Health Plans Have A Unique Role To Play Making AI Work In Healthcare
Kumar Srinivas is the tech evangelist for the health plan group at NTT DATA, working to democratize AI for healthcare through co-innovation.
How often does a health plan initiate a call to a member ? More importantly, what is it likely calling about?
The answers are “not often” and if they do, it is probably about a claim or payment issue.
What if health plans got proactive about contacting members about preventative care? Many people in the United States suffer from chronic diseases and would benefit from early detection. Type 2 diabetes is an illustrative example.
According to the CDC, 21.4% of adults with diabetes don’t know they have it. Another 88 million of American adults are prediabetic, but only 16% of them know it. The CDC’s “National Diabetes Statistics Report, 2020” estimated that the direct and indirect costs of diagnosed diabetes in 2017 exceeded $327 billion. That doesn’t include the costs associated with undiagnosed diabetes or prediabetes. The many comorbidities associated with Type 2 diabetes, including heart disease, chronic kidney disease and stroke, compound poor health outcomes and healthcare costs.
Health plans that use AI as a predictive tool can improve both patient outcomes and cost containment.
Encouraging Clinical Application Of AI
Using AI to predict early detection of chronic diseases faces numerous challenges. A threshold challenge is building trust of AI among clinicians. The difficulties faced by IBM’s Watson to earn that trust is a reason why the doctors and hospitals have been slow to adopt any practical application of AI beyond increasing administrative efficiencies.
The problem is that traditional AI works using a neural network approach that generates conclusions without context. Doctors aren’t going to respond clinically to a list of potentially prediabetic people spit out from a black box. Healthcare professionals need an explainable AI, what the AMA calls “augmented intelligence.” Clinicians need the AI to provide an explanation and foundation for the conclusions it’s making. They won’t defer to AI-made decisions.
To encourage adoption of AI-supported healthcare by clinicians, the AI needs to operate on a Bayesian model instead of a deep learning neural network. The Bayesian model produces a graphical representation of variables, dependencies, and probabilities that healthcare professionals can use to make their own decisions regarding diagnosis and care. With the clinician in control, explainable AI closes the loop from data analysis to practical use of data that leads to meaningful patient interventions.
Scaling AI Across The Healthcare Ecosystem
Numerous healthcare organizations are exploring how AI can help them execute their mission. However, each organization, whether it is a hospital, lab, pharmacy or a health plan, is operating with its own silo of data. Data silos present a variety of challenges to productive use of AI.
First, the low volume of data in a single organization’s data set makes it difficult to draw actionable, explainable insights. The more information the AI model can analyze, the more refined the analysis is. Bayesian networks rely on prior real world knowledge to generate their probabilities graphs. For this reason, effective AI tools also require a high variety of data points. Medical nonprofits, pharmacies, hospitals and health plans all collect different types of patient information. If the full scope of this data isn’t combined, the AI operates like the three blind mice. Each one sees only its small portion of the bigger picture, leaving it with an inevitably distorted view.
Sharing data across organizations also minimizes the potential for AI bias. AI bias can show up in data sets in different ways. Hospitals or health plans that work primarily with homogenous populations, say Medicare patients or within a narrow geographic region, will have a population bias. AI bias also occurs based on dissimilar treatment standards and protocols used by practices and hospitals. Without combining data from hospitals that have different clinical pathways to address similar diseases, an AI model has only a limited foundation on which to analyze likely outcomes.
Changes in CMS regulations and initiative from certain healthcare organizations are leading to more anonymized data sharing. Data sharing must continue to expand for AI to reach its potential in healthcare. Massive data sets are needed so data scientists can design data maps that produce an accurate, explainable output for clinicians.
Health Plans Are The Natural Hub For Industry-wide Collaboration
Data silos aren’t the only negative consequence of an atomistic approach to AI. The various generic AI engines or homegrown AI tools used by separate organizations also reinforce AI bias. It’s also a challenge for organizations, whose missions are healthcare, to maintain the technological talent needed to keep up with all the changes in AI approaches and tools.
Industry-wide collaboration and co-innovation is a more effective paradigm to building the comprehensive AI tool and data sets healthcare organizations will need. As the financial center to patient healthcare, health plans are the natural hub to facilitate collaboration and communication among healthcare entities servicing patients. Each patient potentially touches multiple pharmacies, practices, hospitals, and medical device vendors and all these entities already have communication channels set up with the health plans.
Because health plans are so uniquely placed, they have a unique opportunity to lead in building the AI and pushing it out directly to members and clinicians. The need for early diagnosis and prevention is key to improving patient outcomes and managing healthcare costs, and it’s where health plans can start using AI with a huge impact.
Every health plan can take away these action items:
• Initiate calls to members and patients. Follow the Hollywood Principle – “Don’t call us, we’ll call you”. They would love to hear from you if you have specific, personalized information that can encourage preventative check-up.
• Look into whether your organization is using explainable AI with your care management and clinicians as they engage with members with chronic diseases.
• Find out what steps your organization is taking to eliminate AI bias, such as combining data from multiple health plans.
AI presents vast potential for expanding the ways health plans can better serve clinicians and members. In my next article, I’ll dig into how health plans can use AI to improve the experience of their large employer groups.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?