Racial/ethnic concordance between patients and researchers as a predictor of study attrition.