Oncology App - Clinical Decision Support System

Hi, this is Ola form Cognitum. Today I will show you yet another problem which can be solved efficiently by CNL semantic technology.

The problem

Nowadays the medical knowledge is greater than ever. There are official medical guidelines that every doctor is obliged to follow. There are plenty of study cases and a lot of statistical data which help evaluate the probability of successful treatment. Every patient should have right to take advantage of most recent medical research. However, it is hard to do on daily basis. How can the doctor analyze all the books with medical data during patient's visit?

Our solution

Our Oncology Application guides the doctor throughout the patient’s visit. It suggests next steps of the therapy in accordance with up to date medical standards.
When the physician enters the essential information into the application, the system analyzes the clinical situation and tries to generate recommendations for the given condition. The recommendations are in accordance with European Society for Medical Oncology standards. It is the doctor who is still the most important in the decision making process but now he is aided with a professional tool analyzing the wealth of in-depth knowledge amassed in the field of medical oncology.

The core of our application is ontology. It is a set of sentences written in English CNL which contain rules for reasoning a next step of therapy for patient. The data for patients is stored in an Ontorion database.The outer layer is a C# web application with dynamically updated forms.
Let’s see how it is done in detail. Our case study is medical treatment of GIST (gastrointestinal stomach tumor.) First visit in our clinic aims to gain information about patient’s medical history and to conduct general physical examination (H&P).
The first form we deal with is called – not surprisingly – a patient basic information form. Or patient-basic-information-form in CNL style. It is stated in the ontology that every such form must have a section for basic information about patient.

Every patient-basic-information-form is a form.

Every patient-basic-information-form must concern a patient.

We query our ontology for constraints for a patient – things that every patient must or can have.

Every patient must have-date (some datetime value).
Every patient must have-pid (some integer value).
Every patient must have-name (some string value).
Every patient must have-age (some integer value).
Every patient must have-sex a sex.
Every patient must have-referred-by (some string value).
Every patient must have-ecog an ecog.
Every patient must have-anamnesis (some string value).
Every patient must have-hp (some string value).
Every patient must have-concomitant-medication a concomitant-medication.
Every patient must have-comorbidity a comorbidity.
Every patient must have-tumor-rupture a tumor-rupture.
Every patient must have-radiological-assessment a radiological-assessment.
Every patient must have-histopathological-diagnosis-status a histopathological-diagnosis-status.

Every patient can have-comment (some string value).

Based on the sentences from ontology our application is able to prepare a form which should be filled in during patient’s visit. Naturally some fields will be required (must modal expression) and some will be optional (can modal expression). In our first form all the fields are required except for the comment.

We are also asked about additional information about past medical examinations that may be relevant to our treatment. Imagine we enter:

Patient-810126 have-radiological-assessment Done.

Due to the following rule in our ontology…

If a patient-basic-information-form concerns a patient and the patient has-radiological-assessment Done then the patient-basic-information-form is a primary-tumor-form.

… our application extends the form with an extra section about the existing tumors. As you could have noticed the core words that give a grammatical strucutre to a CNL sentence are always highlighted in blue.

Every primary-tumor-section must have-tumor-localization-t-1 a tumor-localization-t-1.
Every primary-tumor-section must have-tumor-size-t-1 (some integer value).

Every primary-tumor-section must have-metastasis a metastasis.

Something is a tumor-localization-t-1 if-and-only-if-it is either Nonrectal, Oesophagus, Stomach, Jejunum, Duodenum, Colon, Sigmoid, Peritoneum or Rectum.

In this case, the information gathered about our patient’s first tumor was as follows:

Primary-Tumor-Section-P-810126-D-20140815-T-155933 have-tumor-localization-t-1 Colon.

Primary-Tumor-Section-P-810126-D-20140815-T-155933 have-tumor-size-t-1 equal-to 8.

It is worth mentioning that CNL is compatible with OWL/XML. Here is an example of two sentences above translated to OWL/XML. By no doubt the CNL is nicer to read. And in contrast to OWL/XML not only computer specialists can understand CNL.

<?xml version = '1.0' encoding = 'UTF-8'?>

<Ontology xmlns="http://www.w3.org/2002/07/owl#"xml:base="http://www.ontorion.com/ontologies/Ontologyd8e40ac4f075439080cfc99fb4f0c78f" xmlns:rdfs="http://www.w3.org/2000/01/rdf-schema#"xmlns:xsd="http://www.w3.org/2001/XMLSchema#" xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#" ontologyIRI="http://www.ontorion.com/ontologies/Ontologyd8e40ac4f075439080cfc99fb4f0c78f">

  <Prefix name="" IRI="http://www.ontorion.com/ontologies/Ontologyd8e40ac4f075439080cfc99fb4f0c78f#" />

  <Prefix name="rdf" IRI="http://www.w3.org/1999/02/22-rdf-syntax-ns#" />

  <Prefix name="rdfs" IRI="http://www.w3.org/2000/01/rdf-schema#" />

  <Prefix name="xsd" IRI="http://www.w3.org/2001/XMLSchema#" />

  <Prefix name="owl" IRI="http://www.w3.org/2002/07/owl#" />

  <ObjectPropertyAssertion>

    <ObjectProperty IRI="#hasTumorLocalizationT1" />

    <NamedIndividual IRI="#PrimaryTumorSectionP810126D20140815T155933" />

    <NamedIndividual IRI="#Colon" />

  </ObjectPropertyAssertion>

  <DataPropertyAssertion>

    <DataProperty IRI="#hasTumorSizeT1" />

    <NamedIndividual IRI="#PrimaryTumorSectionP810126D20140815T155933" />

    <Literal datatypeIRI="http://www.w3.org/2001/XMLSchema#integer">8</Literal>

  </DataPropertyAssertion>

</Ontology>

That was a digression. Let's focus again on the application. How is the information connected to the patient?

Patient-Basic-Information-Form-P-810126-D-20140815-T-155749 is a patient-basic-information-form.

Patient-Basic-Information-Form-P-810126-D-20140815-T-155749 concern Patient-810126.

Patient-Basic-Information-Form-P-810126-D-20140815-T-155749 concern Primary-Tumor-Section-P-810126-D-20140815-T-155933.

Primary-Tumor-Section-P-810126-D-20140815-T-155933 is a primary-tumor-section.

It is worth to pay attention to spelling and punctuation. We can write a name with lower case primary-tumor-section – when we want to introduce general rules about the form or a section in the form. Or upper case Primary-Tumor-Section-P-810126-D-20140815-T-155933 – for talking about a given form/section.

Extensive examination goes on and the form is dynamically updated with new fields. The fields are reasoned based on the rules in the ontology.

What is worth noticing, throughout all this updates the result of the form are sentences in CNL. They are by no doubt human readable. CNL sentences are available as an onmouseover text.

In our case study the patient has been already diagnosed with GIST so the system suggests performing a range of additional diagnostic tests before taking the decision about the form of therapy. Everything will be contained in a histopathological report for the patient.

If a patient-basic-information-form concerns a patient and the patient has-histopathological-diagnosis-status Gist then the patient-basic-information-form is followed by Histopathological-Report-Form.

Let me pay your attention to the way in which the articles a/an/the intrinsic to English language are naturally incorporated into CNL - especially when it comes to introducing general rules into the ontology. You just have to be a little bit more precise in using it compared to your everyday use of English.
The doctor was asked to decide between performing two types of diagnostic surgery tests – tumor biopsy and tumor excision. He chose tumor excision.

Every histopathological-diagnosis-section must have-surgery a surgery.

Something is a surgery if-and-only-if-it is either Tumor-Biopsy or Tumor-Excision.

Here is the information entered into the ontology about the tumor excision procedure performed on the patient:

Menu-For-Tumor-Excision-Section-P-810126-D-20140815-T-160755 have-procedure Excisional-Biopsy.

Menu-For-Tumor-Excision-Section-P-810126-D-20140815-T-160755 have-tumor-focality Multifocal.

Menu-For-Tumor-Excision-Section-P-810126-D-20140815-T-160755 have-preresection-treatment Systemic-Therapy.

Menu-For-Tumor-Excision-Section-P-810126-D-20140815-T-160755 have-margin Negative-For-Gist.

Menu-For-Tumor-Excision-Section-P-810126-D-20140815-T-160755 have-treatment-effect equal-to 43.

Menu-For-Tumor-Excision-Section-P-810126-D-20140815-T-160755 have-distance-of-tumor-from-closest-margin equal-to 45.

Menu-For-Tumor-Excision-Section-P-810126-D-20140815-T-160755 have-margin-description equal-to 'good'.

The application suggested performing molecular genetics studies, crucial in sorting out risk levels for patient.

Every immunohistochemical-studies-section must have-cd-117 a cd-117.

Every immunohistochemical-studies-section must have-dog-1 a dog-1.

Every immunohistochemical-studies-section must have-molecular-genetic-studies a molecular-genetic-studies.

Something is a molecular-genetic-studies if-and-only-if-it is either Wide-Type, Kit-Exon-9, Kit-Exon-11, Pdgfra-V-842, Pdgfra-Non-V-842, Other or Not-Performed.

Another important indicator in oncologist treatment is so called mitotic index.

Pathological-Report-Section-P-810126-D-20140815-T-160539 have-mitotic-index equal-to 2.

The subsequent steps of the medical consultation described here can be seen in detail in the picture. All the new fields were reasoned based on rules in the ontology.

After additional diagnostics, the doctor decided that the patient should be operated. There were also other possible options. You can see that those options correspond to a dropdown list field displayed in the form.

Something is a surgeon-decision if-and-only-if-it is either Operable, No-Operable, Post-Primary-Site-Excision or Pending.
Surgeon-Consultation-Section-P-810126-D-20140815-T-161335 have-surgeon-decision Operable.

The doctor was forwarded to a form where he was reminded to ask for patient’s agreement for surgery and then he could input the data about the outcome of the operation. The operation was successful.
Remember that our patient had a tumor in colon, 8 mm large. He also had mitotic index equal to 2. How can application help the physician decide on the form of further medical treatment?
A medical expert with basic knowledge of CNL could write simple rules below. Our application could process the rules and display the appropriate recommendation to the doctor.

If a patient has-mitotic-index lower-or-equal-to 5 and the patient has-tumor-size-t-1 lower-or-equal-to 20 and the patient has-tumor-localization-t-1 Colon then the patient has-risk-group Risk-Group-1-C.

Risk-Group-1-C has-description equal-to 'very low (group 1); percentage of patients due to relapse = 0'.

Risk-Group-1-C is a risk-group-1.

Every risk-group-1 is a risk-group.

If a patient has-risk-group a risk-group and the risk-group is a risk-group-1 then the patient has-recommendation Ref-1-Adj-No.

Ref-1-Adj-No has-description equal-to 'Adjuvant therapy should not be considered when the risk of relapse is low [http://www.ncbi.nlm.nih.gov/pubmed/19286368]'

Now we can see it is no surprise that the operation was successful. It is stunning to realize that we are able to write medical guidelines in a language that is understood both by humans and by the computer.

---

If you want to learn more about Ontorion™ Server, visit this link.

*) Ontorion™ Server, a very powerful and scalable solution that recognize and extract relevant items of information hidden in plain text or reports. It completely changes access to unstructured data and boots the efficiency and scalability of all processes involving the management,distribution, access and analysis of large amounts of textual content. With this toolset user can manage free-text taxonomies in Microsoft Excel and export them to the semantic high-performance server. Server side components allows to collect data directly from Internet (social-media, webpages, blogs, etc) and other text-free unstructured sources.