A general method for handling missing binary outcome data in randomized controlled trials