By Jonathan D. Gold, MD MHA MSc FAMIA FHIMSS
April 8, 2024
Challenge yourself:
How does temporality vary between report dates and unstructured text records regarding event dates?
Explain the rationale behind including both a specific point in time and a range when capturing event timing.
What strategy is proposed to address rounding errors in approximate event dates?
Why are brief time spans essential in medical records despite not often being plotted on a health timeline?
Generally, temporality may either be presented as highly defined or an approximation. If an exact date (or time) is given by a trusted source (for example the date on a radiological study), there is no need for including a range of when the event may have occurred. More commonly though, text records present an estimate as to when the event occurred.[1]
To capture the timing of an event, we typically will need to include both the specific point in time referenced by the text and a range (lower to upper limit) that might also contain the event if the source is only approximating when it occurred. Precision varies between measurement units, such that describing an event in terms of days is a more sensitive measurement than weeks, weeks more than months, etc.
For comparative purposes, “14 days ago” and “two weeks ago” reference the same point in time; however, the source is assumed to approximate when the event occurred—the “rounding error” for weeks is greater than that for days. To address this potential rounding error, we can take the exact time or date deduced from the source and add a range that is ± ½-measurement unit (i.e., the measurement).[2] In the above example, the range for “14 days” equals 13.5 - 14.5 days ago, whereas the range for “two weeks” equals 1½ - 2½ weeks (i.e., 10.5 – 17.5 days)[3] ago. This allows for both an exact date and a range of dates to be determined using the time/date stamp on the entry.
The following are general rules about this approach and some examples:
Input equals an event phrase (element + temporality, e.g., “sore throat” + “beginning 3 days ago”), so output = element and date with time range in days (± ½-time unit) (i.e., sore throat start date = dse minus 2.5-3.5 days).
Determine date as though exact (e.g., 4 weeks ago = Time-Date Stamp of Entry minus 28 days) ± ½-time unit (i.e., using this example, dse minus 24.5-31.5 days).
For a period of time (“between 2-4 weeks ago”), median equals 21 days, lower limit 31.5 days (i.e., 28 days [4 weeks] plus 3.5 days), upper limit equals 10.5 days (i.e., 14 days [2 weeks] minus 3.5 days). In the example "2-4 weeks ago", point in time = dse - 21d, a lower delimiter = dse - 31.5d, and an upper delimiter = dse - 10.5d.
If the value is the fraction "½", like "½ day", "½ week", etc., then use ± ½ of the fraction as the upper and lower bounds for the time unit (e.g., "½ year ago" = dse minus 183 days ± 91 days[4] [¼ year] which equals dse minus 92-274 days).
Lower delimiter values are usually not prior to the patient’s date of birth. Some dates prior to conception and birth are important though, for example, birth defects, prenatal exposures, or pregnancy-related issues.[5]
Minimum values are commonly no smaller than a value of minutes from time of entry (albeit length of an event may span seconds or even fractions of a second).[6] An exception to this relates to ECG measurements, n.b., these relate to observables. (See the section ‘Brief Time Spans and Timed Events’.)
Common measurement units and physiological or clinical phases (like “postoperative day 1”) undergo conversion to their day equivalents when rendering a date. [7] For building a health timeline, values are recorded in day units.
Sequence and Relative Time
While our focus is aimed at generating and plotting a date or date span, the sequence of events may be detailed, but dates for plotting the sequence indiscernible. Relative time for events, i.e., when an event occurred in relation to another event, provides important detail in understanding the sequence.
Let us consider three similar, but vague statements. While none of these three is plottable in its entirety, they all provide important information—allowing the reader to understand that one event occurred followed by a specific second event.
“I had rheumatic fever. Later I was diagnosed with heart valve disease.”
[Sequence of rheumatic fever first, then heart valve disease]
“When I was a kid, I had rheumatic fever. Years later I was diagnosed with heart valve disease.”
[Start date before age 18 years old, therefore, point in time for age 9 years old with delimiters of +/- 9 years (by convention)]
[Sequence of rheumatic fever first, then heart valve disease greater than one year later]
“When I was 15, I had rheumatic fever. Years later I was diagnosed with heart valve disease.”
[Start date point in time at age 15.5 years old with delimiters +/- ½ year (by convention)]
[Sequence of rheumatic fever first, then heart valve disease greater than one year later]
The first statement is interpretable, but not plottable. While the second statement provides a pair of vague landmarks (“When I was a kid” and “Years later”) that may be used for plotting, the only trustworthy information is the sequence. Similarly, the third statement allows us to plot a single point with confidence (i.e., rheumatic disease at age 15) and recognize the sequence for a second event.
One possibility is to plot the chosen point in time as an anchoring event (in the second statement that would be rheumatic heart disease at age 9 [with delimiters +/- 9 years]; in the third, that would be rheumatic heart disease at age 15.5 years [with delimiters +/- ½ year]). The events would include a comment stating, ‘appearance of heart valve disease greater than one year later’. Thus, we can capture the sequence of related events even though only one could be plotted. Optionally, the phrase "years later" could be plotted with an anchoring point in time ('9 years' or '15.5 years') and a lower delimiter for heart valve disease of point in time plus 1.5 years ('>10.5 years' and '>17 years', respectively).
Brief Time Spans and Timed Events
Brief time periods are key measurements for a range of events, e.g., various ECG measures (PR interval, QT interval, QRS complex duration, ST duration, etc.), laboratory measurements (prothrombin time-INR, activated partial thromboplastin time, clotting time), procedure milestones (start time, completion time, etc.), and resuscitation efforts. These range from units smaller than a second to those greater than hours. For the most part, larger time units (equal to or greater than a day) will be captured using the schema discussed in this paper. Brief spans, like the above examples must be captured, but rarely need to be plotted on a patient’s health timeline, other than to say that the event (e.g., “ECG performed”) occurred on a specific date.
Conclusion
The methodology for determining event dates involves assessing temporality, which can be either exact or approximate. Specific trusted sources may provide exact dates, while text records often estimate the timing of events. To capture event timing accurately, both the extrapolated point in time mentioned and a range are considered, with precision varying between measurement units. Specific rules are outlined for different types of event descriptions, ensuring consistency and accuracy. Additionally, sequences of events and relative time are considered, allowing for the plotting of related events even when exact dates are unavailable. Brief time spans, such as those involving medical measurements (like ECG intervals) or procedures, are also accounted for in the methodology, ensuring comprehensive event capture.
[1] Hripcsak G, Elhadad N, Chen YH, Zhou L, Morrison FP. Using empiric semantic correlation to interpret temporal assertions in clinical texts. J Am Med Inform Assoc. 2009 Mar-Apr;16(2):220-7. doi: 10.1197/jamia.M3007. Epub 2008 Dec 11.
[2] The choice of including a range that is ± ½-time unit is by convention since the source would be expected to provide a different predicating number for the measurement were that what was meant.
[3] For calculations where the measurement units are larger than a day (week, month, year), use day units to calculate the date such that ± 3.5 days is used for weeks, ± 15 days is used for months, ± 182.5 days is used for years. As noted previously, the inherent rounding error using these calculations is deemed acceptable for our purposes.
[4] 91 days equals one quarter of a year. Other quarter units for use when the unit is one half of a measurement:
½ week = 2 days, ½ month = 7.5 days, ½ year = 91 days
These are specifically used for “0.5 weeks”, “0.5 months”, and “0.5 years”. They are not used with other concepts like “3.5 weeks” or “10.5 months”; those use the unit delimiters, week and month, respectively.
[5] For example, maternal risk factors like prolonged maternal exposure to a known cause of birth defects.
[6] Like “loss of consciousness for seconds”. We can assign a date to the event, but the length of the event itself is not plotted.
[7] To plot on a timeline, most measurement units are translated into day units where one day equals a unit of ‘1’. Accordingly, the length of certain time units follows:
1/2 day = 0.5, day = 1, week = 7, month = 30.5, trimester = 91, year = 365.25 (These are then rounded to the nearest day prior to representation on the health timeline.)
Comments