There is no clear advantage, at least on the outcomes measured in this study, to limiting medical school entry to either those candidates from secondary school or those with tertiary backgrounds. Medical schools could reasonably broaden their selection criteria to include more graduate entry candidates in addition to secondary school leavers without compromising medical school and practice outcomes.

The letter by Lee and McKenzie 1 in response to our article 2 has drawn readers' attention to the challenges of interpreting and comparing findings from observational studies. The authors quote highly variable outcomes from two observational studies examining the risk of cardiotoxicity in trastuzumab-treated patients, with incidence rates ranging from 3% 3 to 39%. 4 Conflicting observational evidence is not surprising, given the heterogeneity in study designs, including variations in patterns of care of the study population, outcome definitions and the characteristics of populations under study. For example, the differences observed in the studies cited by Lee and McKenzie may, in part, be attributable to exposure to different treatment regimens (37% of the first study cohort were exposed to anthracyclines in contrast with 87% of the second cohort).Our population-based study 2 found increased cardiac testing in metastatic breast cancer patients treated with trastuzumab, a spillover effect of the mandatory cardiac testing for early stage disease. The strength of our study is that we used longitudinal data on the entire population receiving trastuzumab through the National Herceptin Programme, more than 3500 women. Healthcare administrative data allow researchers to examine pharmaceutical and health services use in patients who are often excluded in randomised, controlled trials. Moreover, the longitudinal nature of such data enables us to assess trends in practice over time and to follow patients who have received therapy for much longer periods than the typical duration of trials. The key limitation to our own work was our inability to incorporate any of the outcomes reported from the testing as they are not collected routinely.The benefits of population-level healthcare administrative data have also been illustrated by a recent US study 5 of 12 500 women, 554 of whom received trastuzumab therapy. Bowles et al. 5 reported that the risk of heart failure and/or cardiomyopathy (defined by International Classification of Diseases, Ninth Revision codes: 398. 91, 402.x1, 402.x3, 404.x1, 404.x3, 422.90, 425.4, 425.9, 428.xx) increased significantly in patients treated with trastuzumab alone (adjusted hazard ratio = 4.1, 95% confidence interval (CI) = 2.3-7.4) or with anthracycline plus trastuzumab (adjusted hazard ratio = 7.2, 95% CI = 5.0-10.4) compared with no chemotherapy. While claims data are valuable for estimating risks of drugs in the 'real-world' setting, health outcomes are typically defined by diagnosis codes recorded in claims. In contrast, some observational cohort studies collect detailed patient clinical information, such as left ventricular ejection fraction results and development of symptomatic heart failure associated with trastuzumab, 6 which enables more precise understanding of cardiac safety of trastuzumab, although such studies often have much smaller sample sizes.Based on best evidence available at the time of recommendation, the Pharmaceutical Benefits Advisory Committee determined that rout...