Experiencer Detection

The core function of experiencer detection is to resolve the semantic role of the experiencer in a sentence. It distinguishes between statements like 'patient is homeless' (patient as experiencer) and 'patient's mother is homeless' (family member as experiencer). This prevents the erroneous attribution of a family member's social risk to the patient's structured record, which would corrupt downstream analytics and trigger inappropriate referrals.

This task relies heavily on dependency parsing to map the grammatical relationships between words. The system analyzes the syntactic tree to identify the subject of a risk-related predicate. For example, in the phrase 'brother lost his job,' the parser identifies 'brother' as the nominal subject of 'lost,' enabling the model to correctly classify the experiencer as a non-patient family member rather than the patient themselves.

Experiencer detection is often implemented as a downstream component of a relation extraction pipeline. After a named entity recognition model identifies a social risk mention, a relation classifier links it to a person entity. The system then assigns a label—such as 'Experiencer=Patient' or 'Experiencer=Family_Member'—to the extracted relation triple, creating a structured, queryable fact for the patient's social history.

Accurate experiencer detection requires robust coreference resolution to track entities across sentences. If a note states, 'The patient's daughter is struggling with addiction. She has been unemployed for six months,' the system must resolve that 'She' refers to the 'daughter' and not the patient. Failure to resolve this coreference chain leads to a false positive SDOH flag on the patient's record.

General-domain language models often fail at this nuanced task. High-performance experiencer detection requires fine-tuning clinical BERT models on annotated corpora that explicitly label experiencer roles. These specialized models learn the lexical and syntactic patterns unique to clinical narratives, such as distinguishing a patient's own history from a family health history section where risks are routinely documented for relatives.

Experiencer detection is a critical gatekeeper for SDOH phenotyping accuracy. Without it, a patient could be incorrectly flagged for a housing intervention that their sibling needs, wasting care coordinator resources and eroding trust. By ensuring only patient-experienced risks populate structured data, this task directly improves the positive predictive value of computable phenotypes used for population health equity analytics.

A contextual analysis technique that distinguishes whether a social risk factor is present or absent in clinical text. This is a critical prerequisite to experiencer detection, as a negated mention ('patient denies homelessness') must be ruled out before determining who the experiencer is. Negation scope often interacts with experiencer scope, requiring joint modeling to correctly interpret sentences like 'patient's brother is not employed.'

An NLP task that determines the chronological status of a social risk mention—whether a housing crisis is a current, historical, or future concern. Experiencer detection and temporality are often jointly resolved, as the experiencer may change over time. For example, 'patient reports past food insecurity in childhood' requires both temporal classification (historical) and experiencer attribution (patient as a child).

An NLP subtask that identifies and categorizes specific mentions of social risk factors, such as 'homeless' or 'unemployed', within free-text clinical documents. Experiencer detection is a downstream relation extraction task that links these identified entities to the correct person. A robust NER system must first tag the risk factor before experiencer resolution can assign it to the patient or a family member.

The process of grounding ambiguous medical mentions to unique identifiers in standardized knowledge bases. For SDOH, this involves mapping extracted social risk mentions and their experiencers to ICD-10-CM Z-Codes (Z55-Z65) or SNOMED CT concepts. Accurate experiencer detection ensures the linked code is attributed to the correct individual in the patient's record for billing and population health analytics.

A detailed instruction manual for human annotators that defines the scope, entity types, and edge cases for labeling social determinants of health. Experiencer annotation requires explicit rules for handling ambiguous pronouns, family member references, and multi-person scenarios. High-quality guidelines ensure inter-annotator agreement (IAA) above 0.85 Cohen's Kappa, which is essential for creating a reliable gold standard corpus.

A quality assurance workflow where a clinical reviewer audits and corrects AI-extracted SDOH data, typically focusing on low-confidence predictions. Experiencer misattribution is a high-priority error type in this workflow, as incorrectly assigning a family member's housing instability to the patient can trigger inappropriate referrals and waste care coordinator resources.