Glossary for Cohort Definition Using OHDSI ATLAS SOP

Term	Definition
OHDSI ATLAS	A web-based open-source application for designing and executing analyses on OMOP CDM data; includes tools for cohort definition, concept set management, characterization, and population-level effect estimation.
OMOP CDM	The Observational Medical Outcomes Partnership Common Data Model that standardizes healthcare data (e.g., EHR, claims) into common tables (e.g., CONDITION_OCCURRENCE, DRUG_EXPOSURE) with standard vocabularies; CHoRUS B2AI data is stored in OMOP CDM format to enable use of ATLAS and OHDSI tools.
Cohort	A set of persons who satisfy one or more criteria for a duration of time.
Cohort Definition	Specifies how patients enter (start) the cohort and how/when they exit, effectively implementing a computable phenotype representing a disease, exposure, or outcome.
Cohort Entry Event (Index Event)	A specific event that qualifies a person to enter the cohort and sets the cohort start date (index date), e.g., first occurrence of a diagnosis or initiation of a drug; defined in ATLAS under Cohort Entry Events.
Concept Set	A defined list of standard concepts used to represent a clinical idea; attached to cohort criteria to specify which codes/events count (e.g., “Myocardial Infarction” concept set attached to a Condition Occurrence criterion).
Initial Event Criteria	The primary criteria in Cohort Entry Events that a record must meet to qualify as a cohort entry, including the event domain (condition, drug, procedure, etc.), attached concept set(s), and event-level attributes (e.g., first occurrence, date limits).
Inclusion Criteria	Additional conditions that refine the cohort after identifying initial events; in ATLAS, these filter/qualify cohort entries further (e.g., requiring a prior condition, a lab result, or excluding patients) and are defined in the Inclusion Criteria section.
Censoring Event	An event that ends cohort membership early (e.g., death or initiation of a competing treatment); specified in ATLAS in the Cohort Exit section via Add Censoring Event.
Event Persistence (Cohort Exit Strategy)	The rule defining how long a cohort entry remains active in the absence of censoring; ATLAS supports options such as end of continuous observation, fixed duration relative to the index, or end of continuous drug exposure.
Cohort Era / Collapse	The approach to handling multiple cohort entries per person by merging episodes that occur within a specified collapse gap into a single cohort era (e.g., 180-day gap merges episodes within 180 days), producing eras with a start date (first entry) and end date (per exit rules/last episode).
Continuous Observation Window	A requirement that a person has continuous observation (OBSERVATION_PERIOD) for at least X days before and/or Y days after the index event date to ensure sufficient baseline and/or follow-up data completeness.
OBSERVATION_PERIOD	OMOP CDM table defining when a person is considered observable in the dataset; ATLAS uses it to enforce “continuous observation” constraints and to trim events/eras to valid observation time.
Qualifying Event	An event instance that satisfies the cohort entry criteria (and, after applying inclusion rules, remains eligible to define a cohort start).
Cohort Episode / Cohort Entry	A single interval of cohort membership for a person (start = index date; end = persistence/end/censoring); multiple episodes can exist per person unless limited/collapsed.
Index Start Date / Index End Date	The start/end timestamps of the entry event used as the anchor for relative time windows; which one you use depends on event type (e.g., visit has duration, condition often does not).
Baseline (Lookback) Window	Time before the index date used to assess prior history, eligibility, and covariates (commonly enforced via “continuous observation before” and/or prior-event criteria).
Follow-up (Time-at-Risk) Window	Time after the index date during which outcomes are measured; implemented via persistence and/or continuous observation after index and/or censoring.
Washout Period	A predefined lookback window used to confirm “no prior occurrence/exposure” (incident/new-user designs), typically operationalized as 0 occurrences of a concept set before index.
Incidence vs Prevalence	Incidence cohorts aim for first/initial occurrences (often “earliest event per person” + washout); prevalence cohorts include anyone with evidence at any time (often “all events” or “any occurrence”).
CONDITION_OCCURRENCE vs CONDITION_ERA	CONDITION_OCCURRENCE records individual diagnosis events; CONDITION_ERA groups proximate condition occurrences into episodes; selecting one vs the other changes counts and temporal logic.
DRUG_EXPOSURE vs DRUG_ERA	DRUG_EXPOSURE records individual dispensing/administrations; DRUG_ERA collapses exposures into continuous episodes with allowable gaps; impacts persistence and on-treatment definitions.
VISIT_OCCURRENCE	OMOP CDM table representing encounters; used to restrict cohorts to inpatient/ER/outpatient settings and to tie events to a specific encounter.
Visit Type / Visit Concept	Standard concepts describing visit categories (e.g., inpatient, ER); used in visit-based criteria to define care setting context.
Domain Alignment	Ensuring the concept set’s domain matches the criterion (e.g., Condition concepts in a Condition Occurrence criterion) to prevent silent undercounting/misclassification.
Standard vs Non-standard Concepts	Standard concepts are intended for analysis; non-standard are source-coded and should be mapped—mixing them can cause undercounts depending on ETL and ATLAS settings.
Attrition	Stepwise reduction in counts as entry restrictions and inclusion rules are applied; used for QA and interpretation of “why counts drop.”
Cohort Diagnostics	ATLAS diagnostics outputs (e.g., incidence over time, inclusion rule impact, code distribution) used to validate face validity and detect logic/data issues.
Censoring vs Exclusion	Exclusion prevents entry (or removes qualifying events) based on criteria; censoring allows entry but truncates follow-up when a censoring event occurs (e.g., death).
Competing Risk	An event (often death) that precludes observing the primary outcome; handled by censoring and/or analysis strategy (outside ATLAS cohort logic).
Calendar Time Restriction	Restricting index dates to a study period (e.g., after 2020-01-01) to align with data completeness and protocol-defined eras.
Concept Set Expression Options	Common concept-set semantics: include descendants, include mapped concepts, exclude concepts; these materially affect sensitivity/specificity and must be documented.
"Person count" vs "Record count"	Person count = unique persons with ≥1 matching record; record count = total event rows; concept set counts can differ from cohort counts due to added temporal/logic constraints.
Add Initial Event	Defines the primary entry event type and its concept set. Keep minimal; reflects the phenotype "index."
Add attribute	Adds constraints to an event (e.g., occurrence count, timing, demographics). Prefer readable attributes; avoid over-nesting.
Delete Criteria	Removes a criterion/attribute/group. Document why removed if it changes phenotype intent.
Restrict initial events to: having all/any/at least/at most …	Applies boolean/count logic to additional criteria tied to entry events. Use for essential context (e.g., inpatient setting).
Add criteria to group	Adds another criterion under a boolean group. Use groups to express (A AND B) OR (C AND D).
with at least / at most / exactly N occurrences of	Frequency constraint on matching events. "Exactly 0" is key for washout/no-history logic.
using all occurrences of	Evaluate the criterion over all matching events for the person (not just one). Important when counting occurrences across time.
event starts between … days Before/After index start date	Relative time window for event start anchored to index start. Controls temporal alignment; common for baseline/follow-up windows.
event ends between … days Before/After index start date.	Relative time window for event end anchored to index start
All days Before / All days After	No bound in that direction. Useful for “overlap” logic with visits: start ≤ index and end ≥ index.
The index date refers to the event from the Cohort Entry criteria	Relative timing anchors are computed from the entry event. Ensure you understand which event is the anchor in nested groups.
Restrict to the same visit occurrence	Forces matching events to share the same visit_occurrence_id as the index event. Prevents linking unrelated events across visits; can reduce counts sharply.
Allow events from outside observation period	Allows events to qualify even if outside OBSERVATION_PERIOD. Usually keep OFF; enabling can create implausible person-time.
with continuous observation of at least X days before and Y days after event index date	Requires uninterrupted observation around index date. Baseline/follow-up completeness constraint; major driver of attrition.
Limit initial events to: all/earliest/latest events per person	Controls whether a person can enter multiple times and which index is used. Incident cohorts usually “earliest”; episodic studies often “all.”
New inclusion criteria	Adds an inclusion rule block applied after entry events. Use for modular eligibility logic + transparent attrition reporting.
Limit qualifying events to: all/earliest/latest events per person	Final event-per-person selection after inclusion rules. Use to enforce one index per person after complex inclusion logic.
Add Censoring Event	Adds an event that truncates cohort follow-up. Use death here for follow-up censoring (not via concept sets).
Event Persistence: Event will persist until …	Default cohort end rule in absence of censoring. Choose to match estimand/time-at-risk (fixed window vs on-treatment vs observation end).
Fixed duration relative to initial event	Cohort end = index start/end + offset. Offset from start = uniform duration; from end = variable duration for eras/visits.
Event date to offset from: start date / end date	Selects anchor for fixed duration. End-date anchor matters for visit length/drug era duration.
Number of days offset	Length of persistence window. Document rationale (e.g., 30-day risk window).
No censoring events selected	No early truncation beyond persistence/observation trimming. Verify this is intentional; otherwise add death/other censoring.
Specify era collapse gap size	Merges episodes separated by ≤ gap into one cohort era. Large gaps reduce episode counts; affects incidence/recurrence interpretation.