Harnessing the Value of AI in Health & Human Services: Early Strategies for Long-Term Success

If you glance at the headlines on any given day, you’re almost certain to encounter a new and startling advance in the realm of Artificial Intelligence (AI). Without question, AI is a very hot topic, and companies and organizations around the globe are making huge investments as they deploy this groundbreaking technology to power productivity and profitability.

In the healthcare industry, exciting AI applications are now being developed, including virtual assistants for patients, fraud detection systems and even new clinical treatments. For example, Canadian researchers recently devised an AI system that can diagnose Type 2 diabetes just from a person’s voice. However, the complex and dynamic nature of healthcare presents unique challenges when it comes to AI, from the availability of data to privacy and regulatory issues. Here are five recommendations to help you get the greatest value from your investment down the line.

1.    Clearly Define Your Goals

This is the first important step in any AI project: what exactly are you trying to achieve? Before you set out to experiment with AI, be absolutely certain you know where you want to go. Without clearly defined objectives, these projects can consume significant resources without a real payoff.

When speaking to state Medicaid leaders about AI, I like to frame goals in the form of prediction statements. For example, you might say “I would like to predict ‘X’” — replacing “X” with variables like, “which members are likely to transition to LTSS,” or, “which claims have the greatest potential for fraud.”

When you structure the problem this way, you’re more likely to create defined objectives to help your project stay on track.

2. Link AI Initiatives to Business Value and ROI

Speaking of goals, be sure to link them to return on investment (ROI) as well as short-, mid- and long-term business value. Too many AI projects are ill-defined, transient, and correlated with hype cycles.

For Medicaid leaders, focusing AI projects on areas within Medicaid systems that have clear inefficiencies and room for improvements — from readmissions to chronic pain management — increases the likelihood of achieving the kind of cost savings that can make a real bottom-line impact.

AI projects have enormous potential to save taxpayers large sums of money by not only predicting behavior but devising interventions that directly address the problem. Let’s say you develop an AI system that predicts the likelihood a patient will develop COPD in the next 60 days or be readmitted with a chronic disease within 100 days. Armed with that information, states and healthcare providers can develop intervention plans that help prevent these events from happening — thereby elevating both clinical and financial outcomes. Diagnosing common and chronic conditions early and as widely as possible is crucial to creating healthier populations and reducing the cost to Medicaid. 

3. Identify Your Approach

There are multiple ways to approach an AI project, and it’s critical to make the right choice from the beginning. A good first question is: would your project benefit from generative AI, or will more traditional AI approaches suffice? While traditional AI uses clear rules to perform specific tasks, generative AI can actually take the information it’s given to learn and create entirely new data.  

Generative AI is the current “rock star” and shiny object in technology, but it’s a complex solution that requires substantial expertise and investment. More often than not, it’s faster and more productive to start simple with traditional data science/AI methods to achieve your goals. 

When determining your correct approach, multiple factors should be considered. For instance, does your approach need to be transparent and explainable? Is reducing bias important? If the answers to these questions are “yes,” then you should employ data science and traditional AI methods. If required, you can grow into the use of generative AI at a later stage.

4. Responsible AI 

Responsible AI refers to building AI in a fair and transparent manner. In healthcare, it’s very important that your approach allows for both transparency and explainability. That means you should be very upfront about the data you’ve used to obtain the results you’ve achieved. Without that transparency, healthcare professionals may be unlikely to trust the technology. Any result should allow for easy explainability, where users can drill down to see the specific data the prediction was based upon.  

There’s also a potential for bias when using healthcare data, which goes back to transparency. For example, suppose you have a dataset that captures the frequency of healthcare utilization in urban areas and another that captures the same in rural areas. Directly comparing the utilization rate between these two datasets without accounting for factors like healthcare infrastructure, transportation accessibility and other socioeconomic variables may lead to a biased conclusion. AI models should be carefully analyzed for bias and take into account the systemic social and economic inequities that exist, as well as the differences in disease characteristics among different patient populations.

5. Develop Your Data Strategy

While generative AI’s explosive popularity has focused attention on this advanced technology, it has also underscored the criticality of data. Data fuels AI, and generative AI is no exception. In fact, it requires more data than previous methods. With data being a necessary requirement for AI, organizations should think of their data as an asset, and develop a data strategy that encompasses the full data and analytics lifecycle.   

When thinking about data, there are still more questions to ask. Where will the data for your AI project come from? Is there enough data to achieve your goals? How will the data be acquired, cleaned, stored, and secured? 

You’ll need clear answers to each of these questions to achieve real results. States should think of their data as an asset that can be analyzed and monetized to achieve key goals, from more cost-effective treatments to better member outcomes. 

In addition, the restrictions of protected health information (PHI) can complicate healthcare AI projects. Often, data needs to undergo de-identification — stripping out personally identifiable information — before it can be used. As a result, AI developers sometimes need to walk a fine line between respecting regulatory and privacy requirements and being able to generate productive findings. 

Summary

Given the complexity of AI, state Medicaid systems will need a reliable partner to set clear goals, invest wisely and address the unique challenges of healthcare. Gainwell possesses the deep, broad Medicaid experience and demonstrated technological expertise necessary to achieve results that will yield substantial savings while improving millions of lives.