The following is a guest article by Ben Cushing, Chief Architect of Health and Life Sciences at Red Hat
Artificial intelligence (AI) is showing up in more healthcare contexts, from drug discovery to diagnostic imaging to robot-assisted surgery. But one area where AI can have a tremendous impact is in the reduction of provider burden.
For instance, the Veterans Health Administration (VHA) is exploring automated, AI-assisted documentation of patient visits with clinicians. The system records the conversation between clinician and patient, and then automatically produces a clinical note.
The AI system can apply accurate medical terminology, follow the Subjective, Objective, Assessment, and Plan (SOAP) format, and make best-practice treatment recommendations. It produces the clinical note in near real-time, so the clinician can review it and make any necessary edits at the time of the encounter.
AI can then translate the clinical concepts into standard formats. These formats include Systematized Nomenclature of Medicine – Clinical Terms (SNOMED CT), used for the electronic exchange of clinical health information, and Logical Observation Identifiers Names and Codes (LOINC), which provides a common language for laboratory tests and clinical measures.
Such AI use cases can save clinicians and administrative staff significant time. They can also improve patient experiences and contribute to better patient outcomes. It’s no wonder 94% of healthcare companies say they’re investing in AI, and budget allocations are expected to grow from 5.7% in 2022 to 10.5% this year.
As healthcare systems set aside a greater portion of their budgets for AI, they should look to follow a road map that can lead to AI success. In particular, they should consider: adopting an automation mindset, involving the right stakeholders, building on the right technology foundation, and planning ahead to operationalize their AI solutions.
Embrace Process Automation
Industries like manufacturing and financial services have increasingly been optimizing their operations and achieving more predictable results by automating their processes. Investment in process automation by the healthcare industry has the potential to deliver similar advantages.
Today, healthcare AI is used to analyze population data to both uncover trends and support policy and investment decisions. I believe another future for AI in healthcare, however, lies in clinical decision support.
For an AI system to support physician decision-making, it should understand the relevant process. For example, a patient who is pregnant has a 41-week care plan, which specialists know by heart. But for an AI system to be effective in a similar manner, it must be trained on this plan.
Healthcare systems are already undertaking initiatives to represent clinical best-practice guidelines as visual process models. These process models can also be digitized so that an AI system can support the step-by-step program a clinician would follow to provide care for a patient with heart disease, for example, or diabetes.
Process automation holds particular promise for clinicians in training or those working in remote locations without access to specialists. While the clinician remains the ultimate decision-maker, AI assistance can improve decision speed and quality.
Get the Right Stakeholders Collaborating
In a healthcare setting, AI solutions often begin with a health informaticist. That makes sense because informaticists specialize in the application of digital capabilities to improve the delivery of care.
However, informaticists aren’t typically the people who will implement, use, or govern AI systems. So while they can be an invaluable source of AI innovation, they shouldn’t be the only stakeholders involved.
First, the IT professionals who will deploy and maintain the system need to weigh in on whether the proposed solution can be implemented in a cost-efficient, scalable, manner with the appropriate levels of security. In addition, the clinicians who will benefit from the system must determine whether it will support how they work and drive positive patient outcomes.
Finally, the compliance officers responsible for patient safety and regulatory adherence must assess the system as it’s being designed. They must provide input before an AI system is implemented.
Build on the Right Technology Foundation
Because organizations are beginning to use AI in novel ways, they might treat AI projects as one-offs. But for AI tools to scale, they need to integrate with the IT infrastructure. They also must comply with IT department guidelines – and HIPAA regulations – for data privacy and security.
More healthcare organizations are turning to open-source solutions to support their AI innovation strategies. Open source software is developed in a decentralized and collaborative way, relying on peer review and developer community input. Open source is typically cheaper and more flexible while also providing a stronger security posture than proprietary software because it benefits from diverse inputs.
Outside of “traditional” IT, open source is proving its mettle in AI. For instance, many machine learning (ML) models, as well as most of the large language models (LLMs) that rival GPT-3 and GPT-4, have been trained with an open-source approach. In addition, the use of cloud-agnostic platforms enables the serving of AI models to wherever they’re needed, from ambulance to bedside, without vendor lock-in or dependence on network connectivity.
Operationalize the AI Solution
Finally, to achieve success with AI, I believe healthcare organizations should begin with the project’s end goal in mind. Otherwise, they risk investing in an innovative AI concept that they can’t deploy, scale, and manage as an enterprise solution. This failure to operationalize is a major reason some AI projects might fall short.
Teams developing AI models require a deep understanding of clinical workflows to achieve smooth integration with existing processes. They also need to train and validate their models with robust, high-quality datasets.
Finally, the AI solution will need to integrate with the existing technology stack. It must also be able to operate at scale using existing IT resources unless a budget has been allocated for additional compute power. The solution will also have to connect with sensitive data across relevant departments in the healthcare system.
It’s an exciting time to contemplate the potential of AI in healthcare. And the technology is advancing even faster than many experts anticipated. By considering these four steps to success, healthcare systems can better use their AI investments to improve operations while contributing to better patient outcomes.
About Ben Cushing
For two decades, Ben Cushing has been a leader in emerging technology solutions across multiple industries and is committed to radical innovation in healthcare. Before joining Red Hat, he served as the Chief Technology Officer for MDLogix, a behavioral health IT firm supporting Johns Hopkins Medicine. There he architected and brought to market a behavioral health cloud platform for use with employer, healthcare, and education markets. In addition to supporting analytics and operations at the National Institutes of Health for 6 years, Cushing had the opportunity to practice a scaled agile framework with Accenture where he led the technical architecture and design for the Department of Veterans Affairs’ (VA) Electronic Health Management Platform, an industry-leading Health Management and Care Coordination platform serving 9 million patients. His tenure at Accenture began with the acquisition of Agilex, where he designed LSI solutions, developed systems to automate the Post-9/11 GI bill, and supported in-theater data collection and analytics tools. While at Agilex, Ben architected and led the development of a mobile Software Development Kit, still in use today by the VA to produce more than 60 applications.
