Why adopting AI in healthcare is hard
Healthcare is complex and high-stakes, but the core of the challenge is organizational culture
“How are we using AI?”
That is a question being asked in every board room and C-suite in America. Whether the motivation behind that question is bright-eyed optimism about figuring out what is possible or a deep-seated concern of being displaced in the market, organizations of all sizes and types are figuring out what their future looks like in a world where AI plays a sizable role in most domains.
Healthcare companies and hospitals are asking these same questions.
We are witnessing rapid progress in the application of computer vision, large language models, and multi-modal techniques in areas like diagnostic imaging, drug discovery, and automated clinical documentation. But these technological advancements don’t answer the question “How are we using AI?” AI tools are proliferating, and this is exacerbating the feeling among many leaders that they should be doing something… they’re just not sure what that something is.
It’s easy to identify high-level reasons why figuring out AI in healthcare is difficult. The work of hospitals and healthcare providers is complex and essential. The stakes could not be higher: patients put their lives and wellbeing in the hands of the doctors and nurses who care for them. When people go to the hospital, they are experiencing massive uncertainty about the future; that’s a scary place for trying new things.
There’s another layer to this issue caused by organizational culture in healthcare.
In 2018, Harvard Business Review published an essay called The Leader’s Guide to Corporate Culture that described eight styles of culture. The framework organized cultural styles along two axes:
People interactions: the organization’s tendency for people to interact more independently (valuing autonomy) or more interdependently (valuing integration and coordination)
Response to change: the organization’s tendency to emphasize stability (consistency and maintenance of the status quo) or flexibility (receptiveness to change and focus on the long term)
Using these concepts, the authors plotted a chart showing eight cultural styles in relation to each other.
The essay goes into depth about each of these styles, and there are validated instruments for assessing an organization’s culture within this framework. With a brief overview, a few of those styles jump out as immediately recognizable for hospitals and healthcare organizations:
Safety: “predictable places where people are risk-conscious and think things through carefully”
Order: “methodical places where people tend to play by the rules and want to fit in”
Authority: “competitive places where people strive to gain personal advantage”
Results: “outcome-oriented and merit-based places where people aspire to achieve top performance”
Enjoyment: “lighthearted places where people tend to do what makes them happy”
Learning: “inventive and open-minded places where people spark new ideas and explore alternatives”
Purpose: “tolerant, compassionate places where people try to do good for the long-term future of the world”
Caring: “warm, collaborative, and welcoming places where people help and support one another”
I think—if we’re honest with ourselves—we’ll recognize that many hospitals want to be places that have a culture of Learning or Caring… but actually tend to have more characteristics of Safety and Order. There are a lot of good reasons for this; as mentioned above, there is inherent risk and uncertainty for everyone in a hospital, from the patient to the CEO.
Many organizations seeks to reduce the uncertainty and mitigate the risk by implementing processes and procedures with the aim of doing things the same way every time. The challenge is that, once implemented, there are few mechanisms for those processes and procedures to evolve to continue meeting the needs of patients and caregivers. Every time a new process is introduced, clinicians accumulate another responsibility or task, leading to increased workloads and risk of burnout. In organizations characterized by a culture of Safety or Order, team members are expected to follow the status quo, so there is little opportunity to discover new ways of working to mitigate these day-to-day challenges as they grow over time.
In a Learning organization, a different set of expectations are in place. Instead of simply requiring adherence to specified processes and procedures, team members are encouraged to identify opportunities for improvement and design experiments for finding a better way. Individuals have autonomy to take initiative, design and execute experiments, and present their findings to stakeholders who are open-minded about challenging the status quo. Learning organizations develop systems for experimentation and rigorous decision-making that enable change to occur rapidly and responsibly.
Machine learning cannot work well in organizations that aren’t full of human learning already.
If we are to progress in thoughtfully adopting AI in the healthcare industry to change how care is delivered, leaders have to focus their efforts on cultivating a Learning culture among their teams. Machine learning cannot work well in organizations that aren’t full of human learning already. Individuals doing the core work of the organization need to have permission to run experiments and learn. Organizations need to provide tools and platforms to enable those individuals to rigorously form, test, and validate hypotheses.
UnityAI is preparing a professional services offering to support healthcare organizations in adopting AI effectively and responsibly. Want to stay up-to-date?
There are known patterns for how to do this well at a large scale, and healthcare is just as capable of adopting these patterns as any other industry. A recent article from Amazon Web Services on how to build a culture of experimentation describes the first step as “define what an experiment is and is not in your organization” to ensure potential changes are tested and implemented appropriately. Stefan H. Thomke also describes a blueprint for enabling broad business experimentation in a responsible way in Experimentation Works: The Surprising Power of Business Experiments. He describes the talent, skills, tooling, processes, reporting, etc. that contribute to a strong organization-wide approach to steady learning and improvement. By adopting these practices, Learning healthcare organizations can conduct experiments to drive improvement while also prioritizing patient and clinician safety.
Interestingly, Thomke ends his book with an observation that fits our context of rapid technological advancement in healthcare AI: “We found that leading-edge tools did not result in exponential leaps in performance unless they were accompanied by organizational and cultural change” (pp. 208).
If we want to see the AI tools being developed today make real, lasting change in how healthcare is delivered to millions of Americans every year, we need to focus on evolving our healthcare institutions’ culture to embrace learning and view uncertainty as an opportunity to drive the state of the art forward.