From Administration Records to a Dose History
A dose history is a subject-specific, time-ordered account of what the available records say was actually administered. It is not simply a copy of the prescribed regimen: planned doses and actual administrations can differ, and an absent record does not by itself establish that a dose was missed. A reliable reconstruction preserves that distinction while making the timing, amount, route, and documentation of each administration inspectable.
The input is usually a set of dose-event records, together with information about the subject, regimen, and the period for which records are expected to be available. Before combining records, ensure that event times use the same time origin and units, amounts use compatible units, and the records refer to the same analyte or treatment. The earlier material on dose-event records, dose units, route and formulation, bolus encoding, infusion duration and rate, dose-time conventions, regimen changes, and missed, held, and partial doses supplies the field-level conventions used here.
For a bolus, the administration is an event at a time \(t_i\). For an infusion, it occupies an interval from start time \(s_i\) to end time \(e_i\), with duration \(e_i-s_i\). A useful consistent representation is the half-open interval \([s_i,e_i)\): it includes the start but not the end. This convention lets an infusion ending at the instant another begins be represented without falsely making the two occupy the same instant. It does not change what the source says was administered.
A Reconstruction and Checking Sequence
Keep the source records and the derived history traceably linked. A practical sequence is to establish the coverage boundary, normalize fields, order events, then check intervals and compare the actual history with the regimen. Do not resolve an ambiguous event by silently choosing the most convenient interpretation.
Record the source-supported start and end of administration documentation, including whether each boundary is inclusive. Treat an unknown boundary as unknown, not as proof that no dose occurred before or after it.
Use a common time origin and compatible time and amount units. Retain the source event identifier, event type, actual administration status, route or formulation, and infusion start, duration, or end when applicable.
Order events within subject by actual administration time. Inspect equal timestamps and overlapping infusion intervals, comparing the source identifiers and documentation before labeling records duplicate or distinct.
Compare the documented events with the expected schedule within the known coverage window. Record confirmed omissions, unexplained gaps, deviations, and unresolved uncertainty as different findings.
Suppose documented infusion intervals are \([s_i,e_i)\), sorted by start time. To find uncovered time between consecutive intervals, compare the next start \(s_{i+1}\) with the latest covered end \(E_i\), where \(E_i\) is the greatest end time among intervals processed so far. If \(s_{i+1}>E_i\), the interval \([E_i,s_{i+1})\) is an uncovered gap in the infusion records. If \(s_{i+1}=E_i\), the intervals meet; if \(s_{i+1}<E_i\), they overlap. This follows directly from the interval boundaries: a time after \(E_i\) and before \(s_{i+1}\) belongs to neither recorded interval, while a start before \(E_i\) lies within time already covered by an earlier interval. This check describes infusion-time coverage only. Bolus events are points and do not create a continuous covered interval.
Worked Example: Reconstructing a Bolus History
Consider a hypothetical subject with a documented regimen of 100 mg by bolus every 12 hours. The source covers \([0,60)\) hours and explicitly records actual administrations at 0 hours (100 mg), 12 hours (100 mg), and 36 hours (50 mg, documented as partial), and a held dose at 24 hours. Assume the administration log is complete for this example’s coverage window and confirms that the 24-hour dose was held.
Order the records by time: 0 h, 12 h, 24 h, and 36 h. The events at 0 and 12 hours are full administrations. At 24 hours the record documents a held dose, not an administered dose. At 36 hours the documented administered amount is 50 mg. The interval from the previous administration at 12 hours to the next one at 36 hours is \(36-12=24\) hours; it includes the held scheduled dose at 24 hours.
The total documented amount administered is \(100+100+50=250\) mg. The schedule has expected administration times at 0, 12, 24, 36, and 48 hours within \([0,60)\). Of these, the 24-hour dose is confirmed held, and the 48-hour administration is not represented in the stated records. Because this example stipulates a complete log and an explicit held status at 24 hours, the 48-hour absence is an unexplained record inconsistency, not evidence that the dose was held. In real data, completeness must be established from source documentation rather than assumed from a blank event list.
Checking Coverage, Repetition, and Regimen Agreement
A regimen is a reference for checking the actual history, not a substitute for it. For a fixed-interval regimen with reference start time \(t_0\) and interval \(\tau\), expected scheduled times are \(t_k=t_0+k\tau\) for integer \(k\geq0\), restricted to the documented coverage window. Actual events may occur before or after those times, may be partial, or may have a documented held or missed status. A regimen change requires a new reference schedule from its documented effective time; applying one interval across a change can create false discrepancies.
For each expected administration, classify the record using the available evidence: documented full administration, documented partial administration, documented held or missed dose, administration at a discrepant time, or unresolved/no matching record. These categories must not be collapsed. In particular, a gap between recorded administration times can reflect an explicitly held dose, an unrecorded event, a source-coverage boundary, or an actual delay. The records and coverage evidence determine which interpretation is justified.
Repeated timestamps deserve a targeted check. Two records at the same time may be duplicate imports, distinct administrations recorded at coarse time precision, or separate administrations that genuinely occurred together. Compare source event identifiers, amounts, routes, formulations, and original documentation. Do not automatically sum repeated rows or delete one. For infusions, compare intervals: an overlap may be a valid concurrent administration, a duplicated record, or a recording error, and timing alone cannot decide which.
Worked Example: Infusion Intervals and an Uncovered Gap
In a hypothetical record, an infusion of 100 mg starts at 0 hours and lasts 2 hours. A second infusion of 50 mg starts at 8 hours and lasts 1 hour. The documented source window is \([0,16)\) hours. Both events are recorded as completed, and no other infusion is documented.
The first infusion has end time \(0+2=2\) h and interval \([0,2)\). Its rate is \(100\text{ mg}/2\text{ h}=50\) mg/h. The second ends at \(8+1=9\) h and occupies \([8,9)\), also at \(50\) mg/h. Since the second start is later than the first end, the uncovered interval between the recorded infusions is \([2,8)\), of duration \(8-2=6\) h. The source window also has no recorded infusion during \([9,16)\), a 7-hour span.
These are gaps in recorded infusion intervals, not proof of inadequate treatment or missing records. If the documented regimen specifies infusions every 8 hours by start time, starts at 0 and 8 hours agree with those two expected times. Whether the absence of another infusion before the end of the coverage window is a discrepancy depends on the regimen’s timing and boundary rules. In any case, the 6-hour interval between infusions is not itself a missed dose: it is simply time when neither infusion was running.
Worked Example: Resolving a Repeated Administration Time
A hypothetical subject has a documented 100 mg bolus regimen every 24 hours, with complete source coverage over \([0,72)\) hours. The source contains records at 0 h (100 mg, event A), 24 h (100 mg, event B), 24 h (100 mg, event C), and 48 h (100 mg, event D). Events B and C have different source identifiers, but no administration narrative is available to establish whether both were given.
The schedule expects administrations at 0, 24, and 48 hours: \(t_k=0+24k\), for \(k=0,1,2\). The 0- and 48-hour records match expected times. At 24 hours, two records share a timestamp and amount. This is a repeated-time discrepancy requiring reconciliation, not enough evidence to conclude that 200 mg was administered or that one row is a duplicate.
Preserve both source records and flag the 24-hour event as unresolved. If source review confirms one was a duplicate import, retain the traceable correction and count one 100 mg administration. If it confirms two distinct administrations, represent both and document that the total at that time was 200 mg. If the question cannot be resolved, do not invent a definite dose total; report the ambiguity. This example also shows why checking expected schedule times alone cannot validate the dose history.
Diagnostics, Alternatives, and Limitations
Useful checks include verifying that each event belongs to one subject, timestamps fall within plausible source boundaries, amounts and units are compatible, infusion end times follow start times, and recorded rates agree with amount divided by duration when both are specified. Flag identical records, equal-time events, overlapping infusion intervals, unexpectedly long gaps, and events that conflict with documented regimen changes. These are diagnostic signals for review, not automatic grounds for correction.
A chronological event table is the most auditable representation: it preserves each source event and its status. A derived interval summary can make infusion overlap and gap checks easier, while a regimen-based schedule can show expected versus documented administration times. These are complementary views. A compact summary alone can hide repeated records, changes in status, or the evidence behind a correction; retain the event-level history as the reference. Automated validation can identify inconsistencies reproducibly, but cannot determine whether an ambiguous entry reflects a real administration without suitable source evidence.
The reconstruction is limited by what the source records support. Unknown coverage boundaries, coarse time precision, inconsistent units, missing administration status, and undocumented regimen changes can prevent a definitive history. Do not extend the observation window by assuming that no administration occurred outside the supplied records. Similarly, do not infer that an unexplained gap is a held or missed dose, and do not use a planned schedule as if it were an actual administration log. Keep planned and actual information distinct, and preserve uncertainty in the derived dataset.
Common Mistake
Check Your Understanding
Answer each question using the distinction between actual administrations, source coverage, and the planned regimen.
- A complete log covers \([0,48)\) hours for a regimen scheduled at 0, 12, 24, and 36 hours. Records show administrations at 0, 12, and 36 hours, and an explicit held status at 24 hours. What is established, and what is the interval between the 12- and 36-hour administrations?
- An infusion starts at 3 hours, lasts 4 hours, and a second starts at 9 hours. Give both intervals and the uncovered gap between them.
- Two 25 mg records have the same subject and timestamp but different event identifiers. What should the reconstructed history do before deciding the amount administered at that time?
- A dose is absent from an event list, but the source coverage for that subject on that date is unknown. Can the dose be classified as missed? Explain.
Solutions: (1) The 24-hour dose is confirmed held, and administrations are documented at 0, 12, and 36 hours. The elapsed time between the latter two administrations is \(36-12=24\) hours. The held status does not turn the 24-hour event into an administration. (2) The first interval is \([3,7)\) hours because \(3+4=7\). The second ends at \(9\) hours, so its interval is \([9,\ldots)\); its duration is not supplied. The gap is \([7,9)\), or \(9-7=2\) hours. (3) Preserve and inspect both records and reconcile them against source evidence, including identifiers and administration documentation. Do not automatically sum them or delete one; until resolved, mark the amount as ambiguous. (4) No. Without known coverage or explicit status, absence from the list does not establish a missed dose. Record the uncertainty and seek evidence about source completeness or administration status.