A Survey Of Large Language Models Accelerating Healthcare Businesses Today

Over the past 18 months, large language models (LLMs) have emerged as a breakthrough technology, promising step changes in efficiency for numerous moderately complex but repetitive workplace tasks. In December 2023, we surveyed friends across our network and portfolio to identify early adopters of LLMs in healthcare, and to discern any emerging patterns about their application and business impact.

While many industries are racing to take advantage of these new technologies, healthcare is particularly ripe, due to the long tail of fragmented and semi-structured data lurking beneath the surface, as well as the tens of billions of dollars spent annually on manual, administrative interactions. Mario Schlosser, co-founder of Oscar Health, says it well: “LLMs are uniquely capable at going from structured data into unstructured data, and the other direction. That makes them exceptionally powerful in healthcare, because healthcare might be the industry vertical with the highest intersection of natural language and rules-based systems.” 

The stakes are enormous. For decades, the story of healthcare in the United States has been soaring costs, shrinking productivity, burned-out clinicians, and sicker, less satisfied patients. Our society as well as our population are in acute danger, as the specter of insufficient healthcare has led us to accept more debt and more bureaucracy, while discouraging many talented students from pursuing medicine. Yet a better way is evident. As healthcare builders and investors, our mission is to support those striving towards better health outcomes, better patient (and clinician) experiences, and care that is truly affordable, as opposed to just passing along costs. LLMs, as we will see, are already helping on all three fronts. 

What follows are a few illustrative case studies drawn from those conversations, which speak to the range of exciting innovations underway.

Financial Performance Levers In Healthcare Services

To anchor our case studies, let’s begin with a simple framework for modeling healthcare services companies, and the most important levers driving their financial performance.

We can model healthcare businesses as comprising two levels: 1) the corporate organization, containing e.g. engineering, recruiting, growth, payer contracting, and executive headcount, which supports 2) discrete clinic or care pod units with clinicians and clinical support staff, and generates revenue through insurance billing, risk contracts, and patient collections. Each level includes different levers to improve fundamental financial performance, as measured by e.g. internal rate of return (IRR).

Clinic / Care Pod Level

De novo clinics or care pods usually start off with a negative contribution margin, reflecting spending during the upfront buildout and panel establishment phase, then ideally grow into a healthy margin surplus as staffing ratios saturate and quality impact is realized. When we track cumulative cash outlays, we get what seasoned healthcare operators refer to as the “J-curve”:

Along this notional clinic lifecycle, there are different levers affecting the shape of the curve, and ultimately IRR.

Corporate-Level Levers

Sitting above the clinics is the corporate level, responsible for sustaining growth and supporting existing clinics. Over time, the ratio of selling, general and administrative (SG&A) spend growth rate to clinical revenue growth rate stabilizes, as clinics and revenue mature. The cost burden of the corporate level is the difference between clinic contribution margin and overall operating margin.

Key Levers

• I. Growth Capacity: The corporate level’s capacity for adding new clinics or care pods in a given year.
• J. Operating Leverage: The degree to which a company can increase operating income by increasing revenue.

The higher the (I) growth capacity, the steeper the slope of revenue growth in out years.

Meanwhile, the slower SG&A spend grows relative to revenue, the higher the (J) operating leverage of a given business. The chart above illustrates two different scenarios, one with higher operating leverage, and one with lower. The higher operating leverage business is much more profitable than the lower operating leverage business, and the greater the scale, the greater the advantage.

Current LLM Applications in Healthcare and Their Early Impact

Next, let’s come back to our survey. 

Below are some illustrative applications of LLMs already being deployed today, categorized and aligned with the financial levers we discussed above.

Conclusion

We are in the early innings of a massive and exciting transformation, with the introduction of AI into services businesses of all stripes. It is especially encouraging to see so much experimentation and adoption already occurring in healthcare.

One early observation is that many LLM applications drive operating leverage improvement, allowing a smaller corporate back office to serve more clinics and care pods, with lower costs and higher quality. This implies that LLMs will drive higher returns to scale and furthers the case for consolidation, particularly in a world where clinic contribution margin is increasingly compressed by wage growth.

In many ways, our initial investigation has yielded more questions than answers. As future directions, it will be fruitful to explore:

• Will certain frameworks and foundational models be better suited to the unique challenges of clinical fine-tuning and navigating PHI? How will LLMs be deployed in healthcare businesses as the technology and use cases mature?
• What operational or process changes are required to maximize possible value creation from LLMs? Which use cases are relatively harder or relatively easier to implement? 
• Relatedly, are there advantages that new healthcare businesses will have over incumbents by virtue of being able to design processes from scratch?

We have captured a snapshot here. Beyond the above areas, we will be excited to observe how use cases evolve, as technology develops and operators have more time and space to envision what is newly possible.

‍