What Three Wise Men have to say about diagnosisBMJ 2011; 343 doi: http://dx.doi.org/10.1136/bmj.d7769 (Published 20 December 2011) Cite this as: BMJ 2011;343:d7769
- Navin Mani, specialty registrar, otolaryngology—head and neck surgery1,
- Nick Slevin, consultant, clinical oncology2,
- Andrew Hudson, specialty registrar, clinical oncology2
- 1Department of Head and Neck Surgical Oncology, Christie Hospital, Manchester M20 4BX, UK
- 2Department of Head and Neck Clinical Oncology, Christie Hospital, Manchester
- Correspondence to: N Mani
Most clinicians will be familiar with Occam’s razor, a guiding principle that is interpreted as “the simplest explanation is to be preferred.” In relation to medical diagnosis this is taken to mean that when a patient has multiple symptoms and signs, a single diagnosis should be sought that accounts for all the clinical features rather than attributing a different diagnosis to each.
To what extent should we rely on Occam’s wisdom? All clinicians can recall patients who have not conformed to this principle, and over reliance on it has the potential for misdiagnosis. Accurate diagnosis underpins good clinical practice; figures suggest that rates of diagnostic errors in the emergency department setting are as high as 10-15%1 and that misdiagnosis is responsible for around 10% of adverse events in hospitals.2 With this in mind, we present examples for and against the use of Occam’s razor from our area of interest: head and neck oncology, the management of which provides a good template to study these diagnostic dilemmas. Most patients undergo imaging of distant sites at diagnosis, and spread to the regional lymph nodes usually occurs in a predictable pattern before distant metastases. We also examine the views of other “wise men” who have proposed principles that caution against over reliance on Occam.
Occam: “entities must not be multiplied beyond necessity”
A 60 year old female smoker presented with haemoptysis. Chest radiography showed an opacity in the middle lobe of her right lung. She underwent lobectomy. Histology showed adenocarcinoma with clear resection margins (stage T2 N0 M0). No adjuvant therapy was given. One year later she presented with a left tonsillar mass and left cervical lympadenopathy (fig 1⇓). Biopsy of the left tonsil was reported as adenocarcinoma (stage T2 N2b M0) with similar histological features and an identical immunohistochemical panel of markers to the previous lung tumour. She received radiotherapy to the tonsil and neck. Five years after treatment of the lung and four after treatment of the tonsil there were no signs of recurrent disease.
William of Occam (1285-1348) was a Franciscan friar born in the village of Occam in Surrey. He studied theology at Oxford University but was unable to complete his studies as he was summoned by the papal court on charges of heresy. Occam escaped to Munich, where he continued his writing and is thought to have died from the black death. He was one of the major philosophers of his day. Although the principle of simplicity or parsimony dates back to Ancient Greece, Occam has been credited with its popularisation. His approach is summed up by the Latin phrase entia non sunt multiplicanda praeter necessitate, “entities must not be multiplied beyond necessity.”3 4
The medical interpretation of Occam’s razor is often called diagnostic parsimony. This states that when a patient with multiple symptoms is being investigated the clinician should seek a single diagnosis rather than diagnosing two or more unrelated ones. To favour the existence of multiple diagnoses would involve making additional assumptions. Occam asserts that the best theory should fit the observations without making any assumptions. The use of Occam’s razor is not restricted to medicine; another example is in the discipline of quantum theory. Both Einstein and Lorentz had similar theories of relativity. However, Lorentz’s theory depended on the existence of an “ether” for which observational evidence was lacking.5 As Lorentz’s theory required a further unproved assumption, Einstein’s theory is preferred (for now at least).
Based on clinical experience, the most likely diagnosis on presentation of the tonsillar mass was a separate primary squamous cell carcinoma of the tonsil (over 90% of tonsillar cancers are squamous cell). Using Occam’s razor the unifying single diagnosis would be either a primary tonsillar carcinoma that has presented with metastatic lung disease or a primary lung carcinoma that has metastasised to the tonsil. However, primary adenocarcinoma of the tonsil is rare and there are only a few reported cases of lung adenocarcinoma metastasising to the tonsil.6 7 So although a separate primary tonsil carcinoma was clinically most likely, the relatively short time scales between tumours, the use of Occam’s razor, and the similar histology point to a different conclusion.
Hickam: “patients can have as many diseases as they damn well please”
A 58 year old man presented with dysphagia and hoarseness. Investigations confirmed a hypopharyngeal squamous cell carcinoma (stage T3 N0). A computed tomogram showed bilateral apical lung lesions and a positron emission tomogram showed all the three lesions (fig 2⇓). The patient received induction chemotherapy followed by chemoradiotherapy to the hypopharyngeal primary. After a good response he underwent sequential left and right lung upper lobectomies. Histology from these showed adenocarcinoma (T2 N0 and T1a N0, respectively). No adjuvant treatment was required. He remains disease-free one year after final treatment.
John Hickam (1914-70) graduated from Harvard University School of Medicine in 1940 and became departmental head of medicine at Indiana University in 1958.8 Hickam’s dictum provides a counterbalance to Occam’s razor, stating that multiple symptoms and signs may be due to more than one disease, or succinctly put “patients can have as many diseases as they damn well please.”9 This may also have statistical validity, as it is more likely that a patient has several common diseases rather than one rare disease that explains multiple symptoms. Additionally, independent of statistical likelihood, some patients do in fact turn out to have multiple diseases.
Diagnosis followed Hickam’s dictum. The initial presentation of a primary head and neck cancer with lung lesions was suggestive of metastatic disease to the lung and this would be the preferred unifying diagnosis using Occam’s razor. However, the absence of lymph node metastases cast doubt on the unifying diagnosis, as it is likely that a head and neck primary would have metastasised to cervical lymph nodes before the lung. Further investigation and treatment confirmed different histology for the lung lesions, and the eventual diagnosis was three synchronous primary tumours. Had Occam’s razor been followed with a presumed diagnosis of metastatic lung disease, the patient would have received palliative treatment only, rather than radical management.
Crabtree: “no set of mutually inconsistent observations can exist for which some human intellect cannot conceive a coherent explanation, however complicated”
A 60 year old male former smoker presented with extensive fixed right upper neck lymphadenopathy (fig 3⇓). Biopsy of the neck mass confirmed metastatic squamous cell carcinoma. No primary site for the tumour could be found despite clinical and radiological examinations, including positron emission tomography and computed tomography. His tumour was classed as unknown primary (stage T0 N3 M0), with neck node metastases. The most likely site for the primary, at this stage, was in the head and neck region. He was treated with chemoradiotherapy, receiving irradiation to the neck and to the mucosal sites of likely primaries in the head and neck. Three months after treatment positron emission tomography and computed tomography scanning revealed a left upper lobe lung lesion. The patient underwent left upper lobectomy. Histology confirmed squamous cell carcinoma (stage T3 N0) but with different morphological features to the neck metastases. He received adjuvant radiotherapy to the chest. He remains disease free one year after final treatment.
Joseph Crabtree (1754-1854), a poet, was born in Gloucestershire and attended Eton College and Oxford University. He influenced contemporary poets of his day, including Wordsworth and Tennyson. In fact, Crabtree was a fictitious figure created in 1954 by a group of academics with the intention of injecting some humour and satire into their academic proceedings. The success of this concept led to the establishment of the annual Crabtree Orations, given by members of the Crabtree Foundation, each aiming to provide new insights into Crabtree’s life. The 1957 oration, “Crabtree and science” was given by Professor Reginald Jones.10 11 Jones was MI6’s principal scientific adviser during the second world war and later became professor of natural philosophy at Aberdeen University. In his writings about the intelligence community he outlined Crabtree’s bludgeon, which states that “no set of mutually inconsistent observations can exist for which some human intellect cannot conceive a coherent explanation, however complicated.”12 13 14
In the present case, the primary disease site was unknown at presentation and two potential unifying diagnoses exist, each involving elaborate explanations. One is that the lung was the primary site but unusually metastasised to the upper neck nodes only and uncharacteristically presented after the neck metastases had appeared. Alternatively, the patient had an unproved head and neck primary, which subsequently developed lung metastases. Both scenarios involve an elaborate series of events that fit the findings to create a seemingly logical explanation. With the histology from the lung lesion it becomes clearer that two separate primary cancers might be involved. The unifying theories, although based on logical links, would have been wrong.
Therefore in contrast to Hickam’s dictum, which acts as a counterbalance to Occam’s razor, Crabtree’s bludgeon contributes to blunting it. It emphasises the human tendency to try and link observations with unifying theories. If we become too focused on finding a unifying diagnosis then there is a danger of directing efforts to look for evidence to support our claim and ignoring information telling us the contrary. It is important that on finding a unifying diagnosis we test its validity and not accept it simply because it can explain all the symptoms.
These three frameworks may each be used to guide diagnosis. We have selected complex cases illustrating real life diagnostic uncertainty. Diagnostic errors in the United States now account for over a quarter of medicolegal claims and have overtaken surgical incidents as the leading cause of litigation.15 The cause of these errors is multifactorial but is mainly related to cognitive factors involving the clinician, usually due to inaccurate synthesis of information, incomplete data gathering, or insufficient knowledge.16
Knowledge of a specific diagnosis may not always alter initial patient management. However, in oncology it is important to distinguish between metastases that represent incurable systemic spread compared with curable multiple isolated separate tumours, to ensure patients are not incorrectly consigned to palliative treatment. Head and neck cancer is a special case owing to the increased risk of second primary cancers in the upper aerodigestive tract associated with cigarette smoke.17
William Osler (1849-1919), professor of medicine at Johns Hopkins Hospital and later the regius chair of medicine at Oxford University, is credited with applying the teaching of Occam’s razor to clinical medicine. During his time the prevalence of infectious disease was much greater than that of chronic illnesses,18 with a single diagnosis therefore more likely to link clinical features. Many patients now live to older age with multiple chronic diseases and this must be taken into account when choosing between Occam’s razor and Hickam’s dictum. As a consequence, for older patients the scales may tip in favour of Hickam,19 and for younger patients in favour of Occam. As medical knowledge advances, seemingly unrelated conditions may be found to share common causes, especially at molecular biological level, testing the validity of Hickam’s dictum.
So what of probability? We sometimes hear phrases such as “common things are common” and “when you hear the sound of hooves think of horses not zebra.” We use our previous observations to help predict what is most likely based on probability (such as our observation of the relative frequencies of horses and zebra in the British countryside). This is the basis of bayesian probability first described by Thomas Bayes (1701-61), an English Presbyterian minister whose theorem was published posthumously.20 In bayesian probability one starts with an estimate of probability and then refines that probability based on subsequent observations to create a new updated probability. Repeating this process gives a more accurate probability for the cause of an observation. Experienced clinicians intuitively use this method to elucidate the cause of symptoms in their patients. They use their experience and the clinical features and results of diagnostic investigations to refine the likelihood of each cause and then choose the most likely.
Surely then Bayes must be the wisest of all? Bayesian principles were used in all three cases to help elucidate the diagnosis because they were used to help accept or reject theories. However, to use Bayes effectively, adequate experience is required to be able to make an assessment of the initial situation. If we are inexperienced or find ourselves in an unfamiliar situation then we do not have the previous observations required to help refine the initial probability. Also, as the first case illustrates, the most probable is sometimes not the correct answer. We should therefore remain open minded and be prepared to challenge our diagnosis. Knowing when to continue testing and when to stop and accept the current theory is the diagnostic art of the clinician.
So how should we balance the differing approaches of these three wise men in a particular clinical situation? The cases highlight that probability and Bayes’s theorem, although useful, can have limitations in clinical practice. We should therefore also consider the guidance of the three wise men to help lead us to the correct diagnosis. Occam is a good general principle to follow, Hickam reminds us that Occam is not a rule, and Crabtree cautions against over elaborate explanations. Ultimately, to combine and use this knowledge effectively, the wisest of men and women needs to be the clinician.
Cite this as: BMJ 2011;343:d7769
Competing interests: All authors have completed the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declare: no support from any organisation for the submitted work; no financial relationships with any organisations that might have an interest in the submitted work in the previous three years; and no other relationships or activities that could appear to have influenced the submitted work.
Patients’ consent obtained.
Provenance and peer review: Not commissioned; externally peer reviewed.