Variables, concept sets, and anchors

Concept sets

The Concept sets section lets you create reusable sets of medical codes. A concept set groups codes from standardized terminologies (ICD-10, SNOMED CT, LOINC, ATC…) that describe the same clinical concept — for example, all codes for a diagnosis of “sepsis,” or all antibiotic prescription codes. These terminologies were introduced in the article on terminologies.

These sets are then used in the selection criteria (previous article) and in the definition of variables (below).

Creating a concept set

Three methods are available to create a concept set:

• Browse catalog — opens a catalog of concept set dictionaries. A dictionary groups predefined, validated, and ready-to-use concept sets. You select the relevant sets for your study and import them in one click.
• Import from URL — imports a concept set from an external link, useful for accessing dictionaries not yet listed in the built-in catalog
• Create manually — creates an empty set to which you add concepts one by one

Managing concepts

Once a concept set is created, click on it to open it. You can then:

• View the list of included concepts, with for each: the name, vocabulary (ICD-10, LOINC…), code, and identifier
• Include or exclude each concept from the set (an excluded concept will not be used in queries)
• Add new concepts (by searching on ATHENA or ATLAS)
• Remove existing concepts

Each concept set also lets you define:

• Unit — selected from a dropdown list of UCUM (Unified Code for Units of Measure) codes, the international standard for units of measurement. For example: mg/dL, mmol/L, bpm.
• Retained min and max values — define a plausible value range to automatically exclude outliers during data extraction.

Temporal anchors

Temporal anchors are the reference dates around which variables are collected. For example, if you want to measure creatinine “within 24 hours of admission,” the admission date is the temporal anchor.

Anchors are defined in the Anchors tab of the Variables section.

Anchor types

Several anchor types are available:

• Hospital admission — the hospital admission date. You specify whether it is the first, last, or each admission. You can also filter by hospital name.
• Hospital discharge — the hospital discharge date (first, last, or each discharge). You can filter by hospital name.
• Care unit admission — the date of entry into a care unit. You specify the occurrence (first, last, or each) and can filter by unit name.
• Care unit discharge — the date of discharge from a care unit (first, last, or each). You can filter by unit name.
• Concept set event — a date linked to a concept set, for example the date of the first sepsis diagnosis. You select the relevant concept set and the occurrence (first, last, or each).
• Free text — for describing an anchor that doesn’t match any of the above types

Each anchor has a name (e.g., “ICU admission”, “Sepsis diagnosis”) and optional details.

Variables

The Variables tab in the same section lets you define each data point to extract. A variable corresponds to a measurement, result, or characteristic that you want to obtain for each individual in your cohort. The concepts of concept, temporal anchor, collection window, and aggregation function were introduced in detail in the article on defining variables.

Creating a variable

Click Add variable to open the creation form. You fill in:

• Name — the display name of the variable (e.g., “Creatinine at admission”)
• Identifier — the technical name used in exports and generated scripts (e.g., creatinine_admission)
• Description — a free-form description of the variable
• Data type — continuous, categorical, binary, ordinal, date, or text

Variable source

Two types of variables are available:

• Concept set — the variable is extracted from a previously defined concept set. You select the set from a searchable dropdown.
• Computed variable — the variable is derived from structural visit data or patient demographics: age, sex, hospitalization duration, or unit stay duration.

Temporal anchor and collection window

Each variable is linked to a temporal anchor. This is the reference date from which the collection window is calculated.

The collection window specifies the interval around the anchor during which data is searched:

• Start — offset relative to the anchor (a negative number means “before the anchor”)
• End — offset relative to the anchor (can also be negative for a window entirely before the anchor)
• Time unit — hours, days, weeks, months, or years

For example, for “the highest creatinine within 24 hours of admission”:

• Anchor: admission
• Start: 0
• End: 24
• Unit: hours

Aggregation function

When multiple values exist within the collection window, you choose how to summarize them:

• First / Last — the first or last value chronologically
• Maximum / Minimum — the highest or lowest value
• Mean / Median — the average or median of the values
• Presence — indicates whether at least one value exists (yes/no)
• Duration — the duration between the first and last value
• Count — the number of values found
• Sum — the sum of the values

Table and timeline

Defined variables are displayed in a summary table that shows for each variable: the name, unit, temporal anchor, collection window, and aggregation function. You can edit or delete each variable from this table.

A timeline view is also available. It visually represents the temporal anchors and collection windows of each variable as horizontal bars. This is a good way to check at a glance that all windows are consistent.

Key takeaways