From Testiwiki
Jump to: navigation, search

<section begin=glossary />

Class is a set of items (objects) that share the same property or properties. The membership in a class is determined by an inclusion criterion. The property is utilised as a part of all objects that fulfill the criterion. Classes can be used in describing general information that is shared by more than one object. Class efficiently reduces the redundancy of information in the open assessment system. This improves the inter-assessment efficiency of the assessment work.

<section end=glossary />

Research question about the class structure
What is a structure for a class such that it
  • unambiguously describes the common property,
  • unambiguously describes the inclusion criterion, i.e. the rule to find out whether an object has the property and belongs to the class or not,
  • inherits the main structure from universal objects,
  • complies with the Set theory,
  • complies with the PSSP ontology.

The attributes of a classD↷ closely resemble those of a variable. However, the interpretation is slightly different, as can be seen from the table. In addition, the usage of data is not clear at the moment.

Table 5. The attributes of a class.
Attribute Subattributes Comments
Name Identifier for the class.
Scope Description of a property or properties, which are shared by all the items in the class.
Definition An inclusion criterion that unambiguously distinguishes whether a particular object has the defined properties or not. In other words, definition separates objects that belong to the class from those that do not belong. The definition also contains the discussion about memberships.
Result List of items (formally structured objects) that belong to the class.


Examples of use

A class may contain information about e.g. a good function that should be used to calculate the result, or a range of plausible values for a certain type of variable. Examples of these are given below.

A general dose-response function can be a class. For example, the multi-stage model for cancer dose-responses can be defined as

  P(d) = 1-exp(-q0 -q1d -q2d2)

This function has four parameters: q0 (the "background"), q1 (the "slope" at low doses), q2 (the "curvature" parameter) and d (lifetime daily dose of the chemical of interest). The function can be applied to a particular chemical among a wide range of chemical carcinogens, if the chemical-specific parameters q0, q1, and q2 are known. The result attribute of this class is equal to the general form of the multi-stage function. The function is used in the definition/formula attribute of a dose-response variable of a particular chemical, together with the chemical-specific parameters. The result attribute of this variable is the dose-response of the particular chemical, with one parameter, d. This variable can then be applied in a case-specific risk assessment, when the parameter d is replaced by the dose in an exposure scenario in that assessment.

This is an efficient way of organising information: all discussion about the plausibility of the multi-stage model in general is located in the class. Therefore, this discussion is held only once, for all chemicals and all assessments. Also the discussion whether the multi-stage function applies to a particular chemical is located there. The resolution of that discussion applies to all risk assessments on that chemical. The chemical-specific dose-response variable contains the discussion about the best estimates of the chemical-specific parameters. And again, the variable is applicable to all risk assessments on that chemical. A particular risk assessment can focus on estimating the exposures. The whole dose-response part of the assessment is ready-made.

Prior values for variables can also be located in classes. For example, imagine a class "Plausible range of PM2.5 annual average mass concentrations in ambient air." This is a uniform probability distribution of concentrations ranging possibly from 3 µg/m3 (in Antarctic) to 300 µg/m3 (in downtown Delhi). This can be applied in PM2.5 variables for checking for implausible values. The range (i.e., the value of the result attribute of the class) can be located in the definition/data attribute of e.g. a variable "PM2.5 annual average concentration in downtown Kuopio." If we do have measurements from Kuopio, we can do Bayesian updating using the range as the prior. This way, we can operationalise the use of both the case-specific measurements and the general knowledge from the class.

In practice, when new variables are created, they can partly be described using the results of existing classes, as long as the variable belongs to these classes. Possibly there should be a possibility to overrule the class information with case-specific information, if this is explicitly mentioned. However, deviations from the general rule should be defended.