Background for a Clinical Phenotype Ontology
These notes are preparatory for the Dallas Workshop: "Signs, Symptoms and Findings: First Steps Toward an Ontology of Clinical Phenotypes"
I will also attempt at adopting, at the same time, a more realist or more epistemological perspective.
Diagnostic Evidences, Physiological Phenomena, Pathological Patient Classes
The meta-model of the NEUROWEB Reference Ontology can be generalized. It relies on this analytical deconstruction of the general workflow in medical diagnosis:
collect diagnostic evidences, which are stored in the medical record in the form of clinical indicators
Note: the term diagnostic evidence puts the accent on the inference of a certain phenomenon from certain observations, and according to a varying mix of general theory and expert, thus is more related to the context (diagnostic process); the term clinical indicator puts the accent on a storable item of informationinfer the occurrence of (Patho)Physiological Phenomena, referring to the standing Anatomic-Physiological model of the human organism
Note: how the inference is formulated is pretty much of a hard matter; eraly AI approaches attempted at replacing the physician in this task, but substantially failed; here the accent is mostly on the content theory, rather than the inference mechanismassign the patient to a formal Pathological Class, useful for standardized treatment decision, clinical trials, genetic or other kind of association studies, etc...
Example: simplified diagnosis of a stroke:
1. identify symptoms of cognitive and motory impairment,
2. perform brain imaging and identify a darker area possibly corresponding to a brain lesion,
3. infer that a brain lesion occurred because of the formation of an obstructing body (different possible causes), which eventually obstructed a brain artery,
4. collect other diagnostic evidences relating to the cause of the stroke
5. finalize the diagnosis and assign the patient to a class
Different areas of medicine are expected to have different taxonomies of Pathological Classes, but most of the content theory concerning (Patho)Physiological Phenomena is expected to be shared, to a certain degree (although there are probably problems of granularity, and different perspectives).
Test case: how shared is the notion of Stenosis?
Degree of Confidence in a Diagnosis
A Pathological Class may be defined also on the basis of a confidence degree. The confidence degree must not be interpreted as the intensity of the pathology, but rather as the probability of correctness of the diagnosis. However, patients belonging to the same degree-of-confidence class may share a similar clinical state, and in that sense the degree-of-confidence may be interpreted as having an additional meaning (cf. association studies); the opposite scenario is in case the variability/dishomogeneity in the patients' clinical state just increases with the decreasing of the degree-of-confidence.
How to handle this feature? Is any component in the meta-model required to handle confidence degrees?
Different confidence degrees may be due to different factors; if these factors are formulated at the diagnostic-evidence level:
- quantitative strength of the diagnostic evidences (e.g. mild impairment may )
- presence of confounding/conflicting evidences
- insufficient availability of diagnostic evidences
Taxonomical Criteria for the Pathological Classes
Possible criteria for specifying a clinical class are:
- etiology (causal factors)
- anatomical location
- severity
- degree of certainty