During a Featured Session sponsored by Health Catalyst, as part of Becker’s Hospital Review 11th Annual Meeting in May, Jason Jones, PhD, chief analytics and data science officer at Health Catalyst, shared the framework of Healthcare.AI™ and how it demonstrates improved uses for AI and broader applications in healthcare.
Four key takeaways from the session:
1. AI can clarify data to improve day-to-day analytic activities. Healthcare.AI™ integrates analytics into daily challenges, especially leadership decision-making. There are typically many measures and dashboards, but correct, consistent, and transparent insights are difficult to ensure. People draw different conclusions from the same data. Using modern algorithms and visualization techniques, data can lead to better conclusions.
2. Most predictive models should be built with greater simplicity. “The problem is not whether we can build a technically excellent predictive model,” Dr. Jones said. “The challenge is facilitating the change to act on predictive models.” Health Catalyst recommends tailoring the model and emphasizing efficiency and simplicity. Too many parameters result in a less flexible model because of interdependencies in the parameters. As models become more complex, they are harder to understand and risk losing the trust of providers.
3. Optimizing the deployment of predictive models will boost confidence and understanding. This presents the greatest opportunity to enhance the value of predictive and prescriptive models in healthcare, according to Dr. Jones. Healthcare.AI™ can grow confidence and drive change in workflows. Jones’ analysis relies on three types of understanding to instill justifiable confidence for decisions: 1) Functional model understanding is how the value of an input parameter changes the value of an output parameter, and whether it matches expert understanding; 2) Operational model understanding sets the thresholds for action, the value tradeoffs that rely on human decision-making and cannot be replaced by a computer; and 3) Contextual model understanding asks human experts whether they can do better when the models are wrong. When people agree they could not have done better, predictive and prescriptive models gain trust. When models are trusted, they get used and support the change we wished to see.
4. Use of AI should be extended from transactional predictive models to leadership decision-making. There is an opportunity to “dramatically broaden how and where” AI is applied in healthcare organizations, according to Dr. Jones. Healthcare.AI™ has been used to enhance transactional predictive models, such as who might quit smoking or who is at risk of hospital readmission. Expanding the use of AI in healthcare to our most common analytic tasks and leadership decision making results in better decisions and broader change.
To view the session on demand, click here.