Gray, Robert J.

Sensitivity analysis is an important statistical technique that assesses whether the results of phase III trials are robust and likely to be generalizable. Until recently, sensitivity analyses were rarely included in phase III trials, and they remain poorly understood by many oncologists. Sensitivity analyses are critical to understanding the strength of conclusions made in the primary analysis of a late-stage clinical trial. They examine the influence of protocol design errors, unintended biases, deviations from assumptions underlying statistical models, and any unanticipated treatment delivery or practice patterns on trial results. In trials with complex or subjective end points, they also allow an understanding of the extent to which a positive outcome is driven by a single, possibly subjective, and therefore biased, element of an end point. The purposes of this article are to explain how sensitivity analyses are performed, to discuss areas of a clinical trial where sensitivity analyses should focus, and to illuminate the importance of this technique in the rigorous evaluation of late-stage clinical trial data, using specific examples. This article focuses on late-stage trials that use progression-free survival or time to progression as their primary end point, because sensitivity analyses are particularly important in these cases for which the end point is potentially subject to bias. Three sources of potential bias are explored: assessment time, symptomatic (ie, nonradiologic) disease progression, and missing data. For each source of potential bias, case studies are presented to highlight the role that sensitivity analyses play in determining whether the trial's conclusions are robust.

PurposeTo determine the risk of ipsilateral breast events in patients with ductal carcinoma in situ (DCIS) treated with local excision without irradiation.Patients and MethodsPatients with either low- or intermediate-grade DCIS measuring 2.5 cm or smaller, or high-grade DCIS measuring 1 cm or smaller who had microscopic margin widths of 3 mm or wider and no residual calcifications on postoperative mammograms were eligible for a prospective trial conducted from 1997 to 2002 by the Eastern Cooperative Oncology Group and North Central Cancer Treatment Group. Patients entered in 2000 and later could take tamoxifen if they wished. Median age at last surgery for the entire population was 60 years (range, 28 to 88 years), and median tumor sizes in the two strata were 6 mm and 5 mm, respectively.ResultsWith a median follow-up of 6.2 years, the 5-year rate of ipsilateral breast events in the 565 eligible patients in the low/intermediate grade stratum was 6.1% (95% CI, 4.1% to 8.2%). With a median follow-up of 6.7 years, this incidence for the 105 eligible patients in the high-grade stratum was 15.3% (95% CI, 8.2% to 22.5%).ConclusionRigorously evaluated and selected patients with low- to intermediate-grade DCIS with margins 3 mm or wider had an acceptably low rate of ipsilateral breast events at 5 years after excision without irradiation. Patients with high-grade lesions had a much higher rate, suggesting that excision alone is inadequate treatment. Further follow-up is necessary to document long-term results.