What Structured Data Changes for Your Study Team

It is the week before database lock. The study is finished, the animals are off study, and the data manager is trying to build one clean dataset out of the data collected by the team. There are four copies of the same spreadsheet. Two were emailed in from sites, one is the master, and someone saved a fourth as final_v2.

Then the biostatistician asks the obvious question: which file is the current one? Nothing in the file answers that question and now the next several days are spent on data reconstruction rather than analysis. None of that work is captured in the protocol, but it is necessary because the data was recorded on paper and rebuilt in spreadsheets.

TL;DR

Structured data capture lets a study team spend the end of a study reviewing and resolving data instead of transcribing it and reconciling spreadsheet versions.
Data is entered once, at the point of observation, and validated as it is recorded.
One record with a tracked history replaces competing spreadsheet files.
The result is an earlier database lock and lower data-collection cost.
Under veterinary GCP, that structured entry is the raw data a regulator reviews, not a copy that has to be reconciled to an original.

How Reliable Is Unstructured Data?

Manual paper-and-spreadsheet workflows are unreliable in any field that runs on data, and the evidence is well documented. A review by Panko found that 94% of audited operational spreadsheets contained errors. While the error-per-cell rate is small, in the low single digits, those errors add up in large documents substantially increasing the likelihood of incorrect results.

Recent work has confirmed these results, put the average cell error rate at nearly 4% and also found that people are consistently more confident in the spreadsheet they are building than the error rate warrants (Panko 2015).

Even peer-reviewed science is not immune: when researchers screened the supplementary data files attached to published genetics papers, they found that Excel had silently converted gene names into dates in roughly a fifth to a third of cases. That error went undetected until that analysis was performed (Ziemann; Abeysooriya).

It gets worse when the data is copied by hand before analysis. A field study on humans that replaced paper forms with direct electronic entry saw erroneous data drop from 7% to 1%. Missing entries could be reduced to zero because the system would not save a record until every required field was answered (Thriemer et al, 2012).

Animal studies are recorded the same way often under challenging conditions in barns or pens. Data is first recorded on paper and later copied into Excel. Each copy is a chance for a value to change or be omitted. By the time the database is locked, the dataset is the only record of what happened, and any value lost along the way is gone with it.

What Does Structured Data Change?

Most animal study data is still captured in an unstructured fashion. It is written down, moved between formats, and assembled into a dataset by hand often long after the original data was recorded. A production lead at an animal health manufacturer described an aquaculture workflow they ran in three steps. A measurement was calculated in Excel on a laptop, the result was written onto a paper form, and the form was later typed into a separate system for analysis. The same value was handled three times before anyone analyzed it. None of the three steps added anything to the value but each one could introduce errors.

Structured capture changes how the first record gets made. The observation is entered once into a structured record at the time it is generated. At no point is the data copied into another format and the value entered is the original record.

The system also defines what a valid entry looks like as the data is entered. Issues such as values outside the possible range or a required field that’s left blank are caught while the animal and the record are still in front of the person entering the data. On paper, those same problems are found later by a data manager who is reading the form long after it was recorded and when checking the source means a return to the field or is simply not possible.

Recording data in a structured way also means that there is only one record rather than several, and edits are tracked with a history of what changed, when and who made those changes generating a single source of truth.

Structured data capture is especially important in studies with large animal counts. A data manager at a top-10 animal health company described scoring several hundred animals in a single session. On paper this means several hundred lines that need to be written and then retyped. However, entered once at the point of scoring, it is recorded as the session happens.

What Does Structured Data Really Buy?

The benefits of structured capture show up in the parts of the study the team is measured on: the lock date, the volume of end-of-study cleanup, and the regulatory milestone.

Validation at entry reduces the work necessary at the end. For example, out-of-range values, impossible dates, and required fields that are left blank are resolved during the session rather than during database lock. A comparison of field data collection methods found that the main advantage of direct electronic entry was the reduction in time from data collection to database lock, because entered data reached the database without a separate transcription step (Walther et al, 2011). That study was in human clinical research, but the same mechanism applies to animal studies.

With structured data capture, transcription and version reconciliation largely disappear. The data manager is no longer typing forms into spreadsheets or working out which copy is current. This time can then be used for data review and query resolution, which can then start earlier in the study rather than accumulating before lock.

Capturing data in a structured way comes with a cost as well: the forms and validation rules have to be built and tested before enrollment starts, and field staff have to learn to record and check data on the system rather than on paper.

But even with that setup cost, the total cost of collecting the data in a structured manner tends to be lower. A cost simulation based on clinical trial data put the reduction at about 55%, depending on study design (Pavlovic et al, 2009).

The study leader is measured on the regulatory milestone. A study that closes on schedule, with a complete dataset and a documented record of how the data was collected, reaches its regulatory date without a repeat study.

Structured capture solves a specific set of problems. It removes the transcription steps and the competing versions, and it catches the entry errors a constrained field detects. It does not fix a flawed protocol or make up for untrained staff, and those problems show up in the data no matter how it is captured.

How Important Are Structured Data in the Regulatory Context?

Animal studies submitted to regulators are conducted under veterinary Good Clinical Practice. The governing standard is VICH GL9, the Good Clinical Practice guideline issued by the International Cooperation on Harmonization of Technical Requirements for Registration of Veterinary Medicinal Products, the program through which the FDA, the EU, and Japan align their requirements for approving veterinary medicines. FDA adopted it as Guidance for Industry No. 85.

GL9 defines raw data as the original record of what was observed, and is explicit on one point: transcribed data is not raw data. The original is the field sheet the data is first recorded on, the value later typed into a spreadsheet is a copy.

This is where structured capture matters beyond the lock date. When the observation is entered once, into the system, that entry is the raw data, and the record of who entered it and when is associated with it. When the same observation is written on paper and later keyed into Excel, the raw data continues to be the paper form, and the spreadsheet a reviewer or inspector sees is a copy whose link to that paper form has to be demonstrated.

What regulators do with that distinction, and what FDA’s Center for Veterinary Medicine cites most often when it reviews animal study data, is the subject of a separate piece. The point here is narrower. Structured capture does not just close the study faster it also produces the record the study is judged on.

The distinction has a practical consequence. When the record under review is the original entry, there is nothing to reconcile it against. When the record is a copy, its reliability depends on matching it back to an original held elsewhere, and the gap between the two is where discrepancies arise. CVM’s published review findings include cases where a final study report did not fully and accurately reflect the raw data behind it. Structured capture therefore makes the record a regulator examines the same record the study produced and does not have to be reconstructed to be trusted.

For a study team this means that data integrity is settled at capture rather than defended after the fact.

See Where Your Study Stands

Prelude’s CVM Submission Readiness Check compares an animal study against the issues CVM cites most often and returns a gap summary to share with the study team. It is a free self-assessment and a planning aid, not a compliance determination.

Run the readiness check →

FAQs

Is electronic data capture always more accurate than paper?

Not automatically. In a controlled field study, electronic capture on the stronger devices matched the accuracy of paper that was double-entered into a database, and on weaker setups (a handheld and a phone interview) it was worse. Its consistent advantage was speed, because the data reached the database without a separate entry step. Two things drive accuracy more than the technology itself. The first is the comparison: electronic capture clearly beats single-entry paper, but double data entry closes most of that gap. The second is form design. In the same study, free-text and date fields produced more errors than numeric, single-select, and skip fields, so the accuracy gain comes from a structured form with constrained, validated fields, on screen or on paper. (Walther et al, 2011)

Does an EDC on its own remove transcription errors?

No. When data still starts on a paper source and is keyed into the system afterward, the paper-to-system transcription step remains, and studies of EDC databases have found that these simple transcription errors from the paper source are the main errors that appear. The error is removed by capturing the observation directly into the structured record at the point it is generated, which also lets range and consistency checks flag a problem at entry rather than weeks later. The benefit comes from where the data is first recorded, not from the system being electronic. (Mitchel et al, 2011; Nahm et al, 2008)

Does the accuracy gap between paper and structured capture grow with the size of the study?

Yes. In a randomized comparison, the odds of an error on paper rose with every additional question on the form, about 1.5% per question relative to electronic capture, and overall 42% of paper records carried at least one data-quality issue against 31% of electronic records. Because the penalty accumulates with each field and each subject, it grows with the size of the study. A form scored across several hundred animals, with many fields each, is exactly where the difference is largest. (Zeleke et al, 2019)

References

Panko, R. “What We Know About Spreadsheet Errors.” J. End-User Computing 10(2), 1998 (rev. 2008). http://panko.shidler.hawaii.edu/SSR/Mypapers/whatknow.htm
Panko, R. “What We Don’t Know About Spreadsheet Errors Today.” Proceedings of the EuSpRIG 2015 Conference, European Spreadsheet Risks Interest Group, 2015. ISBN 978-1-905404-52-0. Preprint: https://arxiv.org/abs/1602.02601
Ziemann M, et al. Genome Biology 2016;17:177. https://genomebiology.biomedcentral.com/articles/10.1186/s13059-016-1044-7
Abeysooriya M, et al. PLOS Comput Biol 2021. https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1008984
Thriemer K, et al. BMC Research Notes 2012;5:113. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3392743/
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. https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0025348
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 Inf J 2011;45(4):421–430. https://doi.org/10.1177/009286151104500404
Nahm ML, Pieper CF, Cunningham MM. “Quantifying Data Quality for Clinical Trials Using Electronic Data Capture.” PLoS ONE 2008;3(8):e3049. https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0003049
Zeleke AA, Worku AG, Demissie A, Otto-Sobotka F, Wilken M, Lipprandt M, Tilahun B, Röhrig R. “Evaluation of Electronic and Paper-Pen Data Capturing Tools for Data Quality in a Public Health Survey, Ethiopia.” JMIR Mhealth Uhealth 2019;7(2):e10995. https://mhealth.jmir.org/2019/2/e10995/
Pavlović I, Kern T, Miklavčič D. “Comparison of paper-based and electronic data collection process in clinical trials: Costs simulation study.” Contemporary Clinical Trials 2009;30(4):300-316. https://www.sciencedirect.com/science/article/abs/pii/S1551714409000445 (paywalled)
Lazo A, et al. “QASR Hot Topics.” FDA Center for Veterinary Medicine, Office of New Animal Drug Evaluation. Society of Quality Assurance Annual Meeting, March 2023. https://www.fda.gov/media/185616/download