Metroline Step: Analyse data semantics
- Sander de Ridder
- Ana Konrad
- Hannah Neikes
status: in development
Short description
Understanding and clearly defining the meaning (semantics) of (meta)data is an important preparation for creating the semantic model, as well as for data collection via, for example, electronic case report forms (eCRFs). In this step, the aim is to ensure you gain a clear and unambiguous understanding of the (meta)data. The step provides guidance for both existing data and data that is to be collected.
To illustrate the issue, consider the example where you receive a dataset with a variable called “date”. Without clearly defined semantics, it is unclear whether this means “date of data collection”, “admission date”, “date of birth”, or something else. This must be resolved before you can start with the semantic (meta)data model. In the How to section of this page we provide instructions to achieve this.
Thus, the outcome of this step is a set of data elements (variables) with clear and unambiguous semantics, known as a codebook. For metadata, the outcome is a […?]. Note that finding machine-actionable items from ontologies for the data elements is not yet part of this step, but is described in Create or reuse a semantic (meta)data model.
Why is this step important
Several of the Metroline steps that follow rely on being familiar with your data. For example, in order to create or reuse your semantic (meta)data model, it is crucial to understand the meaning and relationships of variables.
While performing this step, keep your FAIRification goals in mind. If you have a clear idea of your FAIRification goals, it might be easier to define what elements should be present in your (meta)data and how these elements should be represented.
For example, in a dataset a variable to collect sex-related data might be called ‘sex’. If the semantics of such variable are not provided or not analysed, it would be unclear if it means ‘biological sex at birth', ‘phenotypic sex’, or ‘gender’. These issues have to be solved before you start with the semantic (meta)data model.
How to
This “How to” described 5 steps to take in order to make your (meta)data clear and unambiguous. Since projects vary in their levels of (meta)data semantics, we created two diagrams to help you navigate which steps are relevant for you:
Use the first - green diagram for analysing data semantics.
Use the second - yellow diagram for analysing metadata semantics.
By completing these steps, you will end up with clear and unambiguous semantics for your (meta)data.
To ensure this, it is essential to analyse various aspects of the data elements and variables involved, such as:
The definition/description of data elements. For example, a variable called “sex” could refer to “Biological sex” or “Administrative gender”.
Value ranges for data elements. For example, in system A, sex allows for male and female, while in system B, sex also allows for intersex.
Relationship between data elements. For example, the “sex” variable is one attribute of “patient”, which may imply that the semantics of this “sex” variable is “sex of patient”.
To see which steps are relevant for your (meta)data, please follow the diagrams below.
For analysing data semantics:
For analysing metadata semantics:
For easier understanding, we will follow the example dataset containing patient information with the following metadata:
Metadata Field | Value |
|---|---|
Dataset Name | Health Data |
Date of Upload | 01/02/2023 |
Keywords | BP, HR, Conditions |
Creator | Dr. Smith |
Description | Patient health data including BP and HR |
Format | CSV |
Source | Hospital A |
Rights | Open |
In this example, we are working with existing metadata. According to the diagram, we should start with Step 1 - Compile information.
Step 1 - Compile information
Compile all the information of data elements, data values, and data structure. Examine the data in whatever format and structure it is available. This step helps to identify inconsistencies, ambiguities, and errors in the data.
a) For existing (meta)data: Locate all relevant sources in which the (meta)data is stored. Compile information about the following:
Which variables are present in the (meta)data (i.e. in the eCRFs)?
What are the value ranges for each variable?
In our example, we are working on FAIRifying an already existing metadata of a dataset. Let’s compile and examine the information we have.
Value | Metadata field / Variable | Description of the field | Value range |
|---|---|---|---|
Health Data | Dataset Name | The name of the dataset. | Text |
01/02/2023 | Date of Upload | The date on which the dataset was uploaded | Date values in the format MM/DD/YYYY |
BP, HR, Conditions | Keywords | Terms that describe the main topics of the dataset | Text |
Dr. Smith | Creator | The person or organisation that created the dataset | Text, in our example title and last name |
Patient health data including BP and HR | Description | A brief summary of the dataset | Text |
CSV | Format | A file format of the dataset | Text, in our example a short string indicating the file format |
Hospital A | Source | The origin of the dataset | Text, in our case the name of the institution |
Open | Rights | The usage rights or licence of the dataset | Text |
b) In case you are aiming to collect FAIR (meta)data from the start:
Which data elements/variables are you planning to collect? For this, the competency questions (QCs) might provide some guidance.
If possible, determine the value range for each data element (e.g. for ‘biological sex at birth', values could be ‘male’, ‘female’; while for 'age’, the value range might be 0-110).
Existing data (1a)
Existing metadata (1a)
New data (1b)
New metadata (1b)
Step 2 - Check for an existing standard/codebook
a) For existing (meta)data: check if it comes with a codebook or metadata standard. In case it does and it is clear, you can use it for your (meta)data and this step is done.
If the codebook is not helpful, you should contact the owner of the data and get the semantics cleared up, so you don’t misinterpret the data. If you see you still need to do additional work in order to make the data clearer, follow the steps below.
b) For new (meta)data: check if there is a codebook or metadata standard you can use. A domain expert (for data) or FAIR data steward/semantic expert (for metadata) can help you find out if and where a codebook or standard might be available.
In case there is a codebook or standard, you can use it. If there is no codebook or standard available, proceed to step 3.
If you find a codebook or metadata standard that fits partially, use the elements that are included and follow the steps below for the remaining elements.
Health-RI, together with domain representatives, will be aiming to develop domain-specific national data standards in the future.
You can find more about metadata standards and ontologies at the following link: https://howtofair.dk/links-additional-reading/#more-on-metadata-standards-and-ontologies-
Step 3 - Check (meta)data semantics
Check the data semantics. Is the meaning of the data elements clear and unambiguous? For data elements with ambiguous meaning, try to improve their definition. For this, it might help to look into the intended value range of a variable - is the exact range known and is it clear enough?
In the example of collecting data on a patient’s 'sex', it might be unclear if it means ‘biological sex at birth' or ‘gender’. In another example, 'age' of a subject can be expressed in years, but in some cases, such as studies with small children, could also be expressed in months. It should therefore be clearly stated if the value range for age should be expressed in years or months.
In the below spreadsheet we can see what the issues are with our current metadata and suggested improvements in order to make the meaning of them clearer.
This recipe in the FAIR cookbook gives some additional guidance on specifying the semantics of elements of your data.
Metadata Field | Original Value | Issue | New variable description | New Value Range | New Value |
|---|---|---|---|---|---|
Dataset Name | Health Data | Generic and not descriptive. | The name of the dataset. | Text | Patient Health Records 2023 |
Date of Upload | 01/02/2023 | Ambiguous format | Date when the dataset was uploaded, in ISO 8601 format (YYYY-MM-DD). | Date in ISO 8601 format (YYYY-MM-DD) | 2023-01-02 |
Keywords | BP, HR, Conditions | Abbreviations used without context. | Keywords describing the main topics covered by the dataset. | Text | Blood Pressure, Heart Rate, Hypertension |
Creator | Dr. Smith | Generic name without additional identifying information. | Full name and affiliation of the dataset creator, as well as ORCID. | Text and ORCID identifier | Dr. John Smith, Hospital A ORCID: 0001-0002-3456-7890 |
Description | Patient health data including BP and HR | Lacks detail. | Extended description providing context and details about the dataset. | Text | Detailed patient health records including measurements of blood pressure (BP) and heart rate (HR), along with diagnosed medical conditions and prescribed medications. |
Format | CSV | Broad category, can be more detailed. | Data format and version. | Text | CSV, version 1.0 |
Source | Hospital A | Lacks detail, too generic. | Specific department and institution where the data was sourced. | Text and ROR identifier | Hospital A, Department of Cardiology, https://ror.org/example |
Rights | Open | Too broad. | Licensing terms specifying the rights for data usage. | URL to CC License |
Step 4 - Check relationships
Compile information about the relationships between (meta)data elements. For example, if the dataset is in a relational database, the relational schema provides information about the dataset structure, the types involved (the field names), cardinality, etc.
For example, the variable 'biological sex at birth' in a dataset is an attribute of ‘patient’.
In our metadata example above, the Creator (Dr. John Smith) is an employee of the Source (Hospital A, Department of Cardiology) of the dataset.
Step 5 - Check for FAIR features
In addition, check whether the data already contains FAIR features, such as persistent unique identifiers for data elements (for more information, see pre-FAIR assessment).
In our example above, the ORCID of the creator is a unique persistent identifier (F1) for a person.
After having performed the relevant parts of this Metroline step, proceed to the next: Metroline Step: Create or reuse a semantic (meta)data model
Expertise requirements for this step
Below are experts that may need to be involved, as described in Metroline Step: Build the Team.
Semantic data / modelling specialists: Can help understanding the data’s structure and ambiguity.
Domain expert: Can help understanding the data’s elements and their potential ambiguity.
FAIR data steward: Can help identifying FAIR features in the dataset, or data to be collected, as well as ambiguity in the data elements.
Practical examples from the community
Examples of how this step is applied in a project (link to demonstrator projects).
Training
Relevant training will be added in the future if available.
Suggestions
Visit our How to contribute page for information on how to get in touch if you have any suggestions about this page.