Prakash Menon, Chief Technology Officer, BaseHealth

home

authors

prakash-menon-chief-technology-officer-basehealth

The healthcare industry has a problem with predictive population health analytics.

Healthcare organizations (HCOs) too frequently settle for cookie-cutter population health analytics prediction solutions that are built on narrowly defined data sets that lack accuracy, comprehensiveness and reliability. For the HCOs that invest in these inadequate solutions, the result is missed opportunities — to improve patient care, properly account for risk, reduce costs and receive appropriate reimbursement.

Best-in-class predictive population health analytics must do more than just identify patients who are likely to be the high utilizers who account for high costs in the future. Analytics must go a step beyond, helping HCOs identify which patients are likely to benefit from certain interventions today that will prevent them from entering higher-cost disease states tomorrow — and conversely, recognize which patients are unlikely to improve from those interventions.

To better understand what HCOs now expect from predictive population health analytics, look to Target. The retail giant 

 a few years ago when it mined data around a teenage girl’s recent purchases and, in turn, identified that she was pregnant before her father knew. Target had analyzed the girl’s purchasing history and found a high correlation between some of the products she had recently bought with those purchased by pregnant women, so the company sent the girl coupons for baby-related items.

This was a testament to the retailer’s ability to leverage analytics at the population level to make specific, accurate forecasts at the individual level. This is now the benchmark against which all predictive analytics solutions are measured, and many current population health analytics approaches fall short.

The best path to improve population health analytics starts with a diverse set of data that enables HCOs to develop a 360-degree view of patients, accurately predict risk and focus their resources on the patients who will most benefit from early interventions that would reduce future utilization, reduce costs and improve outcomes.

The following is a discussion of a few key healthcare data types, what each data type reveals about individual patients and how each one contributes to an overall better approach to population health management.

Unlike enthusiastic Target shoppers who may visit the store multiple times per week, most people see their physician once or twice per year. As a result, healthcare data sets are comparatively less comprehensive and of lower quality than retail data.

This contrast illustrates a clear bias in claims data. While this type of data has value as a record of the care that actually took place, it oversamples high utilizers, people who are already sick and those who are really into taking care of themselves. Who gets left out? Those people who have been quietly ignoring their health and are suddenly going to show up at a doctor’s office with any number of serious chronic conditions.

The traditional justification for using claims data to predict population health risk has been that the top 5% of utilizers 

. While that is accurate, our analysis reveals that, of that 5%, just 1% of those patients are most responsible for driving utilization and costs over the following year. The other 4% comes from portions of the population that today have been identified only as low- or moderate-risk and have yet to transition to the high-risk cohort.

This inevitably leads to the conclusion that healthcare needs more data, and the next logical place to look is in electronic health records (EHRs).

Clearly, EHR data will not provide insight into a patient who never goes to the doctor. However, the information does help us understand infrequent users of healthcare who might otherwise show up as a single line in claims data with a code for a regular health assessment, for example. EHR data are qualitatively richer than claims, which helps predict rising risk, especially when employing models that have been trained using clinical risk factors.

Nonetheless, relying on EHR data comes with a real and substantial cost, as historically there have been no standards implemented that ensure semantic interoperability. Effectively, this means that, even though you could read what is stored in an EHR database with some pain, there is no guarantee that the terms mean the same across providers. Many of the problems associated with using EHR data in population health analytics can be traced to the effort required to do this mapping.

Pharmacy data have probably been analyzed longer than most other healthcare data in the U.S. Historically, the information has been used to understand physicians’ prescription patterns and demand analysis. Its use in the population health space to understand patient behavior is somewhat more recent.

A disappointingly large number of patients either don’t fill, or partially fill, prescriptions for economic reasons, with obvious effects on their health. Additionally, observed long-term use of certain drugs without expected change in health status is also a signal for intervention at an individual level.

An often-overlooked source of quality data for population health is lab providers. Some may wonder: If lab test results are also found in the EHR, what is the value of sourcing data from lab providers? In a system where data do not follow patients when they move between health plans and doctors, large lab data providers become a significant source of longitudinal data. Historical lab data also enrich EHR data for existing patients, enhancing the ability to predict new conditions that are likely to surface, which then translates to more accurate risk-factor capture and enhanced revenue to address these new conditions.

In healthcare, the power of data comes from its diversity. It takes many different data types — each with its own strengths and weaknesses &mdash; to drive next-generation predictive population analytics that accurately predict risk and identify interventions that will prevent longer-term, more costly disease states from developing. Only once that happens will HCOs know their patients as well as Target knows its shoppers.

A Smart, Small, Familial Approach to Population Health

Emory, Kaiser Permanente Join Forces to Leverage Big Data, Population Health

3 Keys to Cost-Conscious Population Health Management

Articles

How 4 different types of data improve population health management.

Prakash Menon, Chief Technology Officer, BaseHealth

Articles

Healthcare Orgs Must Know Their Patients Like Target Knows Its Shoppers