Aligning Doses, Samples, and Clinical Events in Time

Why a Clinical Timeline Must Represent Events

A dosing model uses the timing of events to interpret exposure and response. A concentration collected three hours after a dose is not equivalent to one collected ten hours after it, even if both results are entered into the record at the same time. Similarly, a laboratory result may be released after a dose was administered even though its specimen was collected before that dose. The important task is to reconstruct when the underlying clinical events occurred and express their timing consistently.

A timestamp can refer to several different things: the time a dose was administered, the time an order was placed, the time a specimen was collected, the time a result was verified, or the time a record was transmitted to another system. Treating these as interchangeable can reverse event order or change a calculated dose-to-sample interval. Time-zone conventions introduce another source of error: two systems can display different local clock times for the same instant, or identical clock readings for different instants.

Representing Events on a Shared Time Scale

A model-ready timeline needs a consistent reference for comparing events. A common approach is to convert each timestamp that identifies a unique instant to UTC (Coordinated Universal Time), while retaining the original timestamp, its offset or time-zone identifier, and its source for traceability. UTC is a comparison scale, not a claim that the clinical event occurred at a location in the UTC time zone.

For arithmetic, represent a local timestamp as a number of minutes or seconds from an agreed origin. Let \(t_{\mathrm{local}}\) be the local clock time and let \(o\) be the UTC offset in minutes, signed as local time minus UTC. Then the UTC time is obtained by subtracting that signed offset. For example, an offset of \(-300\) minutes means local time is five hours behind UTC.

Once the relevant event times are on the same scale, the elapsed time from an administration to a sample is the sample time minus the administration time. If both times are represented in minutes, the result is in minutes; dividing by 60 converts it to hours. This elapsed time is a model input when the model requires time since dose. For repeated administrations, the relevant dose history and the model’s dosing convention determine which intervals are needed.

Use the Timestamp That Matches the Clinical Meaning

The appropriate event time depends on what the record describes. For an administered dose, use the time the dose was actually given, not the order time, scheduled time, or time the administration record was signed. For a concentration, use specimen collection time, not laboratory receipt, analysis, verification, or result-release time. For a covariate, use the time the underlying measurement pertains to: for example, the time blood was drawn for a laboratory value or the time a weight was measured.

Some events have a duration rather than a single instant. An infusion is naturally represented by a start and stop time, together with the delivered amount and rate information available to the model. A dose entry that records only a date or a scheduled time may not establish when administration occurred. Do not silently substitute another timestamp; preserve what is known and make the limitation explicit.

A covariate also needs a clinically meaningful temporal interpretation. A creatinine value belongs to its specimen-collection time, not automatically to the time it became visible in the record. A covariate may be treated as constant over a period only if that modeling convention is appropriate for the use case; timestamp alignment alone does not establish that a value remains valid indefinitely.

Daylight-Saving Changes and Imprecise Times

A local clock reading is not always enough to identify an instant. When clocks move backward at the end of daylight-saving time, a local time such as 01:30 can occur twice, with different UTC offsets. When clocks move forward, some local times do not occur at all. A time-zone name and event date can help determine the applicable offset, but only if the location and time-zone rules are known and correctly applied. If the record gives neither an offset nor enough information to resolve an ambiguous time, the instant may not be recoverable from that timestamp alone.

For an ambiguous time, preserve both plausible interpretations until another source resolves the event—for example, an explicit offset, a device timestamp, or a reliable sequence of surrounding events. Do not choose whichever interpretation makes the timeline appear more convenient. A nonexistent local time may indicate a transcription error, an incorrect zone, or a system conversion issue and requires clarification rather than an invented UTC time.

Timestamp precision also limits what can be concluded. If an administration is recorded only to the nearest hour, reporting a dose-to-sample interval to the nearest minute suggests more certainty than the data support. A date-only value identifies a calendar date, not a unique time of day. Where an exact time cannot be established, represent the uncertainty or coarseness explicitly; do not fill in midnight or another conventional time as though it were observed.

Deriving and Checking Elapsed Time

The subtraction used for elapsed time is valid when both event times identify instants on a common, consistently interpreted time scale. To see why a shared offset matters, suppose local times \(L_1\) and \(L_2\) are recorded with the same UTC offset \(o\). Their UTC representations are \(L_1-o\) and \(L_2-o\). Subtracting gives:

Thus, for two events in the same correctly identified offset period, their local clock difference equals their elapsed time. But across a time-zone or daylight-saving offset change, the offsets can differ. If event 1 has offset \(o_1\) and event 2 has offset \(o_2\), then:

The extra offset term explains why naive subtraction of local clock readings can be wrong across a clock change. Converting each event using its own applicable offset handles this change, provided those offsets are known and correctly assigned.

A practical alignment process is:

Diagnostics, Alternatives, and Limitations

Useful diagnostics include checking whether samples appear before or after the administrations that are supposed to inform them, whether repeated event intervals are plausible, and whether timestamp patterns cluster around timezone or daylight-saving boundaries. Compare event time with documentation and ingestion times to identify systematic delays, but do not use those delays as a correction unless their meaning and reliability have been established. Check that conversions preserve event ordering where expected and that the same instant from two systems maps to the same common time.

The main alternative is to sort or calculate intervals directly from raw local timestamps. This can be acceptable when values are known to share the same offset and unambiguous event semantics; the derivation above shows why a common offset cancels. It is unsafe as a general strategy across locations, offset changes, or records with unspecified zones. Converting each known instant to UTC is more robust for comparisons, though it cannot recover information absent from the source.

For uncertain or coarse timestamps, a single point estimate can imply false precision. A more faithful representation may be a time interval, such as “sometime on this date,” or one of multiple possible instants. Whether a particular model can use interval-valued timing is a modeling question; ordinary dose-to-sample calculations require a resolved event time. If it is not resolvable, flag the limitation rather than quietly choosing an assumed instant.

Common failure modes include treating an order as an administered dose, using result-release time for specimen collection, assuming every timestamp is UTC, ignoring duplicated local times during a clock change, and assigning midnight to date-only records. These mistakes can alter the apparent time since dose and thus change the interpretation of a concentration or covariate. Timestamp alignment supports correct model inputs; it does not by itself establish that a dose was correctly documented, that a covariate is clinically appropriate, or that the model is suitable for a patient.