Retrospective Analyses of Cancer Trials or Existing Clinical Databases: Potential for Highly Clinically Relevant Hypothesis Generation

neal cavity as treatment of small volume residual advanced ovarian cancer could only realistically favorably impact patients with microscopic or very limited volume macroscopic cancer (following an attempt at primary maximal surgical cytoreduction) [1] . This conclusion resulted from the observation that the extent of direct tissue penetration of antineoplastic agents (such as cisplatin) following regional delivery was limited to a maximum of only a few millimeters from the surface. However, in a retrospective review of a landmark phase 3 randomized trial that compared intraperitoneal to intravenous cisplatin in this clinical setting it was revealed that regional drug delivery to patients who initiated treatment with a single tumor mass of between 0.5 and 1.5 cm in maximal diameter achieved at least as much relative benefit from this strategy (when compared to systemic administration) as women who had only microscopic residual cancer [1, 2] . (Note: This outcome may be hypothesized to result from the fact patients are treated with multiple drug instillations over several months, rather than a single treatment, with each treatment reducing the total volume of cancer to be treated with subsequent treatments.) Thus, this analysis challenged the existing ‘conventional’ thinking in this area, prompting investigators to explore therapy in the setting of larger volume cancer. The superiority of prospective clinical trial data in the quality of evidence provided to permit a meaningful conclusion is well recognized. In addition to the inherent difficulty of being certain all relevant information has been captured in a retrospective review of trials or previously existing databases when there was no previous intent to obtain the specific data in question, there is always the concern that patients inadvertently or intentionally may not have been included in the analysis or reasons for the decision to employ (or not employ) particular management strategies may not have been clearly discussed or even mentioned in the medical records. Assuming a prospective analysis has been optimally considered and undertaken, the decision regarding what is relevant information to collect and review in a given study will have been decided prior to the initiation of the research. Further, at least in theory, such data will be obtained on all patients, and if not available an appropriate explanation for its absence should be provided. However, despite these serious limitations, it is relevant to acknowledge the potential for retrospective studies to serve a critical role in hypothesis generation or as data to support or refute existing clinical paradigms or explanations for observed outcomes. For example, ‘conventional wisdom’ appeared to declare that administering cisplatin directly into the peritoPublished online: September 18, 2012