The complexity and variability of biomedical language pose a challenge in fields dealing with biomedical data. A vast array of terms, referring to various entities, including living organisms, biochemicals, drugs, biological pathways, diseases, and disease symptoms in patients, is utilized in biomedical research and closely associated healthcare industries. The intricacy inherent in this domain necessitates both a structured representation of knowledge and the use of a standardized language to describe these entities. Medical ontology serves as a valuable tool for fulfilling this crucial purpose.

What is Ontology?

Originally a philosophical study of existence, ontology, in the context of information science, refers to a formal representation of knowledge or information within a specific domain. Ontology provides a structured framework for organizing entities, their properties, and relationships. To illustrate, let’s consider the ontology of a country. As shown below, it encompasses entities such as states, cities, towns, and villages, along with their hierarchy and relationships. Interestingly, a similar ontology focused on geographical information serves as the foundation for the widely used Geographic Information System (GIS).

By extension, medical data is also highly complex—perhaps even more so—due to the vast diversity of biological entities and their interdependent relationships and interactions. This inherent complexity has prompted the development of numerous types of ontologies, which are designed to maintain consistency and accuracy in medical data, records, and research. At present, there are hundreds of medical ontologies, and you can find many of them by following this link.

There are numerous advantages to utilizing ontologies. For biomedical researchers, the use of medical ontologies can assist them in annotating and categorizing data with standardized terminology. This, in turn, leads to clear communication and makes it easier to identify relationships, patterns, and other elements within the data. For medical practitioners, ontologies facilitate the exchange of patient information among healthcare providers and entities, ensuring that data sent and received by different systems are interpreted consistently. This not only enhances the accuracy and consistency of medical information but also promotes interoperability between different healthcare systems. Furthermore, it supports the development of intelligent systems for medical diagnosis and treatment.

Apart from the benefits in research and practice, medical ontology also plays a crucial role in text mining. There is a constant influx of biomedical text from various sources every day. However, due to the intricacies of biomedical concepts and the inherent ambiguity in natural language, precise data extraction from large and heterogeneous data sources is not feasible without a system of standardization. Medical ontology helps mitigate this ambiguity and variability in text data. Furthermore, medical ontology is also pivotal in the Semantic Web, where web data are represented through structured ontological representations, allowing interpretation by both humans and machines. In short, the use of medical ontology leads to more intelligent search, enhanced data integration, and improved knowledge discovery across a wide range of medical and healthcare-related resources.

An overview of commonly used medical ontologies

There are a number of ontologies that are particularly helpful when building text analysis solutions for the medical field. In what follows, we provide an overview of some of the more commonly used ontologies.

MeSH

The Medical Subject Headings (MeSH) dictionary is a hierarchical collection of medical terms describing various conditions, diseases, and symptoms. It was created by the US National Library of Medicine (NLM) at the National Institutes of Health (NIH) in the United States.

MeSH is utilized for indexing and searching biomedical information found in national medical databases, including MEDLINE/PubMed. Many biomedical articles are now indexed with MeSH terms, either manually or automatically by computer, enabling users to retrieve records on a specific subject with accuracy and efficiency.

ICD-10

The International Statistical Classification of Diseases and Related Health Problems (ICD) is a standardized system for coding medical diagnoses and procedures. It contains a description of all known diseases and injuries across languages and allows quick access to disease information, including symptoms, treatment, and mortality rates.

MedDRA

The Medical Dictionary for Regulatory Activities (MedDRA) is an ontology for classifying and sharing regulatory information. This medical ontology is typically used by pharmaceutical and biotechnology companies for FDA pharmacovigilance reporting on drugs and therapies. MedDRA is used for registration, documentation, and safety monitoring of medical products before and after a product has been authorized for sale. This includes pharmaceuticals vaccines and drug-device combination products.

MedDRA contains over 70,000 standard terms for various signs, symptoms, diseases, diagnoses, and indications, and is updated twice a year. It consists of a 5-level taxonomy, arranging symptoms and diseases from very specific to very general. For example, consider the condition known as “chest pain” (stenocardia). The 5-level hierarchical taxonomy describing this is summarized by the following animation.

MedDRA is dictated by the FDA in the United States. The FDA requires the use of MedDRA in the submission of safety reports and other regulatory documents related to drug development and approval. It’s widely recognized as a key standard for the classification and analysis of medical data in the pharmaceutical industry.

RxNorm

RxNorm is a comprehensive database containing over 100,000 medications and chemicals. It serves as a standardized nomenclature system that establishes connections between brand names and generic synonyms for clinical drugs. This mapping enables medical care providers to effectively manage the diverse vocabulary associated with drug and pharmaceutical products across various databases.

Moreover, RxNorm’s primary aim is to facilitate efficient and accurate communication of drug-related information among computer systems. It is diligently maintained and updated on a weekly basis by the National Library of Medicine (NLM) and encompasses all medications for humans available on the US market.

RxNorm serves as an exemplary instance of a highly intricate ontology, as demonstrated in the example below that showcases the connections for Zyrtec® (cetirizine hydrochloride), an over-the-counter antihistamine. This not only provides the standard drug names but also highlights the relationships between different drug entities.

Other sources for ontologies

A few additional sources of coding systems and ontologies include:

• Current Procedural Terminology (CPT) codes: an ontology used to report procedures and diagnostic services to healthcare providers, insurance companies, and accredited institutions.
• Systemized Nomenclature of Medicine – Clinical Terms (SNOMED-CT): an ontology used mainly by healthcare providers to share information in a more standardized manner.
• Diagnosis-Related Group (DRG) codes: a system used to classify hospital cases into approximately 500 groups for the purpose of payment.
• Gene Ontology Consortium: a set of frameworks for modeling complex biological systems and concepts.

How ontologies are used in PolyAnalyst™

PolyAnalyst™ utilizes ontologies to search for and extract pertinent information. Within PolyAnalyst™, ontologies are imported as distinct “semantic dictionaries,” which are then used by PDL (Pattern Definition Language) to extract text patterns that correspond to various medical entities and their relationships as defined within those semantic dictionaries. In addition to the existing ontologies, users have the ability to create their own ontologies.

PolyAnalyst can import and start working with any kinds of ontologies. Some of these ontologies can be used free of charge, others require the user to obtain a license from the ontology owner. Currently, the following ontologies are provided along with PolyAnalyst™ (out-of-the box):

• Human Disease Ontology: provides a framework for organizing and integrating information related to diseases in humans.
• Human Phenotype Ontology: provides a standardized vocabulary of phenotypic abnormalities encountered in human disease.
• MeSH: covers a wide range of topics related to health and medicine.
• RXNorm: contains information on medication names and their active ingredients.
• WordNet: a broad semantic ontology that attempts to build a lexical database of English nouns, verbs, adjectives, and the relationships between them.

Other ontologies can be imported by the user (possibly, after acquiring the necessary license).

The Dictionary Manager in PolyAnalyst™ is the most direct way to access, edit, and view any imported ontologies in PolyAnalyst™. For example, you can import and view MeSH dictionary in PolyAnalyst™, searching for terms such as “CVA” to view all the associated information.

These ontologies can also be accessed within the project nodes. Some key nodes that utilize these types of ontologies include Entity Extraction and Taxonomy. Within these nodes, users have the option to select which ontology to import. Subsequently, they can craft a query to identify semantically related terms, such as synonyms, hyponyms, or hypernyms, or any other semantic relations, as defined within that particular ontology.

The example below illustrates how a query searching for medical conditions within the “influenza and pneumonia” category, as per ICD-10, can be simplified by leveraging the PDL functions integrated into PolyAnalyst™ in conjunction with the MeSH ontology. By employing the “hyponym” function, as shown below, it becomes possible to capture all the subtypes of influenza and pneumonia without the need to list them individually.

Lastly, one of the challenges of using ontologies is the issue of ontology cleaning, which involves removing or correcting errors, inconsistencies and irrelevant or outdated information in the ontology. This is particularly important in specialized domains, such as chemistry or toxicology, where the ontology may contain a large number of concepts and entities that are not relevant to the domain or may cause confusion or misinterpretation (i.e., “milk” or “coffee” being present within a chemical substance ontology). Thankfully, the task of modifying ontologies is quite straight-forward within the Dictionaries module of PolyAnalyst™.

Ontologies and beyond

Ontologies are powerful tools in information science, with potential that far exceeds the medical domain of the topics discussed here. Additional resources on using dictionaries in PolyAnalyst™ and how to import and utilize ontologies and specialized dictionaries can be found in our support and training materials.

Tip 2: Employ a clean layout

Information cannot be placed just anywhere on the dashboard, nor can sections of the display be sized simply to fit the available space. Items that relate to one another should usually be positioned close to one another. — Stephen Few, author of Information Dashboard Design

Dashboard content must be organized in a way that reflects the nature of the information being displayed. It not only needs to create efficiency for the audience, but it should also provide a manageable foundation for the report itself. Building modules is one method that can help you to create a basic structure and skeleton for your design, so that you can achieve an organized data display with effective alignment and consistency. We recommend keeping these DO’s and DON’Ts in mind when assessing your dashboard layout.

A. Plan ahead

Don’t approach your dashboard design without a plan, or you might get a dashboard like this (above).

Before designing the report dashboard, preparation and planning for all the visual components will reduce time wasted on trial and error. That means one should decide on what information should take priority ahead of time. And even then, the flow could be compromised when trying to cram too much information into one dashboard.

We recommend sketching a basic structure or “skeleton” for the design with some possible modules layouts, as seen here.

Once the basic structure is determined, identify what kind of visual components to include in the dashboard. This process will help you judge how effectively the key information is displayed and test the drawn-out framework. When everything looks good on paper or in theory, now it’s time to evaluate effectiveness.

Don’t overload your dashboard with too many components. It occupies a lot of visual space and ends up distracting your audiences from obtaining key information at a glance.

Do choose your key components wisely. Try to use no more than 5-7 different data components to create a view.

Don’t add too many irrelevant accessory elements to the dashboard. For example, data tables have lots of viewing options—like “Zebra” color (shown above), column header background color, aligners, and toolbars—but not all of these effects should be implemented in a single component.

Remove formatting elements that add visual noise. If something feels unnecessary, it probably is and it’s not recommended to keep it in the display. Simplifying the visual aspect of the component makes it more powerful than an element that is over-embellished.

Don’t use an over abundance of colors. Like graphical embellishments, color can detract from the overall look and confuse the audience.

Do simplify the use of color. Fortifying a dashboard with a well-tested color scheme that contains fewer colors actually brings balance to the dashboard.

Example: Sans Serif font.

Example: Serif font.

Don’t use multiple font types in different component headers.

Do use consistent fonts with the same size and the same type.

Don’t use bulky font types in the legend that overwhelm the key information.

Do use a consistent font type between the legend title and legend contents. Keep the font size and font type consistent across your label text.

Don’t use inconsistent label styles between the y-axis and x-axis of your component.

Do use a consistent font type, font size, and font color between the y-axis and x-axis.

Don’t stagger the alignment of the header and content that appears in the same column.  Instead, do use left-justified headers for columns containing left-justified content, and vice versa. Right-aligned numeric columns are easier for users to compare numbers, whereas a left alignment is more natural for text columns. The new design concept for data tables in PolyAnalyst™ web reports allows users easy access to these text alignment features.

Feedback surveys are a tried-and-true way that businesses gain insights into their customers’ opinions. And while a lot of information can be obtained from structured survey questions, there are even more insights hidden in the text responses customers provide in surveys, social media and other online company and product review channels. Nowadays, a majority of customer feedback is in unstructured formats, so it’s important to implement analysis techniques that can help us uncover these insights quickly and effectively.

So here we present a quick guide to some of the data-driven analysis techniques you can use in your text analysis projects. But first, let’s clarify what we actually mean by “data-driven.”

As you can see, the main themes in the unstructured text column are starting to emerge. Some of these terms we could guess very easily, such as toothpaste, tooth, and mouth. And because these terms are merely identifying the physical items and areas the toothpaste affects, we could essentially discount these themes as far as their relevance to our analysis. On the other hand, we see taste, cap, and texture are also prevalent themes in the text. This gives us some insight into the high-level issues that customers cited regarding the toothpaste products.

Auto taxonomy from keywords

Auto taxonomy only

When the results from keyword extraction are used, the system constructs a fairly extensive taxonomy structure of the product review themes. For example, we still see the themes we discovered earlier, like taste, cap, and texture, but we also see more details behind these concepts, such as sensitive teeth and mint flavor.

Using another automated taxonomy method, without the keywords, the system generates a different overall structure for the text data. We see some similarities in the themes compared to the first automated taxonomy the system constructed. For example, we see topics such as cap/tube and flavor/mint appear in both category structures. However, in this second taxonomy, we also see themes like morning/night and good job. When we drill down to the underlying records, we can see that consumers usually provide information about what time of day they use the toothpaste and whether the toothpaste product did a ‘good job’ of cleaning their teeth. However, it will be up to the analyst to decide where these types of topics should appear in the hierarchical classification structure.

The Precursor to Analyst-Driven Analysis

Data-driven techniques are clearly an important method of exploring unstructured text data. As a precursor to analyst-driven approaches, data-driven analysis equips users with a useful starting point so that they can impart a more meaningful structure onto text, improving the overall analysis results. Through this article, we hope you have gained a general overview of some of the unsupervised data-driven analysis techniques you can apply in your own analysis work.

As the saying goes, “mo’ data, mo’ problems.” Well, that’s not quite the saying and that’s not quite what we mean, but it certainly is true that large data comes with some problems that need to be solved. There are two ways that data can be “big.” Traditionally, Big Data is considered large literally by its physical size in terms of the memory space it consumes. In this article, we’re considering another type of “big” data, which we like to call Wide Data. Wide Data is unstacked tabular data that consists of large amounts of columns or variables, usually in the hundreds and above. Wide Data and Big Data are often overlapping but they are not necessarily the same: for example, Wide Data can be shallow, with relatively small number of rows – but large number of columns. Wide Data is common and comes with its own set of problems that need to be addressed in any analytical setting. Next, let’s discuss those problems.

Dimensionality Reduction in PolyAnalyst™

PolyAnalyst™ contains several built-in methods for performing dimensionality reduction. We’ll highlight a couple of these below.

The first method is Data Simplification. This process scans data to identity columns that are nearly uniform. A column with almost no variation is very uninformative and could be dropped entirely. Additionally, Data Simplification compares the columns with one another to see if there are any that are nearly identical. These columns can be triaged so that only one remains in the dataset to avoid unnecessary duplication. In addition to Data Simplification, Correlation Analysis can be performed to discover correlated variables. While not identical, strongly correlated variables serve a common purpose and we could consider cutting down to one in this case as well.

The second method is Factor Analysis, which includes techniques such as principal component analysis. The goal of this method is to find the vectors along which most of the variance occurs in the data, reorient the data along those vectors, and discard the vectors in which little variance occurs. This is best understood with an animation.

In this example, we have measured and plotted only two variables—height and weight. We can see the direction along which most of the variance occurs. This is the principal component. Now we rotate our data such that this direction is our horizontal axis and the orthogonal direction is the vertical axis. This destroys the semantic meaning of these axes of course, but this not a big deal. Where they were once semantically width and height, they now hold no such meaning. We will be able to regain the meaning after performing the analysis in the smaller number of dimensions and rotating the results back to the original axes. Now, if we would like to shrink the size of our data, we can throw out the vertical axis, which exhibits little variance. Thus, we have eliminated an entire variable but preserved the majority of the useful information, which simplifies further analysis.

Stay tuned…

Wide, bulky data is fairly common in the world, and although it comes with some problems, these can be managed with the techniques described here, such as Factor Analysis and Data Simplification. Once our data is manageable, PolyAnalyst has a wide range of modeling solutions. Stay tuned for more discussions about data analysis!