The Ayasdi Machine Intelligence in Healthcare summit concluded with day two dedicated to focused discussions on population health cutting through the hype of population health and focusing on pragmatic initiatives sorely needing the power of machine intelligence.
Bent Christiansen, Senior Consultant at Milliman Medinisght, kicked off the day with a provocative presentation on the elusive “Battle for ROI” and new approaches to realizing value from care management analytics. Bent examined the longstanding challenges to clearly attributing ROI of disease management programs (DMPs) ranging from the impracticality of running randomized longitudinal studies to skepticism on the definitions of evidence of ROI beyond broad financial metrics which are difficult to attribute to DMPs.
Bent suggested a different approach to considering ROI utilizing a “panel of predictive filters”. Such predictive ROI target metrics reflect different dimensions of influence on ROI and as such must be considered in ensemble to yield indisputable ROI with DMP initiatives. The table below is an example of a set of such filters designed to assess 5 dimensions of realizing value
- Clinical Coherence: Is this the right member?
- Impactability: How financially impactable is the DM candidate?
- Actionability: How clinically actionable are the candidate’s conditions?
- Capability: How capable is the candidate of change?
- Engagement: How motivated is the candidate to change?
He quickly pointed out the importance of identifying and targeting the right “microsegments” of the population and the increasing role of Social Determinants of Health (SDoH) in driving these ROI risk targets.
Bent also pointed out the challenge in relying on traditional approaches to developing these metrics. He described the effort required to develop a single metric tapping into decades of subject matter expertise and several cycles of iteration.
Bent concluded his talk with a look to the “New Frontier” and the need for advanced machine intelligence techniques such as Topological Modeling to be able to automatically and unbiasedly identify multi-faceted micro segments of populations to whom different initiatives can be targeted depending on their propensity to impact the totality of ROI dimensions.
Dr. Randall Jotte, MD Associate Professor, Washington University St. Louis and ER Physician at Barnes-Jewish Hospital was up next with an intriguing “Swiss Army Knife Approach to Case Management”. He motivated that such an approach is required for the kinds of patients on whom the health systems seem to have “given up”. Dr. Jotte described the high financial impact of such patients: High ED utilization patients targeted with the Barnes-Jewish “Familiar Faces” program were costing MO Medicaid an average of $177K annually.
Patients in this vulnerable population exhibited a variety of conditions including mental health, substance use disorder, chronic illness and a variety of social issues including becoming “more homeless”, social security payments diverted by relatives and a host of other such issues. Focused intervention through the “familiar faces” program reduced ED utilization from a high of 20+ days in one month to a single visit over the next year!
With MO Medicaid processing over 100 million provider claims annually, and complex cases such as the one above being characterized by several clinical and non-clinical factors, there is immense opportunity to employ machine intelligence to proactively identify these vulnerable patients and discovering other blades in the “swiss army knife” approach.
The third speaker was Dr. Zeev Neuwirth, Medical Director, Population Health, Carolinas Healthcare. Dr. Neuwirth challenged the community with the notion that Transformation is borne of a set of ideas not already part of the core processes. He asserted that effective Patient / Customer engagement is vitally missing from the arsenal of tools being applied to population health. He went on to postulate a framework for “Reframing Patient Engagement with a Marketing Mindset”. The three-pronged framework consists of:
- Re-Branding: Targeting for specific segments and value propositions
- Re-Designing: Demand-side and consumer orientation
- Re-Organizing: Modular business units with distinct focus and value propositions
Dr. Neuwirth shared a vision for how such modular business units may be organized and aligned with the various layers of the risk pyramid. He went on to echo the other speakers on the need for advanced machine intelligence to help with the challenge of prospective patient selection and matching to the branded programs.
Dr. Todd Stewart, VP Clinical Integration, Mercy St. Louis, closed out the speaking session with an alternative viewpoint on “Population health, Vulnerable Populations” focused on missed opportunities “within the 4-walls of our hospitals”. Namely, the limited attention to Hospital Acquired Conditions (HACs) and Hospital Acquired Infections (HAIs) within risk-based contracts.
Dr. Stewart described the “tapeworm of our economy”, the impact of patient safety events: 2M HAIs every year, 90000 deaths per year, $5.7B estimated annual direct cost! The in-patient experience is almost ignored in population health programs. He went on to share that close study of one of the Mercy ACO costs revealed that 1/3 of inpatient cost was for patients with HAIs, HACs
Dr. Stewart described Mercy’s systematic DIVE (Discover, Iterate, Validate, Explain/Enact/Engage) approach utilizing machine intelligence to tackle complex problems such as understanding and managing HAIs. He went on to share early findings from utilizing Topological Data Analysis (TDA) to understand Central Line Associated Bloodstream Infections (CLABSIs). One of the interesting findings showed a strong heretofore unsuspected association between ethnicity differences and CLABSI alongside other associations with specific facilities and medications.
Dr. Stewart extolled the role of machine intelligence in facilitating the iterative “spiral to success”
A succeeding panel discussion featured all four speakers responding to a very engaged audience posing a variety of questions:
- Reconciling the apparent contradiction between personalized medicine and population health: Personalized medicine, while individualized, is derived from what has worked for similar micro-segments.
- Approaches to identifying/labeling “avoidable ED visits”: Efficacy of models such as the RWJF model for preventable admissions.
- Tackling patient privacy issues: Cynthia Burghard raised the hurdle of patient privacy in the use and sharing of large amounts of data required for machine intelligence driven decision support. Bent Christiansen offered perspective on emerging legal frameworks for data sharing.
- Tracking longitudinal patient “Topology of life” trajectories: The need to assemble patient topologies of life across all life attributes that can be predictive of health spending trajectories, health declines, rising risk populations.
In a perfect, unintended curtain closure, Bob Griffin, CEO Ayasdi, drawing from his vast experience, urged the group to consider emulating the intelligence community which utilizes vast amounts of public data on individuals to develop “Pattern of Life Analysis”. He further urged the healthcare organizations present to consider building similar “Intelligence Units” charged with analyzing and reacting to the pattern of life analysis.
As the day wound down, the participants shared their observations on both days – and one remarked that it was interesting that we can now determine how a single cell may impact a tumor, but how we struggle to understand the evolving nature of risk within our patient populations. This, as much as anything, underscored our need to apply new techniques to better understand populations and within those populations find the appropriate treatment protocols for those individual patients. Like the spiral of success noted by Dr. Stewart, our ability to zoom in and zoom out as appropriate will define our ability to deliver value based care.