Readers of my blog posts have probably noticed an ongoing theme: Organizations taking the journey from volume to value need to apply advanced analytics to data to be able to manage risk and make the most out of value-based care. Over my next few blog posts, I’ll stick with that theme, with my focus on a critical aspect of analytics: prediction.
Historically, providers have struggled to accurately identify high-risk patients. Some motivated providers try to identify their sickest patients by manually combing through EMR data to create a list. A valiant effort, no doubt, but such a process proves to be too resource-intensive and almost always inaccurate.
Predictive analytics has long held the promise of solving this problem. Only recently, however, has data become available in sufficient quality and quantity to bring predictive analytics closer to fulfilling that promise. Powered by vast quantities of high-fidelity clinical and claims data, predictive analytics can identify high-risk patients with greater speed and accuracy than ever before.
Predictive analytics uses regression models on underlying data to predict outcomes. Now, this is not new to health care. The challenge, in past years, has been the underlying data. Up until a few years ago, the main source of digital health care data was claims data. But claims data does not get at a patient’s overall health or disease-specific functioning.
What’s new, and what makes this topic so exciting, is the broad availability of clinical data. In the past, clinical data was often handwritten, dictated, or incomplete. Hence, predictive modeling relied on relatively small data sets, often of poor quality, and with limited variables. The result was only marginally predictive models. But because of the recent and rapid adoption of EMRs, large and diverse digital health care data sets are now available.
Technology also exists that can aggregate this clinical data with other data sources, from across care settings and organizations. This data can then be better structured to enable analytics. Additionally, natural language processing (NLP) can be used to access unstructured data. The result is “bigger and better” data with higher predictive ability.
Many other industries have already benefited from today’s inexpensive data storage and massive parallel computing power. Health care has lagged because data availability lagged. But the quantity and quality of digital data that is required now exists.
For more on how predictive analytics rely on comprehensive clinical and claims data, download “Predictive analytics: Poised to drive population health.”
–Jeremy Orr, MD, MPH, Chief Medical Officer, Optum Analytics