Standard/ontology selection

Luckily, in many different clinical data domains, public standards have already been created that can be used for this work. Some of these commonly available standards are listed below:

• Medication data – MeSH, MEdDRA, IDMP, RxNorm
• Diagnoses – ICD-9-CM, ICD-10-CM, SNOMED, NCIT
• Lab data – LOINC
• Genomic data – Gene Ontology

Using CENtree, SciBite’s award-winning ontology management platform, organizations can manage these standards. CENtree enables users to participate in a democratized management process to ensure the vocabularies are up to date. Importantly, CENtree enables organizations to deploy these standards downstream to any tool, including SciBite’s Named Entity Recognition Engine, TERMite.

By integrating CENtree and TERMite together, organizations can realize the full value of flexibly updating their standards in CENtree. Let’s look at a specific data type and how organizations can normalize their data using CENtree and TERMite.

Unlocking real-world patient data within clinical notes

Many clinical notes are difficult to interpret, even while they have a lot of information in them. Some of that information is valuable, and some of that information is not as valuable. In any solution to unlock this data, it is imperative that it is possible to sift through all the information to find what is actually of value. In this example, I asked chatGPT to write example clinical notes for fake patients. In this example, let’s say I was a researcher that wanted to analyze the effects of different drugs in early-stage breast cancer treatment for different patient populations. I’m trying to identify patients that I would want to analyze for my study.

Here is a sample note chatGPT generated for a fake patient:

Patient 1:

• Name: Sarah Thompson
• Age: 45
• Gender: Female
• Medical History: No significant medical history
• Family History: No known family history. Clinical Presentation: Sarah presents with a painless lump in her left breast that she discovered during a routine self-examination. Possible breast cancer. No other symptoms reported.
• Physical Examination:
  • General: Well-nourished and in good overall health.
  • Breasts: Left breast reveals a firm, non-mobile, palpable lump measuring approximately 2 cm in diameter. No nipple discharge or skin changes noted. Right breast examination is unremarkable.
  • Lymph Nodes: No palpable axillary lymphadenopathy. Diagnostic Workup:
  • Mammogram: Shows an irregular mass in the upper outer quadrant of the left breast.
  • Ultrasound: Confirms the presence of a solid mass with irregular borders.
  • Core Needle Biopsy: Reveals invasive ductal carcinoma, estrogen receptor (ER) positive, progesterone receptor (PR) positive, and HER2/neu negative.
  • Stage: T1N0M0 breast cancer
  • Surgical Consultation: Referral for lumpectomy or mastectomy with sentinel lymph node biopsy.
  • Medical Oncology Consultation: Discussion of the need for adjuvant therapy, including endocrine therapy with Kentadex and consideration for radiation therapy.
  • Genetic Counseling: Evaluation for genetic testing due to the patient’s age and lack of family history.

From scanning the note above, I can see that the patient has stage 1 breast cancer. The note doesn’t explicitly say that, but as a human, I can see that mentioned via different synonyms and context clues. The question becomes the following: How can I get a machine to quickly and easily recognize the piece of information that I can see after quick analysis?

In this instance, we are focused on getting diagnosis/indication and medication data from our notes. Using TERMite, as shown by the screenshot below, we can see the important information, TERMite was able to recognize from those standards managed in CENtree:

Creating machine-readable annotations to use downstream

TERMite generates machine-readable annotations to use downstream. Therefore, organizations can feed these machine readable annotations to your NLP and AI teams so they can run a more rigorous analysis of the data for research purposes. Additionally, that researcher we started with can immediately see if it makes sense to analyze this patient in more depth for their study. From this simple example, we can see how TERMite can help standardize information within a chart for use in downstream data analysis.

To reiterate earlier points in this article, even if a synonym isn’t included in the standard initially, with CENtree you can easily augment these vocabularies as needed. Case in point, the phrase “T1N0M0 breast cancer” wasn’t included in our NCIT VOCab as a synonym to Stage I Breast Cancer since it isn’t a common way of saying exactly that. I added it in a matter of minutes:

Figure 2: Description automatically generated with medium confidence

In the above screenshot, you can see how I added an exact synonym to this specific term. After deploying this change to TERMite, it was immediately picked up for use in entity recognition.

In the next blog…

In the later installments of this series, we will see specifically how organizations can take these machine-readable annotations provided by TERMite and powered by CENtree in order to leverage them to enable researchers to do their job quickly and effectively. This efficiency gain not only helps internally in healthcare organizations, but also fosters collaboration to share RWD with pharmaceutical organizations for the betterment of patients everywhere.