We appreciate the thoughtful comments of Drs. Montemurro and Aglietta regarding our recently published article. We agree that it is important to compare time to the development of central nervous system metastases between those patients treated with trastuzumab and those who were not. However, the hazard rate (incidence rate in time-to-event analysis) of the two groups may not be constant over time; consequently, the derived crude hazards ratio may be inaccurate. We believe that a better method with which to estimate the hazards ratio is to use the Cox proportional hazards model. In addition to its ability to adjust for imbalances in baseline prognostic factors, the Cox model works well even when the hazards rates are not constant, provided that the ratio of the hazards rates is constant over time.1
The reason we did not perform a time-to-event analysis in our study was the limit of our sample size. Assuming an increase of the hazards ratio to 1.5 (a 50% increase), α = 0.05, β = 0.2, and a 3-year survival probability in the nontrastuzumab group of 45%,2 the minimum required sample size per group is 157 patients; we only had half of the number required in the trastuzumab-treated group. Moreover, unlike randomized trials, observational studies often require a larger sample size than the calculated minimum because of the presence of confounders. Because we were underpowered to perform such an analysis, we studied the development of CNS metastases as a binary variable that required fewer patients for the same power.
One group of investigators compared the rate of acquiring brain metastases between patients receiving trastuzumab and those who were not and found no difference with regard to the time to the development of CNS disease.3 There is no question that time-to-event analysis is more informative, but as a first step, it still is reasonable to evaluate the overall association between trastuzumab therapy and subsequent disease recurrence in the CNS. It may be possible to examine time-to-brain metastases more accurately using large data sets available from relevant trials of the Cancer and Leukemia Group B (CALGB) or other cooperative groups.