Glossary of Features

Information Extraction Capabilities

Simple concepts: Concepts that are represented in text as a single word or simple phrase.

Complex concepts: Concepts that are represented in text as multiple phrases and clauses and that may also require the synthesis of information from multiple and possibly disjoint sentences.

Fine Details: Information that is very specific.

Closely Related Concepts: Concepts that differ only in minute details.

Overlapping Concepts: Concepts that share some details of meaning but differ in other details.

Large Number of Concepts: More than hundreds; potentially millions.

Vital Signs: Both vital sign and value.

Lab Values: Both lab type and value.

Medication Values: Medication type, dosage, route and timing.

Diagnoses and Findings: Multiple diagnoses and findings per document with the location in the text of each noted, and each normalized to canonical values (codes).

Procedures (Simple Narrative): Procedures that can be named or described in a single word or simple phrase; may be multiple procedures per document with location in the text of each noted and each normalized to canonical values (codes).

Procedures (Complex Narrative): Procedures that are described in a multi-clause and/or multi-sentence (possibly disjoint) narrative that must be analyzed to identify the procedure and normalize it to a canonical value (code); may be multiple and/or overlapping procedure descriptions in the text.

E/M Level: Extraction and normalization of the elements making up the Evaluation and Management codes, accessing external knowledge to determine the values of elements and calculating the E/M Level.

Review of Systems: Extraction and normalization of the elements making up a Review of Systems (ROS).

Physical Examination: Extraction and normalization of the elements making up a Physical Examination.

Medical Risk: The medical risk posed by an injury/disease/treatment constellation as determined by extraction and normalization of the injury/disease/ treatment details and accessing external knowledge to determine the risk level.

Physician Instructions: What, when, where, why, with whom, etc.

Certainty: How certain the speaker is of a statement.

Source: Who is the source of a statement: physician, patient, PA, resident, etc.

Severity: How severe a condition is.

History: Is a condition/event current, pending, past personal history, past family history, etc.

Extent: The physical extent of a condition.

Change: Condition stable, improved, worse, etc.

Conditionality: Interpretation of statements with context sensitivity to conditioning, contra-positives etc.; e.g. "had the patient not shown signs of pneumonia, I would have ordered a stress test immediately." => possible pneumonia; no stress test order.

Categorization and Routing Capabilities

Very Many Categories: More than a few hundred; possibly millions.

Based on Canonical Data: Is the categorization based on an automated (statistical) analysis of examples (validated or not), or is it based on canonical definitions.

Requires Minimal Examples: Can the system be built and tested with a minimal number of hand coded/categorized examples or are many hand coded/categorized examples needed? In many cases there is not a large sample of reliable, hand coded/categorized with sufficient detail to be used for statistical learning techniques or it may be more labor intensive to create the training set than to specify the criteria.

Statistical Training: Automated training based on example documents. Neural Net approaches can also be classified here. LifeCode® is shown as partial because it combines a variety of techniques, not just statistical.

Linguistic Analysis Capabilities

Segment Based Analysis: Analysis that is context sensitive to the section or segment of a document in which a statement is made; e.g. history, exam, impressions etc.

Grammatical Analysis: Analysis based on sentence structure.

Discourse Analysis: Analysis that spans multiple sentences and is based on the structural relation of concepts across the sentences.

Pragmatic Analysis: Analysis that depends on information external to the document in order to make inferences about items that are only partially specified in the document and which presume some world knowledge to understand.

Other Capabilities and Features

Self-Explaining Results: Results that have information about where information came from and why it was interpreted as it was.

Identifies Need for Review: Directives indicating that a human coder or abstractor should investigate a particular result or lack of result.

Identifies Reason for Review: Explanation of why a review is needed and what it should target.

Easily Updated: Ability to change system behavior based on published specifications; e.g. code changes, billing guidelines etc.

Easy for Quality Control: Ability to understand why the system performs as it does and ability to make principled changes to specific behaviors.

Completeness Checking: Ability to determine if a clinical document is complete or if required information is missing.

Addenda Matching: Ability to match addenda to the original documents.

Integrated Knowledge Sources: Ability to use externally specified knowledge sources to guide the NLP process.

Handles Local Task Variations: Ability to customize multiple behaviors according to practice variants that are local to the document source.

Specialty Specific Processing: Ability to customize those system behaviors that vary across medical specialties.

Medicare Cross Checks: Ability to assess whether a procedure description meets Medicare guidelines for reimbursement; e.g. discussion of the appropriate number and type of aspects of an EKG.