The introduction of the cloud and other powerful computing technologies has ushered in the era of big data in medicine. This has had a profound effect on the clinical practice of urology on numerous levels and has been driven, in no small part, by a growing number of studies based on analyses of large clinical and administrative data sets. To understand the staggering number of studies performed in urology using large administrative data sets, consider that a search of PubMed.gov using the key word terms “prostate” and “SEER” (in reference to the Surveillance, Epidemiology, and End Results database) returned over 1,400 unique publications over the past 2 decades.¹ Importantly, this search represents just a small sampling of the total number of studies using large databases in urology, as it only focuses on a single topic and a single database. If we included all urological topics and all large data sets, this number would be considerably larger. While many of these studies have improved our understanding of how to diagnose and treat urological conditions, some of them have reached flawed conclusions and may have had a negative impact on the practice of urology. In this year’s Journal of Urology Lecture, we will discuss the promises and the pitfalls of large database studies with an eye towards improving clinical urological care.

Examples of important studies using large clinical databases will be presented to help the audience understand the power of these analyses. Consider the analysis by Shahinian and colleagues that leveraged the SEER-Medicare Database (a large data set that includes data from men diagnosed with prostate cancer in the SEER program linked to data from the Medicare administrative data set).² This analysis conclusively documented and quantified the risk of various bony fractures for men who were treated with androgen deprivation therapy (ADT) for prostate cancer. In fact, the authors clearly show a dose-response relationship between number of doses of GnRH agonist and unadjusted fracture-free survival. Results from this well-done study using a large linked clinical and administrative database likely impacted ADT utilization for years to follow and serve as a shining illustration of the promise of this type of research.

Conversely, examples of the limitations of large database studies will also be presented. These limitations can result from various causes. There is the “garbage in, garbage out” (GIGO) phenomenon that can result in inaccurate data. A recent example of this is the coding error that occurred in the SEER registry due to incorrect data entry of prostate specific antigen (PSA) values for men newly diagnosed with prostate cancer in 2014.³ While not as bad as originally suspected, this simple rounding error that resulted in incorrect PSA values being entered into this public data set could have negatively affected the validity of numerous population-based studies of prostate cancer. This limitation and several others will be presented so that the clinician better understands the pitfalls of these types of studies.

As computers get more powerful and more large data sets are made publicly available, we will see even more of these types of analyses. The urologist must be aware of both the strengths and weaknesses of these studies and be able to recognize the good from the bad. If we are able to realize the great promise and avoid the potential pitfalls of these data sets, clinical urological care will only get better and better.