|
Activity Number:
|
200
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 10, 2004 : 9:00 AM to 10:50 AM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #301881 |
|
Title:
|
Predictive Values for Combinations of Diagnostic Tests with Multiple Outcomes
|
|
Author(s):
|
Natalie Khutoryansky*+ and John R. Ventre and Simon Roitt
|
|
Companies:
|
Amersham Health and Amersham Health and Amersham Health
|
|
Address:
|
101 Carnegie Center, Princeton, NJ, 08540,
|
|
Keywords:
|
diagnostic tests ; truth standard ; Bayesian method ; predictive value ; sensitivity ; specificity
|
|
Abstract:
|
In many diagnostic efficacy studies the disease status of each patient is determined using a truth standard. However, sometimes a perfect truth (gold) standard test does not exist or may be invasive or expensive. We consider an approach alternative to a gold standard by a combination of imperfect diagnostic tests. Many authors examined this approach in the case of dichotomous diagnostic tests. We address the case when diagnostic tests can have multiple outcomes (>2). For each sequence of outcomes of consecutive diagnostic tests the total probabilities of disease presence or absence (predictive values) are calculated using a Bayes formula that includes the likelihood ratio of outcomes. The subsequent test can be prescribed if one of the predictive values does not reach a pre-specified level. The outcome likelihood ratios for dichotomous tests can be calculated based on their sensitivity and specificity. However, in the case of multiple outcomes the calculations are more complicated and include probabilities of outcomes under conditions of disease presence or absence. This approach is implemented to calculate predictive values for selected combinations of imaging diagnostic tests.
|
