A Doctor's Guide to Using Artificial Intelligence to Diagnose Patients

A slow change is taking place in the way medical care is delivered in doctor's offices and hospitals as a result of artificial intelligence.

‍

Doctors at Mayo Clinic's cardiology department are able to detect any new heart problems by using an artificial intelligence program that uses artificial intelligence. There is also a group of primary-care doctors who are using this technology to identify an eye condition that can result in blindness elsewhere in the country. It has been used by a number of hospitals to help identify and monitor patients who are at risk for developing sepsis.  

‍

Using artificial intelligence (AI) tools, it is possible to identify patients in a better way that may be at risk for certain conditions or diseases. There are two main ways that doctors diagnose patients, and they don't rely solely on technology to do so, and their use of generative AI, such as ChatGPT, is mostly confined to paperwork and reports, although some are experimenting to see if they can play a more useful role in this area. There are some doctors, however, who are using artificial intelligence to assist them in arriving at a diagnosis, often earlier than they would otherwise do. 

‍

Although physicians say AI is a promising technology, they are also wary of yielding to machines, not just because AI is still a relatively new technology, but also because research has shown that biases within AI can make certain people's care less effective. IBM's Watson Health effort was one of the big bets that were made in the early 1990s that artificial intelligence would transform healthcare but ultimately proved to be a failure, or at least premature. 

‍

As Michael Pencina, the director of Duke AI Health, a department within Duke University School of Medicine that focuses on research in areas such as Artificial Intelligence and Machine Learning, asserts, "I do not think that we are at a point where we are able to just let algorithms run and make the decisions.". Medical AI programs generally employ an algorithm or a set of algorithms that learn over time with input and become more efficient over time. 

‍

In order to develop artificial intelligence tools, one challenge lies in the way the technology is developed, said John Halamka, president of Mayo Clinic Platform, which works with health technology companies to develop AI tools. There are a number of algorithms on the market now that are designed to detect whether a patient might have certain health issues based on information collected from the patient's electronic health records, such as demographic and health histories, vital signs, and lab results. 

‍

The technology becomes more accurate the more that it is used, as doctors report to the algorithm whether or not its assessment was accurate or not the more that it is used. The algorithm might be applicable to patients with different demographics, but that would require some modification, at least if it was developed on the basis of data from patients in Minnesota, for example. 

‍

According to a high-profile study published in Science in 2019, an algorithm had a racial bias, which prevented Black patients from receiving additional care they should have received. The development of guidelines for fair, unbiased use of AI in healthcare was started by a group of tech and healthcare professionals last year to address these issues. 

‍

So far, there has been much research on the promise of artificial intelligence, and a number of companies are working on products for doctors and health systems, but it has been slow to bring about widespread changes in medical practices as a result of the technology. The following are three examples of how artificial intelligence has been used to diagnose patients in the past.

‍

Detecting Heart Conditions

As part of the Mayo cardiology department's artificial intelligence tool, doctors have been able to diagnose new cases of heart failure and irregular heart rhythms, known as atrial fibrillation, in Rochester, Minnesota, years before they would have been detected otherwise, according to Dr. Paul Friedman, who chairs the department's cardiology section.

‍

Outside of the electrocardiogram, or ECG, when a patient has a normal electrocardiogram or ECG, doctors cannot tell whether or not he or she might have atrial fibrillation based on the results of the ECG. The AI, on the other hand, can identify red flags in the ECGs that humans are not able to identify since they are too subtle.

‍

A study published in the Lancet in 2022 by Mayo Clinic researchers used algorithms to analyze more than 600,000 ECGs of patients to determine whether they might be at risk of silent atrial fibrillation, or an irregular heartbeat. The artificial intelligence program identified 1,000 people and asked them to wear a heart monitor for one month as part of a study.

‍

These patients were five times more likely to be diagnosed with atrial fibrillation after a month of heart monitoring, as compared with patients in a control group, according to a study by researchers.

‍

Early Eye Problems 

There were a few attempts last year by Cano Health, a group of primary-care physicians in eight states and Puerto Rico, to use artificial intelligence to analyze images taken with a special eye camera that could detect diabetic retinopathy, one of the leading causes of blindness that can affect people with diabetes. Cano Health's senior medical director, Robert Emmet Kenney, said that after carrying out a small test in four offices in the Chicago area, the group is now looking at expanding its use of the program.

‍

Dr. Kenney said that patients with diabetic retinopathy who test positive for the AI must still see an ophthalmologist for a diagnosis and treatment plan. The program helps primary-care doctors identify patients with diabetic retinopathy right in their office, so they do not have to go to a specialist for a diagnosis. 

‍

“In this patient population, there are a lot of cases where patients fall through the cracks and do not receive the care that they deserve,” said Dr. As far as the largely Spanish-speaking, elderly population of the group is concerned, Kenney estimates that over a third of them have diabetes and need to see an eye specialist once a year.

‍

Catching Sepsis Risk

A hospital in Baltimore, known as Sinai Hospital, uses an algorithm to identify hospitalized patients who are at risk for sepsis, a rapid-acting response to an infection that is one of the leading causes of death in hospitals.

‍

In order to figure out if the algorithm is suitable for the patient, it examines more than 250 factors, including vital signs, demographic data, health history, and lab results, according to Suchi Saria, a professor of artificial intelligence at Johns Hopkins and chief executive of health AI company Bayesian Health.

‍

As soon as the system detects a patient is septic or deteriorating, it alerts the doctors to the situation. Following this, the doctor will evaluate the patient and begin antibiotic treatment if he or she agrees with the diagnosis. Esti Schabelman, the hospital's chief medical officer, said that the hospital's system will automatically adjust over time based on the feedback of its doctors.

‍

Using the algorithm in hospitals could result in patients getting sepsis treatment as early as two hours earlier on average, according to a study published in Nature Medicine last year, resulting in an 18% reduction in the hospital mortality rate associated with the condition.

‍

"Each hour of antibiotic treatment is associated with an improvement in mortality," Dr. Schabelman said. “In sepsis, every hour counts.” 

‍