Abstract:
|
A new screening test is often evaluated with a cross-sectional study. However, programmatic performance of the test, that is, its benefits and risks with repeated testing at regular intervals, is also important to characterize. For programmatic performance, we consider cumulative probabilities of false positive and true positive test results. For subjects testing negative at baseline, discrimination (sensitivity, specificity) and predictive value (positive and negative) at the next screen are also considered. Typically, not all subjects are verified for disease status, especially when verification is invasive (e.g., colonoscopy for a colon cancer screening test). Assuming verification is missing at random, inverse probability weighted estimators are considered. Cumulative risk of disease by baseline test result (negative, positive) is also useful but difficult to estimate because time to disease onset is censored in an imprecise interval due to potential false positive and false negative test and verification errors at the screening times. Augmenting the data with predictive draws of time to disease onset simplifies estimation.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.