Hospitals are a critical component of any modern society, but too many hospitals in the U.S. are deeply struggling in many ways.
First, hospitals are struggling financially. The COVID19 pandemic shook the finances of many hospital systems while the people at these organizations rose to the moment to prevent a once-in-a-century pandemic from being worse than it was. Reimbursement rates have not kept up in wages or in the additional administrative burdens that “payers” (or “deniers” depending upon whom you’re speaking with) have placed on them.
Second, hospitals are also struggling to remain adequately staffed. Depending upon the source, between 100,000 and 200,000 nurses and physicians left the workforce during the COVID19 pandemic. There simply aren’t enough clinicians to take care of people. Unfortunately, this problem is only going to exacerbate over the next decade as the baby boomer population ages into their 80s, which may also double Medicare expenditures.
AI alone will not replace nurses or doctors over the next 5-10 years, but we believe that AI can help to create a significant boost in productivity to ease some of the strain.
Many people who have been in the healthcare industry for more than a decade may look at AI and be tempted to cite the false promise made to reduce work when Electronic Health Records (EHRs) were rolled out en masse starting in the late 2000s. It’s now widely agreed that this generation of EHRs ended up contributing more to the current issue (burnout) than it helped to reduce work. What’s maybe less commonly understood is that the majority of the data that our clinicians enter into EHRs is effectively unusable for any type of analysis because it’s buried in the record in a non-standardized and unstructured format.
Until it is structured, these unstructured data are effectively useless for quality improvement efforts or for clinical research purposes. Left unstructured, too many of these words represent time that is wasted by our most valuable clinically-trained people.
Newer generations of AI, however, have the potential to help solve some of this data structuring problem. In one collaboration that is still confidential, our team demonstrated the ability to abstract unstructured quality data and to structure in a way that it would be ready to be submitted to a major quality registry, achieving higher levels of accuracy than our partner saw with previous technologies they had tested such as NLP.
When it comes to saving clinician’s time, we believe that this generation of AI will have a much more positive impact than migrating from paper records to electronic records, and even from siloed records to more interoperable ones. When applied by somebody with the right skills, the core technology is now there not only to capture and store the data, but to reduce the effort required to do so. It’s finally time for clinician’s to stop serving their EHRs, and for their EHRs to start serving them, with the help of AI.
Our team’s unique background puts us in a unique position to be the first to solve this data abstraction problem with AI. Our CTO, Russell, was the previous CTO of the market-leading tech-enabled services company in the quality reporting data abstraction space, Q-Centrix. He then went on to lead machine learning projects at Amazon, and to spend several years deep diving on AI at the Paul Allen Institute for AI in Seattle, where he was also published for his role in developing an open-sourced LLM. Our Head of Engineering, Alyssa, worked with Russell at Q-Centrix, where she built certified tools for quality data collection for several different quality registries. She brings additional practical knowledge in the quality registry space as well as some serious engineering chops. They were joined by Jeff, who has designed software for remote nurses and other clinical staff. Jeff proceeded to grow revenue over 4x year-on-year, leading to an oversubscribed $15.5M Series A round. Jeff also worked on product and data strategy at Freenome, a machine learning company in the diagnostics space that has raised over $1.4 billion from organizations including a16z, Roche, Kaiser Permanente and others.
The cost of healthcare in the U.S. has risen dramatically over the past two decades, with no sign of slowing down, to the point where it’s undermining the financial health of our nation. The U.S. spends about $4.5 trillion per year on healthcare and the federal government has a forecasted budget deficit of about $2 trillion per year. We also spend about twice on healthcare vs. what other developed nations do on a per capita basis. So if we were able to bring our healthcare expenditures in line to what other countries spend we could have close to a balanced budget. McKinsey & Company, attributed almost $1 trillion per year of the total U.S. healthcare expenditures to administration, and identified $250B in potential cost savings, and this did not even factor in the potential benefits of AI [link].
There are many reasons why healthcare is so expensive, but a major factor is the legions of back office clinical staff that are employed by hospitals to review charts and structure data for quality reporting, billing and coding and other tasks. Payers also employ clinical staff to replicate and double check much of this same work in an effort to find a reason to reject as many claims as possible. An average health system may spend millions or tens of millions of dollars per year on staff that does this work. Despite these efforts, they may also be losing millions or tens of millions of dollars per year in the form of sub-optimized reimbursement processes that are either outright missed or are rejected by payers.
This administrative version of a Rube Goldberg machine has also undermined the transition to value-based care. Value-based care sounds great in theory, but in practice it requires an even more complicated type of quality-based accounting system for health systems. Implementing this in practice requires standardized measurable definitions of quality and even controlling for differences to enable compensation to be adjusted based on the population’s risk levels. Today, too much of the value that is created by aligning incentives in the right way is lost in additional complexity of the administration.
The type of measurement and reporting required to keep the entire value-based care moving is already difficult, but it’s impossible to compare apples to apples without accurate and structured quality data. This is one reason why most health systems today employ dozens or even hundreds of people in their quality departments, including many people (often nurses) whose main job function is to “abstract” data, meaning that they look in the patient’s EHR and create structured data reports from a combination of structured and unstructured data. These abstraction tasks alone cost the U.S. health system billions of dollars per year, but it’s also common that the needed staff simply cannot be hired. In addition to the well known nursing staffing shortage, this type of work can be monotonous and solitary, which are not commonly cited reasons that most nurses cite for entering the nursing profession.
In the next decade, health systems that successfully tap into the power of AI will have a fundamental advantage. Given the tight margins in this industry, those that fail tap into the cost-saving potential of AI might even struggle to sustain themselves financially. A part of our mission is to democratize access to AI to make it available to all types of hospitals and health systems, including the smaller and more regionally-focused ones, so that they are not as overpowered as payers also adopt these tools. We believe that well-functioning hospitals are a critical part of society, and we will do what we can to help and support them.
Over the next 10 years the promise of the value to be delivered by a modern EHR will finally be realized, with AI playing the leading role in delivering on that promise. The administrative efforts around data entry and abstraction will gradually subside and our doctors and nurses will be able to spend more of their time with patients again. AI will not fix healthcare’s inefficiencies overnight, but it should be able to start to deliver an impact just in time to catch up to relieve some of the strain that our already fragile healthcare system will face as more and more baby boomers age into their 80s and 90s.
These are exciting times in healthcare and AI. We’re thrilled to be building something that we believe will reduce the time that clinicians have to spend in EHR while enabling the system to operate at a higher level more efficiently.
- Russell, Alyssa, and Jeff