Failures of Diagnosis

Of late, a confusion has emerged in the diagnostic and philosophical literature concerning the ways in which diagnostic failure can occur. It’s rare to see clear attempts at a taxonomy of diagnostic failure. Here, I want to disambiguate a number of ways in which clinicians (and computer diagnostic aids) can fail in their diagnostic capacity. This breakdown should help us to pinpoint the responsibilities (both ethical and legal) of clinicians—and maybe even of computer programmers, or computer diagnostic AI itself—and to better understand the risks and principles involved in diagnosis.

I outline a range of categories of diagnostic failure. Some categories are proper subsets of other categories, or have overlap, explained below.

Misdiagnosis is the act of giving an incorrect diagnosis—that is, of telling a patient they have or may have a condition which they do not, in fact, have. Misdiagnosis covers a large proportion of the landscape we will consider here.
A subtler point within misdiagnosis is the question of complex diagnoses—diagnoses which involve the clinician postulating to the patient that a number of conditions are or may be present. For example, the clinician might say: “You have had transient ischemic attack, or you are suffering from multiple sclerosis.” Suppose the patient was indeed suffering from multiple sclerosis, and had not had any TIAs. In this case, it is unclear whether there was misdiagnosis occurring. The result depends on how we break down the diagnosis. The diagnostic statement was true—so, if we take the diagnosis at face-value, there was no misdiagnosis here. However, if we take the clinician as offering two putative diagnoses, then one of these was a misdiagnosis.

Undiagnosis refers to the act of failing to provide a diagnosis, leaving the patient undiagnosed. Non-diagnosis (below) should be considered a form of undiagnosis (as long as the nondiagnostic statement is not accompanied by any diagnostic statement). But undiagnosis includes both cases in which in a statement which is not diagnostic is offered as if it were, and cases in which no diagnosis is attempted, or no diagnosis is offered: “We don’t know what’s wrong with you”. One sub-case of undiagnosis involves the practitioner diagnosing the patient and then failing to communicate the diagnosis to them, for whatever reason.

Non-diagnosis will here refer to the act of giving a something which is not a diagnosis as if it were a diagnosis. Non-diagnoses are statements that are, for whatever reason, not diagnostic statements. For example, the statement: “You have blond hair” is not a diagnosis, though it fits the normal logical form that diagnosis-statements take (ascribing a physiological or psychological predicate to the patient). It is an interesting question, which I take up elsewhere, whether some statements are non-diagnoses for some patients, but diagnoses for others. For instance, the statement “You have lung cancer” is clearly diagnostic for some patients. But for others, for instance patients with all the symptoms of a broken leg but none of the symptoms of lung cancer, the statement may be considering non-diagnostic (as opposed to simply a very bad case of misdiagnosis). See also ‘illegitimate diagnosis’, below.

An inelegantly gerrymandered term (we soon run out of prefixes) for the complement to misdiagnosis: anti-diagnosis is the act of excluding or ruling out the correct diagnosis for a patient. For instance, a clinician who tells his patient “You don’t have cancer” is performing anti-diagnosis just if the patient does in fact have cancer. It is an often-severe form of misdiagnosis (if an alternative diagnosis is provided) or undiagnosis (if no alternative is offered).

Illegitimate Diagnosis:
Illegitimate diagnosis involves what could have been a genuine diagnostic statement being offered illegitimately, in one of two situations:

  1. The clinician provides a diagnosis when there was insufficient information to justify providing that diagnosis. We might call this unjustified diagnosis. This might occur even when the clinician is correct in her diagnosis.
  2. The clinician provides a diagnosis which, while not necessarily a non-diagnosis (at least, not in every situation), is not plausible for this patient. For instance, providing the diagnosis “You have lung cancer” to a patient with all of the symptoms of a broken leg and none of the symptoms of lung cancer. We might call these unjustifiable diagnoses, to indicate that there is evidence that the diagnosis is not correct, as opposed to merely an absence of evidence that it is correct. It’s not clear whether all unjustifiable diagnoses are non-diagnoses – this depends on the details of the account of diagnosticity.

Over- and Under-Diagnosis:
Over and underdiagnosis can refer to general practices within the healthcare system of overusing certain diagnostic labels. Here, I use it to refer specifically to giving an diagnosis which is accurate broadly construed, but either under or over-estimates the severity, extent or scope of the condition. For instance, the diagnosis of headache might underdiagnose a migraine; the diagnosis of shoulder impingement might under-diagnose a full-thickness rotator cuff tear. The range of these terms is fuzzy at the edges – does stomach pain underdiagnose or misdiagnose appendicitis? – but fuzziness does not preclude the category from being useful.

Partial Diagnosis:
A partial diagnosis definitionally includes true components. It may accurately diagnose part of the symptom set which the patient presents, without diagnosing the rest, or might identify a condition which the patient suffers without identifying all of the relevant conditions.

Shallow Diagnosis:
A shallow diagnosis does not go deep enough to reach the underlying causes of a condition. Hypertension might explain a patient’s symptoms, but what explains the hypertension? Shallow diagnosis is only a diagnostic failure where the shallowness prevents what might be superior treatment, leads to incorrect prognostic statements, or detracts from the patient’s ability to understand their condition and make informed decisions.



Two final categories may be useful to attempt to spell out some of the nuances at the border between undiagnosis and non-diagnosis – those cases in which practitioners offer diagnoses which, for a range of reasons, cannot be considered part of clinical diagnostics. These are not illegitimate diagnoses, in the sense outlined above, as they aren’t recognised diagnoses. But they stand as an interesting category as they may be recognised by non-scientific or pseudoscientific authorities, or by patients, and sit at the complex boundary of un- and non-diagnosis.

A form of undiagnosis similar to non-diagnosis in which the clinician persuades (or attempts to persuade) the patient that they have diagnosed them when they have not in fact done so. This can include practices such as giving an elaborate medicalised or jargon name to the lack of a diagnosis. For instance, offering the diagnosis “You have idiopathic disease” is taken to mean a set of unexplained symptoms. Debates continue about whether certain conditions are true diagnoses or pseudo-diagnoses (see for instance the vocal debates around chronic fatigue syndrome, sudden infant death syndrome, etc.). One prominent form of pseudo-diagnosis is to restate the patient’s symptom or symptoms in a way which makes them believe that more has been learnt about the condition, when in fact it is a mere synonym. For example, a patient presenting with high blood pressure is told “You have hypertension”, and takes this to be an underlying condition which causes the high blood pressure symptom, as opposed to a more medicalised naming of the symptom. Pseudo-diagnostic statements may be entirely true, but are not diagnostic, and go beyond simpler cases of undiagnosis in that they also necessarily involve persuasion or intend to persuade.

Giving an alt-diagnosis consists of offering a diagnosis which is not accepted to exist as a condition in scientific medicine. Alt-diagnoses are most typically offered by complementary and alternative medicine practitioners. They are not diagnoses insofar as the definition of a diagnosis requires a “real” or medically accepted condition be ascribed to the patient. From that perspective, alt-diagnosis is another species of undiagnosis. It may also be considered a subset of pseudo-diagnosis, insofar as the purpose of an alt-diagnosis is to make the patient believe that they have been diagnosed when in fact they have not. The scope of alt-diagnosis will be a highly contentious issue between proponents and opponents of alternative medicine, but it is important to be clear that accepting the use of alternative therapies (homeopathy, chiropractic, acupuncture, crystal healing, etc.) in practice, and even accepting that such therapies could be effective for some patients, does not entail that corresponding diagnoses and alt-conditions from the same traditions are legitimate. Alt-diagnoses might include imbalances in qi, depletion of vital force, or even demonic possession. Some diagnoses which have some basis in medical systems may take on the force or mantle of alt-diagnoses when they applied liberally and far beyond the extent to which they could be accepted as practically occurring. Take for instance, the practice of alt-diagnosing candida overgrowth. Candida is a real fungus, and overgrowth of candida is possible – so candida overgrowth could be a legitimate medical diagnosis. However, in some alternative medicine traditions, it has been claimed that candida overgrowth underlies many (or even most) medical conditions, and candida overgrowth is applied as a widespread diagnosis for conditions which (according to the body of medical science), it cannot cause. As such, candida overgrowth could be an alt-diagnosis in some cases whilst not being so in others.