Propensity Scores in Receiver Operating Characteristic (ROC) Analysis

• Activity Number: 35
• Type: Topic Contributed
• Date/Time: Sunday, August 11, 2002 : 4:00 PM to 5:50 PM
• Sponsor: Biopharmaceutical Section*
• Author(s): Marina Kondratovich*+
• Affiliation(s): Food and Drug Administration
• Keywords: Receiver Operating Characteristic Analysis ; ROC ; Propensity Scores ; Sensitivity ; Specificity ; Diagnostic Accuracy

Abstract

The area under ROC curve (AUC) is often used to summarize the diagnostic capability of the diagnostic test which result is a continuous variable. In many clinical studies, confounding factors (for example, age, body mass index) influence the distributions of the test variable. For example, the mean value of the ultrasound test for the bone status, expressed as a T-score, becomes lower with the increase of age. The diseased population is generally older than the non-diseased population and, therefore, the ability of the diagnostic test (for example, ultrasound test for the bone status) to discriminate these populations is usually overstated because some portion of the difference in the test distributions is due to the difference in age. The author developed weighted ROC analysis based on the propensity-scores methodology, which allows one to calculate the AUC of the diagnostic test without the effect of other potentially confounding variables, and, thus, to estimate correctly the diagnostic capability of the medical device.