AI Could Slash Diagnosis Costs in Stable Chest Pain Patients

Liam Davenport

September 06, 2019

PARIS — An artificial intelligence decision support system (AI DSS) to help decide which diagnostic tests to perform in individuals with stable chest pain (SCP) could lead to substantial reductions in healthcare costs versus standard human-led care, suggests research presented as two posters at the ESC Congress 2019.

The researchers, led by investigators at the Royal Brompton Hospital, London, examined the potential cost effectiveness of an AI DSS derived from clinical guidelines for the assessment of SCP.

Humans vs AI

They looked at data from the cooperative ARTICA registry database on individuals who underwent a standard evaluation by clinicians, alongside assessment via artificial intelligence, all on the same day.

In the first study, the two approaches were compared in more than 1700 individuals who underwent coronary tomographic angiography (CTA) to verify the presence of coronary artery disease but for whom a range of non-invasive and imaging tests were available.

If it had been followed, the AI DSS could have led to potential reductions in lifetime costs compared with standard human evaluation for all possible diagnostic test permutations, regardless of the patients' pretest likelihood of coronary artery disease.

In the second study, the team compared AI DSS-led and standard care evaluations in more than 1000 individuals with SCP, all of whom underwent CTA but none of whom had coronary artery disease.

They found that, compared to blanket CTA, the AI DSS would have chosen follow-up with no tests for the majority of individuals with a low or intermediate pretest likelihood of coronary artery disease, and reserved invasive coronary angiography (ICA) for only those few people with a high pretest likelihood.

Consequently, healthcare costs could have been cut by 65%.

Now, the AI DSS is accessible via a smartphone or a computer, with the aim that it not only be available for clinicians to aid in decision making but can also feedback the results to the system.

Dr Marco Mazzanti, International Research Framework on Artificial Intelligence in Cardiology, Royal Brompton Hospital and Harefield NHS Foundation Trust, London, was the senior investigator for the two studies.

He told Medscape News UK that feeding back the data to the AI DSS is "very important", as it will make the system "more precise" as more and more information is accumulated.

AI in Clinical Use: A Good Fit

Alison Pottle, a cardiology nurse consultant at Royal Brompton Hospital and Harefield NHS Foundation Trust, London, who led the second study, added that the AI DSS will fit in "very well" with current clinical care.

She said that the National Institute for Health and Care Excellence (NICE) has produced guidance on the assessment and diagnosis of recent onset chest pain, but that the studies have shown that, compared with the guidelines, "we tend to over-investigate people. More I think to reassure us than to reassure the patient, we send patients for tests, and a lot of those tests prove that there is nothing wrong with the patient.

"So by having a means of selecting the right patients to go forward with tests, you are using resources better but also you are not putting patients through additional stress," as well as reducing exposure to radiation, she noted.

Another aspect is that not all hospitals can afford, for example, CTA.

Alison Pottle said that she and Dr Mazzanti work in an institution "where we have everything available to us but that isn't always the case".

Consequently, a tool like the AI DSS "does help when you're having to think I've got to make a decision about a test that is quite expensive, that is not readily available to me, that is going to involve the patient going to another centre".

Helping Clinicians

Dr Juhani Knuuti, University of Turku and Turku University Hospital, Turku, Finland, who was not involved in the studies, welcomed the research, saying that, "in the future, we are going to really utilise these kind of approaches….choose appropriate diagnostic testing".

He told Medscape News UK that there are "a lot of parameters which influence what tests to choose in which patients, and to integrate all this information in an easy way would be then, of course, helping the clinician".

Dr Knuuti underlined, however, that this was not a "real trial" but a simulation based on assumptions about the performance of the tests.

He said that the "real data" will come when patients are randomised to test selection based on artificial intelligence or standard diagnostic approaches, "and then you compare the real costs, and the outcome of the patient".

For Dr Knuuti the "important thing is not only the cost of the diagnosis, because, if you miss the disease, that may have consequences in terms of outcomes or, if you over-diagnose, unnecessary testing and, further, therapies".

The researches say that CTA has potential as a 'gatekeeper' for ICA to diagnose coronary artery disease in a patient with SCP.

However, there are a number of other options available to clinicians, such as exercise treadmill testing (EET), stress echocardiography (SE) and gated myocardial perfusion scintigraphy (gMPS).

They note that, while non-­invasive cardiac imaging testing is typically favoured in symptomatic patients with at least an intermediate pre­test likelihood of obstructive CAD, the optimal strategy from a cost-effectiveness standpoint is not clear.

Study Details

The team therefore used an AI DSS developed from ESC and other guidelines on the criteria for the appropriate use of CTA and other tests in patients with SCP.

The AI DSS comprises a clinical decision support that integrates diagnostic algorithms with prior data to identify the best choice for a given patient from the clinical possibilities.

For the first study, the team looked at 1725 individuals, including 982 males, from the cooperative ARTICA registry database with SCP and without known coronary artery disease, who were referred for clinical evaluation over a 2-year period.

On the same day, the individuals, whose average age was 61 years, underwent standard human evaluation and assessment using the AI DSS, with potential strategies including EET, SE, gMPS, or their combination, and follow-up with no tests.

All participants underwent CTA to verify the presence of coronary artery disease, which was defined as stenosis in at least one major epicardial artery vessel of at least 70%.

The researchers used a Markov model to compare the costs of the standard care and AI DSS-led approaches over a follow-up period of 16 months, taking into account individuals who were diagnosed with coronary artery disease.

From this, a Monte Carlo simulation was performed to determine the average lifetime costs of each diagnostic strategy at different pretest likelihoods of the patient having coronary artery disease.

Study Results

The results showed that, across all pretest likelihoods, AI DSS-led care was associated with lower lifetime costs than standard care.

For example, follow-up with no tests in individuals with a pretest likelihood of coronary artery disease of 15% was associated with a lifetime cost with AI DSS-led care of €350 versus €1015 for standard care.

At the other end of the spectrum, performing CTA with ICA, alongside SE and gMPS, in individuals with a pretest likelihood of coronary artery disease of 80% attracted a lifetime cost of €31,765 with AI DSS-led care and €35,660 with standard care.

The team writes that the AI DSS resulted in "improved costs and enhanced effectiveness when compared with human standard care in subjects with stable chest pain".

Artificial intelligence, they say, "has the potential to markedly improve productivity, efficiency, work flow, accuracy and speed," which will allow a "great transformation" in healthcare systems.

For the second study, the team looked at 1017 individuals, including 620 males, with SCP from the same database, who were referred for CTA between October 2016 and June 2019.

As before, the participants, who had a mean age of 62 years, were evaluated with standard care and, on the same day, with the AI DSS.

CTA demonstrated that none of the individuals had significant coronary artery disease.

Various testing strategies were compared, with the costs calculated as the sum of the technical and professional components, and the sensitivity and specificity of each non-invasive test derived from published multicentre trials.

The team classified the participants by their pretest likelihood of coronary artery disease, and conducted probabilistic sensitivity analysis to assess the impact of uncertainty in the parameters of the model.

In all, 60.7% of the participants were deemed to have a low pretest likelihood of coronary artery disease, while 36.5% had an intermediate likelihood and 2.8% had a high likelihood.

The AI DSS recommended that 89.6% of individuals with a high pretest likelihood of coronary artery disease undergo ICA, compared with just one person with an intermediate likelihood and none of those with a low pretest likelihood.

In contrast, it suggested that 97.2% of those with a low pretest likelihood have follow-up with no tests, alongside 69.8% of those with an intermediate likelihood and none of those with a high pretest likelihood of coronary artery disease.

Individuals with an intermediate pretest likelihood were most likely to be recommended to undergo SE and gMPS, at 13.7% and 12.9%, respectively.

The consequence was that, compared with standard care of CTA in all patients, the AI DSS resulted in a 65% cost saving, at a total of €146,030 versus €406,800.

The team writes that the results "seem to demonstrate that AI DSS is extremely cost-saving in subjects with stable chest pain across the whole range of pretest likelihood of coronary artery disease".

ESC Congress 2019: Abstract P839 . Presented August 31st.

ESC Congress 2019: Abstract P5245 . Presented September 3rd.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.
Post as: