“It isn’t equipment that wins the battles; it is the quality and the determination of the people fighting for a cause in which they believe.” Gene Kranz

This is an overview of artificial intelligence in the U.S. health care system in the New England Journal of Medicine by two authors and consultants from the Center for U.S. Healthcare Improvement at McKinsey, neither a clinician. The authors elucidated on why AI adoption has been slow in health care delivery:

  • the majority of clinical data is both unstructured and heterogenous
  • the potential for AI tools to be disruptive to workflow
  • the lack of financial incentive in the fee-for-service model
  • privacy and regulatory concerns
  • the lack of longer term vision for AI as an investment.

The authors concomitantly revealed that there should no longer be significant doubt about the value of AI in health care, especially with the exponential increase in medical knowledge as well as the emergence of “health care services and technology” to reduce health care burden. Most of the AI adoption to date has been focused on medical image interpretation, but the authors present a table of the range of health care delivery domains (from consumers to clinical operations as well as reimbursement) in which AI is either implemented or there are plans to increase its use.

The authors further focused on three of these areas: reimbursement, clinical operations, and quality and safety: First, in reimbursements, a predictive model can find attributes that are most correlated (or associated) with denied billing claims so that a health system can address these issues prior to submission. Of course this AI-enabled strategy on the provider side can be matched by a similar AI-enabled strategy on the payor side. Second, in clinical operations, focus on optimization of operating room capacity can improve patient experience as well as increase revenue; this AI-enabled process can be accomplished in three steps in which AI is used increasingly more: improving operating-room management, predicting operating-room use, and using operating-room analytics in realtime. In short, the process involves descriptive, predictive, and then prescriptive analytics, the latter enabled by AI. The authors briefly mentions clinician burnout as an area that can be improved with AI (and probably should have discussed this issue more- an understandable oversight given that neither author is a clinician). Lastly, in quality and safety, AI adoption remains minimal but can be effective in two areas: reduction of major adverse events (specifically adverse drug events, decompensation, and diagnostic errors) and improvement of patient experience in conjunction with Consumer Assessment of Healthcare Providers and Systems (CAHPS).

One key takeaway from this paper is the emphasis on both financial and non-financial returns on AI in health care delivery. Another key message is the importance of data management so that AI efforts can be more easily enabled. Lastly, the authors pointed out that AI adoption is about change management and cultural transformation as well as work flow integration. All require time and education.

Read the full paper here.

The adoption of AI within healthcare will be discussed in detail at the annual Ai-Med Global Summit, scheduled for May 29-31 2024 in Orlando. Book your place now!