From W3C Wiki
Jump to: navigation, search

Different viewpoints on Evidence across the HCLS Spectrum

Question: Can we come up with a uniform way of representing and reasoning with evidence across various HCLS contexts?

Biological View of Evidence

  • Simple Pragmatic approach: Create a generic OWL annotation property (Matthias Samwald)
    • Observation of Existence of A. e.g., experiment, XML file with results of the experiment, paper interpreting results
    • Provides rationale/reason to believe in existence of B
    • Focus on physical, biological and clinical reality, e.g., preference to experiment over
    XML file or opinion of scientist

Clinical Research View of Evidence

  • Clinical Trials Inclusion or Exclusion Criteria: Can we view these criteria as evidence (Karen Skinner)
    • E.g., definition of a smoker, Fagerstorm Definition for Nicotine Dependence
  • Evidence based View of Medicine: Classification of evidences (Dirk Coalert)
    • Depends on the Source of Statement. The scale/weight of the evidence depends on various factors
    • Opinion has the lowest ranking
    • Results substantiated by a Randomized Clinical Trial with senstivity, specificity and confidence interval has the highest ranking

Clinical View of Evidence

  • Radiology: Evidence is a function of the following (Daniel Rubin):
    • The facts or observations
    • The analysis method
    • The inference
    • The observer
  • Clinical Obsevations and Judgements: Similar to the above and seems to be an instantation (Vipul Kashyap):
    • The clinical observations which sometimes are not recorded
    • Clinical Judgement, e.g., the pain suffered by the patient is moderate to severe
    • Scores, e.g., Sometimes judgements are scored on a scale
    • Mappings between the observations and the assessments.
    • The inference or a set of aggregation and classification rules
    • The observer, e.g., Patient, Nurse, Physician, etc.
  • The instability of facts: Facts are not necessarily stable or consistent (Matt Williams):
    • Mutliple physicians can have conflicting evaluations of patients
    • Evidence may or may not be linked to probabilities, e.g. Wigmore Charts, Braden Risk Assessment Scale.
    • But: Is there an underlying equivalence between these approaches and Probability Theory? (Vipul)

Experimental Data View of Evidence

  • Use of provenance information, regarding 'How', by 'What method' and by 'Which agent' was a piece of data generated, as evidence to derive conclusions about the identity of data. (Satya Sahoo)
  • When a list of peptides is derived from a 'biochemical sample' using mass spectrometry (ms) the evidence that accompany these results are:
    1. The details of the original sample (organism, type of cells, cleavage enzyme used etc.)
    2. The ms instrument used, the settings of the instruments, the algorithms used in processing the ms data, the databases used by the search algorithms etc.
  • Hence, these types of evidence, for example a tryptic enzyme (trypsin or chymotrypsin) was used as cleaving agent for proteolysis, decides that the given data is specifically a list of 'tryptic' peptide.
  • Evidence may be considered as subset of Provenance (Satya Sahoo and Cory Henson)
    1. Specific provenance parts, those that have a direct role to play in identifying a piece of data (with some amount of trust), may be classified as 'evidence'
    2. In some scenarios, evidence may also be a proper subset of provenance

Current Models of Evidential Reasoning

  • Wigmore Charts (Wigmore, Tillers, Anderson, Twining)
  • Toulmin-style (informal) argumentation (Toulmin)
  • Bayes Net based approaches (Schum, Dawid)

There are a few scenarios that were discussed on the mailing list. These were X-rays, physical examination findings, clinical trial results and classification of patients according to some criteria.

= A quick Toulmin-style Model of Evidence

  • Some terms:

Toulmin described the different parts of an argument as the datum, the warrant, the backing, claim and qualifier; I will ignore the last one and attempt to apply the others to this domain. However, as we shall see I think we need to refine some of these.

  • The datum is the information we have about this case/ instance/ individual
  • The warrant is a statement about some general case that we may apply in this situation, to this datum
  • The claim is the result of the application of the warrant
  • The backing is the more technical basis on which the warrant is given

In the examples above, therefore:

  • The X-ray:

The datum is the patient's X-ray; the warrant is the interpretation of this x-ray; the backing is the radiologist; the claim is the radiologist's report

  • The Classification of a smoker:

The datum is the patient's questionnaire score; the warrant is the mapping of scores to categories of dependence; the backing is the published justification of such a mapping; the claim is the classification category that the patient is put in.