Can we trust strong recommendations based on low quality evidence?