Why HCOs Need a Portfolio Management Approach to AI

Featured in MedCityNews

No one AI technology will address everything and solve all problems, so systems should prioritize those that promise the greatest value and ROI.

AI is a short acronym that encompasses a long list of technologies.

As healthcare organizations assess and adopt artificial intelligence, there is a lot of confusion about what exactly it includes. All the attention around Large Language Models like ChatGPT and generative AI has overshadowed other types of AI, some of which healthcare organizations have been using for years, perhaps without even realizing they are using artificial intelligence.

Given the onrush of AI spending in healthcare, it’s important that healthcare systems understand the different AI technologies, how they’re used, and which offer the best value and return on investment (ROI). Systems are already spending fortunes on AI software. AI spending in healthcare and life sciences is projected to grow from $11.6 billion in 2024 to $19 billion by 2027, with a five-year CAGR of 16.6%, per Gartner. That’s way too much money to spend without knowing exactly what you’re getting in return.

AI is best understood as a portfolio of complementary technologies and capabilities, some of which simply automate manual and often repetitive administrative tasks, while others deliver in-depth analysis, predictions, and courses of action to optimize outcomes and value. Here’s a guide to the various technologies that can be grouped under AI:

• Machine learning – This is the most mature technology in the AI portfolio and the one with which most systems are familiar. It uses data and algorithms to allow AI to imitate the way that humans learn, gradually improving its accuracy. In general, machine learning algorithms are used to make a prediction or classification based on patterns in the data. This is commonly used for such things as stratifying patients based on risk, identifying gaps in care, and delivering personalized healthcare to improve patient outcomes, particularly for high-risk patients. It is also used for automated scanning of medical images to help radiologists proactively identify patients at risk of a stroke or heart attack for intervention well before an acute event happens.
• Deep learning – This is a subset of machine learning, one that comes close to human reasoning. It uses multilayered neural networks, called deep neural networks, to simulate human decision-making. Unlike machine learning models, which require structured and labeled input data to be effective, deep learning models can make accurate outputs from raw, unstructured data. One of the most common uses for this in healthcare is for image analysis.
• Natural language processing (NLP) and natural language generation (NLG) – This uses machine learning to allow computers to understand and communicate with human language. It allows computers and digital devices to recognize, understand, and generate text and speech by combining computational linguistics with statistical modeling, machine learning, and deep learning. In healthcare, it is used for computer-assisted coding to translate medical records into plain English, to analyze health records, and present a summary of the patient’s chart at the bedside/point-of-care (POC) for physicians and nurses to enhance productivity.
• Generative AI/Large language models (LLMs) – Similar to NLP, this AI can create original content including text, images, video, audio, and software code in response to a user query. It can perform such tasks as powering online chatbots for scheduling appointments, analyzing patient sentiment from different sources, and more. One of the most compelling use cases of Gen AI/LLMs, amply in evidence at HIMSS 24, is seamlessly capturing nurse and clinician notes via a cell phone running the application, turning it into text while editing out non-relevant content, with the ability to do final edits before automatically entering these notes into Epic’s electronic health records (EHRs).

There are other technologies not always thought of as AI, but which, in fact, are. This includes medical robotics and its subfields of:

• Robotic process automation (RPA) – Also known as software robotics, this employs intelligent automation technologies to perform such repetitive tasks as extracting data, completing forms, and moving files, freeing up humans to do other work. It also can be used to improve call center operations and to enable customer and patient self-service across multiple channels.
• Machine vision – This gives medical equipment the ability to “see” a task it is performing and make real-time decisions based on that input. It can help with everything from identifying injuries and interpreting medical images, to medication management and making diagnoses. Advances in this arena are paving the way for Virtual Reality (VR) and Augmented Reality (AR), both of which have enormous potential for robotics-assisted surgeries. This field also includes what we commonly think of as medical robots– semi-autonomous machines that can deliver medications, assist in surgeries and rehabilitation, monitor patients and even serve as companions to those who would benefit from it.
• Robotics-assisted surgeries – Medical robots deployed for surgeries today have 3D cameras that record operations. The video streams to a computer screen somewhere and aids the surgeon to perform the operation using surgical robotic arms, such as the Da Vinci surgical system, which enables minimally invasive surgery and rapid patient recuperation that lowers length-of-stay (LOS) while ensuring superior patient outcomes.

Building an AI Portfolio

Faced with such pressing needs and such promising technology, how is a healthcare organization supposed to know which AI to invest in?

There is no single answer. It’s an individual decision depending on each organization’s resources, needs, and priorities. No one AI technology will address everything and solve all problems, so systems should prioritize those that promise the greatest value and ROI.

There are a large number of factors for organizations to consider, including the cost of the technology, ease of adoption, potential resistance from the providers and payers who will use it, disruption to existing workflows, compatibility with existing systems, potential savings, and more. Organizations also must weigh whether to build or buy AI technology. Building provides greater transparency into operations but can require resources and expertise that systems lack.

Thoughtfully building a portfolio of the most useful and impactful AI technologies is the best way for organizations to ensure that they get the maximum benefit from this amazing innovation.

Identity Data Management (IDM) for AI data fidelity and readiness

Of course, critical to the success of any AI or analytics data program is the caliber of the patient/member/consumer identity data being used, beginning with Identity Data Management (IDM). Faulty and missing data or duplicate data interferes with AI performance, which can make it difficult for organizations to achieve the desired ROI and deliver value from their AI initiatives. Organizations need high-quality IDM processes and resources in place for their AI investments to have a meaningful impact.

Organizations uncertain of their IDM abilities should partner with experts who can evaluate, benchmark, and enhance their operations to maximize the return from AI technology.