We started by reviewing all possible cancer cases in the cohort and found that those recorded in the tumor registry were quite accurate.  We then reviewed only the hospital records until complete tumor registry coverage was available.