While the healthcare industry becomes increasingly adept at applying clinical and claims data to improve care, it has largely ignored other data sources that provide the greatest opportunity to positively impact health and cost at scale. To understand the relationship between consumers’ demographic data, daily-living choices, health and interactions with the health care system, researchers created a proprietary database drawn from multiple private and public sources that represents 275 million consumers across America. The resulting models illuminate the behavioral contributors to health, allowing the identification of interventions and messages that will have the greatest impact in shaping consumers’ behavior to improve health.
While the healthcare industry becomes increasingly adept at applying clinical and claims data to improve care, it has largely ignored other data sources that provide the greatest opportunity to positively impact health and cost at scale. The dependence on this limited data set originates in the system’s orientation toward “sick care” — treating illness. To radically improve health care, we need to apply consumer demographic and lifestyle data in ways that help the health care industry shift its focus from providing sick care to partnering with people (rather than “patients”) to help them stay well.
The government and private sector have dedicated enormous capital and energy to building electronic health record and claims systems to automate and record sick-care transactions. This digitization supports consistent quality of care and payment accuracy, but the data is primarily retrospective – it tells a story of what has been. To most effectively predict future health and how a person will interact with health resources, the health care industry must learn to integrate selected consumer data with medical and claims data.
To understand the relationship between consumers’ demographic data, daily-living choices, health and interactions with the health care system, we created a proprietary database drawn from multiple private and public sources that represents 275 million consumers across America. The data includes disparate demographic and lifestyle information such as income, home ownership, household composition, buying patterns, voting history, transportation choices, social networking activity and physical activity as measured by tracking devices. It also includes clinical, claims and health-resources use data made available through our partnerships with client providers and insurers. We have identified approximately 800 variables that are indicators of consumer behavior and intent that robustly and meaningfully supplement existing clinical and claims data.
Our machine learning algorithms and models combine these data and learn from each combination of data points analyzed. For example, we might analyze the relationship between a diagnosis, income level and commuting methods to identify predictable patterns. Through this sort of data-crunching, we generate a new set of predictive models every ten seconds. These models illuminate the behavioral contributors to health and allow us to identify the interventions and messages that will have the greatest impact in shaping consumers’ behavior to improve health.
We have found, for instance, that household composition and voting history can be leading indicators for emergency room usage. Childless or only-child households are more likely to use the ER inappropriately, which suggests educational interventions and incentives that could reduce such use. Our predictive models can also reveal which health engagement channels will be most effective for a given individual — text, automated voice, email, mail, phone, coaching or a combination approach. (Just for fun, we used the model to determine that Philadelphia Eagles fans are more likely to join condition-management programs if contacted by text than by email.)
In one case, by combining key consumer attributes including socioeconomic status, voting history, education level, medication adherence rates and demographics with a national health plan client’s data, we generated a 16% increase in medical cost savings and a 10 percent improvement in care-management program engagement. This was achieved by identifying individuals who were both at elevated risk of being readmitted to the hospital within 30 days and who would likely be receptive to outreach and respond well to a care management program.
In another case, we determined the likelihood of an individual not recertifying, but still eligible for, a Medicaid program. Our predictive models successfully identified the top 25% who were 1.8 times more likely to not recertify and the optimal outreach channels for engaging them (automated voice and live agent calls). This targeted and coordinated outreach resulted in a 39% decrease in recertification failure rate compared to the control group.
Another way we’re putting predictive models to work is to fill the gaps in cases where clinical data is unavailable or inadequate. One payer client, for example, had a dearth of information for a largely overlooked population segment. The analysis of lifestyle and demographic information revealed that 14 percent of those consumers were likely to be obese and/or diabetic and would be receptive to targeted interventions, giving the client a head start on engaging that population.
Failing to take consumer data into account is to ignore the most powerful change-agents in health care — consumers themselves. Clinical data and expertise are vital, but the only way the healthcare industry will fundamentally improve care is to understand consumers at an individual level – by leveraging information about every aspect of their lives — to create personalized interventions that ultimately drive behavior change and improve outcomes.
Powered by WPeMatico