4 ways AI is improving healthcare, according to one of the world's largest medical groups

Artificial Intelligence (AI) holds great potential for improving the world's overburdened healthcare systems. An insight report from the World Economic Forum’s Digital Healthcare Transformation Initiative has identified the potential for digital, data and AI to tackle three of the most pressing healthcare challenges: the increasing burden of chronic illnesses, inequitable patient outcomes and healthcare access worldwide, and resource constraints.

The Mayo Clinic has been an early adopter of AI, and at the Forum's Sustainable Development Impact Meetings this year, its president, Dr John Halamka, highlighted the clinic’s experiences with the technology. He identified four AI applications that are already helping to improve healthcare provision and make it more equitable.

The Mayo Clinic’s Housing-Based Socioeconomic Status (Houses) index uses housing data from public records to indicate socioeconomic status at an individual level. The goal is to go beyond traditional socioeconomic parameters and create tailored healthcare plans to eliminate health disparities.

He highlighted that using traditional social determinants of health and commercial algorithms can lead to skewed results, which could put patients at risk.

“We infer based on where you live - what is your exposure to toxins in the environment and crime? What opportunities do you have for primary care, colonoscopy screening and the like?” explained Halamka.

This approach also enables payers and healthcare providers to distribute resources more equitably, and clinicians can more easily identify where interventions would have the greatest impact.

Identifying heart disease as early as possible can substantially impact a patient’s quality of life and life expectancy. However, it is complex to diagnose.

The AI-enabled ECG can predict the likelihood of a patient having heart conditions such as atrial fibrillation, amyloidosis or hypertrophic cardiomyopathy. This could, in turn, enable clinicians to diagnose heart disease earlier and monitor its progression – something Halamka himself has benefitted from.

“I happen to have a super ventricular tachycardia, which means my heart rate sometimes goes from 50 to 170.”

“Whenever I get an ECG, it is run through 14 algorithms. The results are shown to a clinician and describe my risk of atrial fibrillation. These kinds of things will help the clinician give me the right treatment.”

Improving global health outcomes and achieving universal health coverage are central to the United Nations (UN) Sustainable Development Goals.

This is one area where AI can make a vital impact. In some parts of the world, tuberculosis is still rife, but because of a lack of affordable healthcare, it is rarely diagnosed. Google, for example, is collaborating with Salcit Technologies in India to improve lung health assessments by analyzing cough sounds. With the help of AI and machine learning, the partners hope to create a faster, safer and more affordable way to diagnose tuberculosis (TB) and other lung diseases.

To reach vulnerable communities, you have to look no further than the modern smartphone, Halamka told the Forum: “I have done a fair amount of work in northern India, in the diagnosis of diseases like tuberculosis.”

“Often, we use a cartridge to do a rapid diagnostic test in the field. But interpreting the result is actually done on your smartphone, using a computer vision app that can read a cartridge and say ‘TB’ or ‘No TB’. In effect, you're bringing a clinical laboratory.”

“AI and image analysis is quite advanced today,” adds Halamka, pointing to an algorithm developed by Mayo Clinic that analyzes CT scans to predict pancreatic cancer – on average, around 16 months before a clinical diagnosis.

Pancreatic cancer is notoriously hard to detect, with most patients only finding out at an advanced stage when treatment options are typically limited. Three in four don’t survive the first year following their diagnosis. Earlier detection could improve patients’ chances of survival.

“You could ‘bake’ this algorithm into the scanner so that the scanner itself could flag when it detects a high risk for cancer,” he explains.

While the potential benefits of AI for patient outcomes and more equitable access to healthcare are evident, Halamka also stressed the need for caution.

One concerns bias. It is important to analyze a representative number of health records that also represent the underlying population: “One has to be careful that you also have spread or heterogeneity. Do you have low income, high income, educated, not educated, male, female, young, old?”

Another major concern is data security, specifically whether health records can be truly anonymised or “de-identified.”

“De-identification of data is hard. Sure, I could remove the name, the phone number or the address. But what if – and I'm, of course, making this up – the first line of a medical record said this was a former president of the United States?”

A job role like this, geographic data or a familial relationship could all inadvertently give away somebody’s identity, he added.

“So, one has to be very careful that the data is hard to re-identify to preserve privacy, and then it can be used for societal good.”