Health care providers have struggled in the past to accurately identify their full cohort of high-risk patients. Doctors can have an accurate sense of whether patients they see will become high-risk, but what about the patients within their population that don’t present? Too often, doctors only become aware of such patients’ conditions after an emergency department visit or a hospitalization.
But within the last decade, data sufficient for greater in-depth analysis has become available to make the art and science of health care predictive modeling better than it ever has been.
Predictive analytics have become highly predictive. Enabled by health care clinical and claims data that can be classified not just as “big,” but also as high-quality, predictive analytic engines can identify high-risk patients with greater speed and accuracy.
Predictive analytics typically work by using regression analysis of health care data. And while such analysis is not new to this industry, health care data hasn’t been robust enough to make predictions with a high degree of accuracy. Claims data, the standard information source used in most predictive modeling solutions, is limited in its predictive capacity. The data is stale, as it’s most often available months after the episode of care it refers to. Also, it’s limited in scope, since most providers only add information to claims that are necessary for reimbursement.
Clinical data is the key to making predictive analytics more accurate. Some legacy predictive modeling solutions used clinical data, but relied on relatively small data sets, often of poor quality, and with limited variables. The reason for this was that most clinical data was handwritten, dictated or incomplete. The result was predictive models that didn’t offer much insight. But because EMRs have been adopted so rapidly, large and diverse digital health care data sets are now available.
New technology can now assemble this digital clinical data with data sources that span care settings and organizations. Such data sets can then be structured in such a way that the data becomes more easily analyzed. Additionally, natural language processing (NLP) technology can aggregate unstructured data. The result is data that is not only bigger but better, giving it higher predictive ability.
Health care lags behind other industries in taking advantage of parallel computing and inexpensive storage, not because there weren’t attempts to utilize data, but because the best data just wasn’t available. Now, however, such data is within the grasp of healthcare providers and payers.
For more on how predictive analytics rely on comprehensive clinical and claims data, download this Optum white paper.