Pathognomy, Sine Qua Non and Constitutive Matching (Philosophy of Diagnosis, Part 2)

This paper continues the series Philosophy of Diagnosis. See also Part 1 and Part 3.

A goal throughout this framework will be to infer and defend rules which constrain how a structure for diagnostics must work. These constraints will allow us to critique existing accounts of diagnostics and to build a new model which does justice to practitioners’ intuitions, behavioural norms and goals.

To begin, this paper will lay down some extreme cases in which the constraints form clear and indisputable rules. There are four main elements to this guiding work: pathognomy, sine qua non, constitutive matching and the caveat versions of each of these. We will be able to lay out three sets of rules which will built into a strong basis for preliminary evaluation of philosophical accounts of diagnostics.

Pathognomy

When the presence of a symptom entails the presence of an underlying condition, we will say the symptom is pathognomonic for the condition. In other words, a pathognomonic symptom s is sufficient to infer the presence of a condition c. For convenience, we will call this condition c the indicated condition and the symptom s the indicator. In cases of pathognomy, the symptom is a pathognomonic indicator which requires the presence of the indicated condition.

There are many pathognomonic signs and symptoms across a wide range of specialisms. For instance, Koplik’s spots are pathognomonic for measles – every patient who has Koplik’s spots has measles (see Zenner & Nacul, 2012). Similarly, Kayser-Fleischer rings are pathognomonic for Wilson’s disease (Walsche, 1976), while pill-rolling tremors are pathognomonic for Parkinson’s disease (e.g. Findley, Gresty & Halmagyi, 1981). Test results are particularly apt to pathognomy, exactly when the test conclusively rules in the disease. For instance, observation of Reed-Sternberg cells on microscopy is a pathognomonic test for Hodgkin’s lymphoma (Reed, 1902), while detection of Aschoff bodies in the heart is pathognomonic for rheumatic fever (Aschoff, 1904). In some cases, a pair of symptoms might be jointly sufficient, so form a pathognomonic pair (or triad, etc.).

However, while pathognomonic symptoms are sufficient for inferring the presence of the disease, they may not be necessary. For instance, while every patient who exhibits Kayser-Fleischer rings has Wilson’s disease, not every patient with Wilson’s disease has the rings – indeed, around a third of Wilson’s disease patients present without them (Frommer et al. 1977). So, while the presence of a pathognomonic indicator entails the presence of the indicated condition, the absence of the symptom does not entail the absence of the condition.

Pathognomy creates several constraints which any defensible framework for diagnosis must respect. These are:

Pathognomy Constraint 1:

If a patient p exhibits symptom s which is pathognomonic for condition c, then a diagnosis which does not include c is not a candidate diagnosis, a legitimate diagnosis, or a fully accurate diagnosisfor p.

Pathognomy Constraint 2:

If a patient p exhibits symptom s which is pathognomonic for condition c, then a diagnosis which includes c is at least to some degree an accurate diagnosis for p.

Constraint 1 proposes absolute proscriptions. We do not even consider any diagnoses which omit a pathognomonically indicated condition. Hence, diagnoses omitting the indicated condition are not candidates. Similarly, because a practitioner informed of the pathognomy and the presence of the symptom can conclusively exclude any diagnosis which omits c, it would be illegitimate to offer that diagnosis to the patient. We should not offer diagnoses we know to be incomplete where we also know for sure a way to render the diagnose more complete. Finally, we know that the diagnosis cannot be fully accurate because it omits a condition which we know the patient has. Any account of candidacy, legitimacy and accuracy must respect and fulfil this criterion.

Furthermore, per Constraint 2, if we create a diagnosis which includes c then we are saying something which we know to be correct about the patient, namely that they have c. So, no diagnosis which invokes c can be completely inaccurate. Note that this constraint presupposes a fuzzy notion of accuracy which admits of degree. If we required a binary classification of diagnoses as accurate and inaccurate, then we can only rule a diagnosis accurate where it is completely accurate, and therefore Constraint 2 would not hold.

We should also consider a somewhat broader version of pathognomy in which a symptom is sufficient to infer that at least one condition from some set {c1, c2, c3 …} is present. For example, risus sardonicus (the “rictus grin”) is sufficient to conclude that the patient either has tetanus, strychnine poisoning, or Wilson’s disease (see e.g. Mackenzie, 1979). They must have one, but which one is not certain from the sign alone. Here, we can see that the constraints given above can be broadened to incorporate any member from the indicated category.

We should also consider circumstances in which a set of symptoms S experienced in combination, rather than a single symptom, is pathognomonic for a condition. This is often the case with syndromes, a subgroup of conditions which are sometimes defined extensionally by the concurrence of a specified set of symptoms. Note, though, that not all conditions which have been termed syndromes are purely extensional (many have well-known causes and causal relationships amongst the symptoms, which would be omitted in a simply extensional syndrome, such as Down syndrome – see below for further discussion of purely constitutive syndromes). Examples of such syndromes include exploding head syndrome, extensionally defined through auditory hallucination of loud noises on waking or falling asleep combined with a strong fear reaction but no associated pain. Because of the lack of known etiology, the existence and diagnosis of extensionally defined syndromes are often controversial and arguments to attribute the physical manifestations to psychological disorders or mass hysteria are often widespread in the medical literature (e.g. chronic fatigue syndrome, Gulf War syndrome, fibromyalgia and multiple chemical sensitivity syndrome). At this stage, there is no need for this framework to make any rulings on these topics: the framework can accommodate or exclude such syndromes depending on the underlying philosophical commitments introduced.

Pathognomy Constraint 1*:

If a patient p exhibits a set of symptoms σ ∈ S which is jointly pathognomonic for a non-empty set of conditions C, then a diagnosis which does not include at least one condition c C is not a candidate diagnosis, not a legitimate diagnosis, and is not a fully accurate diagnosis for p.

Pathognomy Constraint 2*:

If a patient p exhibits a set of symptoms σ ∈ S which is jointly pathognomonic for a non-empty set of conditions C, then there is a condition c C for which any diagnosis which includes c is at least to some degree an accurate diagnosis for p.

These revised constraints do in fact entail the previous versions of the constraints for the special case where the sets σ and C each have exactly one member. As such, we will take the revised constraints as foundational and the single-condition versions as a special case. However, as most pathognomy in practice relates to a single indicated condition, we will continue to discuss pathognomy through the simplified single condition case and use the simpler version of the rule in most examples.

As it stands, the two sets involved in the definition function somewhat differently. The set of conditions, C, resembles a disjunction – the presence of the pathognomonic indicator set entails that (at least) one of the elements of the set is present. But it is possible (if rare) for a single indicator set to conjunctively indicate the presence of two or more conditions simultaneously (i.e. the presence of σentails that both c1 and c2 are present). This is best treated by viewing these as two separate sets C1 = {c1} and C2 = {c2} and applying the constraint twice over. This will also allow for additional levels of complexity (e.g. a symptom which is pathognomonic for c1 and one of c2 or c3 can be rendered through two sets C1 = {c1} and C2 = {c2, c3}). Meanwhile, the symptom set σ is treated analogously to conjunction. The constraint stipulates that all members must be concurrently present for the condition to apply. Of course, there can be conditions for which there are multiple pathognomonic indicators, each of which are individually sufficient to infer the presence of the condition. Again, the most harmonious way to accommodate this is to apply the constraint several times, with each individual pathognomonic indicator forming its own set. In combination with the machinery for conjunctive and disjunctive condition indication, this already provides quite a powerful set of tools to handle deterministic relationships from symptom to condition.

Sine Qua Non

While pathognomy is sufficient for a condition, a sine qua non is necessary for a condition. Sine qua non always occur when the underlying condition is present. That is, if s is a sine qua non of c, then the presence of c entails the presence of s. Consequently, the absence of s entails the absence of c.

The absence of a sine qua non rules out the underlying condition, much as the presence of a pathognomonic symptom rules it in. For example, a vaginal pH reading greater than 4.5 is a sine qua non for bacterial vaginosis. A lower pH would rule the condition out. But a higher pH does not rule it in (Manka et al. 2002). A sine qua non might not guarantee that the indicated condition is present.

Sine qua non requires a complementary rule to those for pathognomy:

Sine qua non Constraint:

If s is a sine qua non for condition c and for patient p’s symptom set S, s S, then any diagnosis which includes c is not a candidate diagnosis, legitimate diagnosis, or a fully accurate diagnosis for p.

The Sine qua non Constraint resembles Pathognomy Constraint 1. It stipulates a diagnosis which invokes a condition for which a sine qua non of that condition is missing from the patient’s symptom set cannot be a candidate or legitimate diagnosis. We do not need to consider this diagnosis because we can already rule out the presence of a condition diagnosed. Hence, it is not a candidate diagnosis. Similarly, it is illegitimate to offer a diagnosis that includes a condition which one knows cannot be present. Finally, we can conclude that the diagnosis cannot be entirely accurate because it proposes that the patient has a condition which we know they do not have.

There is no complementary second constraint with respect to sine qua non, because the presence of a symptom which is a sine qua non of some condition does not entail that the condition is present. Sine qua non can be extremely non-specific (but see the discussion of constitutive matching, below).

Of course, we may see test results or symptoms which are both pathognomonic and sine qua non for a single condition. That is, they are necessary and sufficient to confirm that the patient has the condition. Mostly, this happens where a specific disease or condition has been defined in terms of a specific test (e.g. diabetes mellitus as sometimes characterised simply is persistent high blood sugar levels, so this symptom is both necessary and sufficient, entailing that some form of diabetes must be included in the diagnosis – cf. Alberti & Zimmet, 1998), or when there is a unique cause of the condition which is known, such as a genetic mutation which guarantees the presence of the condition (and only that condition). In these cases, using the two Pathognomy conditions and the Sine qua non condition simultaneously captures this relationship.

Similarly to pathognomy, we can generalise sine qua non reasoning to cover the possibility of a disjunctive sine qua non relationship. Whereas in the case of pathognomy the machinery we needed used a disjunction of conditions and a conjunction of symptoms, to govern sine qua non we need the reverse. We are interested in cases in which all patients with a condition c exhibit at least one symptom from a set {s1, s2, …}. We can define the set σas the set of symptoms such that any patient who has c must have at least one member of σin their symptom set. This will again be particularly useful to capture certain kinds of syndrome – those in which patients exhibit a number of symptoms from a specified set but no one individual symptom is necessary. For example, Gulf War syndrome is often defined as exhibiting some but not necessarily all of chronic fatigue, musculoskeletal pain, insomnia and cognitive dysfunction. The absence of all of these symptoms would be considered sufficient to rule out Gulf War syndrome, so they constitute a disjunctive sine qua non (see e.g. Ismail et al. 1999).

It might be possible, if unusual, for the combination of a given set of conditions to produce its own sine qua non. For instance, neither c1 nor c2 individually have s as a sine qua non, but if c1 and c2 are experienced simultaneously, s must always occur. To account for such a case, we can also generalise the account to include the sine qua non of a setof conditions C = {c1, c2, …} considered conjunctively. As before, we will denote this as σ. As in the case of pathognomy, this modified constraint yields the simpler version as a special case where C contains only a single condition and σ only a single symptom.

Sine qua non Constraint*:

If σ is a non-empty set of sine qua non for a non-empty set of conditions C,and for patient p’s symptom set S, σ ∩ S = ∅, then any a diagnosis which includes all elements of C is not a candidate diagnosis, legitimate diagnosis, or fully accurate diagnosis for p.

This constraint says that if there are no symptoms from the set of sine qua non of C exhibited by a patient, then any diagnosis which ascribes all of C to the patient cannot be a candidate, legitimate or fully accurate diagnosis.

Again, we can anticipate that there may be cases in which one condition has multiple sine qua non which all must be present. This is covered by applied the constraint repeatedly, once for each sine qua non symptom. One symptom can similarly be a sine qua non of multiple conditions, and again this is covered by repeat application of the constraint.

Constitutive Matching

A final specific version of these constraints concerns constitutive matching. Constitutive matching can occur when a purely constitutive condition is ascribed to a patient. A condition is purely constitutive when it is extensionally defined, solely through the presence of a specified set of symptoms. A purely constitutive condition does not stipulate any causal mechanisms or associations among the symptoms, or any causal mechanism from the condition to the symptoms. Saying that a patient has a purely constitutive condition c simply is saying that they exhibit the requisite symptoms and nothing more. For such a condition to be recognised at all, it must be the case that the chance concurrence of these symptoms with the observed prevalence is highly unlikely, but the etiology is unknown. As discussed above, some syndromes have this characteristic.

The set of symptoms which defines a purely constitutive condition may be simple (e.g. a patient has c if and only if they have symptoms s1, s2 and s3), or more complex (e.g. a patient has c if and only if they have symptom s1, and either symptom s2 or s3). Sometimes, a constitutive condition will refer to a list of symptoms and stipulate that a patient must experience at least a minimum number of these to have the condition. While many conditions invoke such lists of symptoms as diagnostic criteria, what distinguishes the purely constitutive conditions is the absence of any further causal structure beyond the mere presence of the symptoms together. At the simplest level, we can state that any symptom trivially defines a purely constitutive condition which is the condition of having that symptom. For example, we could stipulate that chronic pain syndrome is a condition which occurs precisely when a patient has the symptom of chronic pain.

Purely constitutive conditions can be quite controversial (see e.g. Ismail et al. 1999). Some purely constitutive conditions may betray the absence of knowledge about the causal relationships involved. The absence of causal understanding leads to their usefulness, legitimacy and existence (insofar as conditions are believed to ‘exist’) being questioned. However, we will not presume any judgment on this question at this point. Questions of the value and legitimacy of purely constitutive conditions within diagnoses will rise to the forefront in our construction of diagnostic legitimacy and quality and in responding to the problem of diagnostic effect.

When a patient satisfies all the requirements of a constitutive condition, we will say that they have a constitutive match. A constitutive match is sufficient to establish that including that condition in a diagnosis is accurate. This is the case even if we are able to provide a diagnosis which ascribes causal structure for and amongst the conditions which goes beyond that achievable by offering the constitutive condition as a diagnosis. For instance, if we diagnose that a tumour is causing the chronic pain, it remains accurate to state that the patient has chronic pain syndrome. So, the accuracy of that diagnosis will depend on the accuracy of the claim that the tumour is causing the pain (and anything else which is included), and the inclusion of chronic pain syndrome cannot lower the accuracy level. Thus we have the constraint:

Constitutive Matching Constraint:

If c is a purely constitutive condition and a patient’s symptom set S constitutively matches c, then including c in a diagnosis for p does not decrease the accuracy of the diagnosis.

Moreover, insofar as accuracy might be negatively defined (that is, a diagnosis is accurate to the extent that it includes nothing inaccurate), a diagnosis which only invokes a purely constitutive condition is fully accurate. However, we will not presuppose this approach to defining accuracy for the meantime. It may be possible to offer alternative accounts of accuracy which are not defined in this way. We retain the question of whether a diagnosis solely of chronic pain syndrome for a patient who has chronic pain due to a tumour would be considered accurate, even if potentially illegitimate.

Constitutive matching is subsumed under the category of sine qua non. A purely constitutive condition describes a set of sine qua non symptoms (often disjunctively) for the condition, and nothing more. Thus, constitutive matching is a special case of sine qua non. But it merits special discussion because the Constitutive Matching Constraint uniquely applies only to those conditions which are purely constitutive, and goes beyond the scope of the Sine qua non Constraint. We might regard the Constitutive Matching Constraint as the analogue of Pathognomy Constraint 2 for sine qua non, but as it only applies to a subset of sine qua non situations, it has been presented separately here.

Caveating

Finally, we must consider what I will call the caveat versions of each of these three clear-cut cases. A caveat case is one in which a symptom is pathognomonic or a sine qua non unless there is something more, generally something unusual, amongst the symptom set which invalidates the relationship.

Consider whether leg pain is a sine qua non of a broken femur. We expect patients with a broken femur will be experiencing pain in that leg. Unless, that is, there is something else in their symptom set working against this, such as being unconscious, anesthetised, or insensitive to pain due to a condition such as CIPA. This means that as long as certain other symptoms are absent, the symptom of leg pain is a sine qua non for broken leg. This is a sine qua non with caveat.

Fortunately, the existing machinery of the Sine qua non Constraint can handle caveats without modification. In the broken femur case, we know that the patient will either experience leg pain or is insensitive to pain. Thus, where c = broken femur, we construct a disjunctive set of pathognomonic symptoms: σ= {leg pain, insensitivity to leg pain}. By applying the Sine qua non Constraint, we know that a diagnosis which includes a broken femur can only be a candidate, legitimate and fully accurate diagnosis if the patient’s symptom set includes either leg pain or insensitivity to leg pain.

The same goes for pathognomy with caveat – this occurs when a symptom is sufficient for the presence of a disease just when some other symptom(s) is absent. This is most likely to occur where a specific test result is pathognomonic for a disease as long as certain criteria about the patient are met (e.g. a blood test result proves that the patient has the disease as long as they are not also pregnant). Again, the rules of pathognomy are able to handle these cases. We include the negation of the symptom which must be absent in the set σ. In the case of a test which is pathognomonic for a condition unless the patient is pregnant we construct σ = {positive test result, not pregnant}.

Again, these caveats could be considerably more complex. There could be multiple caveats to satisfy (e.g. a positive test result is pathognomonic for some condition just if the patient is not pregnant, suffering from measles, over 75 or fasting). This is handled easily by adding more members to σ, as only one of those negated symptoms or properties need be met. But we could also consider a caveat in which multiple circumstances transpiring at once forms the caveat. For example, a positive test result is pathognomonic for some condition as long as the patient isn’t both pregnant and taking St. John’s wort. In this case, we include the symptoms together as a subset within the set σ, i.e. σ = {positive test result, {pregnant, taking St. John’s wort}}. Such subsets are treated conjunctively.

Summary

Pathognomonic symptoms only occur when an indicated condition is present. So, we can rule in the condition when we see the pathognomonic symptom in a patient’s symptom set. Sine qua non always occur when an indicated condition is present. So, we can rule out the condition if the sine qua non is missing.

The four constraints proposed here allow us to impose minimal requirements on any theory of diagnostics and any response to the problems of diagnostic candidacy, legitimacy and accuracy. To be able to handle these comparatively simple deterministic relationships between symptoms and conditions, an account of diagnostics must satisfy the criteria presented here. Hence, any account of candidacy, legitimacy or accuracy which violates any of these criteria in even a single case is invalidated.

Most conditions will not have simple one-to-one pathognomonic or sine qua non relationships to symptoms. But for those few which do, those relationships can be captured through the four constraints here. In particular, as far as we wish to include purely constitutive syndromes in our diagnostic processes, these can be captured by the pathognomy and constitutive matching constraints. But the caveated and expanded versions of these criteria are able to handle substantially more complex relationships which expands the territory of condition-symptom relationships governed by these rules. Indeed, these rules alone cover a lot of ground for the best-understood areas of diagnostics. Most conditions, though, will require a more sophisticated set of tools to capture complex statistical and mechanistic dependencies between symptoms and conditions, which will only be available once we develop a detailed account of the structure of diagnoses.

These constraints have left open an important question: whether a diagnosis consists of ascribing a condition or conditions to the patient, or whether it can include more than that, including postulating causal relationships and dependencies between conditions and symptoms or amongst symptoms. Sadegh-Zadeh’s well-known account of diagnostics (2000; 2011) considers only the ascription of conditions to patients, but the framework developed here will go beyond this to include causal claims within the remit of diagnosis. This allows us to move beyond the relatively simple mechanics of purely constitutive conditions (where there is no further causal structure to be offered beyond the condition(s) alone). We will turn to the structure of diagnosis in the next parts of this series.

Bibliography:

  • Alberti, K.G.M.M. & Zimmet, P.Z., 1998. Definition, Diagnosis and Classification of Diabetes Mellitus and its Complications. Part 1: Diagnosis and Classification of Diabetes Mellitus. Provisional report of a WHO Consultation. Diabetic Medicine, 15: pp.539-553
  • Aschoff, L., 1904. Verh. dtsch. path. Ges., 8: 46.
  • Findley, L.J., Gresty, M.A. & Halmagyi, G.M., 1981. Tremor, the cogwheel phenomenon and clonus in Parkinson’s disease. J Neuro, Neurosur, Psych, 44: pp.534-546.
  • Frommer, D., Morris, J., Sherlock, S., Abrams, J. & Newman, S., 1977. Kayser-Fleischer-Like rings in patients without Wilson’s Disease. Gastroenterology, 72(6): pp.1331-1335.
  • Ismail, K., Everitt, B., Blatchley, N. et al., 1999. Is there a Gulf War syndrome? Lancet, 353(9148): pp.179-182
  • Mackenzie, I., 1979. Risus Sardonicus. French’s Index of Differential Diagnosis, 11th ed. John Wright & Sons Ltd. p.697.
  • Manka, W., Adrianowicz, L., Wesolek, Z. & Adrianowicz, K., 2002. The value of determining vaginal secretion reaction (pH) as a screening test of bacterial vaginosis. Wiad Lek., 55(1-2): pp.51-55.
  • Reed, D., 1902. On the pathological changes in Hodgkin’s disease, with especial reference to its relation to tuberculosis. Johns Hopkins Hosp Rep, 10: pp.133-196
  • Sadegh-Zadeh, K., 2000. Funda-mentals of clinical methodology: 4. Diagnosis. Artificial Intelligence in Medicine20(3), pp.227-241.
  • Sadegh-Zadeh, K., 2011. The logic of diagnosis. In Philosophy of Medicine, pp.357-424, North-Holland: Amsterdam.
  • Walshe, J.M., 1976. Wilson’s disease (hepatolenticular degeneration). In: Vinken PJ, Bruyn GW (eds) Handbook of clinical neurology, 27, part I. North-Holland: Amsterdam, pp 379-414.
  • Zenner, D. & Nacul, L., 2012. Predictive power of Koplik’s spots for the diagnosis of measles. J Infect Dev Ctries, 6: pp.271-275.
Updated 01/10/20