Limited information is available from large clinical investigations about the agreement among sources of diagnoses for endpoints. The authors used data from the Women’s Health Initiative clinical trials and observational study from January 1994 to November 2000 to evaluate the agreement among self-report, hospital discharge codes, and two different levels of physician review of medical records for cardiovascular endpoints. For myocardial infarction, stroke, pulmonary embolism, and venous thrombosis, the agreement of hospital discharge codes or self-report with review by study...

Limited information is available from large clinical investigations about the agreement among sources of diagnoses for endpoints. The authors used data from the Women’s Health Initiative clinical trials and observational study from January 1994 to November 2000 to evaluate the agreement among self-report, hospital discharge codes, and two different levels of physician review of medical records for cardiovascular endpoints. For myocardial infarction, stroke, pulmonary embolism, and venous thrombosis, the agreement of hospital discharge codes or self-report with review by study physicians at clinical centers was substantial (kappa = 0.64–0.84). For coronary revascularization, agreement among these sources of information was substantial to almost perfect (kappa = 0.79–0.92), but for angina, congestive heart failure, and peripheral vascular disease, concordance was only fair to moderate (kappa = 0.37–0.56), indicating that these endpoints remain difficult to classify reliably. Agreement between physician adjudicators at clinical centers and central physician adjudicators was substantial to almost perfect (kappa = 0.67–0.94). The findings also suggest that, for the endpoint of myocardial infarction, physician review of events with hospital discharge codes for angina and congestive heart failure is an important source of validated events, and for stroke, review of all events with cerebrovascular codes is important.