A Novel Data Mining Approach to the Identification of Effective Drugs or Combinations for Targeted Endpoints?Application to Chronic Heart Failure as a New Form of Evidence-based Medicine

Abstract

Data mining is a technique for discovering useful information hidden in a database, which has recently been used by the chemical, financial, pharmaceutical, and insurance industries. It may enable us to detect the interesting and hidden data on useful drugs especially in the field of cardiovascular disease. We evaluated the current treatments for chronic heart failure (CHF) in our institute using a decision tree method of data mining and compared the results with those of large-scale clinical trials. We enrolled 1,100 patients with CHF (NYHA classes II-IV and EF < 40%) who were hospitalized at the National Cardiovascular Center during the past 31 months. Drugs prescribed at discharge were extracted from the clinical database. Both echocardiograms and plasma BNP level at 6-12 months after discharge were determined prospectively. It was found that beta-blockers, angiotensin converting enzyme inhibitors, and angiotensin II receptor antagonists independently improve both the plasma BNP level and %fractional shortening (FS), while oral inotropic agents increased the plasma BNP level and decreased %FS. These findings agree with evidence accumulated from several large-scale trials. Interestingly, statins, histamine receptor blockers, and alpha-glucosidase inhibitors also attenuated the severity of CHF, suggesting the possibility of new treatment of CHF. Clinical data mining using Japanese CHF patients yielded almost identical data to the results of large-scale trials, and also suggested novel and unexpected candidates for CHF therapy. Further validation of the data mining approved in the cardiovascular field is warranted.