Utilization of multiple imperfect assessments of the dependent variable ina logistic regression analysis

Often, in biomedical research, there are multiple sources of imperfect info rmation regarding a dichotomous variable of interest. For example, in a stu dy we are conducting on the relationship between cocaine use and stroke ris k, information on the cocaine use of each study patient is available from t hree fallible sources: patient interviews; urine toxicology testing, and me dical record review. Regression analyses based on a rule for classifying pa tients from this information can result in biased estimation of association s and variances due to the misclassification of some subjects and to the as sumption of certainty. We describe a likelihood-based method that directly incorporates multiple sources of information regarding an outcome variable into a regression analysis and takes into account the uncertainty in the cl assification. The method can be applied when some sources of information ar e missing for some subjects. We show how the availability of multiple sourc es can be exploited to generate estimates of the quality (for example, sens itivity and specificity) of each source and to model the degree to which mi ssing data are informative. A fitting algorithm and issues of identifiabili ty are discussed. We illustrate the method using data from our study. Copyr ight (C) 2000 John Wiley & Sons, Ltd.