A Mixture Model for Determining Diagnostic Cut Points for Levels of Serum Antibodies to Pertussis Toxin

• Activity Number: 425
• Type: Contributed
• Date/Time: Wednesday, August 6, 2003 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Andrew L. Baughman*+ and Kristine Bisgard and Lynn Freyja and Bruce Mead
• Companies: Center for Disease Control and Prevention and Centers for Disease Control and Prevention and Food and Drug Administration and Food and Drug Administration
• Keywords: mixture models ; NHANES III ; pertussis ; serum antibodies

Abstract

No national cut point has been established to diagnose acute pertussis infection using a single serum sample. The IgG antibody (Ab) level to pertussis toxin was measured by enzyme-linked immunosorbent assay in a national sample of 5,420 persons aged 10-50 years who participated in the Third National Health and Nutrition Examination Survey. To cluster Ab levels into hypothesized disease groups (e.g., not exposed, immunized, acute infection), we developed a mixture model that included a point distribution located < = the assay's lower limit of quantitation (LLQ), and three component populations located > LLQ that were represented as normal distributions with unrestricted variances. Parameter estimates were weighted to represent the U.S. population. The jackknife variance estimator was used to estimate standard errors. A serum Ab level for diagnosing acute pertussis infection was chosen to have a 99% specificity based on the two overlapping normal distributions assumed for the two component populations with the highest Ab levels. This method may be more sensitive for diagnosing acute pertussis infection than other methods that assume a single distribution for Ab level.