![[ Visit Client Website ]](images/banner.gif)
Wednesday, October 22, 2008
Columbus A-C (Hyatt Regency Penns Landing)
Purpose: With the rising importance of evidence-based medicine, there is a need to provide accurate estimates of test characteristics for diagnostic tests. The purpose of this study was to develop and evaluate Bayesian generalized linear models (BGLMs) for meta-analysis of diagnostic tests.
Methods: Three BGLMs were developed. Two of these models, a Bayesian bivariate normal model (BBNM) and a Bayesian bivariate binomial model (BBBM), were created to analyze pairs of sensitivity and specificity values while incorporating the correlation between these two outcome variables. Noninformative independent uniform priors were used for the variance of sensitivity, specificity and correlation. We also applied an inverse Wishart prior to check the sensitivity of the results. The third model was a multinomial model (BMNM) in which the test results were modeled as multinomial random variables. Extensive simulation studies were conducted to investigate the properties of the three proposed models. In order to make direct comparisons between these Bayesian models and frequentist approaches, we applied the same eight simulation scenarios described in a published study. All datasets were simulated using the binomial model. Statistical computations were conducted in WinBUGS 1.4.2 using the ‘Bayesian inference using Gibbs sampling’ implementation of Markov chain Monte Carlo techniques.
Results: Point estimates of sensitivity and specificity from BBNM and BBBM were close to the true values and insensitive to change in the priors on the covariance matrix. The standard errors obtained from BBBM were smaller than those of BBNM. BBBM with uniform priors provided good estimates of correlation in eight different scenarios. Both point estimates of sensitivity and specificity from the BBNM were better than the results obtained by its frequentist counterpart, while the BMNM underestimated the outcomes consistently.
Conclusions: The BBBM provides the most flexible framework for future applications because of its following strengths: (1) it accounts for all sources of uncertainty and provides probability intervals for all parameters; (2) it facilitates direct comparison between different tests; (3) it captures the variability in both sensitivity and specificity simultaneously as well as the intercorrelation between the two; and (4) it can be directly applied to sparse data without ad hoc correction.
Methods: Three BGLMs were developed. Two of these models, a Bayesian bivariate normal model (BBNM) and a Bayesian bivariate binomial model (BBBM), were created to analyze pairs of sensitivity and specificity values while incorporating the correlation between these two outcome variables. Noninformative independent uniform priors were used for the variance of sensitivity, specificity and correlation. We also applied an inverse Wishart prior to check the sensitivity of the results. The third model was a multinomial model (BMNM) in which the test results were modeled as multinomial random variables. Extensive simulation studies were conducted to investigate the properties of the three proposed models. In order to make direct comparisons between these Bayesian models and frequentist approaches, we applied the same eight simulation scenarios described in a published study. All datasets were simulated using the binomial model. Statistical computations were conducted in WinBUGS 1.4.2 using the ‘Bayesian inference using Gibbs sampling’ implementation of Markov chain Monte Carlo techniques.
Results: Point estimates of sensitivity and specificity from BBNM and BBBM were close to the true values and insensitive to change in the priors on the covariance matrix. The standard errors obtained from BBBM were smaller than those of BBNM. BBBM with uniform priors provided good estimates of correlation in eight different scenarios. Both point estimates of sensitivity and specificity from the BBNM were better than the results obtained by its frequentist counterpart, while the BMNM underestimated the outcomes consistently.
Conclusions: The BBBM provides the most flexible framework for future applications because of its following strengths: (1) it accounts for all sources of uncertainty and provides probability intervals for all parameters; (2) it facilitates direct comparison between different tests; (3) it captures the variability in both sensitivity and specificity simultaneously as well as the intercorrelation between the two; and (4) it can be directly applied to sparse data without ad hoc correction.
See more of: Poster Session V
See more of: 30th Annual Meeting of the Society for Medical Decision Making (October 19-22, 2008)