by Dr Jessica Watson
Research Fellow
Centre for Academic Primary Care
‘If you can afford to have your blood tested for everything available, do it quarterly so you have a baseline of your own personal health’
– Mark Cuban, Twitter
With this one tweet Mark Cuban started a twitter storm. Advocates argued that the future of medicine lies with new technologies, with online companies and even future smartphone dongles allowing patients’ to test their own blood for everything from allergies to zinc.
This brave new world overlooks the responsibility of doctors to ‘first do no harm’. I have been qualified as a doctor for over 10 years but am still in training as a GP registrar and ‘junior doctor’. I have gradually learned that my well-meaning enthusiasm to do more and more tests to try to find out what is wrong with patients can quickly lead to difficulties.
Keith Marton describes this vividly in a blog ‘How the CA-125 became a $50,000 blood test’. His story starts with his wife getting a simple blood test, looking for any early signs of ovarian cancer. Unexpectedly the result was slightly raised, leading to cascades of further tests. Finally after five months of worry, surgery to remove her ovaries and uterus; which were found to be entirely cancer free. She had become a victim of over-testing and over-diagnosis.
Why does this happen? Well no test is perfect – they all cause ‘false positives’ – people with positive tests who turn out to have no disease, and ‘false negatives’ – people with negative tests who do have disease – both can cause harm.
Having grappled with the decisions about when to perform blood tests on my own patients, my research has been exploring the use of certain tests called ‘inflammatory markers’ through interviews with GPs and nurse practitioners. These blood tests detect inflammation in the blood which can be caused by infections, auto-immune conditions, or cancers. Sometimes GPs use them to try to help them feel more confident that they are not missing anything.
‘So if we had a test that, a single blood test, that doctors could do which would reassure the patient there was nothing wrong at all, then that would be a very popular test… the [inflammatory marker] is probably the closest thing that we’ve got to a ‘nothing wrong at all’ test.’
— GP
The idea of a ‘nothing wrong at all’ test is certainly appealing, especially in today’s culture of litigation and defensive medicine. The problem lies when we are trying to prove that there is ‘nothing wrong’ and the test comes back slightly raised (see text box below).
It seems counter-intuitive, but the exact same test will perform very differently in different circumstances. This is why it is not a good idea to follow Mark Cuban’s advice to ‘have your blood tested for everything’, and is why it is important for doctors to be careful about selecting the right people for blood testing. If you are having tests done, think about asking your doctor questions first such as, ‘Do I really need this test? What are the risks? What are the alternatives? What happens if I do nothing?’
As a GP registrar wanting to make sure I don’t ‘miss anything’ I might check someone’s inflammatory markers, hoping that as a ‘nothing wrong at all test’ it will help me exclude anything serious. But for a person with mild or non-specific symptoms, the risk of serious disease will be low, making the chance of false positives high. This is tricky for a test which might be raised due to something as mild as a viral infection or as serious as cancer; which tests should I do next, how far should I look? Or as one of the GPs interviewed said: ‘I sometimes think I shouldn’t have done that test because it’s just complicated things.’
For doctors like myself this means trying to get the balance right. As I learn how to make careful diagnoses, I need to know when to perform tests and I also need to learn when to watch and wait, when to listen and notice, when to think and when to prevent harm; or as Iona Heath beautifully described in her essay ‘the art of doing nothing’.
The danger of false positives – the maths
Imagine the serum-rhubarb test: let’s suppose it has a ‘false-positive rate’ of 1 in 100, meaning that one in every hundred patients without rhubarb syndrome tests positive. If we use this test in a population where the prevalence of rhubarb syndrome is 1 in 100, then for every one patient correctly diagnosed with rhubarb syndrome, another will test positive who does not have the disease; the test will be correct only 50% of the time.
Supposed we use this same test in as a screening tool in a population of patients where the prevalence of rhubarb syndrome is only 1 in 1000 – now for every one patient correctly diagnosed we will have 10 patients with ‘false positives’ – in other words ten people, like Keith Marton’s wife, will have a positive test without any underlying disease, leading to anxiety, further tests and treatments.
How can this be avoided? Well let’s imagine instead of using this test to check for rhubarb in the general population, GPs instead carefully select patients who they think are at high risk and only test this group. The prevalence of rhubarb syndrome in the tested group is now much higher, say 1 in 10. Now if we test 100 people we will pick up 10 cases of rhubarb syndrome we will only have 1 ‘false positive’, so the test will be right 90% of the time.
How can you determine a personalised prevalence for a specific condition or alternatively each condition in a differential diagnosis for a specific test in order to determine that test’s positive or negative predictive value?
Good question! In theory we can do this using Bayes theorum, that is ‘the pre-test odds of a hypothesis being true (ie the patient has the condition) multiplied by the rate of new evidence (the test result) generates the post-test odds of the hypothesis being true’. An excellent resource for this is JAMA rational clinical examination series (https://medicinainternaucv.files.wordpress.com/2013/02/jama-the-rational-clinical-examination.pdf) which systematically reviews the diagnostic accuracy of history, examination findings and investigations for a range of target conditions. In reality this approach is not always feasible, and instead the pre-test probability may often be based on the clinicians’ judgement. What we must remember then is that if a patient is judged to be at low risk of disease, then the chance of a positive test result being a false positive will be greater. This is why screening programmes for healthy people can have unintended harmful consequences, and why we when we order batteries of tests without thinking we sometimes ‘just complicated things’. If you want to read more I highly recommend Gird Gigerenzer’s book ‘reckoning with risk’.