False Negative Rate

The failure of a system to correctly link a negation cue to a target clinical entity when they are separated by many tokens or clause boundaries. Rule-based systems like Negex often fail when the scope window is too short, while even transformer models can lose attention over long sequences.

• Example: 'The patient denies any history of chest pain, shortness of breath, or the severe, crushing substernal pressure that would indicate an acute myocardial infarction.'
• A system may correctly negate 'chest pain' but fail to apply negation to 'acute myocardial infarction' due to the intervening descriptive clause.

A linguistic construction where a negation trigger word appears but does not semantically reverse a clinical condition. Systems that rely solely on trigger word lists without contextual disambiguation are highly susceptible to these false negatives, as they may learn to ignore legitimate negation cues that resemble pseudo-negation patterns.

• Example: 'Not only does the patient have diabetes, but also hypertension.'
• A model overtrained to avoid pseudo-negation false positives may fail to detect true negation in similar syntactic structures, such as 'The patient does not have diabetes.'

Clinical statements that semantically rule out a condition without using standard negation triggers like 'no' or 'denies'. These require world knowledge and reasoning beyond pattern matching, making them a primary failure mode for both rule-based and purely distributional models.

• Example: 'The chest x-ray was unremarkable for pneumonia.'
• Example: 'The patient's troponin levels ruled out a myocardial infarction.'
• The absence of explicit negation vocabulary causes systems to incorrectly assert the presence of 'pneumonia' or 'myocardial infarction'.

Uncertainty regarding whether a single negation cue applies to all elements in a conjunctive or disjunctive list. Systems must correctly resolve the syntactic structure to determine if negation distributes across coordinated clinical entities.

• Example: 'The patient denies fever and chills and reports a mild cough.'
• A false negative occurs if the system fails to apply 'denies' to 'chills' due to the ambiguous coordination structure, incorrectly attributing chills to the patient.

The failure to recognize negation cues or clinical entities that were absent or underrepresented in the training data. This is especially prevalent in specialized subdomains or when models trained on general biomedical text are applied to niche clinical specialties.

• Example: In oncology, 'The margins were clear of malignant cells' indicates negation of malignancy at the surgical margin.
• A model unfamiliar with the phrase 'clear of' as a negation trigger will incorrectly assert the presence of malignant cells at the margin.

The misclassification of a finding that is negated only in the present but was true historically, or vice versa. Systems that perform binary polarity classification without modeling temporal negation or historical negation will generate false negatives for currently inactive conditions.

• Example: 'The patient's pneumonia has resolved.'
• A system that does not recognize 'resolved' as a temporal negation cue will assert that the patient currently has pneumonia, representing a false negative for the negated present state.

The evaluation metric measuring the proportion of correctly identified negated findings out of all findings flagged as negated by the system. While the False Negative Rate focuses on missed negations, precision focuses on false alarms. A system with high precision but a high False Negative Rate is overly conservative, failing to detect many true negations to avoid making mistakes. The two metrics must be balanced against the clinical cost of each error type.

The specific span of tokens within a sentence whose meaning is inverted by a negation cue. A False Negative often occurs when the system correctly identifies a negation trigger like 'no' but fails to extend the semantic inversion to the correct clinical entity. For example, in 'no evidence of acute chest pain or shortness of breath', the scope must cover both findings; a truncated scope leads to a missed negation and a False Negative.

A linguistic construction containing a negation trigger word that does not actually negate a clinical condition. Phrases like 'not only pneumonia but also...' or 'did not respond to antibiotics but improved with steroids' contain negation words used in non-negating contexts. Failure to disambiguate pseudo-negation leads to False Positives, but overly aggressive rules to avoid them can cause the system to ignore real negation cues, directly increasing the False Negative Rate.

A pattern where two negation elements cancel each other to form an affirmative statement, such as 'not unlikely' or 'cannot rule out the absence of'. These constructions require logical resolution to avoid misclassification. A naive system might classify 'not unlikely' as negated, but a sophisticated system that fails to resolve the double negation will miss the affirmed finding, contributing to the False Negative Rate for uncertainty detection.

A probabilistic output associated with a negation or uncertainty prediction that allows downstream clinical systems to threshold results or prioritize ambiguous cases for human review. A high False Negative Rate can often be mitigated by lowering the confidence threshold for flagging a finding as negated, but this increases False Positives. The confidence score distribution reveals the model's certainty and is essential for tuning the operational point on the precision-recall curve.

A classification label assigned to a clinical named entity indicating whether the concept is present, absent, or uncertain in the patient record. This is the core output of factuality detection systems. A False Negative in negation detection means an entity that should have an 'absent' assertion status is incorrectly labeled as 'present', leading to a ghost condition being added to the patient's problem list—a critical patient safety error.