Experiencer Negation

Standard negation asserts the absence of a finding (e.g., 'no chest pain'). Experiencer negation asserts the presence of a finding but redirects the subject. The condition is real, but the experiencer is a family member or contact. This requires the system to resolve semantic roles, not just polarity.

• Standard Negation: 'Patient denies cough.' (Cough is absent).
• Experiencer Negation: 'Mother has a cough.' (Cough is present, but not for the patient).

Detection relies on identifying familial entities and possessive constructions that shift the subject away from the patient. Common triggers include kinship terms and relational phrases.

• Triggers: 'mother', 'father', 'sister', 'brother', 'son', 'daughter', 'spouse', 'family history of'.
• Possessives: 'His father's diabetes' or 'Her son was diagnosed with'.
• Context: The ConText algorithm explicitly extends negation detection to include an 'experiencer' dimension to handle these cases.

Failure to detect experiencer negation causes false positives in the patient's structured data. This leads to incorrect problem lists, skewed risk scores, and flawed clinical decision support.

• Phenotyping Errors: A patient may be incorrectly flagged for a genetic condition based on a relative's diagnosis.
• Billing Inaccuracies: False attribution of major comorbidities can lead to incorrect risk adjustment and reimbursement.
• Patient Safety: Clinicians may make treatment decisions based on a condition the patient does not actually have.

Modern systems move beyond simple keyword matching to contextual embedding models that understand the syntactic relationship between the medical entity and the subject.

• Rule-Based (ConText): Uses regular expressions to capture 'family' keywords and link them to clinical findings within a defined scope.
• Deep Learning (NegBERT/SpanBERT): Fine-tuned transformer models learn to classify the experiencer status of a clinical entity based on the full sentence context, distinguishing 'patient's mother with cancer' from 'patient with cancer'.

Experiencer negation is the first step in structured family history extraction. Once a finding is identified as belonging to a relative, a secondary process links the diagnosis to the specific family member and establishes the genetic relationship.

• Step 1: Detect 'father' and 'myocardial infarction'.
• Step 2: Assert that the myocardial infarction is negated for the patient.
• Step 3: Create a separate family history observation linking 'father' to 'myocardial infarction'.

Ambiguity arises when the same lexical trigger can indicate the patient or a relative. The phrase 'family history of diabetes' is a clear experiencer shift, but 'mother denies chest pain' requires the system to understand that the mother is the patient in this specific sentence context.

• Ambiguous: 'She has a family history of breast cancer.' (She = patient, cancer = relative).
• Clear: 'Her mother had breast cancer.' (Patient's mother = experiencer).
• Resolution: Requires co-reference resolution to determine if 'she' refers to the patient or the mother.

The categorical label assigned to each clinical named entity that defines its relationship to the patient. Beyond simple present/absent polarity, assertion status includes the critical 'associated with someone else' category. Modern deep learning models like NegBERT and clinical transformers fine-tune on assertion classification tasks to distinguish:

• Present in patient
• Absent
• Present in family member
• Historical
• Hypothetical

The broader clinical NLP task that experiencer negation supports. Family history extraction identifies not just that a condition belongs to a relative, but which relative and the specific diagnosis. This requires:

• Named entity recognition for family members
• Relation extraction to link relative to condition
• Negation detection to rule out conditions
• Temporal reasoning for age of onset Accurate experiencer negation is the gatekeeper preventing family history from contaminating the patient's active problem list.

The computational process of determining exactly which tokens in a sentence are semantically inverted by an experiencer trigger. For example, in 'mother with breast cancer and diabetes', the scope must extend across the conjunction to negate both conditions for the patient. Scope resolution errors cause false negatives where family conditions leak into patient records. Modern span-level classifiers using contextual embeddings have significantly improved boundary detection over fixed-window approaches.

The primary clinical risk that experiencer negation mitigates. When a system fails to detect that 'father died of myocardial infarction' refers to a family member, the myocardial infarction diagnosis is incorrectly attributed to the patient. This contaminates:

• Problem lists
• Clinical decision support inputs
• Quality measure calculations
• Research cohort selection Precision metrics for experiencer detection directly measure the system's ability to prevent these dangerous false attributions.