Causal inference in paired two-arm experimental studies under noncompliance with application to prognosis of myocardial infarction