Introduction to clinical reasoning

How do doctors make decisions? Alison Round explains some of the thought processes that lead to a diagnosis

Doctors make decisions all the time - what the problem is, what the diagnosis is, whether to do anything, what to do. What facts do doctors take into account when they come to a decision, and what processes do they use to decide on a course of action? Where does intuition come from? These are the basics of clinical reasoning. When decisions are made in conjunction with the patient, doctors need to have an understanding of the "building blocks" of their thinking in order to explain this to the patient and to explore areas where differences in values and opinion may occur.

In all fields, not just medicine, experts make decisions in very different ways from students or beginners. Traditional bedside teaching guides students to take a history and perform an examination before constructing a differential diagnosis. In real life, however, experienced doctors do not work like this. They utilise a number of shortcuts (heuristics), based on knowledge and previous experience, which enable them to work much more quickly and, in general, more accurately than students.¹ There are many advantages to heuristics, such as very rapid processing and an ability to handle complex information without overload. There are, however, also a number of biases incorporated in the heuristics that may lead to poor decision making.² This article aims to discuss the processes and biases, using making a diagnosis as an example, and considers how improvements could be made.

Clinical reasoning in differential diagnosis

Experts use three main methods, or a combination of these, in making a diagnosis. Probably the most common is the hypothetico-deductive approach. An initial hypothesis or hypotheses are generated very early during the initial presentation of the problem, from existing knowledge, associations, and experience. Further questions or examination are oriented towards supporting or refuting these first ideas. If an hypoth- esis is discarded, an alternative one is considered and treated in the same way. Several hypotheses can be actively considered at any one time. Both awareness of probabilities (prevalence) of disease and knowledge of causal pathways are important.³

Pattern recognition is also common. A particular combination of symptoms, or even certain phrases used to describe a symptom, can suggest a diagnosis very strongly. People build up their own internal library of patterns on the basis of their experience and existing knowledge.

Finally, pathognomonic signs and symptoms exist where a particular finding almost guarantees a certain diagnosis. Ulnar deviation in rheumatoid arthritis, Kaiser-Fleischer rings in Wilson's disease, and the slow relaxing jerks of hypothyroidism are examples. Unfortunately, most of these findings are rare and of little help in day to day practice.

All diagnostic methods depend on breadth and depth of knowledge, but the application of knowledge is not as straightforward as it seems. The use of algorithms (following a structured guideline to reach a diagnosis) is not welcomed by many doctors, despite their accuracy and relative freedom from bias. Professionals may consider they have enough confidence in their unaided decision making, describing algorithms as too time consuming.⁴

Biases in thinking

Several common features of thinking can occur during the clinical reasoning process. These are known as cognitive biases. Firstly, there is the difficulty in estimating probabilities accurately, giving undue weight to small samples, or overestimating the similarity between people or events (often known as representativeness bias). Secondly, there is a tendency to attribute too much weight to easily available information, or to an event that is easily remem- bered because of particularly salient features; an example is overestimation of the probability of death by lightning. Thirdly, some people, when asked to estimate the probability of an event, place the initial probability at too extreme a figure and then make insufficient adjustment for subsequent information. Finally, there is also a bias towards positive and confirming evidence at the expense of negative evidence.

The case studies illustrate these points. When you read case study 1 (table), imagine you are seeing a patient in an accident and emergency department, and think what is going through your mind at each point.

When doctors are not certain about a diagnosis they search for more information, either from the history or examination or by performing investigations or tests. One important source of bias in handling information is incorrect application or interpretation of tests. Case study 2 (box 1) considers these points (see below for answers).

Bayes' theorem and its uses

Bayes' theorem is a simple formula which suggests how one should modify one's original idea in the light of new information. In the case above, the new information is in the form of test results. In order to work out how much modification is necessary with the positive exercise electrocardiogram, one needs to know what the chance of having a positive test result is when the disease is present and what the chance of having a positive test result is when the disease is absent. This gives an idea of the diagnostic value of a test, is called the likelihood ratio, and can be calculated from the sensitivity and specificity of tests (see box 2).

The sensitivity of a test is equivalent to the chance of having a positive test result when the disease is present. The specificity of a test is equivalent to the chance of having a negative test result when the disease is absent, and therefore (12specificity) is the chance of having a positive test result when the disease is absent.

Bayes' theorem is perceived to be complicated and difficult to understand. If, however, the initial probability of disease can be estimated (which one does whether or not Bayes' theorem is being used), estimates of the sensitivity and specificity of tests can be used easily to arrive at a final (posterior) probability of a diagnosis, given a particular test result. Such data are available for many investigations, but the method is not generally taught when differential diagnosis is being discussed.

In the case study above, most people estimate the probabilities as about 5% initially, rising to about 70% with the positive exercise test and back to about 50% with a negative thallium scan. In fact the actual probabilities are 1%, 6%, and 2% respectively. So whereas the initial probability is not far wrong, the final estimate is markedly inaccurate. The positive exercise test seems much more important than it is, partly because the initial probability of disease is so low that more positive tests are false positives than true positives. What Bayes' theorem demonstrates is that for diseases that are initially unlikely, a positive test of reasonable sensitivity and specificity increases the absolute chance of disease by only a small amount. Conversely, for diseases where the prior probability is very high, a positive test may add very little, or a negative test may not substantially reduce the chances of disease. Box 3 shows some tips to improve clinical reasoning.

This article has covered only a basic introduction. There are several good texts that provide more information, such as Clinical Epidemiology by Sackett et al, and Professional Judgement, edited by Dowie et al.^{5 6}

For particularly interested readers, the publications by Elstein et al⁷ and Kahnemann et al⁸ are strongly recommended, as are articles discussing other theoretical approaches to clinical reasoning.⁹

Alison Round, consultant in public health medicine, North and East Devon Health Authority, Southernhay East, Exeter


studentBMJ 2000;08:1-44 February ISSN 0966-6494

1. Kassirer JP, Kuipers BJ, Gorry GA. Toward a theory of clinical expertise. Am J Med 1982; 73:251-9.
2. Tversky A, Kahnemann D. Judgement under uncertainty: heuristics and biases. Science 1974; 185:1124-31.
3. Kassirer JP. Diagnostic reasoning. Ann Intern Med 1989;110:893-900.
4. Pearson SD, Goldman L, Garcia TB, Cook EF, Lee TH. Physician response to a prediction rule for the triage of emergency department patients with chest pain. J Gen Int Med 1994 9(5):241-7.
5. Sackett DL, Haynes RB, Guyatt GH, Tugwell P. Clinical epidemiology. A basic science for clinical medicine. Boston: Little, Brown, 1991
6. Dowie J, Elstein A. Professional judgement. A reader in clinical decision making. Cambridge, Cambridge University Press, 1988.
7. Elstein AS, Shulman LS, Sprafka SA. Medical problem solving-an analysis of clinical reasoning. Cambridge, Massachusetts, Harvard University Press, 1978.
8. Kahnemann D, Tversky A, eds. Judgement under uncertainty: heuristics and biases. New York: Cambridge University Press, 1982:359-69.
9. Greenwood J. Theoretical approaches to the study of nurses' clinical reasoning: getting things clear. Contemp Nurse 1998;7:110-6.