Is there another way to take account of noncompliance in randomized controlled trials?