By: Navdeep Alam, Chief Technology Officer, Abacus Insights

 Since I wrote this post a year ago, two big changes have occurred.

First, while my premise that we have a data problem that is holding back AI’s potential in healthcare remains true, Abacus Insights has better defined the type of data needed. While much of the industry remains focused on achieving quality data—that is, data that are correct and complete—our experience with the needed applications for data tells us that quality is not enough to power significant breakthroughs in AI or other key changemakers for healthcare.

Rather, we need to achieve a higher standard for healthcare data: Usability. Usable data are correct, complete, up-to-date, relevant, versatile, and use case and application-agnostic while being digitally interoperable. Usable data are key to all tech-driven solutions aimed at reinventing care, equity, experiences and costs.

Second, I said in my original post: “Once we can paint a longitudinal picture of the patient journey that pulls information from every corner of the healthcare system, then we can begin using AI/ML in ways that make decisively positive impacts on the healthcare experience.” Abacus Insights has reached this critical milestone. In the fall of 2022, we launched our Clinical Data Solution, which is now being used by payers to create longitudinal patient records.

To reflect the evolution of our technology, I’ve updated my post to reflect the need for usable data to drive AI.

For a decade or more, healthcare leaders across the industry have been teased with grand promises of artificial intelligence and machine learning and the role these technologies will play in the future of medicine. From automating the creation of digital clinical documentation and advanced clinical analytics to delivering real-time clinical guidance and amplifying patient engagement, AI/ML have been discussed for years as the catalysts we need to realize a more intelligent healthcare future.

Much like children over the course of a long road trip, every now and then healthcare leaders beg the question: Are we there yet?

On the surface, this question makes sense. In social media, for example, AI is used to dictate content, to steer users in their thinking, to guide advertising — because social media companies know what their users read, what they vote for, what they like and consume.

But the road to a similarly intelligent healthcare industry is long, and we have just recently started on this journey. Some tech firms have been able to implement AI/ML tools successfully on a limited scale. Amazon’s Comprehend Medical, for example, uses machine learning and NLP to extract health data from doctors’ notes, clinical trial reports, and patient health records. However, unlike other industries, we have yet to see any widely impactful application of AI/ML in healthcare.

That’s because healthcare data is much, much more fragmented. Our interactions with the healthcare system — our provider visits, pharmacy interactions, lab work — are logged as individual, time-separated events. That information is sensitive, unstructured, and often non-interoperable, making it extraordinarily difficult to visualize the healthcare journey for any given patient. Without a truly complete, correct and up-to-date story of the patient’s journey, told with longitudinal data reaching across all facets of the healthcare experience, AI/ML cannot reach its full potential.

Usable data is the foundation

Any and every accurate and effective AI/ML model is trained on data — and a lot of it. While healthcare, which is brimming with data, may seem ripe for AI/ML innovation, ambitious and forward-looking AI projects from tech giants like IBM and Google quickly learn that healthcare has a serious data problem.

Most healthcare data is unstructured and semi-structured, coming from different sources and in different formats, and is often incomplete and/or duplicative. It is also extremely sensitive and highly regulated. This is why, for example, so few AI solutions in medicine have been given the greenlight by medical device regulators: at the end of the day, the efficacy of these tools is hampered by the quality of the data they’re built upon and the depth of consent administrators are able to garner.

Ideally, for health plans, AI can one day be used not only to inform patients that they are sick today but predict their healthcare journey down the road by detecting the patterns and preconditions leading patients toward chronic conditions like diabetes or diseases like Alzheimer’s. When can AI/ML truly be realized to let us know what will happen tomorrow so we can act today? There’s a reason why we haven’t seen this level of innovation. It all comes down to data.

Ultimately, you cannot train AI to be intelligent about the healthcare outcomes of a patient if the model does not meet standards for usable data about that patient and millions more. With healthcare data in its current state as it exists within most health plans, AI models cannot be used in any medically meaningful manner.

Realizing the promise of AI in healthcare

Currently, AI’s role in healthcare is much more nascent than many may understand, but its limited current applications are still important steps toward realizing the big AI dreams many of us in the healthcare industry hold. In order to deliver the full promise, we must start by rethinking data management and data engineering.

Our roadmap has Abacus Insights harnessing AI/ML to improve data quality, specifically data observability, and improving our ability to detect metric drift early in our data pipelines as we process structured and semi-structured data such as claims. This application will ensure we deliver increasingly better data to improve usability in downstream systems that impact the everyday experience for health plan members and other operational and analytical uses.

The future of healthcare begins with usable data. This is a prerequisite for AI/ML, but also for many of the other grand visions we have for the future of healthcare.

Without usable data, success in value-based care models will continue to be difficult to achieve. For the U.S. healthcare system to incentivize higher-quality, cost-effective care, payers and providers need usable data that can help them find the signal in the noise — just the right information that can be used to properly risk-adjust, guide clinical decision-making, reduce unnecessary care while promoting proactive preventive care, and ultimately be leveraged to improve patient outcomes while reducing costs.

And, as healthcare companies make inroads in data usability, precision medicine can be used to activate from the “wisdom of the crowd,” drawing from the insights of many to guide the care of any given patient.

As an industry, we understand that the promise of AI/ML relies heavily on long-term data management strategy, and we’re in the early stages of implementation. To move forward requires the continual application of advanced data engineering concepts and an undying willingness to create and harness usable data. Once we can paint a longitudinal picture of the patient journey that pulls information from every corner of the healthcare system, then we can begin using AI/ML in ways that make decisively positive impacts on the healthcare experience.

So, are we there yet?

In reality, a year ago when I wrote this blog, I said we may have just left on our journey to usable data.  I will now say we’ve made major strides in only 12 months, unlocking the standard for a new era of healthcare data that will fuel needed changes and goals to produce better access, patient and provider experiences, outcomes and affordability. All we need now is recognition and adoption of these new standards. The rest of our goals for healthcare improvement will follow.  What are we waiting for?