In an interview with InformationWeek Healthcare, Weiner, who also founded and directs the Center for Population Health Information Technology at Hopkins, said clinical data generated by electronic health records is insufficient for analytic purposes because of shortcomings in that data and the lack of interoperability among EHRs.
"To get a full picture of a population, you need hundreds of doctors, many hospitals, and many payers, and it's going to be a long time until they're all interoperable," he noted.
Health information exchanges (HIEs) can be used to connect providers that have disparate EHRs. But, even in the geographical areas where HIEs are operational, they're nowhere near being able to give ACOs and other healthcare organizations the kind of data they need to manage population health, Weiner said. "Right now, there are only two or three HIEs in the country that are really functional," he added.
[ What kind of tech is gaining the most favor in healthcare? Read Most Wired Hospitals: Health IT's Hot Trends. ]
The paucity of structured data in EHRs is another major barrier to the use of clinical data in analytics. A recent study found that 65% of the data elements required for Meaningful Use quality reporting were unstructured in EHRs, making them unavailable to analytic applications. Weiner predicted that this situation will improve, but only when natural language processing and other "machine learning" tools become more effective and reliable.
"The alternative is very smart and very complete workflows," he said. "But doctors hate workflows; they call it 'click-through hell.' They don't like having to provide all the structure [for data], versus the structure coming in afterwards."
Loosely adapting Abraham Maslow's hierarchy of psychological needs, Weiner said there is a "hierarchy of EHR needs" that must be traversed before healthcare providers can start using health IT to manage population health. This schema can be broken down into five elements following implementation of an EHR:
1. Make sure that the EHR doesn't harm patients.
2. Get clinical workflow back to normal.
3. Figure out how the EHR can pay for itself.
4. Make the doctor's life easier by using functions like clinical decision support, computerized physician order entry, and electronic prescribing.
5. Proceed to quality improvement and population health management.
This five-step sequence cannot be completed overnight, Weiner noted. Yet he's encouraged by the fact that 75% of physicians now have EHRs. Most of that adoption has occurred within the past three years, so we have to be patient and await the outcomes of this large-scale change, he said.
Claims data is not the only route to ACO success. Jonathan Niloff, CEO of MedVentive, said in a recent webinar that a combination of billing data, Surescripts prescription fill data, and lab data can give ACOs the essential information they need, and Weiner agrees.
"An organization that fully uses all of those is off to a good start," he said. "But I'd also say that this is stage 1 of a several-step process. The things that we care about are only partially measured by those data sources."
Weiner wasn't surprised by the finding, in a recent HIMSS Analytics survey, that 35% of providers who had clinical and business intelligence applications were using them to do predictive modeling of their patient populations. Some EHR/HIS vendors have improved their analytic tools enough for providers to begin doing this, he said. However, the EHR predictive modeling tools are designed to work with claims data, not clinical data, he noted.
Weiner has considerable expertise in this field, having led the team that, many years ago, devised the Johns Hopkins adjusted clinical groups (ACG) model, which is used for both risk adjustment and predictive analysis. Most of the leading EHR vendors, he noted, have recently contacted Hopkins and DST Health Solutions, the vendor that disseminates the ACG software.
As provider interest in ACOs and predictive modeling has grown, some EHR vendors have added modules that apply risk stratification algorithms to claims data, said Amy Salls, DST's business solutions director, decision support, in an interview. The vendors are using these analytics with ACO-contracted data from the Centers for Medicare and Medicaid Services (CMS) and third-party payers, she said.