Thank you for Subscribing to CIO Applications Weekly Brief
Applying Deep Learning to Streamline Healthcare Administration
Sanji Fernando, SVP Artficial Intelligence & Analytics Platforms, Optum
First, applying artificial intelligence to clinical decisions typically has greater regulatory burdens and implications to patient care. Administrative processes typically do not directly impact clinical care. As a result, we can put these tools in place for administrative improvements far more quickly and broadly. Beyond that, there are so many areas in health care administration with complex and time-consuming processes –coding, risk adjustment, prior authorization and provider directories – that could benefit from the automation deep learning brings that even small improvements can have a big impact.
Secondly, health care administrative processes are well suited for deep learning because many interactions between a care provider and an insurer are a dialogue on the level of care and payment for that care. When a health care system can come to an automated decision that both payers and providers are satisfied with, it eliminates a lot of inefficiency and volume by changing the process to require human review only in cases of exception.
Many health care executives already recognize the potential of AI driven applications.
Advanced technology like deep learning allows us to harness the vast amounts of health care data generated daily and put it to work making a better system for everyone
According to a 2018 Optum survey of 500 senior-level leaders across the industry, 43% have begun implementing deep learning by automating business processes such as administrative tasks and customer service. Thirty-six percent are employing it to detect patterns in health care fraud, waste and abuse.
Revolutionizing case reviews
Let’s look more closely at case reviews for medical necessity as an example of a critical administrative process ripe for deep learning. The traditional method involves physicians manually reviewing lengthy patient medical records to determine that inpatient admission, for instance, is justified. If every record is reviewed, this becomes a very time-consuming and costly process.
In contrast, a deep learning neural network can be trained using hundreds of thousands of complex decisions made by physicians in past case reviews. By analyzing past decisions, the neural network can determine which cases are complex enough to require a physician advisor review. The more cases it reviews, the more accurate it becomes further reducing the time, cost and number of denials.
Avoiding the ‘black box’ problem
Deep learning works by essentially turning text into numeric scores, but the process can seem like something of a “black box “if clinicians aren’t given insight into the “why” behind a prediction or classification.
To overcome this, it’s important to help staff understand and have some say into how data is scored. For example, work with clinicians and demonstrate how notes and even which particular words are scored so they can confirm that they would attribute the same importance to those words. Bringing more transparency to the process is vital for achieving buy-in from stakeholders and trust in the model.
Impact on costs and care
Developing and deploying innovative and efficient solutions to improve coding, claims processing and case reviews and prior authorizations promises to save millions. In fact, U.S. insurers can unlock $7 billion in total value — 10% to 15% of operating expenses — in 18 months by using artificial intelligence to automate certain core administrative functions, according to a study from Accenture.
Lowering per capita costs is just one aspect of the Triple Aim. But, as health care leaders, tackling costs also enables us to better address making better care more available to more people. Advanced technology like deep learning allows us to harness the vast amounts of health care data generated daily and put it to work making a better system for everyone.