This article is a portion of a book titled "Challenges, Risks and Opportunities in Today's Spine World" edited by Stephen Hochschuler, MD, Frank Phillips, MD, and Richard Fessler, MD. You can find links to the previous chapters at the end of this article.
These two “catch phrases”, which define new fields of information analysis and application, often strike fear and loathing in the hearts of physicians. Numerous questions immediately present themselves to the physician who is presented with them. What is “big data”? What is “predictive analytics”? Why is it being applied to medicine? Who is pushing this? Insurance carriers? Administrators? Government? And most significantly, how is this going to impact my practice? Will a computer now decide my patient’s healthcare plan rather then me? With questions such as these, it’s easy to understand why many physicians are skeptical. Given the “newness” of these fields, at least in their application to medicine, it’s probably worth taking a look at what they are, where they are likely to be applied to medicine, and how that will impact our practices as physicians.
In medicine, “big data” refers to data sets that have been enabled by the electronic medical record, and share three characteristics: volume (massive data availability), velocity (flood-like arrival of data), and variety (arrival of data from multiple sources with differing formats). “Predictive analytics” is the branch of advanced analytics that is used to make predictions about unknown future events. When combined, the hope is that by analyzing previously unattainable massive data sets, accuracy of prediction about the probability of future events is enhanced (for example the risk of infection in a specific patient or the probability of a successful outcome of surgery in a specific patient). This maximizes the concept of “personalized medicine” which hopefully would result in improved patient care. In this sense, it is not fundamentally different from the way we practice medicine now, just bigger and (hopefully) better.
One way to think of it is like this: Today, I read the literature to learn the latest methods of treatment for my patients. Given my busy schedule, I can read a limited number of manuscripts, which analyze limited data points on at most a few hundred patients. Among these patients, it’s likely that only a handful will match the demographics of the patient I am caring for. Thus, application of the knowledge I acquire to my specific patient is, at best, imperfect. Imagine being handed a summary of thousands of manuscripts, analyzing essentially every important parameter (physiologic, demographic, treatment history, etc., etc.) on tens of thousands of patients, among whom thousands exactly match the details of my patient. Predictions on probabilities of events for my patient, such as infection, DVT, or successful outcome, now become vastly more accurate. This enables me to have a “fact based” discussion with my patients and give them realistic probabilities of events/outcome relevant to them, upon which they/we can make decisions. This is real “personalized” medicine, and can only improve my performance as a physician and the overall outcome of my patients.
There are many forces pushing the application of these tools into medicine. Not necessarily in order of importance, one factor weighing in is the success seen in other fields in which it has been applied. Increased sales resulting from personalized marketing such as that seen on Facebook is one example. The second factor is the prospect of improved patient care. As physicians and hospitals become increasingly more at risk for payment (or lack thereof) based upon outcome, the drive to find ways to increase the probability of good outcomes is significant. Finally, as the cost of medicine continues to increase, in the face of decreasing reimbursement, hospitals are exploring multiple alternatives to decrease overhead. Using big-data to predict OR utilization behavior, for example, and making adjustments based on that information is one way to maximize efficiency.
Given the success seen in other fields, the impact on hospital and physician efficiency, and the possible improvement in patient care, it seems inevitable that big data and predictive analytics will be applied to medicine. The next questions then are “where and how soon will it be applied?” We have already alluded to one place where it can be applied: efficient utilization of the OR. With accurate knowledge of what types of operations surgeons perform, how frequently they are performed, how long they take, if they vary seasonally or by time of day (such as specific types of trauma), and what type of patients they are performed on, room utilization can be improved. Moreover, knowing the accurate probability of complications (e.g. infection) in specific sub-populations of patients could lead to pre-emptive alterations in care designed to minimize their occurrence. Thus, hospital/physician efficiency is improved, cost of delivering care is decreased, and overall patient outcome is improved. Win, win, win.
Another application has been demonstrated at the Cleveland Clinic. They have developed a tool called RAPT (Risk Assessment and Prediction Tool) which in only six questions predicts the probability of early discharge to home versus the necessity of skilled nursing in total joint replacement patients. With accurate prediction of length of hospital stay, patient flow and room utilization can be maximized, and the most appropriate, patient specific, care can be planned prior to the administration of any treatments. Again, efficiency and outcome are maximized. It also should be obvious that the hospitals and physicians who are able to achieve these goals most effectively, are going to experience a competitive edge over those who do not.
However, it is important to realize that there are huge barriers to the widespread utilization of big data and predictive analytics in medicine. First and foremost, the necessary investments in infrastructure, analytical tools, personnel, and data governance will be huge. It is uncertain at this time where the money for this will come from, and if there will be a large enough pool of appropriately trained analysts available to enable widespread implementation. Obviously, an EHR (electronic health record) which is fully functional between all parties must be in place. On site data warehousing will be necessary. Currently, because data is collected in multiple disparate formats, much if which is unstructured, few hospitals or even health care systems are capable of this level of data management.
Second, many problems fall under the general challenge of turning “information overload” into “actionable insight”. For example, to effectively use this technology physicians (not just IT personnel) must be able to fully understand, interpret, and utilize the information, models, and predictions. That will require lengthy and detailed training, in an area of knowledge with which most physicians will not be familiar. Models of workflow must be accurate. For accurate predictive algorithms, questions must be specific, detailed, and relevant to topic. The target need must be clearly defined, computational questions created which are accurate to that need, and specific variables gathered. To achieve these ends, participants at all levels of the organization must be willing to actively participate.
Finally, the last “v” of health care data (see paragraph 2) is of paramount importance: Veracity! Ultimately, the accuracy of data input will determine our ability at accurately “predict”. It once again comes down to “garbage in, garbage out”.
In summary, then, it is clear that the potential benefit of big data and predictive analytics at all levels of the health care spectrum is huge, but the barriers to their widespread implementation are equally large. Although focused and limited programs are currently being trialed in selected areas of medicine, widespread implementation in the near future is highly unlikely.
Previous chapters:
Challenges, risks and opportunities in today's spine world
Spine care - Balancing cost with innovation