Health insurers continue to use their claims data to build predictive models. Because claims have been processed digitally for years, health plans often have more mature predictive analytics capabilities than health systems, Powell says. Furthermore, insurers’ larger size gives them the necessary scale to afford investments in analytics.
But despite what has been accomplished, there is much more to do. “The healthcare industry is very early in its journey of employing predictive analytics,” Clarke says. “The potential is immense, but turning that potential into tangible business value has not been proven at scale. This is partly due to the technical challenge of having data that are large and complicated; true data experts are often not facile with new approaches and techniques to turn data into insights. However, an even bigger challenge to adoption is proving impact. Many of the most exciting use cases involve long-term changes in care patterns that make it difficult to prove impact. Therefore, there’s a great focus on revenue capture and other areas where the direct link to impact is clearer.”
McIntyre says that many health insurers have recognized that claims-based analytics alone are insufficient to fully optimize the quality and affordability of healthcare. As a result, many payers have invested in advanced analytics capabilities that leverage claim, clinical, and nonclaim/nonclinical data wherever possible. “Many larger payers are building these capabilities internally, while others are leveraging healthcare analytics vendors in the marketplace,” he says. “Progress in leveraging claim-based analytics is generally well developed with most payers and/or healthcare analytics vendors. However, only in recent years has much traction been made at integrating clinical and psycho-socioeconomic data in a meaningful way.
Again, using the example of a member diagnosed with diabetes, Anthem can use the member’s digital interactions with an Anthem health plan along with medical records information from a provider partner to analyze the member’s body weight, demographics, physical activity, and other information to help make the connection to a treatment or wellness plan that may not be as invasive or expensive for the patient. “Perhaps it’s providing them with information about healthy nutrition for people with diabetes, or sharing information about physical fitness or directing them to a nutritionist. We can work to find the easiest and least-expensive option,” McIntyre says.
McIntyre says this is where the majority of capital investments by payers will occur over the next several years. This will require a combination of attracting the right talent, such as experts in big data and data science, and investing time and resources to build new value-based relationships with providers and engage with consumers in a way that payers have not done in the past.
As far as specific gaps, Vijay Murugappan, vice president, analytics and process transformation, Health Care Service Corporation, a customer-owned health insurer in Chicago, says typical lags in predictive analytics occur when the volume and granularity of coverage data (across members or conditions, for example) are not enough to derive and accept data-driven insight. For example, laboratory and pharmacy data provide good insights for obtainable data, but because such data is not available for most or all of a population, applying these insights broadly can be tricky.
For independent general practitioners, Fox says the benefits of analytics are effectively out of reach due to cost. “The smaller the entity, the more this is true,” he says. “With a single patient’s information being spread over disparate systems—customer relationship management, billing management, and EHRs—smaller providers experience the same obstacles presented by data in silos, but they lack the scale to readily deploy cost-effective data management and analytics solutions. Instead, valuable patient information sits unused and sometimes even abandoned in documents, content management systems, and even worse—in filing cabinets and boxes. The result of this stark reality is that the lack of access to essential technologies is driving consolidation among these smaller provider entities that need to compete in this era of outcomes-based reimbursement, which could potentially benefit patients.”
Marshall Greene, analytics manager, Mosaic Medical, a community healthcare organization with clinics across Central Oregon, says one challenge in predicting health outcomes is that the timeline for realizing a return on investment can be quite long. For instance, in trying to predict a patient’s claims costs, forgoing preventative care and screenings may lead to lower costs in the short and even medium term. However, if it means that a chronic condition goes unmanaged or cancer goes undiagnosed, the short-term savings turn into a very expensive patient decades later.
Healthcare organizations frequently struggle to operationalize the clinical insights gleaned from predictive analytics for a couple of reasons, Greene says. First, there is an inherent disconnect between the analysts/data scientists who create the predictive models and practitioners who need to use the information to guide an intervention. “Analysts generally don’t have a formal clinical background and often have very little context for the disease pathways and health outcomes that they are trying to predict,” he says. “Hence, once the analytics team has developed a model that they think can bring value to an organization, they don’t have the clinical background to make a strong recommendation as to what an appropriate clinical intervention would be.”
On the other side of the fence, clinicians have been trained to treat the patient in the exam room. “Thinking about how to actively engage an at-risk population prior to a problem occurring is a big paradigm shift for a provider—one that many analytical organizations like Mosaic Medical are still grappling with,” Greene says.
In addition, trying to incorporate new work flows—especially ones based on predictive models that have an inherent level of uncertainty—is difficult in any industry. “At a time when many providers are being asked to see more patients, add relative value units, and check more boxes to meet quality measures, it is hard for clinicians to find time to think strategically, let alone collaborate on predictive model development,” Greene says. “If the analytics team works in silos without frequent input from practitioners, their predictive clinical efforts are bound to fail.”
Donald Fry, MD, executive vice president, Clinical Outcomes Management at MPA Healthcare Solutions, an analytic healthcare consultancy, says that hospitals generally do not have internally risk-adjusted outcomes or cost evaluations of specific clinical services because it is too expensive. In addition, the interpretation of risk-adjusted outcomes needs to be benchmarked to a national or regional level of performance. “Hospitals find little value in conducting risk-adjusted outcomes without a reference group to gauge performance,” he says. “State or regionally based programs of risk-adjusted outcomes use comparative effectiveness and efficiency evaluations of hospitals so that common reporting and common evaluation metrics can be used to drive improvement. Insurers could be key players in the development of state-wide databases.”
Karen Appold is a medical writer in Lehigh Valley, Pennsylvania.