Predictive analytics are revolutionizing healthcare, as wearable IoT and sophisticated mobile apps collect data on individuals to inform them and their caregivers about their health. This data is invaluable for spotting incipient conditions of concern and nipping them in the bud. But IoT, wearables and mobile apps aren't the only sources of data that can predict health issues: Old healthcare claims data is a great source, and potentially a richer one.
Health snapshots, health documentaries
While IoT and mobile apps capture an individual's health parameters in the moment, claims data can contain a staggeringly detailed health history. That historical claims data aggregates more than vital signs.
Wearable IoT and mobile apps that track pulse, respiration and blood pressure in the moment and record weight, diet, footsteps logged and other activities over time, are fine, in and of themselves. Added together, they amount to a scrapbook, but of snapshots only.
A patient's claims history contains snapshots like those above and much more: diagnostic data, prescriptions consumed, test results, and often rich demographic details.
It's more than a scrapbook of daily activity snapshots; it's a full-blown health documentary that includes all the other players, such as physicians, care managers and family members. There is more to work with because this data fleshes out the context of the patient's life in ways daily health snapshots can't. This provides a rich and informative backdrop that can shine light on the path to an emerging pathology or risky condition when mined for data.
When a provider or care management organization uses healthcare claims data for predictive analysis, they can study their entire patient population to identify indicators that can flag the possibility of problems in a current patient's health.
Seeing red flags
Any given population of patients will include groups that have presented specific ICD diagnostic codes for any number of diseases and disorders, and each person with a positive diagnosis will carry a rich history of peripheral details. Identifying those groups, and then analyzing their claims histories for those peripheral details -- some of which may be commonalities shared with others -- can potentially produce a set of indicators that serve as warning flags to look for in future patients.
How is this achieved? Cluster analysis is the preferred analytical approach (principal component analysis and K-means cluster analysis are two highly effective methods). The idea is to isolate particular risk conditions or known groups of conditions, into a subset of data to be analyzed --for instance, all patients in the population who have been diagnosed with hypertension -- and apply cluster analysis to that data set to see what preconditions might consistently present.
Once this occurs, demographic data can then be examined in the positive results. Many socioeconomic factors -- socioeconomic status, marital status, race, domestic violence, history of alcohol or substance abuse -- may commonly appear, as clusters, with enough frequency to be considered red flags in the subject population. Assuming statistical significance, it becomes possible to watch for these factors in the claims stream, flagging those of concern -- and then alert physicians and care managers early on to watch for signs of anxiety or depression, addressing them before they lead to a potentially fatal case of hypertension.
This style of analysis can establish a risk ratio -- a specific, numeric likelihood of a risk condition's emergence -- based on its historical occurrence in the population.
This is just one of many thousands of possible scenarios.
Prevention, not cure
This predictive approach is already being employed by care management organizations servicing Medicaid populations in several states and can potentially save billions of dollars in healthcare expenses annually. If analytics can detect a risk condition before it becomes an illness or disorder, the cost of intervention is far less than the eventual cost of treatment.
How is this better than the existing system, where physicians and care managers have patient history in front of them as they are evaluating and advising a patient under care, catching risk conditions as they emerge? It's better because a digital system can review the data before a human performs the evaluation. Healthcare claims data can be reviewed automatically, day in and day out, and every single patient is reevaluated with each new day's additional data. IT can place patients under constant scrutiny, while human beings can only evaluate one patient at a time, on the day they are in the office.
Healthcare has always been an odds problem: What is the likelihood that any particular person is going to develop any particular condition, given any particular circumstances? A rich body of healthcare claims data can build out that entire landscape and build a superb predictive model that spots emerging conditions before they become serious. This allows healthcare providers to intervene early to suppress the problem before it even occurs.
A new layer of metadata
These analytics can create a new layer of metadata around every patient, facilitating more effective care and tremendous efficiencies in the healthcare system. The patient is no longer defined as "age x," "gender y," or "income level z;" now the patient is defined as "incipient diabetic" or "at risk of cardiac event" with a specific severity level. That's something the physician and care manager can work with.
It's no longer a question of demographics and the surface pattern of patient behavior that defines a patient; it's a deeper set of features that can trigger preemptive care and attention, to the benefit of the patient and the system overall.
For many organizations, the data is sitting there, waiting to be explored -- and the patients are there, ready to be treated more effectively. That means money saved, workload reduced, value realized and lives changed.