By Jennifer Skillen, Gordon Rugg and Sue Gerrard
Some diagnoses are clear and simple, and give you a crisp either/or categorisation; either the case in question has a positive diagnosis, or it doesn’t. Diagnosis involves identification of the cause of a problem; this may occur in medicine, or in mechanical fault finding, or in other fields. This article focuses on medicine, but the concepts involved are relevant to diagnosis in other fields.
The diagram below shows crisp either/or diagnosis visually; cases either go in box A or box B, with no other options.
Not all diagnoses are this simple. Medical syndromes are an example. A syndrome in the medical sense involves a pattern of features that tend to co-occur, but where the cause is unknown. They typically involve multiple signs and symptoms, each of which may or may not be present in a particular case. (Signs are features that can be observed by someone other than the patient; symptoms can only be observed by the patient.) Making sense of syndromes, and of how to diagnose syndromes, is difficult, and often encounters problems with misunderstandings and miscommunications.
This article discusses ways of defining syndromes and related issues. Its main focus is on medical diagnosis, but the same principles apply to problems in other fields that have the same underlying deep structure.
Some cases involve a greyscale, such as the one shown below, rather than a crisp division into clear-cut categories, as shown in the “A or B” example above.
An example of a greyscale might be the severity of a fracture, ranging from very severe at one end of the scale, to minor at the other.
It’s often possible for a case to involve more than one form of categorisation. For instance, a particular fracture might be described both by its category (e.g. compound versus simple) and its severity (e.g. very severe to minor).
It’s also possible for a diagnosis to involve a combination of crisp and fuzzy sets, as shown below. In this diagram, some cases are definitely A; some are definitely B; others are on a greyscale between A and B. Using the fracture example again, it’s possible to think of greenstick fractures (C) as fitting on a greyscale between completely fractured (A) and not fractured at all (B).
An advantage of using diagrams in this way is that many people find them easier to understand than verbal descriptions, which can help communication.
In the examples above, the diagnosis involves a single condition with a known cause which definitely exists in the world. However, syndromes are more conceptually complex. They have unknown causes, and typically involve several different diagnostic features, which may not all be present.
This means that there is a risk of reification, i.e. a risk of inventing something which doesn’t exist in the real world. The pair of images below show how this can happen.
In the image below, Syndrome X may be caused by A, and/or B and/or C, and leads to outcomes 1 and/or 2, and/or 3.
However, in the next image, the same causes and outcomes are mapped directly onto each other, without using an intermediate Syndrome X. In this mapping, not all causes lead to all outcomes; Cause A, for instance, leads to Outcome 1 and Outcome 3, but not to Outcome 2.
The first of the two images above uses a syndrome as part of the causal modelling between causes and outcomes; the second models the same set of causes and outcomes without using a syndrome. Both handle the same evidence. So, how does involving a syndrome help make sense of what’s going on?
One key feature of a proposed syndrome is that it involves features that co-occur at above chance levels, for whatever reason. If you are sure that Syndrome X is present, and you’ve already seen Outcome 1 and Outcome 2 in a given case, then you know to check for Outcome 3 as well, and to be ready for it if it isn’t already present.
However, there is the risk of clusters of correlations occurring just by chance, or because of other factors that don’t actually justify invoking a syndrome (e.g. co-morbidity, or a shared deeper cause for two features). A possible example is Gerstmann Syndrome, which involves several loosely related problems, such as inability to distinguish the fingers on the hand from each other. There is debate in the literature about whether there is any medical advantage in treating these problems as a syndrome.
One way of representing the problem more systematically is to use a representation that shows all the logically possible permutations for a given set off diagnostic signs and symptoms, as shown below. This format is described in more detail in Skillen (2017). On the left, in Column A, all four are present. The next four columns show the different ways in which three of the four can be present. The following six columns show the different ways in which two of the four can be present, and the final four columns show each attribute occurring on its own.
With this representation, it’s possible to show how often each permutation occurs. For instance, Permutation B (Attributes 1, 2 and 3 together) might be much more common than Permutation C or Permutation D, but all three of those permutations might occur at levels well above chance. This provides a finer-grained way of handling syndromes than a simple binary “either the syndrome is present or it isn’t” diagnosis. Differences in frequency of occurrence of permutations would help identify different sub-groups of people within the syndrome, or different versions of the syndrome; occurrence of a permutation at above chance levels would support the argument that there is a real syndrome, rather than just a coincidence.
So, in summary, using systematic representations can help clarify terminology and definitions relating to syndromes and similar concepts.
Notes and links
https://en.wikipedia.org/wiki/Gerstmann_syndrome
Skillen, J.D., 2017. The symptom matrix: Using a formalism‐based approach to address complex syndromes systematically. Musculoskeletal Care, 15(3), pp.253-256.
https://onlinelibrary.wiley.com/doi/abs/10.1002/msc.1168
Jennifer Skillen can be contacted at her Keele University email address: j.d.skillen@keele.ac.uk
Gordon Rugg and Sue Gerrard can be contacted at Hyde & Rugg: gordon@hydeandrugg.com
Other resources that you may find useful:
The Hyde & Rugg website contains material relating to ways of handling knowledge, including elicitation, representation, testing/error and education.
The Hyde & Rugg blog contains articles that discuss concepts and methods for handling knowledge.
The Knowledge Modelling Handbook covers a wide range of concepts and representations, including the ones in this article.
Gordon Rugg & Joe D’Agnese’s book, Blind Spot tells the story of our work on identifying and reducing error in expert reasoning about difficult problems:
The Unwritten Rules of PhD Research, by Gordon Rugg and Marian Petre, is a book written to help PhD students understand the world of research, and how to survive and do well in it.
hyde and rugg 