One of the key takeaways from this year’s HIMSS meeting is that a lot of healthcare executives are dismayed by the underwhelming results of their massive investments in health IT.
Many of them blame the analytic features in their EHRs and enterprise data warehouses, which have been unable to meet the requirements of population health management and financial risk management.
At the same time, the show featured a plethora of AI applications that apply machine learning techniques to analytics, including predictive modeling. About 80% of these programs were vaporware, but some of the other AI solutions had racked up impressive wins in areas as diverse as population risk stratification, revenue cycle management and patient acquisition.
AI is particularly well-suited to population risk stratification and clinical variation management, both of which are key to managing financial risk. With the convergence that is now occurring between payers and providers coupled with the move to value-based care, these capabilities are now mission critical for every player in healthcare.
Population risk stratification can help healthcare organizations, health plans and self-insured employers to:
• Gain a holistic, prospective assessment of their at-risk populations across multiple risk vectors, including disease progression, cost and utilization of resources
• Develop a clear understanding of the clinical and non-clinical drivers of risk for each subpopulation, which can inform targeted interventions
• Match patients’ predicted health risks with targeted care protocols for each stage in their health journeys, from preventive and chronic care to acute and post-acute care
In addition, healthcare providers and payers need analytics to figure out how to eliminate unnecessary care variations for expensive and highly variable episodes of care, such as joint replacements, cardiac surgery, and hospitalizations for congestive heart failure.
Population risk stratification is often considered a predictive modeling problem, but it is more than that – it requires the ability to construct, in an unsupervised (and unbiased) way, groups of patients that share similar characteristics. These characteristics are not univariate such as cancer, diabetes or chronic pain. They are multifaceted disease states that incorporate multiple diseases. Traditional approaches miss the multivariate nature of diseases and as a result, craft care plans that treat a single dimension of a multi-dimensional problem.
This is where AI is so valuable. For population risk stratification, AI solutions can examine a virtually unlimited set of patient, member or employee attributes contributing to the health of individuals, detect systematic patterns predictive of health degradation, and provide precise, high-resolution stratification of at-risk populations.
To guide payer and provider programs aimed at reducing unwarranted clinical variations, AI solutions can examine longitudinal clinical data to surface treatment regimens that have yielded the best outcomes at the lower cost for specific subpopulations.
AI use cases
AI’s predictive capabilities, based on its ability to detect hidden patterns in diverse data, is what makes it superior to traditional analytics. While an EHR might be able to provide a list of patients with diagnosed orthopedic problems, for example, an AI-powered risk stratification solution can uncover a subpopulation with recurring and worsening orthopedic issues, and predict which ones are likely to need joint replacement surgery. The risk-bearing entity can then intervene or prompt the patient’s provider to intervene to modify their conditions so that surgery may not be needed in the long run. If the patient does need a joint replacement, the AI solution can steer him or her toward the local facilities and surgeons that demonstrate the best outcomes at the lowest costs.
Similarly, by identifying “rising risk” patients in early stages of chronic diseases, AI can help providers and payers ensure that these individuals receive proper care so that their conditions don’t worsen and lead to ER visits or hospitalization. AI can also use both clinical and non-clinical data to continuously monitor population health risks and innovations in clinical practice that can reduce these risks.
While AI can illuminate potential pathways for clinicians, physicians and administrators must ultimately decide whether or not to act on its advice. Critical to this process is the ability for AI to explain its recommendations. The “black box” algorithm may work against test data, but it will be difficult to find traction in day to day practice. Successful AI will be able to explain to the doctor and nurse how it arrived at the recommendation – in ways that they can understand.
Technology is just one factor in a broad-ranging strategy that requires healthcare organizations and payers to change how they’re organized, how they deliver or influence care, and how they try to reduce costs. Nevertheless, AI solutions, which use machine learning to derive new insights from large amounts of data, are much more adaptable and capable than traditional analytics that have been used for the same purposes in healthcare. Because of their applicability to population risk management and clinical variation management, AI solutions are bound to become key players in the quest to manage financial risk.
###
By Gurjeet Singh, CEO of Ayasdi