“Large collections of electronic patient records have long provided abundant, but under-explored information on the real-world use of medicines. But when used properly these records can provide longitudinal observational data which is perfect for data mining,” Duan said. “Although such records are maintained for patient administration, they could provide a broad range of clinical information for data analysis. A growing interest has been drug safety.”
In this paper, the researchers proposed two novel algorithms—a likelihood ratio model and a Bayesian network model—for adverse drug effect discovery. Although the performance of these two algorithms is comparable to the state-of-the-art algorithm, Bayesian confidence propagation neural network, by combining three works, the researchers say one can get better, more diverse results.
I saw this a few weeks ago, and while I haven't had the time to delve deep into the details of this particular advance, it did at least give me more reason for hope with respect to the big picture of which it is a part.
It brought to mind the controversy over Vioxx starting a dozen or so years ago, documented in a 2004 article in the Cleveland Clinic Journal of Medicine. Vioxx, released in 1999, was a godsend to patients suffering from rheumatoid arthritic pain, but a longitudinal study published in 2000 unexpectedly showed a higher incidence of myocardial infarctions among Vioxx users compared with the former standard-of-care drug, naproxen. Merck, the patent holder, responded that the difference was due to a "protective effect" it attributed to naproxen rather than a causative adverse effect of Vioxx.
One of the sources of empirical evidence that eventually discredited Merck's defense of Vioxx's safety was a pioneering data mining epidemiological study conducted by Graham et al. using the live electronic medical records of 1.4 million Kaiser Permanente of California patients. Their findings were presented first in a poster in 2004 and then in the Lancet in 2005. Two or three other contemporaneous epidemiological studies of smaller non-overlapping populations showed similar results. A rigorous 18-month prospective study of the efficacy of Vioxx's generic form in relieving colon polyps showed an "unanticipated" significant increase in heart attacks among study participants.
Merck's withdrawal of Vioxx was an early victory for Big Data, though it did not win the battle alone. What the controversy did do was demonstrate the power of data mining in live electronic medical records. Graham and his colleagues were able to retrospectively construct what was effectively a clinical trial based on over 2 million patient-years of data. The fact that EMR records are not as rigorously accurate as clinical trial data capture was rendered moot by the huge volume of data analyzed.
Today, the value of Big Data in epidemiology is unquestioned, and the current focus is on developing better analytics and in parallel addressing concerns about patient privacy. The HITECH Act and Obamacare are increasing the rate of electronic biomedical data capture, and improving the utility of such data by requiring the adoption of standardized data structures and controlled vocabularies.
We are witnessing the dawning of an era, and hopefully the start of the transformation of our broken healthcare system into a learning organization.