Skip Internal Navigation
Biomarker Data Analysis
Methodological issues inherent in designing and analyzing studies with biomarker measurements occur in most of the Division's research. These issues often motivate many of the independent methodological research projects in the Branch such as those listed below.
Design and Analysis of Studies Using Pooled Biomarker Studies
BBB investigators, in collaboration with EB investigators, are designing new ways to pool biomarker specimens in epidemiologic and clinical studies. Specifically, they are developing approaches for cohort studies in which the exposure is based on pooled biomarker samples. Design and analysis approaches for longitudinal studies, such as BioCycle, in which the longitudinal outcomes may be pooled are also under investigation. These new analytic approaches for both the design and statistical analyses will provide investigators with the ability to extract more information with fewer resources in subsequent studies.
Analyzing Data with Detection Limits
BBB investigators, again in collaboration with EB investigators, are developing new methods for analyzing both exposure and outcome data with detection limits, a particularly important issue for environmental epidemiologic research in which environmentally relevant exposures are at or below the laboratory detection limits. Methods research, much of it led by the Division, revealed that automatic substitution of values below the limits of detection introduces bias when estimating health effects. BBB investigators are developing new semi-parametric methods, which will rely on fewer assumptions to make statistical inferences in this situation.
New Methods for Assessing Inter-rater Agreement and Diagnostic Accuracy
One important objective in many epidemiologic and clinical studies is to assess the agreement between different categorical or ordinal ratings. The "kappa coefficient" is a widely used index for inter-observer agreement that is more appealing than percent-agreement because it corrects for the proportion of agreement expected by chance. Studies often collect longitudinal assessments of agreement. BBB investigators are developing new methodology for assessing agreement of longitudinally collected ratings or scores in these situations. In addition to assessing agreement, researchers are often interested in assessing the accuracy of ratings or tests when there is no "gold standard" test available. Other new methods, developed by BBB staff, include those for making inference about the diagnostic accuracy of these tests when no such standard exists. In fact, many of the methods were developed from collaborative research in the Endometriosis: Natural History, Diagnosis, and Outcomes (ENDO) Study, which is focused on comparing and evaluating different measures for diagnosing endometriosis in the absence of a gold standard.