Records of antidepressant dispensings are often used in public health research as a surrogate measure of depression. In 1⁄3 to 1⁄2 of cases however, antidepressants are prescribed for indications other than depression, which results in misclassification.
Medicare Benefits Schedule (MBS) data contain the dispensings of antidepressants and other pharmaceuticals, and data from the 45 and Up Study provide a gold standard response variable (a yes/no answer to “Have you been treated for depression?”).
How can these data be used to design a predictive algorithm identifying antidepressant users with depression?