The broad objectives of this grant application are (1) to provide a mechanism by which """"""""Healthy People 2000"""""""" objective 16.16 might be more fully realized via the evaluation of the variability in the accuracy and consistency of mammographic image interpretation in the United States; (2) establish a statistical model by which mammographic centers might be evaluated against a """"""""national average"""""""" standard screening center; (3) investigate possible inequalities in mammographic screening """"""""delivery"""""""" in the United States.
The specific aims of this application are (1) to establish a statistical model which describes the probability of screening diagnosis as a function of characteristics of the patient and the radiologist and which treats mammographic centers and radiologists as random effects; (2) compare the diagnostic efficacy and consistency in mammographic screening between rural and non-rural mammographic screening centers and between screening given white and black women. Study Design: All radiologists at twenty-five rural and twenty-five non-rural randomly selected mammographic screening centers will be asked to participate. A stratified random sample of 96 patients will be selected from Duke University's mammography screening program. Images from these patients will be mailed to each center in randomly-ordered succession. Images will be read in random order and each reader will give each image a 5-point rating indicating degree of concern. Analysis: A latent mixed logistic regression model with a random component reflecting variability between rural and non-rural centers and a random component reflecting variability between radiologists will be fit. Covariates will be included which will account for the effects of the variables used for the stratification of the patients and for the effects of certain characteristics of the radiologists. This model will be used to evaluate the dependence of the sensitivity and specificity of screening mammography upon characteristics of patients and radiologists. The model will be used to compare rural and non-rural manu-nographic centers with respect to their mean diagnostic efficacies. Homogeneity in diagnosis within rural and non-rural areas will be compared. The model will provide an accurate tool by which the performance of mammographic centers may be assessed.
