Temporal Constructs: Building the Foundations of Time

By Jonathan D. Gold, MD MHA MSc FAMIA FHIMSS

March 18, 2024

Challenge yourself:

1. Why is temporality essential for deriving meaningful associations in health data interpretation?

2. How does natural language processing aid in codifying temporal narratives from medical records?

3. Why are unlinked dates considered more reliable for events occurring much earlier compared to tethered dates?

Temporality fills a critical gap in the interpretation of health data, providing agency for deriving meaningful associations between observables, treatments and outcomes. Text fields appearing throughout the medical record, contain essential narratives describing health events. Natural language processing (NLP) can highlight and codify these narratives, including their temporal qualities and be used to construct a health timeline. Temporality together with entities (findings, problems, procedures, orders, etc.) are requisite components for events. Events may also be linked to spatiality.

Event = Entity + Temporality (+/- Spatiality)

The basic word unit for temporality is the ‘concept’ (the primary, default phrase). Precise, synonymous phrases, serve as alternate ways for expressing the specific concept. Concepts should be associated with standard medical codes like SNOMED CT to the closest degree of accuracy. Additionally, concepts may be mapped to other concepts (utilizing concept-to-concept mapping). By allowing concept-to-concept mapping, a pre-coordinated concept (like “21 days ago”) can be associated with a concept that defines the term as a mathematical formula (in this case, “date stamp of entry minus 21 days”) and its upper and lower limits (“plus/minus one-half day”).

Temporal concepts provide building blocks to derive or specify as definite a timeframe as possible. As discussed later, temporal objects cover many different aspects related to time—from the level of certainty to numbers to units of measurements. The approach to adding appropriate concepts is to include both clear cut temporal phrases (like, “January 7, 1952” and “12:53pm”), components of phrases (e.g., “minutes”, “weeks”, “times per day”, “4”), and supporting idioms (“probably”, “currently”, “next”).

By mapping the concepts to SNOMED CT, it becomes possible to group these into temporal “parts of speech”. This is of particular importance in natural language processing when determining whether a phrase has the correct, time-associated components to be interpreted and plotted on a timeline.

Due to its temporal qualities, an event may have an uncertain beginning or conclusion, may be ongoing, have a relationship with other events, have a sequence, may be momentary or have a span, have parts, have a cause, have a result, or have a recurrency pattern.

Temporality may describe slices of an event or events (e.g., “bleeding noted during the first trimester of each pregnancy”) or the event in its entirety. It may designate a period between or across events; it may occur in the past, present, future, or conditionally. Temporal objects may be nested within additional temporal relationships; they may be assigned an extrinsic measure (time-date) or a relation interval (age, age at occurrence, event span, time between events).

Recording events, such as in medical records, introduces metadata related to the capture of the event (when, where, author, data source, etc.). Dates may be tethered or unlinked. Tethered dates are those in which a historic, current, future or conditional event is deduced from the date of record entry (for example, “four days of low-grade fever”) or a different event; unlinked dates for an event are specific (i.e., time/date, date, month/year, or year) and the date of documenting the event is inconsequential to understanding its temporality (for example, “Status Post MI, 1985”). Unless an unlinked date is fully specified (mm/dd/yyyy), a method is used to convert the partially defined date to a specific, derived date.

Unlinked dates for events occurring much earlier may be more reliable than tethered ones since these are given ‘absolute’ temporal values and do not require calculation to determine when events took place.

Glossary	Value Type	Example
Entity	Findings, problems, procedures, orders, observables, etc.	“Serum sodium: 135mg/dl”    “Diabetic neuropathy”.     “Tonsillectomy”
Relation interval	(e.g., age, age at occurrence, event span)	Age at Occurrence:    “First menstrual period at age 12.” Event Span:    “Migraine headaches recurring for 9 years from age 16.”
Nested temporality	A temporal statement embedded in an additional temporal relationship (e.g., duration and sequence)	“Smoked 2 packs/day for 20 years prior to lung cancer dx.” (Example of a measured unit (“2pk/d”) nested in a duration (“20y”) nested in a sequence (“prior to”).)
Tethered date	Links the date of record entry (metadata) to a past, present, future, or conditional event. Derived date calculated from metadata and relation interval.	Past:    “Cough for 5 days”. “TB 20 years ago” Present:    “Dx = Pneumonia” Future:    “Colonoscopy to be performed in the next 2 years.” Conditional:    “If bloody stools occur in next 2 weeks, consider polypectomy.”
Unlinked date	Specific date assigned to event (time/date, date, month/year, year)	“Left lower Pneumonia, Pneumococcal on 7/22/2011”

Conclusion

Not only does temporality enable the type of phenotypic associations needed for precision medicine and longitudinal records, but it is vital for deriving meaningful links between medical history, treatment and health outcomes, and for developing capabilities such as advanced decision support systems. We can use building block concepts to construct a means for interpreting when events occurred and how long they lasted and to map them on a timeline.