AI-driven demand for medical tests and CT scans pose health risks
Sometimes, having a medical test can lead to consequences worse than the original symptom. Doctors need to take that into account because AI does not.
The changes to the longstanding power imbalance in the doctor-patient relationship in the past generation or so have largely been positive. It is healthy that patients are empowered to ask questions about their care. Increasingly, however, “health consumers”, fuelled by information available online and on social media, present to doctors with demands for tests that may or may not be appropriate.
It seems that in the past year or so the community’s preference for using “Dr Google” has been supplanted by reference to ChatGPT, Gemini or Claude as their preferred source of health information. Artificial intelligence-powered large language models are powerful tools. Used appropriately, they can contribute positively to health literacy and shared decision-making. However, the risks of interventions and tests must be considered.
Some tests might be regarded as being painless and harmless. But if, for example, a full blood picture, renal function tests or liver function tests are abnormal, there may be a duty of care for the requesting practitioner to investigate further. Ultrasound scans are very safe. However, careless referral for tests can lead to further investigations that do carry potential harm.
An article earlier this year in the Journal of the American Medical Association (JAMA Internal Medicine) measured the potential harms of the 93 million CT scans performed on 62 million Americans in 2023. Ionising radiation is a known carcinogen. The study estimated that these tests themselves will result in an additional 100,000 cancer cases. The paper concluded: “If current practices persist, CT-associated cancer could eventually account for 5 per cent of all new cancer diagnoses annually.”
The younger a patient is, the greater the lifetime risk of a CT-associated cancer. Chest pain is a common symptom in pregnancy. It is usually musculoskeletal in origin. However, CT pulmonary angiograms may be performed when there is a suspicion of a pulmonary embolism. A single test is estimated to increase the lifetime risk of breast cancer by 0.1-0.2 per cent. These are small numbers for individuals, but at population level it all adds up.
There are various drivers of the increased use of diagnostic imaging. If doctors are behaving defensively because of fears of plaintiff lawyer activity or complaints to the medical regulator about delay in diagnosis of serious conditions, this can result in harm to individual patients.
It is ironic that a scan performed in investigating the possibility of cancer could end up contributing to a malignancy. If patients are getting sick from tests designed to prevent disease, we have a problem.
There have been advances with the development of lower-dose radiation scans. Thoughtful use of clinical guidelines should further reduce the number of unnecessary tests. If these same clinical guidelines are used by payers (including governments, private health insurers, health maintenance organisations in the US) to ration care, that can create its own harms. Strong communication between referring doctors and radiologists increases the proportion of times the right test is done on the right patient at the right time.
The 2022 Reducing Overuse of Diagnostic Imaging Report included a randomised controlled trial, where GPs who were requesting musculoskeletal imaging at a rate higher than 80 per cent of their peers were subjected to an intervention.
They received a letter from the Department of Health stating: “Most people who present with musculoskeletal pain in the absence of worrying features do not need any imaging as it does not help management. Pain can improve rapidly; for example, around 50 per cent of people who experience an episode of back pain will recover within two weeks.” They were invited to reflect on their own clinical practice.
The project showed that GPs who received feedback on their rates of referral had a statistically significant lower requesting rate compared to GPs in the control group over the subsequent 12 months.
Another harm from unnecessary tests is the expense, whether that be to the individual, the taxpayer or the private insurance pool. Certainly, someone pays, and ultimately any health dollar spent on low-value or no-value care is a dollar not spent where it is desperately needed.
The drivers of overuse of diagnostic imaging are many and varied. I know from personal experience that many patients equate “more care with good care”. Life itself is increasingly being “medicalised”. Again, my experience is that Australians are increasingly anxious about and intolerant of uncertainty. Getting an answer to a question often leads to another question that does not necessarily advance the care of the patient.
The media’s fascination with “medical breakthroughs”, the promotion of tests by industry, and the understandable fascination with the potential benefits of new technology all act as drivers of the public’s appetite for tests.
Of course, a referring doctor has an ethical responsibility to consider the pros and cons of an investigation before requesting it. This is the basis of good medical practice. However, a reset is required when health consumers think there is an answer for every single symptom and a therapeutic solution to every diagnosed problem. Primum non nocere (“first, do no harm”) sits at the very heart of ethical medical practice.
Michael Gannon is a consultant obstetrician and gynaecologist with 18 years’ experience as a specialist. He served as president of the Australian Medical Association from 2016 to 2018. He is president of leading professional indemnity provider MDA National.
This column is published for information purposes only. It is not intended to be used as medical advice and should not be relied on as a substitute for independent professional advice about your personal health or a medical condition from your doctor or other qualified health professional.
To join the conversation, please log in. Don't have an account? Register
Join the conversation, you are commenting as Logout