data. We show that the weighted empirical likelihood ratio always follows an asymptotically standard chi-squared distribution no matter which working weight matrix that we have chosen, but a well chosen working weight matrix can improve the efficiency of statistical inference. Simulations are conducted to demonstrate the accuracy and efficiency of our proposed WEL method, and a real data set is used to illustrate the proposed method.
    Journal of Statistical Planning and Inference 11/2010; 140(11):3446-3456. DOI:10.1016/j.jspi.2010.05.007 · 0.68 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Health authorities in several countries recently recommended the expansion of human immunodeficiency virus (HIV) antibody testing, including the use of rapid tests. Several HIV rapid tests are now licensed in Europe but their sensitivity on total blood and/or oral fluid in routine healthcare settings is not known. 200 adults with documented HIV-1 (n=194) or HIV-2 infection (n=6) were prospectively screened with five HIV rapid tests using either oral fluid (OF) or finger-stick whole blood (FSB). The OraQuick Advance rapid HIV1/2 was first applied to OF and then to FSB, while the other tests were applied to FSB, in the following order: Vikia HIV 1/2, Determine HIV 1-2, Determine HIV-1/2 Ag/Ab Combo and INSTI HIV-1/HIV-2. Tests negative on FSB were repeated on paired serum samples. Twenty randomly selected HIV-seronegative subjects served as controls, and the results were read blindly. Most patients had HIV-1 subtype B infection (63.3%) and most were on antiretroviral therapy (68.5%). Sensitivity was 86.5%, 94.5%, 98.5%, 94.9%, 95.8% and 99% respectively, with OraQuick OF, OraQuick FSB, Vikia, Determine, Determine Ag/Ab Combo and INSTI (p<0.0001). OraQuick was less sensitive on OF than on FSB (p=0.008). Among the six patients with three or more negative tests, two had recent HIV infection and four patients on antiretroviral therapy had undetectable plasma viral load. When patients positive in all the tests were compared with patients who had at least one negative test, only a plasma HIV RNA level<200 cp/ml was significantly associated with a false-negative result (p=0.009). When the 33 rapid tests negative on FSB were repeated on serum, all but six (5 negative with OraQuick, 1 with INSTI) were positive. The sensitivity of OraQuick, Determine and Determine Ag/Ab Combo was significantly better on serum than on FSB (97.5%, p=0.04; 100%, p=0.004; and 100%, p=0.02, respectively). When evaluated in a healthcare setting, rapid HIV tests were less sensitive on oral fluid than on finger-stick whole blood and less sensitive on finger-stick whole blood than on serum.
    PLoS ONE 07/2010; 5(7):e11581. DOI:10.1371/journal.pone.0011581 · 3.23 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: The impact of screening coronary computed tomographic angiography (CCTA) on physician and patient behavior is unclear. We studied asymptomatic patients from a health-screening program. Our study population comprised 1000 patients who underwent CCTA as part of a prior study and a matched control group of 1000 patients who did not. We assessed medication use, secondary test referrals, revascularizations, and cardiovascular events at 90 days and 18 months. A total of 215 patients in the CCTA group had coronary atherosclerosis (CCTA positive). Medication use was increased in the CCTA-positive group compared with both the CCTA-negative (no atherosclerosis) and control groups at 90 days (statin use, 34% vs 5% vs 8%, respectively; aspirin use, 40% vs 5% vs 8%, respectively), and 18 months (statin use, 20% vs 3% vs 6%, respectively; aspirin use, 26% vs 3% vs 6%, respectively). After multivariable risk adjustment, the odds ratios for statin and aspirin use in the CCTA-positive group at 18 months were 3.3 (95% confidence interval [CI], 1.3-8.3) and 4.2 (95% CI, 1.8-9.6), respectively. At 90 days, in the total CCTA group vs controls, there were more secondary tests (55 [5%] vs 22 [2%]; P < .001) and revascularizations (13 [1%] vs 1 [0.1%]; P < .001). One cardiovascular event occurred in each group over 18 months. An abnormal screening CCTA result was predictive of increased aspirin and statin use at 90 days and 18 months, although medication use lessened over time. Screening CCTA was associated with increased invasive testing, without any difference in events at 18 months. Screening CCTA should not be considered a justifiable test at this time.
    Archives of internal medicine 05/2011; 171(14):1260-8. DOI:10.1001/archinternmed.2011.204 · 17.33 Impact Factor
Show more