Abstract:
|
Diagnostic studies often focus on the use of the combination of longitudinal biomarkers for predicting a subsequent binary disease status or its risk. Given the disease status, the longitudinal biomarkers are usually statistically handled in a linear mixed effects model. However, it is the rule rather than the exception that these longitudinal data suffer from missing values and/or dropout, furthermore, its underlying mechanism is usually informative, or in general, complex and even unverifiable. The motivation of our work is to impose a less restrictive, hence more flexible, missing data mechanism for the longitudinal biomarkers in predicting a binary event. Under this generally applicable missing data mechanism, we introduce two approaches to estimate the unknown parameters in the linear mixed effects model for the longitudinal biomarkers. Afterwards, we propose to compute the test statistic and use it as the combination rule. We also derive the individual disease risk score and its confidence interval. We conduct simulation studies to illustrate our method and also apply it to a real data analysis.
|