Indicator Ontology - JohnTigue/idots GitHub Wiki

This wiki page is about defining an ontology of infectious disease outbreak indicators. Such an ontology is a deliverable of this Outbreak Time Series project. Examples of the kind of indicators the ontology should address include:

  • "All deaths in the last 7 days"
  • "Cumulative suspected cases to date, in women"
  • "Cumulative deaths, percentage of which are from the last 3 weeks"

Table of Contents

Introduction

So, the task this page is addressing is to map out and name well various topics embedded in such indicator names. For example, seemingly confidence is a dimension/property with values suspected, probable, and confirmed. Another property seems to be deaths versus cases. There are populations and sub-populations. Etc. This ontology will be mapped to RDF and serialized as CSVW table annotations, see CSVW Namespace Vocabulary Terms for further information on where RDF and CSVW meet.

This page is currently broken out from the Specs in order to provide easy focus on just the ontology during its development, without having to wade into formal specification language of a work in progress.

The Outbreak Time Series Specification (the Spec) will use this ontology. This page may well get folded into the Spec. Alternatively, there is precedence in web standards to have "well known vocabularies" as separate specs, both of which get released simultaneous. The idea being that the core spec is designed to be extensible. Inevitably, there will be a need for indicators that are not covered in this "well known vocabulary" spec. So make the core extensible, and separately use those extension mechanism to inject a vocabulary of frequently seen indicators, as defined in this vocabulary spec for an ontology of common epidemiological indicators.

Target audience and author

The target audience for this document is epidemiology domain experts who would like to contribute to the discussion of this ontology. By separating this page from the Spec, such experts can quickly read this page and easily leave comments. If you are such a person, please just hit the edit button and type away in this wiki page. This page is under version control; any changes can be rolled back, so not to worry. If you do not have and do not want a (free) GitHub account you cannot edit this page. In that case feel free to email comment to John Tigue ([email protected]).

The original author of this document is John Tigue; hopefully others will contribute. I (John Tigue) am not an expert in the epidemiology field so please feel free to provide feedback which differs radically from is presented here. (Heck, I don't even know if "indicators" is the right term.) I am simply a programmer with standards body experience who saw a place where he could contribute by driving the definition of a spec and building software (visualization, translators, validators, etc.) that works with the spec.

Any existing definitional standards should be used. For example, the Disease Ontology is being used. Is there some already worked out ontology for indicators? That should be used. Please speak up if you know of any thing. Maybe even a good text book?

There is no official standards body which is defining this specification. This is a completely informal process at this time. If this project proves valuable, the end definition of success would be to transfer the work to a standards body. At that time John Tigue simply becomes the spec editor and reference implementation coder. The reference implementation is called Omolumeter.

Meta terminology: ontology versus schema

This ontology is an abstract concept design for consumption by human minds; the ontology provides common understanding between humans. The Spec is designed for consumption by machines. The ontology is to be encoded in the Spec as a schema.

For more background on the distinction made here between ontology and schema, see Ontologies and Database Schema: What’s the Difference?. Simply take "database schema" in those slides to mean CSVW schema.

Goal

The goal of this project is to make outbreak time series data become first class objects of the Web, encoded in a machine readable fashion based on the Web's native data standards such as JSON and CSVW. The specific task of this page is to build a data type for identities of infectious disease outbreak indicators such as cases_confirmed, deaths_suspected, deaths_probable, etc.

In programmer terminology when information is enters or exits a running programs, the thing exchanged is called a serialization. A serialization is often called a document, for example "Outbreak Time Series conformant documents." A serialization may be the contents of a file on a file system, or it may be the payload of an HTTP message sent over the web.

Documents conformant to the Spec contain data typed by the schema. The Spec uses CSVW for the schema. Serialized, this is a small-ish JSON file. The main data of an outbreak time series is a CSV file. So, the ontology defined here is simple a way of saying something like "column 3 of that CSV contains cumulative suspected cases to date."

In CSVW, having a schema of well defined indicators enables expressing assertions such as "column 7 of that CSV contains the same information as that other CSV." The Spec is written such that the set of indicators is not fix; there is an extension mechanism for indicators not defined in this ontology. In this way, the goal is to create a standard that can be easily used across the web on multiple web sites for the exchange of outbreak data.

Potentially relevant prior art

For our application, we used a disease ontology, a database of infectious diseases and associated metadata. An important challenge that we address in our system is the automatic integration of ontology information with document content. For our system, we adapted the ontology used by The Institute for Genomic Research (TIGR) as part of their Gemina project (http://gemina.tigr.org). This ontology is ordered in a hierarchical manner with diseases arranged in order of increasing specificity of disease descriptions. The ontology provides standardized disease names, as well as commonly-used synonyms for the disease as used in medical literature.

Early work from the Spec

v0.0.1 of the Spec was pre-defining the following:

  • otss_cases_all
  • otss_cases_probable
  • otss_cases_suspected
  • otss_cases_confirmed
  • otss_deaths_all
  • otss_deaths_probable
  • otss_deaths_suspected
  • otss_deaths_confirmed
  • otss_basic_reproductive_number

Indicators: types and attribute dimensions

As seen in the above list, even a small set of indicator names shows some internal structure that should be teased out. One dimension seems to be [cases, deaths] and another dimension seems to have values [probable, suspected, confirmed]. Is the latter the same dimension that has values "all" and "discarded" or is there another dimension at play? Etc. What dimensions need to be modeled in the indicator ontological?

These indicators can add up quickly. One WHO CSV file on Ebola 2014 had about thirty indicators.

The follow dimensions are basic and easy to classify:

  • Subject
    • cases
    • deaths
    • (Perhaps this is outcome: death, active, recovery, where cases = active+recovery but not deaths?)
  • Veracity
    • suspected
    • probable
    • confirmed
    • discarded TODO Is this one actually an value of another dimension?
    • all (lame label, and what does it even mean?)
  • Lag, temporal offset
    • notice how these define repeated structures just at different delays
    • now (default
    • X intervals (or Y day?) Ebola data has 7 days and 21 days
    • incubation periods ago?

The following observed indicators are more complex:

  • CFR
    • this is the combo of a subject and attribute; has details
      • count, percentage, cfr (is CFR it's own dimension?)
  • "cumulative deaths, of which % from last 3 weeks
  • What about, for e.g., "cases suspected per 1000" and "cases confirmed per 1000"? The "per 1000" is a different dimension of attribute typing, no?
  • otss_basic_reproductive_number

TODO Can otss_basic_reproductive_number really change per interval? I guess it could if the pathogen evolves. Perhaps this is a property of something else, say the outbreak rather than a specific interval? Perhaps what should happen is that the ontology is a spec unto its own. Then OTSS uses it in two places: on individual intervals or on the outbreak's metadata

Dimension: source

CIDRAP:

The source of his illness is primary, meaning he did not contract the virus from another person... The other patient is a 73-year-old woman from Najran in far southern Saudi Arabia who is hospitalized in critical condition. The MOH said she had indirect contact with camels.

Related definitional needs of the Spec

Diseases need short formal names to be referred to by internally within the CSV & JSON, and optionally in the end-user UI.

  • Textual disease name labels

    • What is this standard naming schema which generates name labels?
      • e.g.s
        • IAV: influenza A
        • MERS-CoV: Middle East respiratory syndrome coronavirus (MERS-CoV)
        • ZIKV, EBOV, CHIKV, HCV
      • what about ZIKV versus ZIKV disease (ZVD)? Which should be the label?

  • URI scheme for diseases

All content in this wiki is licensed freely (CC0 1.0 Universal (CC0 1.0) Public Domain Dedication)

⚠️ **GitHub.com Fallback** ⚠️