Healthcare is full of ideas that are easy to conceptualize but difficult to put into practice. Both clinician network management and patient engagement, for example, make perfect sense on paper but get very complicated, very quickly, when you move beyond the planning stages.
Population health management is the granddaddy of these concepts. Chilmark Research’s recent Population Health Management Insight Report defines the topic – the proactive management of the health of a given population by a defined network of financially linked providers in partnership with community stakeholders – but indicates that a clear head-to-tail PHM solution is still several years off.
The recent Cerner Population Health Summit shed a bit of light on the myriad challenges that healthcare organizations (HCOs) face in their ongoing efforts to address PHM but also offered some hope that progress can, and has, been made. Here are my thoughts on what was presented, discussed and shared at the event.
Risk Stratification Needs “Impactability”
Many of the organizations at the event can be described as PHM hipsters. Advocate Health Care, Intermountain Healthcare and Memorial Hermann, to name three, all began their population health initiatives years before the signing of the Affordable Care Act, when PHM suddenly became mainstream.
These HCOs centered their efforts on readmissions – and with good reason, given the readmissions reduction program (and corresponding penalties) introduced in the ACA. Plus, as David Wennberg of the Northern New England Accountable Care Collaborative pointed out, 65 percent of healthcare spending is associated with inpatient admissions and their aftermath.
Here, most HCOs focus on risk stratification, pinpointing which patients present the highest risk of returning to the hospital and, in addition to negatively impacting the overall readmission score, consuming valuable resources. Therefore, they stratify patients according to cost.
That’s a mistake, according to Dr. Rishi Sikka, senior vice president of clinical transformation for Chicago-based Advocate. Most high-cost patients eventually regress to the mean, so intervening on their behalf is a waste of time and money. That’s why Advocate, working with Cerner, developed a readmissions algorithm that stratifies instead according to “impactability,” or the likelihood that a patient’s condition can be improved through intervention.
According to Advocate, the algorithm predicts readmission 20 percent better than the industry average – provided, of course, that clinical staff actually use it. That points to the need, mentioned by several presenters, to incorporate analytics tools for PHM into clinical workflows. That in turn points to the need, also mentioned by several presenters, to take change management seriously. And that in turn points to the need to ensure that the data input into these evidence-based predictive models is reliable enough that clinical staff will take their output seriously.
The Path from Data to Insight
Based on the aggregate experiences of those organizations at the Population Health Summit, I see an eight-step process from going to raw data to actionable insight. Depending on the data source, this process can range from easy to downright impossible, so HCOs may find themselves on different steps depending on the data type.
- Acquire data. This includes claims, clinical, financial, pharmacy and government data, along with anything else an HCO can get its hands on. Advocate has data from more than 50 sources, including non-Cerner EHR systems such as Allscripts, eClinicalWorks, Epic and Meditech.
- Digitize data.
- Integrate data. More than one PHM hipster alluded to the days of using email and Excel spreadsheets for reporting. This will not work for the purposes of predictive modeling.
- Collect, reconcile and standardize data, all so it can occupy its spot in a clinical data warehouse. (Cerner technology maps this data using LOINC.)
- Apply predictive modeling and analytics.
- Deploy the predictive model into the clinical workflow – ideally, right in the EHR, so clinical staff need not open yet another application.
- Integrate with third-party systems. Nash Health Care in North Carolina integrated the readmissions tool within its Curaspan physician referral system, as this allowed for better readmission tracking by individual provider.
- Take action on the data at the point of care – ideally, again, right in the EHR or third-party system that clinical staff already use. Otherwise, the predictive modeling becomes little more than an academic exercise.
It’s no coincidence that many of the heavy hitters presenting at a PHM conference describe themselves as accountable care organizations. ACOs, by their very nature, cannot achieve shared savings without placing an emphasis on PHM. It helps that ACOs, by their nature, also tend to be large, integrated systems with the financial and technical resources (including data) to embark on PHM in the first place.
Via Twitter, I asked Matthew Swindells, senior vice president of population health and global strategy, if PHM could scale downward. His reply:
Yes. We believe the same strategies and capabilities can be leveraged on a smaller scale. In the same way that we offer a multi-tenant version of our EHR to make it affordable for small community hospitals, we are doing the same thing with our HealtheIntent solutions. The interesting question is how a small organization positions itself to manage financial risk and take advantage of the changing financial regime. We are considering whether we have a role to play in helping our smaller clients collaborate to pool risk and take on capitated contracts.
This financial risk comes in two forms: The risk of not being able to lower the cost of care and the risk of not being able to afford the care management, predictive modeling, and decision support software (among other tools) that helps lower the cost of care. The risk also stems from the fact that PHM remains an acronym in search of technology; there’s no single package that accomplishes everything a PHM strategy encompasses. It’s good to see Cerner focusing on smaller HCOs, but I expect that it (and other vendors) have yet to fully understand what these organizations need.
Overall, the event suggested that PHM is starting to mature but also reinforced the idea that true, honest-to-goodness PHM remains several years in the making. Most celebrations at the Population Health Summit focused on reduced readmissions – a laudable goal, yes, but only one facet of a PHM strategy.
Perhaps that’s why the presentation that struck me most came not from an HCO but, rather, a partnership among Cerner, the Mississippi Division of Medicaid and the nonprofit Delta Health Alliance, which serves an area of Mississippi where PCPs are in short supply and ER visits all too common. The partnership targeted three sobering stats about the state’s overall health – the state ranks first nationally in low birth weight, second in obesity and third in diabetes among adults – by deploying a predictive algorithm that used health and demographic factors to determine who was most as risk of getting diabetes or delivering a baby pre-term birth and then intervene accordingly. The pilot is projected to cut $9 million in pre-term birth costs and $5 million in diabetes treatment costs in a five-county area, compared to a $5 million technology investment.
This case study gets to the core of impactability in population health management: Finding the people who need and want help before they end up in the neonatal ICU or in the emergency room with multiple comorbidities. The work of the PHM hipsters is certainly worth watching – but so, too, is that of nonprofits working with PCPs in some of the nation’s poorest counties.
I agreed Point 1 Population health management is the big goal of healthcare industry, and personal health goal is very important point in the health PHM.
Indeed. It’s important that organizations not neglect the individual goals of personal health as they also pursue the larger goals of population health. It’s certainly a process, but the ability to aggregate data for the purposes of predictive modeling can help on a personal as well as a population level.