Why Cross-Field Consistency Matters
A record can look plausible when each field is reviewed on its own and still contradict another field. A sample time may be valid as a timestamp but fall before the protocol permits sampling. An infusion may have plausible start and end timestamps individually, yet the end may precede the start. A dose may be a positive number but use units that do not agree with the product or medication order. These conflicts matter because population pharmacokinetic analyses depend on linking concentrations to the correct participant, dose history, and event times.
The purpose of a cross-field check is to identify records that cannot all be true under the documented study rules. A flag is a request for verification, not a finding that identifies which field is wrong. Preserve that distinction: the same contradiction may result from a transcription error, an incorrect linkage, a time-zone or date problem, an undocumented protocol exception, or incomplete source metadata.
Specify the Rule Before Applying It
A useful check begins with the meaning of each field, not merely its column name. Establish which records belong to the same participant and event, what “study entry” means in the protocol, whether pre-entry screening samples are allowed, and which time basis the timestamps use. Earlier material on harmonizing dates, clock times, and time zones and reconstructing elapsed time is essential here: comparisons are only meaningful after times are represented on a common basis.
Likewise, distinguish a planned dose from an administered dose, and an amount of drug from a volume of formulation. Product metadata should identify the relevant active ingredient or stated basis, dosage form, strength or concentration, and units. If a product record is incomplete or ambiguous, a numerical conversion may not be defensible.
For each rule, define its scope, the fields it uses, and the result it produces. For example, “sample time must not precede the first time at which the protocol permits collection” is more precise than “sample time must be after study entry.” The latter may be wrong if screening samples are permitted before enrollment. A check should reflect the protocol, not replace it with an assumed universal rule.
Checking Event Order and Time Boundaries
For a sample and a documented entry time on a common clock, let \(t_s\) be the sample time and \(t_e\) the protocol-defined earliest permitted sampling time. The basic order check is \(t_s \ge t_e\). If the protocol permits sampling before formal enrollment, use the applicable protocol boundary instead of the enrollment timestamp. A violation means the records and the chosen rule are inconsistent; it does not prove that the sample time is erroneous.
For a single infusion, let \(t_{\text{start}}\) and \(t_{\text{end}}\) be the start and end timestamps for the same administration event. The recorded duration is \(d=t_{\text{end}}-t_{\text{start}}\). If the record represents a completed infusion, its end should not precede its start, so \(d\ge 0\). This check assumes both timestamps belong to the same event, use a consistent time basis, and are not reversed by a documented data convention.
Timestamp precision affects how strict the comparison should be. Suppose each time is rounded to the nearest minute. Each recorded time can differ from its underlying time by at most half a minute, so the difference between two recorded times can differ from the underlying difference by at most one minute. A small negative recorded duration may therefore be compatible with rounding, whereas a large negative duration cannot be explained by that precision alone. The appropriate tolerance must be declared from the recording process; it should not be chosen after seeing a troublesome record.
Worked Example: Sample Before the Permitted Study Window
In a hypothetical study, the protocol states that research pharmacokinetic samples may be collected only from the documented study-entry time onward. After the timestamps have been harmonized, participant P014 has an entry time of 4 May 2026 at 09:00 and a sample time of 4 May 2026 at 08:42. Determine whether the record passes the rule.
The rule is \(t_s\ge t_e\). The difference is \(08{:}42-09{:}00=-18\) minutes. Thus the sample is recorded 18 minutes before the permitted window begins, and the record fails this check.
This is a contradiction flag, not an automatic correction. Verify participant and visit linkage, the protocol definition of entry, and the original sample and entry timestamps. If the protocol allowed a pre-entry sample or the recorded entry field is not the relevant boundary, the rule must be revised using documented evidence. Otherwise, retain the flag for resolution rather than changing 08:42 to 09:42 or deleting the record by assumption.
Worked Example: Infusion End Earlier Than Start
In a hypothetical administration record, an infusion start is recorded as 16 June 2026 at 10:45 and its end as 16 June 2026 at 10:15. Both timestamps are recorded to the nearest minute and are linked to the same infusion event. Calculate the recorded duration and assess whether rounding could explain it.
Using \(d=t_{\text{end}}-t_{\text{start}}\), \(d=10{:}15-10{:}45=-30\) minutes. The end is recorded 30 minutes before the start. With nearest-minute rounding, the difference between the recorded and underlying duration is at most one minute, so rounding cannot explain a 30-minute reversal.
The event therefore fails the ordering check. Before concluding that either timestamp is wrong, verify the event identifier, date, time-zone handling, and source record. If the source confirms one value, resolve the record according to the study’s data-correction process. A negative duration is not evidence that the infusion actually ran backwards.
Reconcile Dose Amounts With Product Records
Dose checks must preserve the distinction between a drug amount and a formulation volume. A product concentration \(c\), expressed as amount per volume, converts a volume \(v\) to an amount \(a\) by \(a=cv\), provided the concentration and volume use compatible units and refer to the same product and active ingredient basis. This arithmetic can expose a mismatch, but the intended dose must come from an authoritative order or source record.
For example, if a product is documented as 25 mg/mL and an order calls for 500 mg, the corresponding volume is \(500\text{ mg}/(25\text{ mg/mL})=20\text{ mL}\). A dose-event value of “20” without a unit is ambiguous: it could mean 20 mg or 20 mL. Do not infer the unit from whichever interpretation best matches the order. Check the raw administration record and the data dictionary.
Worked Example: Dose Unit Conflicts With the Product and Order
In a hypothetical dataset, the product record states 25 mg/mL, the medication order is 500 mg, and the administration event contains the value 20 with the unit mg. Check whether these fields agree. Assume the product and order refer to the same active ingredient and administration.
If the recorded event is literally 20 mg, it is one twenty-fifth of the ordered amount: \(20\text{ mg}/500\text{ mg}=0.04\), or 4% of the order. If the numeric value 20 was intended to be a volume in milliliters, the implied amount is \(25\text{ mg/mL}\times20\text{ mL}=500\text{ mg}\), which matches the order.
The record therefore has a material unit ambiguity: mg and mL produce different interpretations. It should be flagged and checked against the administration source and field definitions. The arithmetic supports a query; it does not establish that the unit should be changed from mg to mL. If the source confirms 20 mL, document the resolution through the study’s correction process. If it confirms 20 mg, investigate the difference from the 500 mg order rather than silently substituting the ordered amount for the administered amount.
Use Concentration Records Carefully
Concentration records can be checked across fields for explicit incompatibilities: missing or nonnumeric results where a numeric result is required, units that conflict with the assay metadata, or a sample identifier linked to a different participant or collection event. Unit harmonization and below-quantification status require their own documented conventions; a value and its flag should be checked against the relevant laboratory metadata.
A concentration’s magnitude alone does not prove that a dose or time field is wrong. Concentrations depend on the participant, sampling time, prior and current doses, and pharmacokinetic processes. Therefore, do not reject a dose record merely because a concentration “looks too high” or “looks too low.” An unusual concentration can justify review, but deciding whether a dose-concentration sequence is chronologically or pharmacologically anomalous is a distinct investigation. In this check, use concentration fields to identify direct data conflicts, not to impose an unsupported expected concentration.
A Practical Validation Sequence
Confirm participant, visit, sample, and administration identifiers so the compared fields refer to the same person and event.
Use protocol timing rules, administration definitions, product records, and field metadata. Record any allowed exceptions and time precision.
Compute elapsed-time differences or unit conversions explicitly, then test the result against the pre-specified rule and tolerance.
Retain the reported value, note which rule failed, consult source records, and record the evidence for any resolution.
This sequence is more reliable than scanning for implausible values in isolation. It also separates a deterministic contradiction, such as a clearly reversed event order, from a context-dependent concern that requires review.
Diagnostics, Alternatives, and Limitations
Useful diagnostics include counts of failed checks by rule, participant, site, and source system; distributions of signed time differences; and summaries of dose-unit combinations by product. Inspect individual flagged records alongside their linked metadata. A cluster of similar failures may point to a systematic import or mapping problem, while isolated failures may have different causes. Neither pattern identifies the correct value without source verification.
A hard validation rule is appropriate when the protocol or data definition makes a condition impossible and the relevant fields are sufficiently reliable. A tolerance-based rule is preferable when precision or rounding can produce small discrepancies. A review flag is safer when protocol exceptions or metadata ambiguity remain. These approaches are alternatives in how a check is operationalized, not substitutes for defining the underlying meaning of the fields.
The checks have limits. They cannot determine which conflicting field is correct, establish that a recorded dose was actually administered, or prove a biological explanation for a concentration. They can also generate false flags if event linkage is wrong, an exception is undocumented, or time and unit conventions are misread. Keep raw values available, preserve the rule and its result, and do not turn a check into an unrecorded edit.
Common Mistakes
- Assuming every sample must follow enrollment. This fails if the protocol permits screening samples. Use the protocol-defined earliest permitted time.
- Correcting a field to make the record pass. A failed rule shows that fields conflict; it does not identify which value to edit. Verify against source evidence.
- Treating a unit conversion as proof of intent. A product concentration can show what a volume would contain, but only the source can establish whether the event value was recorded as a volume or amount.
- Rejecting a concentration because it seems inconsistent with a dose. Magnitude alone is not a deterministic cross-field contradiction. Flag for appropriate review rather than assuming a data error.
Check Your Understanding
For each question, decide what the check establishes and what still requires verification. Solutions follow the questions.
- A protocol allows samples from 07:00, but a participant’s formal enrollment time is 08:00. A sample is recorded at 07:30. Does it fail the protocol timing rule?
- Two timestamps rounded to the nearest minute give a duration of -0.5 minutes. Can rounding alone account for this recorded difference?
- A product is 10 mg/mL and an event is recorded as 3 mL. What amount does the product information imply, and does that prove the amount administered?
- A concentration is unusually high, but the sample identifier, assay units, and laboratory flag all agree with metadata. Does this cross-field check prove a dose error?
Solutions. (1) No. The applicable boundary is 07:00, not formal enrollment at 08:00; 07:30 is after the permitted start. (2) Yes, potentially. With nearest-minute rounding, the recorded difference may differ from the underlying difference by as much as one minute, so a -0.5-minute recorded duration is not by itself decisive. (3) The implied amount is \(10\text{ mg/mL}\times3\text{ mL}=30\text{ mg}\). This calculation does not prove that 3 mL was actually administered; verify the source record. (4) No. An unusual concentration alone does not establish a dose error. It may merit a separate review, but this check has found no direct metadata conflict.