[The authors respond:]
We appreciate the letter by Jaime Caro and Krista Payne; however, we disagree with their conclusion. Before doctors consider choosing a drug on the basis of real-world compliance, they should ask 2 questions. Is the evidence suggesting a difference in compliance likely to be true? If there is a difference, what is the magnitude of that difference and is that magnitude likely to lead to a difference in morbidity and mortality? The answer to both questions in this case is No.
With regard to the first question, 2 studies[1, 2] suggest that compliance is better with new drug classes than with old drug classes, and 2 studies[3, 4] suggest that there is no difference in compliance. These 4 studies are observational and are subject to bias (i.e., patients prescribed drugs from different classes are not comparable). The most likely bias in the 2 studies claiming a difference is that patients receiving new drugs were more likely to have been given a drug sample in the doctor's office. Old drugs are not available as samples. This sampling would not be captured in the database and would bias the results in the direction seen. The authors should have been aware of this confounder but did not mention it. Lower compliance with the old drugs, thiazides and β-blockers, is highly unlikely to be true; a double-blind randomized controlled trial designed to test this hypothesis demonstrated fewer withdrawals with the old drugs than with the new drugs.5
With regard to the second question, in the study by Caro and colleagues2 the largest absolute difference in nonpersistence was between thiazides and angiotensin-converting-enzyme inhibitors: 9% at 6 months and 13% at 4.5 years. Would this small difference in compliance lead to a difference in morbidity and mortality? We believe it is highly unlikely, and randomized controlled trials would be required to answer this question. It is important that doctors not be fooled into thinking that observational studies measuring compliance are a substitute for randomized controlled trials that are designed to be generalizable and to measure clinically important outcomes.