The Institute of Health Technology Transformation recently released a report which identified data analytics for population health management to be one of the critical capabilities for a successful accountable care organization (ACO). The ability to identify care gaps, categorize patients based on their health risks, and focus on prevention rather than just reacting to health issues has always been considered a key requirement in an ACO model. However, the use of the latest technology advances for large scale, big data analytics across structured and unstructured health data sets has increasingly made the difference for successful ACOs, enabling them to achieve these requirements where others have struggled.
One such organization - Dartmouth-Hitchcock - recently showed just how successful this model can be if relevant patient data is analyzed effectively. It was able to hit all 33 quality benchmarks in the first year of the Pioneer ACO Model, while saving $1.7 million over 17,000 patients enrolled in the ACO. What made them so successful at reigning in cost? It turns out they had a head start of seven years. Dartmouth researchers were looking into the factors contributing to savings under an earlier ACO model already in place before the start of the Pioneer ACO program and gained some valuable insights. One of the key conclusions was that the organization needs to be able to identify patients with multiple chronic diseases and focus their attention on care coordination, prevention and outreach activities customized specifically to that population and those diseases.
The use of technology that can look across and analyze the entirety of patient health data, spotting the equivalent of a needle in a haystack therefore becomes a crucial factor to success. Many organizations are only beginning to realize the need for such large scale data analytics to gain better insights into their patient population. CMS published results on July 16th for all 32 participating organizations for year one of the Pioneer ACO program. While all of them succeeded in meeting quality measures for the first performance year, only 13 of them were able to lower their costs against benchmark accounts. A total of 9 of the participating organizations have indicated to CMS that they won't continue to participate in the second year. 7 of them will apply to the more traditional and lower risk Medicare Shared Savings Program (MSSP), while 2 of them will be dropping out altogether. While there were other factors at play, part of the problem was that these organizations did not achieve the same level of analytics excellence as Dartmouth Hitchcock did.
The Institute of Health Technology Transformation acknowledged in its recent report there is no single roadmap to achieving analytics excellence, but it cited several critical steps for the success of health data analytics in an ACO environment, as summarized by iHealthBeat:
- Identifying care gaps and providing steps to close them.
- Categorizing patients based on their health risks so care teams can intervene with high-risk patients who generate the majority of health costs.
- Changing analytic perspective from episode-based analyses to patient- and population-based analyses.
- Making use of emerging technology to analyze the 80% of electronic health data that is unstructured, rather than solely relying on traditional structured data analytics (e.g. on claims data).
Many prevention and intervention activities depend on early detection of patients at risk of developing serious illnesses. A good example is abdominal aortic aneurysms (AAA). They typically develop in older patients over many years. If detected and tracked early on, preventive measures can be taken (treatment of hypertension, smoking cessation, low-fat diet) or surgery can be performed to repair the aorta long before there is a substantial risk of the aneurysm rupturing, which results in a medical emergency with substantial cost to the health system and less than an 80% chance of survival. Since AAAs are typically only documented in narrative physician notes (that make up a large part of the 80% of unstructured health data) as incidental findings that are easily missed next to the principal diagnosis that is treated and billed for, sophisticated natural language understanding technology that can parse unstructured clinical notes and identify references to AAAs and the size of the abdominal aorta is needed to effectively detect and track such patients.
These are just a few examples of how emerging technologies in support of population health analysis are quickly and vastly improving a health system's ability to identify high-risk patient populations, prevent serious illnesses from developing and to manage chronic diseases in a way that reduces cost and improves patient outcomes.
But we are still in the stone age of using data to drive ACOs and I have no doubt that we will see accelerated successes over the next years, as more and more health systems invest into the technology infrastructure that will allow them to become significantly more successful than the early results of the Pioneer ACO program indicate.