By Jonathan D. Gold, MD MHA MSc FAMIA FHIMSS
April 15, 2024
Challenge yourself:
What are the essential components necessary for linking events temporally in healthcare documentation?
How can spatiality influence event associations, according to the article's findings?
Why is event linking crucial in reconciling conflicting accounts of patient history across multiple records?
What are the key markers used to determine if separate references address the same event in healthcare records?
Connecting an Element to Temporality
As stated previously, events require both an element (finding, problem, procedure, order, observable, etc.) and temporality. Additionally, spatiality may play a role in the event. (See Figure 1). Consider an element such as “exposure to radiation”, and a temporality of “for two weeks”. The spatiality of “at a distance of 5 kilometers from Chernobyl one month after its explosion” is quite different from “at a distance of 5000 kilometers from Chernobyl one month after its explosion”. In specific situations spatiality matters. On the other hand, associating events like diabetes mellitus type 2 and hypertension to spatiality might not be possible or particularly important in most cases.
NLP must both discover the correct element and associate it with the appropriate temporality (and in certain instances like in the “exposure to radiation” example, associate it with spatiality). To do this, NLP must parse a phrase or sentence, identify elements (as targets) and temporal phrases (as modifiers), use a rules-based logic to connect these appropriately, and generate a summary list of event associations (and their generated dates) and an errata list of unassociated elements and temporal phrases.
Temporal Normalization
Temporal Alignment
With ubiquitous electronic medical documentation and multiple provider interpretations of the patient’s history documented in numerous entries and records, conflicting accounts often arise as to when an event occurred and how long it spanned. Identifying when the same event is being addressed in multiple records (“event linking”) becomes a critical step: Is this the same event? A reconciliation methodology must be built that addresses 1) what type of an event occurred, 2) how precise was the time or date assigned to the event, and 3) how trustworthy was the source that reported when the event occurred.[1]
Multiple sources (both internal and external to an organization) may provide accounts around a single event with inconsistent dates and time frames. Additionally, it is not always clear whether similar diagnoses captured by different providers with different degrees of granularity refer to the same event (“Otitis media, left” and “Suppurative left otitis media”), or whether events with the same description that occur around the same period but are captured by different sources are actually the same event or separate instances (e.g., recurrent myocardial infarctions within days of each other or repeat qualitative urine tests within one day of each other.)
Through “temporal alignment”, we attempt to determine which events may have multiple versions and reconcile the versions to extrapolate the closest approximation of when an event occurred. Alignment considers the type of event (non-recurrent, one-time, chronic, acute, etc.), the source for the information, and the highest degree of precision for calculating the date of occurrence within an organization. The three critical components—event type, precision, and source veracity[2]—are crucial in resolving differences across records when building a patient’s health timeline. All events are added to the Patient Event Master List.
Is This the Same Event?
While an event may appear in only one record, often for important events, multiple entries or records from other sites will contain information or reference the same occurrence. Determining which events have multiple versions requires an identification process—event linking—followed by a reconciliation protocol to give the closest approximation of when an event occurred.
A combination of key markers suggests that separate references address the same event. Within each category listed below, options are ranked from highest to lowest likelihood of association. A score is assigned for events.
If element is a “Problem”:
Significance (Criticality)
Near death experience (NDE), Apparent life-threatening event (ALTE)
Organ failure, limb loss
Critical or serious condition
Temporal Classification
One-time event[6]
Chronic
Acute or Finite Duration Event
While attempting to link the same events across records, confounders make this task difficult. For instance, several discrete events can occur within a short period that need to be recognized as distinct rather than a single occurrence (e.g., repeat urinalyses or recurrent ventricular arrhythmias).
Figure 2 captures the three key components for event linking and temporal normalization--event type, precision, and source veracity. These will each be addressed in the next two blog chapters.
Conclusion
The intricate process of event linking and temporal normalization in healthcare data management requires access to numerous data points. The necessity for reconciling conflicting accounts of patient history through temporal alignment is highlighted, emphasizing the significance of event type, precision, and source veracity. Utilizing a systematic approach to determine the similarity of events across records is crucial for establishing a coherent patient health timeline. Through meticulous event linking and temporal alignment, healthcare professionals can attain a clearer understanding of patients' medical histories, aiding in more informed decision-making and improved patient care outcomes.
[1] Dr. Eduard Hovy, from DARPA (the Defense Advanced Research Projects Agency), leads two projects that might support how we determine when an event, that has been given multiple associated dates, most likely occurred—Active Interpretation of Disparate Alternatives (AIDA) and Knowledge-directed Artificial Intelligence Reasoning Over Schemas (KAIROS).
[2] Event type categorizes events in multiple ways—non-recurring, recurring, finite duration, chronic events, etc. The Precision Matrix scores the degree to which the temporality associated with events is considered exact. The Source Veracity Score adjusts that determination based upon the author and source of the date appearing in the record.
[3] May require extrapolation from patient age or time/date of entry.
[4] The convention for this has not yet been determined.
[5] For example, where did a procedure occur? Where was a lab ordered and where was the result sent?
[6] One-time events never recur. For example, problems like “state after amputation below knee, right”, “congenital bicuspid aortic valve”; and procedures like “appendectomy”, “total abdominal hysterectomy”.
תגובות