Generating hypotheses from existing data

Clinical trials are long, hard, and expensive to conduct. So as a fellow, partly out of necessity and mostly out of interest, I’ve thought about how to make use of existing data to generate hypotheses, if not conclusions. As an upcoming “outcomes” researcher, this will be most of what I plan to do with the rest of my life, in any event.

Over the last week, I came across two recent articles that have each illustrated for me an interesting paradigm. The first was an article from the Journal of Clinical Oncology in the December 10th issue that was a single-institution, retrospective review of 190 adult patients with lymphoma who underwent reduced intensity-conditioning allogeneic stem cell transplant. In this article, study author Philippe Armand, MD, PhD, noted that patients who received sirolimus (Rapamune, Wyeth Pharmaceuticals) as part of graft-versus-host–disease prophylaxis had an improved three-year OS compared with those who did not (63% vs 41%, statistically significant). This was not attributable to a decrease in non-relapse mortality (as one might theorize if the effect was to improve graft-versus-host–disease control) but rather a reduction in disease progression. This is particularly interesting because of the biologic plausibility and implications of this observation — which is that mTOR inhibition may improve lymphoma treatment. To me, this study was of course not definitive or conclusive, because of its retrospective nature and possible confounders, but at a minimum hypothesis-generating and very interesting. The paradigm that it illustrated for me was how a thoughtful retrospective review of existing data can lead to an intriguing exploration of unintended effects (ie, that graft-versus-host–disease treatment may actually directly treat the malignancy itself) and an interesting publication.

The second interesting article was also from the Journal of Clinical Oncology, this time from the January 5th issue. In this study, the ACCENT (Adjuvant Colon Cancer Endpoints) investigators pooled individual patient data from 18 phase-3 trials of adjuvant fluorouracil chemotherapy for colon cancer, involving nearly 21,000 patients, to demonstrate that most of the risk reduction for colon cancer recurrence comes in the first two years after treatment. Now clearly, this isn’t really a “fellow’s project” but again does have interesting implications as a paradigm for outcomes research. If one believes the results of this study, then one may be tempted to agree with study author Daniel Sargent, PhD’s contention that, “After five years and particularly after eight years, the risk of the patient’s disease coming back is very, very low, and physicians can then turn their attention to other priorities for that patient.” However, to me, the real promise of pooling individual patient data comes in the tantalizing possibility of data mining a whole host of clinical and genomic variables to generate hypotheses that will craft the future of personalized medicine.