The AI buzz is not subsiding, making headlines across industries. In his summary article about the 2024 HIMSS conference, Chilmark Research’s John Moore III noted, “Good luck trying to get noticed for anything other than AI or cybersecurity (in the wake of the Change Healthcare breach).” One of Moore’s key takeaways was the consensus among attendees that while all the new AI use cases may be compelling, there is still concern about their readiness for broader deployment.
Yet, AI is being hailed as the magic bullet to address the ongoing challenges facing healthcare organizations today – workforce constraints, cost reduction and the need to generate revenue. With 30% of healthcare spending in the U.S. – $935 billion annually – considered to be waste, according to a 2019 study in JAMA, healthcare organizations of all sizes see the potential for AI to address the challenges across clinical, financial and administrative processes.
The path from concept to reality remains uncertain. Dr. Darrick Khor, in an in-depth LinkedIn post about what’s holding AI back in healthcare, summed it up with, “You can’t build a city without roads.” He outlined four fundamental inroads for AI adoption in healthcare: Steady input of high-quality data, efficient methods – and skilled teams – to deploy and evaluate algorithms, and upfront investment.
What’s growing ever clearer is that without robust control over the underlying processes to create a consistent and reliable dataset to build on, AI in care delivery operations will remain little more than a fantasy. Now is the time for healthcare leaders to refocus attention away from all the shiny AI objects and toward what’s needed to apply AI safely and effectively. To begin realizing the potential, healthcare leaders should focus on three important areas.
1. Bridging the process – data gap
AI requires a wealth of data to perform, but we have to be careful that we are training the AI on data that actually makes sense. The core challenge with EMRs is that there is no way of ensuring that the process of care that resulted in the data in the EMR was the right one, nor is there any record of compliance in the data collection process itself. What this means is that the data in the EMR may be incomplete, or at worse actually wrong.
An AI cannot discern this without being able to compare the data collected to a reference process or dataset that should be representative for a given patient, but that, too, results in challenges – notably differences in patient presentation, context, treatment approaches, availability of local resources, funding constraints and simple patient preference all conspire to introduce huge variation in what care is delivered, almost none of which will be explicitly documented in the EMR, particularly as compared to other equivalent choices. The treatment and care process may indeed be right, for this patient, at this time, but without the specific context noting why it was selected, the AI will not be able to create the necessary links between patient outcomes (assuming they are even recorded) and the actions that took place to lead us there. Robust processes and compliance are the only way to achieve this.
2. Normalizing data for reliability and consistency
Having access to internally consistent data is only part of the puzzle. Bringing data together from disparate systems with data that is incomplete, variable, and even conflicting is another core challenge that often requires significant manual effort to make sense of it. For AI to make use of data, it must go through a process of data normalization, meaning standardizing data from multiple sources to reduce ambiguity and make the data useable across systems.
By normalizing data across systems, we can compare apples to apples, and oranges to oranges. This is a long-standing issue in healthcare that is converging toward a solution, but it remains complex, and while much of the patient note is stored in free text, our ability to understand and relate it remains limited.
3. AI requires a firm foundation
As healthcare organizations evaluate the best path forward for adopting AI, it’s important to not lose sight of the foundation required. Without a focus on data and underlying processes, healthcare organizations will experience false starts that only hinder return on investment. Instead, healthcare organizations that take a long-term view and establish a consistent and reliable dataset that can enable AI technologies will be equipped to realize AI’s potential, safely and effectively.
About Robbie Hughes
Robbie is the Founder and CEO of Lumeon. An engineer by training, he started the company after first-hand experience of the impact that fragmented care delivery processes have on patient experience. Taking a step back to develop a fresh approach, he built the award-winning care journey orchestration platform to connect care teams, patients and technology across the care continuum. The platform enables healthcare providers to automate and orchestrate end-to-end processes by creating their own unique care journeys.
Under Robbie’s leadership, Lumeon has grown to an enterprise-level solution, currently in use by 65 major healthcare providers across the US and Europe managing over six million patient lives. Under Robbie’s leadership, Lumeon has grown from his bedroom with a business funded by individual clients, to an enterprise-level solution that is currently in use by 65 major healthcare organizations and growing.
