OBJECTIVES: To investigate patterns in antibiotic prescribing for patients undergoing cardiovascular surgery to determine whether there are differences between physicians with high rates of infection compared to physicians with low rates of infection. METHODS: All cardiovascular surgeries from a mid-sized southern hospital over a 2-year period were examined (approximately 2100 surgeries). A cross-tabulation between physician and patient infections was performed to identify those physicians with a low (under 6%) and a high (over 6%) infection rates. Information from the pharmacy database relating to all patients in the cardiovascular database was extracted, and all antibiotic prescriptions were identified. The data mining techniques of association rules and kernel density estimation were used to investigate prescribing patterns between the two groups of physicians. RESULTS: While the surgeons did not seem to differ in their use of cephalosporins, they differed considerable in their use of fluoroquinolones. Low infection physicians made more frequent use of Cipro while high infection physicians tended to make more use Levaquin. In addition, some patients received up to eight different antiobiotics as inpatients before, during, and after cardiovascular surgery. CONCLUSION: Data mining techniques that have been developed for business applications can be used to investigate physician decisions and their impact on patient outcomes. Variability in physician decisions in the absence of treatment guidelines can be investigated observationally, and meaningful results obtained. In this application, there are differences in antibiotic use related to infection rates. These differences should be examined and a consensus reached for prescribing habits.