Is AI always the answer in healthcare?

While artificial intelligence has already shown value in reducing healthcare costs and improving patient care, certain challenges must be overcome before the next wave of adoption.

AI seems to have permeated every industry, including healthcare. An Accenture study found that 10 specific AI applications—from robot-assisted surgery and fraud detection to dosage error reduction—could create up to $150 billion in annual savings for the U.S. healthcare system by 2026.

The real question is: Can the roles of clinicians and providers be improved through AI?

AI algorithms are trained to look at patterns and general constructs. They can guide the hunt and peck for information and serve it up in a way that can be easily defined. They can also be trained to interpret images.

While humans can certainly perform these functions, overburdened workloads and fatigue can result in great margins of error. Not to mention, AI can process data and detect patterns at a greater scale and speed than humans.

AI offers the promise of revolutionizing healthcare, but three components are needed:

1. Storage and computing power: We need machines powerful enough to store and crunch large amounts of data; and, through algorithms, to find and highlight patterns that are naked to the human eye.
2. Data readiness and interoperability: Large amounts of digital data, sourced from health information technology (such as EHRs, administrative claims, patient-generated data and image digitization), which are intelligently labeled, unified and made shareable.
3. Talent: People who understand clinical medicine or healthcare operations and its data relationships; as well as those having individual or combined skills in programming, math, critical-thinking, problem-solving and communicating.

Four obstacles to AI adoption in healthcare

All three components are beginning to coalesce. However, four obstacles continue to slow the adoption rate of AI in healthcare:

1. Healthcare is complex. This industry requires subject matter domain expertise beyond the bits and bytes that are generated out of a health system, payer, or life sciences organization. Being able to organize, coordinate, and draw out the critical underlying value in held data is a rare skill set. We need more people who are savvy in this space and can inject machine learning into operations in a budget-friendly way. After all, no money means no mission.
2. Seeing the complete picture. No clinician wants to turn their patient relationship and decision-making over to a machine, but most would love to reduce the time spent on unnecessary work each day. Health providers also recognize that low-value work adds no value to the patient, clinician, or the organization’s bottom line. AI principles from other industries that increase productivity would, when framed in totality, be welcomed by most clinicians.
3. There’s a lack of high-quality data available to train AI algorithms. Healthcare, and the data exhaust from it, is heavily influenced by local practice patterns and data policies, so while EHRs and other health information systems generate a lot of data, making it usable and contextual in aggregate across providers and practices can be challenging . This makes it difficult to get a large enough sample size to start drawing out patterns generally applicable to the rest of the population. Having access to reliable data that meets regulatory standards for training these algorithms can be a challenge at times.
4. More work must be done to ensure against bias: AI learns from patterns of data. Intrinsic, implicit, or even unconscious bias, present in our society, organizations and among clinicians, will be propagated in the resulting data. If not recognized, this bias can inadvertently be present when training an AI algorithm, causing such a model to recommend certain things that are artifacts from the data and inappropriate in reality.

The promise of AI today

Regardless of these obstacles, AI still holds promise for provider organizations and patients.

Physicians, for example, know that patients over the age of 45 should have regular colonoscopies. Rather than forcing a clinician to take the time to comb through patient records, AI can do that work and even notify patients via text or email under the physician’s direction.

Natural language processing in combination with AI algorithms that are designed to understand the effectiveness of different communication modalities, can enhance a primary care doctor’s ability to engage with patients in a more meaningful way.

A payer might want some assurance that a patient has tried other options before agreeing to pay for a more expensive procedure. AI can help read through charts and make that process a little easier for the patient, provider and the payer.

Patients also benefit. AI can help create a better channel of communication between doctors and patients. Everyone likes to be communicated with in a personalized style with varying levels of detail. And certainly that is front and center in the growing world of virtual care where Patient preference and limitations of care from a distance may benefit from AI algorithms for optimal care.

Remembering the humanity of healthcare

Though it figuratively sits behind a curtain, AI has become a proven catalyst for improving clinical decision support and diagnostic accuracy, as well as increasing efficiencies in system operations and administrative processes. Its impact is, and will continue to be, an important part of the care experience shared by both physicians and patients.

As powerful, effective, and “shiny” as AI has become, we must honor health care itself as a human-received and human-provided experience.

An engagement that has increasingly become plagued by bureaucratic tasks and growing EHR-related tasks—reported by physicians nationally—as the top driver for their growing stress and burnout.

In a time where results, costs, and outcomes are paramount, we must not forget the physical, emotional, and mental health of our physicians, who carry their own unique journeys in healthcare.