Analysis Of Large Medical Datasets Can Be Both Informative and Misleading

David F. Penson, MD, MPHHamilton and Howd Chair in Urologic Oncology Professor and Chair, Department of Urologic Surgery Director, Center for Surgical Quality and Outcomes Research Vanderbilt University Medical Center Nashville, TN 37232-2765MedicalResearch.com Interview with:
David F. Penson, MD, MPH
Hamilton and Howd Chair in Urologic Oncology
Professor and Chair, Department of Urologic Surgery
Director, Center for Surgical Quality and Outcomes Research
Vanderbilt University Medical Center
Nashville, TN 37232-2765

Medical Research: What is the background for this editorial? What are the main findings?

Response: This editorial discusses the implication of the recent removal of the PSA data from the seer-medicare dataset. It reviews the significance of the action: specifically what it means for prior publications that used this information to address clinical research questions in prostate cancer. It makes the point that, while these datasets are powerful, researchers have stretched the limits of what they can do too far. Simply put, we cant always guarantee that the clinical data collected in administrative datasets will necessarily be accurate so we need to be more selective in how we use these data and not simply run analyses on the data just because it is easy.

Medical Research: What should clinicians and patients take away from your report?

Response: Clinicians and patients need to consider that some of the studies that they are using to aid in decision-making in localized prostate cancer may not be as reliable as they originally thought. Specifically, studies that relied on the PSA data to identify patient groups or to risk adjust may be highly flawed and this may lead to incorrect conclusions.  Patients need to consider this when choosing options for treatment and clinicians need to reconsider which studies they rely on to counsel patients.  It is important for patients and clinicians to remember to consider the weight of the evidence and different types of evidence rather than relying on one or two studies, particularly if they are from datasets that were not specifically designed to answer the research questions at hand.

Medical Research: What recommendations do you have for future research as a result of this study?

Response: Researchers need to reserve the use of large, administrative datasets like seer-medicare for studies that they can reasonably answer using the data. They have to recognize that the ‘garbage in, garbage out” phenomenon may apply.  Tumor registrars do an incredible job but that still doesn’t mean that they get it right all the time and some clinical data points, such as PSA levels, may be a lot tougher to collect than we originally thought.

Citation:

“The Power and the Peril of Large Administrative Databases,” by David F. Penson. DOI: http://dx.doi.org/10.1016/j.juro.2015.05.002 . Published online in advance of The Journal of Urology®, Volume 194, Issue 1 (July 2015)

[wysija_form id=”3″]

 

MedicalResearch.com Interview with: David F. Penson, MD, MPHHamilton and Howd Chair in Urologic Oncology, & Professor and Chair, Department of Urologic Surgery (2015). Analysis Of Large Medical Datasets Can Be Both Informative and Misleading MedicalResearch.com

Last Updated on May 21, 2015 by Marie Benz MD FAAD