In the United States, healthcare expenditures far outpace the rest of the industrialized world, without an improvement in outcomes that matter. Moreover, within the United States, there is wide variation in the performance of providers in caring for complex patients.
The best providers are able to use nuanced decision making to help prevent patient deterioration, reducing emergency department visits and hospitalizations while curbing total costs of care. Often, this involves minimizing the use of “low-value care” --- unnecessary, ineffective or inadequate imaging, diagnostic testing, use of branded drugs and other care decisions sub-optimal for the unique needs of individual patients. Low-value care decreases the resources available for high-value care. It also increases the risk to patients while delivering little benefit.
At the Becker's 8th Annual CEO & CFO Roundtable in Chicago, HEALTH[at]SCALE Technologies, a Silicon Valley company founded by leading machine learning and clinical faculty with ties to MIT, Harvard, Stanford and U-Michigan, hosted a workshop to explore how machine intelligence can drive value-based care in two ways. First, by matching patients to the providers who are most likely to provide high value-based care based on a patient’s unique medical characteristics and needs. Second, by identifying patients at greatest risk of deterioration based on their individual multi-factorial health trajectories, with information about modifiable risk factors and care pathways to reduce this risk.
The limitations of traditional approaches to value-based care
Value-based models of care are meant to reduce costs and support better outcomes. To date, payers have tried to achieve this goal through a variety of approaches. For instance, Medicare and many commercial payers have instituted rules about which treatments are covered and which are not. CMS and many insurance companies have also instituted ratings that reflect provider adherence patterns to evidence-based guidelines and cost containment. However, rules and static provider ratings are a “one-size-fits-all” approach to healthcare, and do not reflect the clinical nuance needed to manage complex patients.
HEALTH[at]SCALE, the leader in machine intelligence for precision care delivery, uses machine intelligence to build predictive models with deep understanding of both patients and providers --- with the goal of identifying specific patients at greatest risk, uncovering their risk factors and pathways to target this risk, and matching patients to the providers that are likely to deliver the best value-based care and outcomes, while lowering costs.
Mohammed Saeed, MD, PhD, CMO of HEALTH[at]SCALE and a faculty member at Michigan Medicine, explained, "The problem with the hard rules-based approach is that such approaches lack clinical nuance, and fail to predict the needs of complex patients. Rules-based solutions create a system of friction between payers, providers and patients."
Rules-based approaches also lead to rigid decision making. Taking clinical nuance into account is important. For example, screening a 30-year-old with no risk factors for colorectal cancer is low value. This same procedure may be high value in someone of the same age who has a young first-degree relative diagnosed with colon cancer.
Machine intelligence leads to precision care delivery and value-based care
Machine intelligence offers healthcare organizations an opportunity to exploit rich, dynamically-evolving longitudinal clinical and claims data to develop a deeply contextualized understanding of patients and providers. Modeling complex variations across patients and providers enables personalized and predictive value-based care delivery to improve outcomes and costs.
Zeeshan Syed, PhD, CEO of HEALTH[at]SCALE and a former faculty member at Stanford University and U-Michigan, expanded on this, "The future of value-based care lies in precision care delivery powered by machine intelligence. For health plans and providers to achieve the best possible outcomes for their populations, they must match the patients in these populations to the right treatments by the right providers at the right times. To do this, we need new kinds of machine intelligence technologies that can holistically integrate data from EHRs, claims, wearables and social determinants of health, and make precise statements about how individual patient outcomes vary across complex care decisions."
Although artificial intelligence and machine learning has been applied to a wide range of domains, the analytical challenges in healthcare are unique in that they are driven by small data problems rather than the challenges of big data. While large amounts of data exist in aggregate in health systems, the quantities of interest are often small, and filtering for relevance, as might be expected when looking at specific providers, diseases or outcomes, greatly reduces the amount of data available for modeling. Conventional artificial intelligence and machine learning struggle in this setting. As a result, a new type of specialized machine intelligence is needed to produce very fine-grained inferences about patient outcomes from datasets that are deeply contextualized and multi-parameter, but where the number of observations might be small.
Precision care delivery can revolutionize the outcomes and cost-effectiveness of many procedures
Individuals who are hospitalized and then receive post-acute care either at skilled nursing facilities (SNFs) or through home health agencies (HHAs) often have high rates of hospital readmissions. This leads to increased patient morbidity and mortality, as well as higher penalties from CMS for organizations with high readmission rates.
Using specialized machine intelligence for precision care delivery, HEALTH[at]SCALE has developed a precision navigation application that identifies the best SNF or HHA for a given patient who is being discharged. The technology company has found that Medicare patients who go to SNFs recommended by the application have 14 percent lower costs, 23 percent lower 30-day readmission rates, and 21 percent lower ED visit rates. For those who receive care from recommended HHAs, there is a 6 percent reduction in costs, 28 percent fewer 30-day readmissions, and 17 percent lower ED visit rates.
HEALTH[at]SCALE’s precision navigation application has also proven to be effective for matching patients with specialists, for example cardiologists and orthopedic surgeons, and primary care physicians. Dr. Saeed observed, "By pairing patients with the right provider for knee replacements, we can reduce ED visits by 2 percent, the 90-day hospitalization rate by 2.5 percent and the 90-day total cost of care by more than $4,000."
The machine intelligence application can also identify the patients at highest risk of going to the ED within the next six months, and identify the specific potential causes associated with future ED visits. With this approach, health systems can achieve value-based care by targeting patients, getting them out of the ED and moving them onto more stable trajectories, aided with precise and actionable data.
Conclusion
To achieve value-based care, health systems must first understand the limitations of traditional models of care as well as the analytical technologies that have supported these traditional models of care but may not be applicable to new use cases for value-based care. Machine intelligence solutions specialized for precision care delivery enable organizations to proactively match every patient to the right treatment by the right provider at the right time. The goal is to navigate patients to providers and networks that will deliver high-value care.
Dr. Syed noted, "Precision care delivery offers tremendous opportunity for value-based care, making payers, providers and patients smarter about choosing providers, treatments, and times and modes of early prevention. Moreover, once you get good at predicting what should happen in terms of patient care, you can compare that to actual patient outcomes. This enables organizations to identify fraud, waste, abuse and errors with a completely different lens. The opportunities of precision care delivery through machine intelligence are endless."