Working For The Machine? AI In Hospital Medicine
Written by Andrew McWilliams
Artificial intelligence (AI) has the potential to fundamentally transform hospital medicine, e.g., by enabling hospitals to deploy and manage their hospitalist resources efficiently, reducing the time hospitalists must spend on data entry and administrative tasks, providing early warning of patient conditions that require attention, and improving care decisions.
For these AI solutions to have the optimal impact, hospitalists need to be involved in their design and implementation. Otherwise, there is a danger that hospitalists could end up working for AI, rather than the other way around, negatively affecting their performance, morale, and retention.
Hospitalist Workforce Optimization
Workforce optimization (WFO) uses data to improve employee and organizational efficiency and decrease operational costs, consistent with an organization’s strategic objectives. The latest, AI-enabled WFO systems use tools such as natural language processing, deep learning, and machine learning to make the process of assigning staff resources such as hospitalists easier and more efficient.
WFO systems can allocate hospitalist resources across a hospital or hospital system, based on need, hospitalists’ personal preferences, and expertise. Such systems can also supplement the in-house workforce with outside or virtual hospitalists as needed, opening the door to “hybrid” hospitalist models that employ a mix of in-house and virtual hospitalists that can be adjusted in real time.
However, if the WFO inadvertently leads to disrupted workflows, personnel dislocations, heavier workloads, and longer/later work hours, it can heighten hospitalists’ perception that hospital management is out of touch. Hospitalists may come to feel they are on a treadmill, working without a clear purpose or real stake in the system.
It is thus essential to involve hospitalists in the design and implementation of the WFO, communicate any changes to them, and provide opportunities for feedback. Without such involvement, hospitalists may come to feel that they are not in control of their work environment, causing them to become disengaged or disaffected and more likely to leave as a result.
Electronic Health Records
Some of the electronic health records (EHR) systems that are currently in use were developed without enough clinician participation and insufficient attention to clinician requirements. Such poorly designed EHR systems can get in the way of hospitalists spending enough meaningful time with their patients and are considered a leading cause of burnout.
A growing number of EHR vendors are now using AI to reduce the administrative burdens and improve the functionality of their systems. New AI capabilities in speech recognition and language processing have resulted in systems that allow hospitalists to record sessions and create documentation without the need to create and edit the notes personally.
However, AI-enabled speech recognition for EHR is still prone to error, and requires close clinician supervision to do its job properly. Thus, a combination of AI and human oversight is still the most effective way to mitigate the burden of documentation for hospitalists.
Clinical Early Warning and Decision Support Systems
A hospital encompasses a wide range of patients with varying conditions and needs. Sicker patients, unexpected clinical deterioration, and lower nurse-to-patient ratios make the clinical staff’s job increasingly difficult. AI-enabled early warning systems (EWS) hold the promise of a fully automated and configurable early warning system for hospital staff to ensure rapid response to deteriorating patients. These systems can help to identify subtle signs of deterioration in a patient’s condition hours before a potential adverse event, so clinicians respond in time to make a difference.
Similarly, AI-enabled clinical decision support systems (CDSS) systems can also inform decisions about the most appropriate care for a patient. AI is capable of providing real-time, data-driven insights that hospitalists can use to enhance the overall quality of the care they provide.
A risk of AI-driven EWS or CDSS systems is that they may function as a kind of “black box” that does not communicate the nuanced details of an alert or diagnostic/treatment recommendation to the hospitalist. It may not be immediately clear to the hospitalist what the likely reason for an alert is until the patient is assessed, delaying a response when time may be critical. Similarly, without enough clinical context it may be difficult for the hospitalist to apply their experience and human judgement in order to modify or override the system’s treatment recommendation if necessary.
There has been significant progress recently in developing methods for “explaining” black-box AI recommendations to the recipient. Some of these explainability algorithms can be used with specific AI models, such as convolutional neural networks, to provide post-hoc explanation of their decision-making process. Other explainability algorithms can be applied to any type of AI model, regardless of its mathematical basis.
Some of these methodologies are promising, but their application to health care EWS and CDSS has been limited to date. There is a pressing need for more research in these areas if AI-enabled EWS and CDSS are to be effectively deployed in the hospital or other clinical setting.
Meanwhile, medical education needs to evolve in order to provide future clinicians with the ability to understand the limitations of AI and understand its communications. Medical education “will need to place less emphasis on threshold decision making and a greater focus on detection, analysis, and the pathophysiologic basis of relational time patterns”. (Lawrence A. Lynn, “Artificial intelligence systems for complex decision-making in acute care medicine: a review” Patient Safety in Surgery 13, 6 (2019). https://doi.org/10.1186/s13037-019-0188-2)
Future Outlook
A recent article (Larry Beresford, “What is AI’s Promise and Potential for the Hospitalist?”, The Hospitalist Vol. 27 No. 1, January 2023) profiles a number of hospitalists who are working on AI applications in hospital medicine, but most of them work in an academic or institutional setting. We know of only a few hospitalists who are working directly with the tech industry in the development of AI applications for hospital medicine.
The time may come when AI has advanced to the point where physician oversight is no longer needed, and the hospitalist’s role is mainly one of providing empathy, comfort, counseling, and end of life care—but that time is still far away, if indeed it ever comes. Meanwhile, more hospitalists should play an active role in developing and implementing AI solutions to ensure that AI meets their needs today and in the future.