An Application of a Bayesian Approach in Diagnostic Testing Problems in the Absence of a Gold Standard
- PDF / 11,851,071 Bytes
- 10 Pages / 612 x 792 pts (letter) Page_size
- 100 Downloads / 137 Views
An Application of a Bayesian Approach in Diagnostic Testing Problems in the Absence of a Gold Standard A number of methods have been proposed to
Faiz Ahmad
GlaxoSrnithKline, PIC. Research Trianale Park. North Carolina
estimate the parameters of diagnostic tests in the absence of a pafect gdd standard. However, these approaches requireat least three diagnostic tests to estimate the parameters and most of these methods are inapplicable in point estimation problems involving - nonidmtifiable parameters. In this article, we describe a method to estimate the properties of various tests for Chla-
Key Words Bayesian analysis; Diagnostic tests; Prevalence; Sensitivity; Specificity
Correspondence Address Faiz Ahmad, GlaxoSmithKline PIC. 5 Moore Drive, MA1 C 3604,3B, Research Triangle Park, NC 27709 (ernail: fai2.f. [email protected]).
INTRODUCTION In many real diagnostic testing situations, perfect gold standard tests either are very expensive or do not exist for detecting a given disease or condition. Therefore, in practice, one or more imperfect tests are used. Employing imperfect diagnostic tests without appropriate statistical adjustment introduces bias in estimating properties relating to the diagnostic tests and the population prevalence, due to misclassification. A number of methods have been proposed to estimate the parameters of diagnostic tests in the absence of a perfect gold standard. Methods such as discrepant analysis (1) and latent class analysis (2-4) have been used in recent literature to estimate disease condition and diagnostic performance in the absence of a perfect gold standard. These approaches require at least three diagnostic tests to estimate the parameters, and most of these methods are inapplicable in point estimation problems involving nonidentifiable parameters. In contrast to these non-Bayesian approaches, the Bayesian approach (5) using proper priors can produce a close form, interpretable point estimate in such problems. In the presence of good prior information, a Bayesian approach can provide an improved inference. The Bayesian method is similar to standard latent class
mydia trachomatis. A general approach has been presented to select a range of suitable pnor distributionsfor the diagnostic testing situation required by the Bayesian approach, and the results jiom this method are compared with those obtained jiom other methods in context to the aamydia example. lhe effects of different classes of p'ordistribution such as skeptical, optimistic, and clinical priors on the estimates of the parameters are demonstrated.
analysis but where the use of prior distributions replaces the need for constraints in the case of nonidentifiable situations. Throughout this article, we assume the tests are conditionally independent, meaning that tests are statistically independent conditional on true disease status. Conditional independence is commonly assumed but it is difficult to verify in practice. However, our main purpose is to demonstrate how the Bayesian approach can be used when there is uncert
Data Loading...
