August 19, 2026

12 min read

A Spreadsheet Is Still Paperwork

A technician writes a weight on a sheet at the chute. Hours later, someone types that number into a spreadsheet. That second step is what is often referred to as “going digital.”

It changed where the number lives. It did not change how the number got there, and it did not change when anyone would find out if it were wrong.

TL;DR

The costliest error is the one caught late. Once an animal is off study, a wrong value can only be dropped because it cannot be checked against the animal.
Spreadsheet QC compares one copy to another. It cannot confirm that the record matches the animal, so a faithfully copied recording error passes clean.
Structured capture records the observation once, directly into the system. That removes the transcription step and surfaces questionable values while the animal is still there.
It is not a compliance add-on. When the entry is the original record rather than a copy, there is nothing to reconcile for a reviewer later.

What Work Does “Going Digital” Actually Save?

Even after going digital, teams running animal studies still spend their days keying data off observation sheets and checking it line by line. A Head of Development running swine studies described his biggest pains as:

“You have study binders full of just observation sheets. First, someone needs to do a QC to make sure everything’s been filled out. And then typically someone has to transcribe that data into some summary format to be able to run statistics on it. There’s an opportunity for transcription error along the way. Plus the amount of data to transcribe can be overwhelming at times.”

Typing and QC do not disappear when the destination changes from a binder to a spreadsheet. Neither do the errors those steps introduce.

Volume makes it worse. A research director running cattle studies illustrated the challenge with a concrete example:

“When you’ve got 400 animals on a study, and maybe there’s 10 different forms per animal, it’s just overwhelming.”

Four hundred animals, ten forms each. Four thousand values, each one hand-entered twice: once at the chute, once at the desk. Moving the destination to a spreadsheet removes none of them.

Multi-site work compounds the problem further. A global research operations manager described the challenges she is facing:

“We have data coming in from all different world areas. Everyone’s recording it a little bit differently, doing their calculations a little bit differently. There’s a lot of manual manipulation of the data that has to go on to get it all into the same format.”

While large amounts of data exacerbate the problem, its roots lie in the transcription. Research points in the same direction: in one analysis of an electronic data capture database, the most common errors were transcription errors introduced when data was copied from paper source into the system. These errors typically surfaced later during cleaning rather than at the point of entry (Mitchel et al, 2011).

A separate study identified a blind spot in how quality is usually measured: checks confirm the final copy is consistent with the source used to create it, but that source itself was never validated against the original observation. That earlier step goes unmeasured, so the true error rate is higher than the number teams track (Nahm et al, 2008).

Both of these studies are human clinical studies rather than animal health studies. Transcription is common to both, and the insights transfer: the transcription step is where errors originate, and digitizing the destination does not address the problem.

Which Errors Can You Actually Catch?

When a value moves from an observation sheet to a spreadsheet, two different errors can occur, and they are not equally catchable.

A transcription error is a value copied wrong. The sheet says 1,240 and the spreadsheet says 1,420. This one is detectable, because the two records disagree, and the sheet settles it.

A recording error is a value written down wrong in the first place. The animal weighed 1,240 lbs and the sheet says 1,420 lbs. Copy that faithfully into the spreadsheet, and the two records agree. There is no disagreement to catch. The sheet and the spreadsheet match, but both are wrong.

Spreadsheet QC only finds the first kind. It confirms that the spreadsheet matches the sheet, but cannot confirm that the sheet matches the animal. A recording error draws attention only if the number is implausible enough to stand out during cleaning.

Why Isn’t Finding the Error Enough?

Suppose a recording error is bad enough to be noticed, for example, the weight is implausible and gets flagged during cleaning. Finding it is not the same as fixing it.

To correct a value, you have to verify it against its source, but for recording errors the source, the animal, is no longer available. Once it is off study, the value cannot be confirmed and cannot be corrected. It can only be kept, unverified, or dropped.

A senior research leader described where that leads:

“When it’s on paper, if somebody doesn’t catch it that day, it’s too late. We just have to throw it out.”

For the sponsor that means an animal you dosed, housed, and observed is now producing no usable data.

The human clinical-research literature reports the same delay: out-of-range values and protocol deviations are typically identified during cleaning rather than at entry (Mitchel et al, 2011), and by that point the time to remeasure has passed.

Wondering how much of your own workflow catches errors too late? The free CVM Submission Readiness Check returns a gap summary in minutes.

What Changes When the First Record Is the Only Record?

Structured capture records the observation on site, directly into the system. There is no observation sheet to transcribe and no second record to reconcile. That does two things:

It removes transcription error entirely. A value that is never copied cannot be copied wrong.
It provides the opportunity to catch a questionable value while it can still be fixed. As the value is entered, the system checks it against the rules for that field so team members notice issues and can catch impossible values or blank fields while the value can still be corrected.

However, capture of data directly into an electronic system does not eliminate all errors: it does not catch wrong values that fall within a possible range. What structured capture changes is timing: if a questionable value is surfaced while the animal can still be re-checked it can be corrected then.

This is the distinction a regulatory lead drew for us: without checks at the point of entry, an electronic system offered her no advantage over paper; with them, her position changed entirely. What mattered to her was not that the data was electronic but that erroneous data and blank fields could be caught in time to fix them.

The advantages of capturing structured data directly into the system are not limited to accuracy. Removing the transcription and QC step saves time, so that review and analysis can start sooner and the database closes sooner. A comparison of field data collection methods found the main advantage of direct entry was shorter time from data collection to database lock (Walther et al, 2011). That analysis was done on human clinical studies, but the time savings also apply to animal health studies.

Does Structured Data Capture Hold Up with Regulators?

The arguments for structured data capture so far are operational: less work, fewer untraceable errors, corrections that can be made on site, shorter time to database lock. The same change matters when the study is reviewed by regulators.

Clinical studies of veterinary products in the target species, the effectiveness and in-use safety studies submitted to regulators, are conducted under Veterinary Good Clinical Practice. The governing standard, VICH GL9, defines raw data as the firsthand record of observations made during a study, and states explicitly that transcribed data is not considered raw data. Record an observation on paper and type it into a spreadsheet later, and the sheet is your raw data while the spreadsheet is a copy of it. Enter the observation directly into the system and, in GL9’s own terms, the electronic record is the raw data.

That distinction has a practical consequence. Original raw data does not need to be reconciled, because it is the original record. Copied data’s reliability depends on matching it back to an original stored somewhere else, and that burden falls on the sponsor. GL9 permits transcription, but it asks you to make an authenticated copy, document why the copying happened, and retain the original, the copy, and that explanation together. The gap this creates is the single most common problem the FDA Center for Veterinary Medicine’s quality assurance study reviewers report: across both GLP and GCP studies, their number one finding is that the final study report does not fully and accurately reflect the raw data.

For more on what CVM cites most often and how it reviews animal study data, see What FDA’s CVM Cites Most in Animal Study Submissions.

Structured capture is not a separate compliance step. It is the same fix, seen from the reviewer’s side: the record under review is the original observation.

See Where Your Process Stands

If your workflow still puts a transcription step between the animal and the record, you already have unrecoverable errors in past datasets. You cannot go back and find them. You can find out where the exposure sits now.

Prelude’s CVM Submission Readiness Check compares your study against the issues CVM cites most often and returns a gap summary you can bring to your team. It is free, it takes about five minutes, and it tells you which gaps to close first.

Run the readiness check →

FAQs

We already double-check our data. Isn’t that enough?

Double data entry checks the keying, not the recording. It is more accurate than single entry and has long been recommended for that reason (Reynolds-Haertle & McBride, 1992). It catches typing mistakes by flagging where two records disagree, and it works best with a second, different operator: same-operator double entry caught about 69% of errors versus roughly 88% with a different one (Kawado et al, 2003). But it cannot check a value against the animal. If the number was written down wrong at the chute and copied correctly twice, both records agree and the error passes. Double entry controls a later step than the one that creates the unrecoverable problem. Both studies cited here are from human clinical studies, but the insights are transferable to animal studies.

Does this mean paper is unacceptable to regulators?

No. Paper source records are permitted under veterinary GCP. The point in VICH GL9 is narrower: the original record is the raw data, and a transcribed copy is not. The issue is not that paper is banned, it is that every transcription adds a copy someone has to reconcile back to an original. Structured capture removes that reconciliation by making the first record the one that gets reviewed.

Is capturing data on a device realistic under field conditions?

Field evidence says yes, with caveats. In rural, resource-limited settings, direct electronic capture reduced error rates and eliminated omissions compared with paper, under real field conditions (Thriemer et al, 2012; Zeleke et al, 2019). The caveats are real: it required upfront software configuration and staff training, and device choice mattered, with weaker handhelds performing worse. These papers establish feasibility in human clinical studies in the field but also show that the solution is not plug-and-play.

Get the Weekly Briefing for Animal Health R&D Leaders

Prelude has powered animal health research for over 20 years. We work with 8 of the top 10 animal health pharma companies and all three top pet nutrition brands, across clinical, preclinical, post-market, and feed and nutrition studies.

Subscribe to Animal Health Weekly for the patterns, decisions, and regulatory shifts shaping how animal health evidence is generated and defended. Short, specific, actionable reading for people who lead R&D in this field.

Subscribe to Animal Health Weekly

Related Resources

References

Mitchel JT, Kim YJ, Choi J, Park G, Cappi S, Horn D, Kist M, D’Agostino RB Jr. Evaluation of Data Entry Errors and Data Changes to an Electronic Data Capture Clinical Trial Database. Drug Information Journal. 2011;45(4):421–430.
Nahm ML, Pieper CF, Cunningham MM. Quantifying Data Quality for Clinical Trials Using Electronic Data Capture. PLoS ONE. 2008;3(8):e3049.
Walther B, Hossin S, Townend J, Abernethy N, Parker D, Jeffries D. Comparison of Electronic Data Capture (EDC) with the Standard Data Capture Method for Clinical Trial Data. PLoS ONE. 2011;6(9):e25348.
Thriemer K, et al. Replacing paper data collection forms with electronic data entry in the field. BMC Research Notes. 2012;5:113.
Zeleke AA, et al. Data quality and cost of paper versus electronic data collection. JMIR mHealth and uHealth. 2019;7(2):e10995.
Reynolds-Haertle RA, McBride R. Single vs. double data entry in CAST. Controlled Clinical Trials. 1992;13:487–494.
Kawado M, Hinotsu S, Matsuyama Y, Yamaguchi T, Hashimoto S, Ohashi Y. A comparison of error detection rates between the reading aloud method and the double data entry method. Controlled Clinical Trials. 2003;24:560–569.
FDA Center for Veterinary Medicine. CVM GFI #85 (VICH GL9) Good Clinical Practice. May 2001. (VICH GL9, CVMP/VICH/595/98.)
Lazo A, Davis D, Desilva J, Smith E, Rhodes L, Cottrell E. QASR Hot Topics. FDA Center for Veterinary Medicine, Office of New Animal Drug Evaluation. Society of Quality Assurance Annual Meeting, March 2023.