Factuality Status

The clinical finding is explicitly stated as present or true for the patient at the time of documentation.

• Polarity: Positive
• Example: 'The patient has pneumonia.'
• Clinical significance: These findings populate the active problem list and drive clinical decision support alerts.
• Detection: Absence of negation or uncertainty cues within the assertion scope.

The clinical finding is explicitly ruled out or stated as absent in the patient.

• Polarity: Negative
• Key triggers: 'no', 'denies', 'without evidence of', 'negative for'
• Example: 'The patient denies chest pain.'
• Critical safety implication: Failure to detect negation leads to false attribution of diseases, triggering unnecessary alerts and compromising data quality.

The finding is expressed with doubt, speculation, or hedging, indicating the clinician lacks full commitment to its presence.

• Key triggers: 'possible', 'suspected', 'cannot rule out', 'suggestive of'
• Example: 'Possible pulmonary embolism.'
• Epistemic modality: Reflects the clinician's degree of certainty rather than the ontological status of the condition.
• Downstream use: Uncertain findings often route to differential diagnosis lists or trigger further diagnostic workup.

The finding occurred in the patient's past medical history but is not active at the time of documentation.

• Key triggers: 'history of', 'prior', 'resolved', 'no longer present'
• Example: 'History of myocardial infarction in 2019.'
• Distinction from negation: The condition is affirmed as having occurred, but temporally contextualized as inactive.
• ConText algorithm: Explicitly handles historical context as a separate dimension from simple negation.

The finding is present but applies to a family member or contact rather than the patient themselves.

• Key triggers: 'mother', 'father', 'sibling', 'family history of'
• Example: 'Patient's father had colon cancer at 62.'
• Risk of misattribution: Without experiencer detection, family history conditions are incorrectly assigned to the patient's active problem list.
• ConText extension: Added experiencer as a distinct contextual dimension beyond negation and temporality.

The finding is discussed in a conditional or future-oriented context, not as an actual occurrence.

• Key triggers: 'if', 'should the patient develop', 'in the event of', 'risk for'
• Example: 'If the patient develops fever, initiate antibiotics.'
• ConText classification: Handled as a distinct context alongside negation, historical, and experiencer dimensions.
• Clinical relevance: Prevents prophylactic or contingency statements from being extracted as actual events.

The computational task of identifying linguistic cues that semantically reverse the existence of a clinical finding. Negation detection transforms a statement like 'no chest pain' into a structured absence of the condition.

• Core Mechanism: Identifies negation cues like 'no', 'denies', 'without'
• Critical Metric: False Negative Rate—missing negation falsely attributes disease to a patient
• Primary Algorithm: Negex uses regular expressions to define the scope of negation triggers

The NLP task of classifying statements that express doubt or hedging regarding a medical condition. Uncertainty detection distinguishes 'possible pneumonia' from confirmed diagnoses.

• Linguistic Targets: Hedging cues like 'suggestive of', 'cannot rule out', 'suspected'
• Semantic Foundation: Classifies epistemic modality—the degree of certainty in a proposition
• Benchmark Corpus: BioScope provides annotated data for training speculation detection models

An extension of Negex that detects not only negation but also historical conditions, hypothetical statements, and the experiencer of a finding. ConText uses lexical triggers to determine if a condition is recent, historical, or attributed to someone other than the patient.

• Temporal Negation: Identifies 'resolved' or 'no longer present' findings
• Experiencer Negation: Prevents attributing a family member's condition to the patient
• Rule-Based Design: Relies on trigger terms and syntactic scope rules rather than machine learning

A transformer-based language model fine-tuned on the BioScope corpus for token-level negation and speculation detection. Unlike rule-based systems, NegBERT leverages contextual embeddings to disambiguate pseudo-negation patterns.

• Architecture: BERT model adapted for biomedical text
• Advantage: Resolves double negation ('not unlikely') and pseudo-negation ('not only pneumonia')
• Output: Token-level classification labels for negation and speculation scope

The assignment of a categorical label to each clinical named entity indicating its veridical context. Assertion status is the structured output of factuality detection pipelines.

• Standard Classes: Present, Absent, Uncertain, Historical, Family History
• Evaluation Metric: Negation Precision measures correct absent classifications vs. false alarms
• Downstream Impact: Incorrect assertion status corrupts patient timeline construction and clinical decision support

The end-to-end process of identifying a negation cue, determining its syntactic scope, and applying semantic inversion to the target clinical entity. Scope defines exactly which tokens are negated within a sentence.

• Challenge: 'No chest pain or shortness of breath'—does negation cover both symptoms?
• Approaches: Rule-based windowing vs. span-level classification with deep learning
• Confidence Scoring: Probabilistic outputs enable thresholding for human-in-the-loop review of ambiguous cases