Initiation of the diagnosis
The diagnosis may suggest itself as a ‘spot diagnosis’ as the patient walks through the door – Parkinson’s, depression. The patient may have already ‘self labelled’ – ‘I have trapped a nerve in my back’, ‘I have sinusitis’. Sometimes there may be a specific presenting complaint – headache. Often in general practice there is a panoply of symptoms and the general practitioner must decide which may indicate significant disease and which are unlikely to. Sometimes it is the recognition of a pattern when the individual features may be relatively subtle – the jaw ache with the headache (giant cell arteritis) or the urinary symptoms with the back pain (cauda equina syndrome).
Refinement
In the process of refinement of that initial impression it is important to ‘second-guess’ oneself. In many cases there are conditions that, though rare and of relatively low probability, are such that the consequences of missing them are too terrible to take the risk. It is worth sending 100 patients into hospital with a sudden onset severe headache to pick up one subarachnoid haemorrhage and avoid the catastrophic stroke in a 30 year old. Murtagh (1990) described this process as using the ‘restricted rule out’.
In a differential diagnosis the condition at the top of the list should not be the most probable but the condition that you cannot afford to miss.
Once the initial impressions have been formed it is necessary to gather information that refines those possibilities – iteratively testing the various diagnoses. This often involves fitting the presentation to a pattern – this aching 70 year old who has difficulty getting out of a chair (‘I’m getting old, doctor’) may fit the pattern of polymyalgia rheumatic. However, it is often at this stage that errors are made – and we will discuss that below.
Often it is necessary, at some vague level of appreciation, to use probabilistic reasoning. A 20 year old with rectal bleeding is unlikely to have colorectal cancer.
In the past it was always extremely difficult to get information about how likely it was that a patient with a particular presentation had a given condition. Medical textbooks were encyclopaedias of diseases such as cryptogenic fibrosing alveolitis. The likelihood was loosely inferred from the incidence of the condition in the population.
In the last 20 years there is far more information about the predictive values of symptoms for particular conditions in particular age groups. Much of this data is from secondary care studies (A&E or specialist clinics). Since these populations are a highly selected group the results will usually not be applicable to general practice. However, increasingly there is primary care data that is obtained from interrogating anonymized computer data uploaded from primary care clinical systems, such as the General Practice Research Database (GPRD). The likelihood that a 65-year-old man with an unexplained haemoglobin of 11.5 g/dl and a low ferritin having colorectal cancer is about 6.5% (Hamilton et al., 2008). The chances that a 70-year-old man with dysphagia has upper gastrointestinal cancer is about 9% (Jones et al., 2007). The data is of course still relatively crude. The predictive value of a symptom is probably an order of magnitude higher if it is volunteered (‘my food sticks, doctor’) as compared with if it is elicited (‘Do you ever notice that your food sticks? Well, yes, actually. A bit.’)
There are now many useful repositories of such information. One of the first examples, starting in 1993, was the excellent series in the Journal of the American Medical Association (JAMA) called ‘The Rational Clinical Examination’ (Sackett, 1992). Now articles on evidence-based diagnosis are frequently published in the British Medical Journal (BMJ) and the British Journal of General Practice (BJGP). This information is usually couched in the language of Bayesian reasoning, and we will discuss that a little further below.
Clinical Prediction Rules (CPR) are increasingly a part of the diagnostic armament. Often they are developed and validated in secondary care (such as the Well’s Scores for DVT and pulmonary embolism, or the Ottawa ankle rules). Secondary care CPRs have to be applied with caution in primary care because the populations are different, the initial likelihood of disease will be different (usually lower) and the clinical features that are discriminatory may be different (the absence of a central prostate sulcus on digital rectal examination is probably of relatively little discriminatory value to a urologist because few patients he sees will have a sulcus, whereas it probably is discriminatory for a general practitioner).
Sometimes the Clinical Prediction Rule may be problematic for the general practitioner if its specificity is poor, yet it finds its way into a national guidance document. An example is the 7-point check list for malignant melanoma that defines urgent referral criteria under the June 2005 NICE Guidelines for Suspected Cancer. The problem is that the specificity of the 7-point check list is only about 30–40% (Whited and Grichnik, 1998). Thus 60 or 70 out of 100 benign lesions in a pigmented lesion clinic will score 3 or more on the 7-point scale. A general practitioner cannot possibly urgently refer 60% of all the pigmented lesions she sees. Yet she may be criticized for not following guidelines if she fails to refer someone who turns out to have a malignant melanoma (see below for tips on how to avoid this problem).
Defining the final ‘diagnosis’
As stated above, the ‘diagnosis’ may merely be a negation – this patient has not got cardiac chest pain. It may be a working strategy – this patient seems to have a peripheral rather than central cause for her vertigo so we will wait and see.
A relatively small number of patients in primary care can be given a confident diagnosis at first presentation. It is also important not to give diagnoses of misleading and spurious accuracy. Not all low back pain to one side of the midline is due to facet joint arthritis. Sometimes a misleading accurate diagnosis (‘hemiplegic migraine’) can be very misleading and harmful because the patient relies on the diagnosis and subsequent clinicians fail to reconsider this.
Sometimes it is necessary to try and refine the diagnosis with further tests, such as X-rays, blood tests or urinalysis. In these circumstances it is important that the general practitioner has some idea of the performance characteristics of the test. A normal chest X-ray does not exclude bronchial carcinoma in a 60-year-old smoker with haemoptysis. The sensitivity of spinal X-rays to detect vertebral metastases (for example, in a patient with a known history of prostate cancer and back pain) is only about 70%. It would be inappropriate (and a common source of diagnostic error) to rule the conditions out on the basis of a normal test result.
Sometimes a ‘test of treatment’ will help to clarify the diagnosis. If the chronic cough responds to a proton pump inhibitor it was likely to be due to gastro oesophageal reflux. However, this approach needs to be viewed with appropriate caution if there are potentially serious causes. It is no longer acceptable to make a diagnosis of angina on the basis of therapeutic response to a GTN spray.
Often in primary care it is necessary to allow some time for the diagnosis to become clear. The natural history of the condition determines the diagnosis. Acute diarrhoea is likely to be due to infective gastroenteritis if it settles over a week or 10 days. However, if the ‘test of time’ is to be used safely (for example, in the child with abdominal pain who does not seem to have appendicitis) then ‘safety netting’ is of critical importance. We will discuss this further below.
Quite commonly it is appropriate that no diagnostic label is applied. This may be unpopular with the patient, who will often welcome spurious diagnostic certainty. However, it may be the more honest and safer course of action. The diagnosis remains ‘open’ rather than ‘closed’ and both doctors and the patient acts accordingly.
If the patient has a long history of multiple symptoms that result in fruitless investigation it may be necessary to discuss with the patient the tradeoff between the risks of over investigation (and over diagnosis) and the risk of missing a diagnosis. If the reasoning is clear, logical, agreed with the patient and recorded, it may be possible to defend a failure in diagnosis in these circumstances. The question though of what constitutes consent on the part of the patient (for a decision, for example, not to be investigated or referred) will be discussed below.
Where do errors occur in diagnosis?
A type of ‘taxonomy’ of diagnostic errors can also be discerned from the diagnostic strategies outlined above by Heneghan et al. (2009). For convenience we will reproduce Figure 1.1 again.
Before discussing where on the diagram above the errors occur, it is worth considering the research into cognitive reasoning that has gone into this area. If we can identify common ‘cognitive biases’ that lead to error we can devise strategies to guard against them.
When diagnoses are missed it is usually assumed that it is due to inadequate data collection. Sometimes this may be the case. Often, however, this is not a consequence of sloppiness or inattention to detail; instead the critical bits of information are missed simply because the general practitioner did not think of the correct diagnosis. It never occurred to him that the patient, who may have had no obvious risk factors, may be breathless because of a pulmonary embolism. It was this failure, of even considering the diagnosis, which meant that he did not detect the swollen calf.
Quite often close analysis will show that the general practitioner has made one or more common cognitive errors.
General practitioners, like all clinicians, use cognitive heuristics that may lead to incorrect weighting of the evidence.
‘Confirmation bias’ leads to information gathering that will confirm rather than refute the diagnosis. If you think the 65-year-old man is hoarse because of gastro oesophageal reflux (GORD) then asking him about the presence of heartburn might appear to confirm the diagnosis. However, heartburn is so common in the population that its presence does not go anyway to excluding the diagnosis of bronchial carcinoma at the left carina. The confirmatory evidence is given too much weight. A more appropriate strategy would have been to refer for a chest X-ray and then referral for laryngoscopy, both to refute the (possibly more likely) diagnosis of GORD. Karl Popper’s one millionth black swan refuted the general law (‘all swans are white’) that had been obtained by induction from the particular observations to the general law. A general practitioner should not carry out a million chest X-rays to detect one cancer (!) but the principle that confirmatory evidence is not necessarily discriminatory is often correct.
‘Premature closure’ is a common cause of diagnostic error in clinical negligence cases. The diagnosis of mechanical low back pain in a somatizing 50-year-old Asian lady may have been reasonable initially but, 3 months later when she was losing weight and sweating at night, the diagnosis needed to be rescrutinized. Unfortunately the general practitioner had already made up her mind that the woman was a depressed lady who ‘catastrophized’ her symptoms.
‘Framing bias’ is common in all areas of medicine. Once the CT scan shows the blood in the subarachnoid space the neurosurgical SHO elicits a perfect history of a subarachnoid haemorrhage, filling in the details not volunteered by the patient by direct questioning. The general practitioner may record a history that is absolutely typical for trigeminal neuralgia once he/she has decided that that is the diagnosis. There are many conditions in medicine in which the way the clinical data is ‘framed’ leads inexorably to a diagnosis – ‘pleuritic chest pain’, ‘dry cough at night and on exertion’. In these circumstances it is important to be very careful about what information is volunteered by the patient and what (often confirmatory) data is elicited by direct questions – ‘Do you cough a lot at night? Yes (but I also cough a lot during the day and sweat a lot at night).’
‘Availability bias’ is the process of giving too much weight to one’s own past experience and easily recalled examples. Usually in primary care this leads to under diagnosis because serious conditions individually are relatively rare.
We will discuss simple strategies below that can help avoid falling into these ‘cognitive traps’. However, first we give a few examples of the types of errors that can be anticipated from the real world structure of the diagnostic process imputed by Heneghan et al. (2009).
‘Premature closure’ on the spot diagnosis can lead to error unless the process is ‘second guessed’. The patient may indeed be depressed, as seems to be the case, but this could be because of serious physical illness. The ‘typical’ left Bell’s palsy may have weakness of grip in the left hand. Just because you have previously seen the ‘pattern’ of sudden onset of conjunctival oedema and periorbital swelling in a patient with allergic conjunctivitis does not mean that this person does not have a pre-septal orbital cellulitis. Beware of pattern recognition and ‘availability bias’. The evidence is very clear that clinicians form their hypotheses very early on in the consultation. To be safe it is necessary to ‘second guess’ them at a later stage.
Patients’ ‘self labelling’ is quite often correct. However, it is important not to put too much reliance on it. ‘It’s my old lumbago, doc’ may turn out to have a vertebral metastasis, as occurs in nearly 1% of patients with back pain in primary care.
Equally, the presenting complain is quite frequently not the important part of the consultation. The throwaway remark about a 6-week history of loose stool with a bit of rectal bleeding in a 55-year-old woman consulting about marital problems is likely to be significant. It is important to realize that, even though the remark comes as she is walking through the door and you have already spent 20 minutes with her, it is crucial to deal safely with the problem. The initial event in many medico-legal cases is the throwaway remark as the patient heads for the door.
It is crucially important to use Murtagh’s restricted rule out method. The first diagnosis on the list, so that you will actually consider it, is not the most likely but the one you must not miss. The difficulty with purely probabilistic reasoning is that the consequence of missing a diagnosis is often in inverse proportion to its probability. The febrile neonate probably has an upper respiratory tract infection but the consequences of delay of diagnosis of septicaemia can be catastrophic. The clinical suspicion and diagnosis of pulmonary embolism is very difficult. Many are missed. If you do not consider it first you will forget it and miss it.
How can we minimize the risks of these errors?
To finalize this section we suggest a few rules to help avoiding the cognitive errors that lead to misdiagnosis. These are the following 6 rules:
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
