Data Quality Assurance in Veterinary Surveillance Systems

By Dr. Zubair Khalid, DVM, MS, PhD · 2026-08-11

Key Takeaways

• Data quality assurance in veterinary surveillance is a continuous process, not a single event, and is essential for validating disease freedom claims, early warning systems, and effective control strategies, as mandated by WOAH standards.
• Data validation encompasses field-level checks against defined rules (e.g., date of birth preceding sampling date), record-level consistency (e.g., male animal not pregnant), and cross-record comparisons to detect anomalies.
• Completeness assessment requires defining expected records and fields, identifying missing data through multi-pass audits (counting records, checking fields, verifying conditional completeness), and understanding whether missingness is random or correlated with disease status.
• Error detection and correction workflows must be systematic, logging every change with reason, date, and responsible person, and prioritizing validation at data entry to prevent errors rather than solely correcting them retrospectively.
• Documented quality procedures, including data dictionaries, validation rule specifications, and correction logs with audit trails, are critical for transparency and acceptance by external assessors and trading partners, mirroring rigorous standards used in laboratory assay validation.
• Common failure modes like systematic under-reporting, diagnostic misclassification due to inconsistent case definition application, and data entry errors necessitate specific discriminating checks, such as comparing reporting rates between regions or verifying laboratory results against field diagnoses.

Veterinary surveillance systems generate data that inform disease detection, trade decisions, and public health policy. The value of those systems depends on the integrity of the data they produce. This article provides veterinary researchers with a structured approach to data quality assurance in surveillance, covering validation methods, completeness assessment, error detection, and correction workflows. It addresses the practical question of how a surveillance unit can verify that its records are accurate, complete, and fit for their intended purpose. Statistical analysis of surveillance data is outside the scope of this article.

Quality assurance in surveillance is not a single activity but a continuous process embedded in the operation of the system itself. The World Organization for Animal Health (WOAH) sets international standards for animal health surveillance and reporting, and those standards explicitly require that surveillance outputs be supported by documented quality procedures WOAH animal health surveillance standards. A surveillance system that cannot demonstrate the quality of its data cannot support claims of disease freedom, early warning, or effective control. The methods described here apply across species and production systems, from wildlife disease monitoring to intensive livestock operations.

At a Glance

Conceptual Foundations of Surveillance Data Quality

Surveillance data quality is defined by fitness for purpose. A dataset may be complete and internally consistent yet still fail to support the surveillance objective if the wrong variables were collected or the case definition was applied inconsistently. Quality assurance therefore begins with a clear statement of what the surveillance system is intended to detect, measure, or demonstrate. The WOAH terrestrial animal health code provides the international framework for defining surveillance objectives and the evidence standards that must be met WOAH terrestrial animal health code. Without an explicit objective, no meaningful judgment of data quality is possible.

The quality of a surveillance dataset can be assessed along several dimensions. Accuracy refers to the agreement between recorded values and true values. Completeness describes the extent to which all required records and fields are present. Timeliness captures the delay between an event and its appearance in the dataset. Consistency requires that related records do not contradict one another. Each dimension demands different checks and different corrective actions.

Quality assurance in animal disease surveillance has been described as requiring an objective, transparent, and systematic approach, with evidence of such high quality that results are acceptable to both the management of the system and external assessors quality assurance applied to animal disease surveillance systems. This standard applies whether the assessment is internal or conducted by trading partners. Repeated negotiation may be necessary when quality judgments affect trade relationships, which makes documentation of quality procedures a practical necessity instead of an administrative formality.

Data Validation Methods

Validation is the process of checking that data conform to defined rules and expectations. It operates at multiple levels. Field-level validation examines individual values against permitted ranges, formats, and code lists. Record-level validation checks relationships between fields within a single record. Cross-record validation compares related records to detect inconsistencies.

Field-Level Validation

Field-level rules are the first line of defense. A date of birth cannot fall after the date of sampling. A species code must appear in the approved list. A latitude value must fall within the geographic bounds of the surveillance area. These rules are implemented at data entry where possible, because preventing an error is cheaper than correcting it later. Where data are entered by multiple people, the rules must be identical across all entry points.

Reference standards play a role in field-level validation. For serological testing, the validation of enzyme-linked immunosorbent assay (ELISA) techniques requires defined reference standards, data expression methods, and quality assurance procedures standardization and validation of ELISA techniques. The same principle applies to any laboratory result entered into a surveillance database. A test result is only interpretable if the assay has been validated and the data expression format is standardized.

Record-Level and Cross-Record Validation

Record-level checks identify impossible combinations. An animal recorded as male cannot have a pregnancy status. A sample collected before the animal was born indicates a date error. Cross-record checks compare entries across visits, animals, or herds. A herd that was negative on a screening test should not appear in the follow-up positive list without an intervening positive result.

Validation rules must be documented in a data dictionary and updated when the surveillance protocol changes. The rules themselves should be tested against historical data to confirm that they flag genuine errors instead of legitimate variation. Overly strict rules generate false flags that erode staff confidence in the validation process.

Completeness Assessment and Missing Data Handling

Completeness is the proportion of expected records and data fields that are actually populated. A surveillance record with a case definition, species, date, and location is complete only if each of those elements is present and legible. Missing data in veterinary surveillance arise from three distinct failure modes: non-submission of reports, partial submission where some fields are left blank, and silent loss where records are transmitted but never entered into the database. Each requires a different corrective action.

The first step is to define what completeness means for each data element. A field such as "date of onset" may be genuinely unknown for a chronically ill animal, while "date of report" is always knowable at the time of submission. The WOAH animal health surveillance standards require reporting of specified diseases within defined timeframes, and completeness of those reports is a contractual obligation between the veterinary authority and the reporting veterinarian. For research databases, completeness thresholds should be set before data collection begins, with explicit rules for which fields are mandatory, which are conditional, and which are optional.

A practical completeness audit proceeds in three passes. The first pass counts records against an external denominator, such as the number of registered premises, the number of laboratory accessions, or the number of slaughter batches inspected. The second pass examines field-level completion rates within the records that do exist. The third pass checks conditional completeness, for example whether every record flagged as "died" has a date of death and whether every positive test result has an associated laboratory accession number.

When missing data are found, the response depends on the mechanism of loss. Missingness that is random with respect to disease status can often be tolerated for descriptive purposes, though it reduces statistical power. Missingness that is correlated with disease status, such as farms with poor record keeping also having higher disease prevalence, introduces bias that cannot be corrected by imputation. The quality assurance methods for animal disease surveillance systems described by Salman and colleagues emphasize that the assessment approach must match the objective of the evaluation, and this applies directly to missing data. If the objective is to certify freedom from disease, even small amounts of non-random missingness may invalidate the conclusion.

Documentation of missing data should include the proportion missing per field, the pattern of missingness across time and geography, and any actions taken to recover missing values. A data quality audit checklist should record whether the missing data team contacted the submitting practice, whether the practice had a reasonable explanation, and whether the missing value was subsequently corrected or permanently flagged as unavailable.

Error Detection and Correction Workflows

Once data are validated and completeness is assessed, the remaining task is to detect and correct errors that passed through the earlier checks. Errors in veterinary surveillance databases fall into recognizable categories. Transcription errors occur when a handwritten laboratory form is entered into an electronic system. Coding errors occur when a diagnosis is assigned an incorrect code, such as using a clinical sign code instead of a confirmed diagnosis code. Unit errors occur when measurements are recorded in the wrong scale, for example milligrams per liter instead of micrograms per liter. Date errors include impossible dates, dates in the future, and dates that contradict other fields in the same record.

The correction workflow should follow a defined sequence. First, the error is logged with a description of what was found and where. Second, the original source document is retrieved and the correct value is confirmed. Third, the correction is applied to the database with a timestamp and the identity of the person making the correction. Fourth, the corrected record is re-validated to ensure the correction did not introduce a new error. This sequence is analogous to the standardization and validation procedures described for ELISA testing, where the FAO and IAEA consultant group recommended explicit procedures for data expression and quality assurance to achieve international conformity. The same principle of documented, repeatable procedures applies to data correction.

A critical decision point in error correction is whether to correct the record or to exclude it. Correction is appropriate when the true value can be determined with confidence from the source document. Exclusion is appropriate when the error cannot be resolved, when the record is a duplicate, or when the error affects a field that is essential to the analysis. Excluded records should be retained in a separate quarantine table with the reason for exclusion, instead of deleted outright. This preserves the ability to re-examine exclusions if the analysis plan changes.

Automated error detection can be implemented at the point of data entry. Range checks reject values outside biologically plausible limits, such as a body temperature of 45 degrees Celsius in a cow. Consistency checks compare fields within a record, such as verifying that the date of death is not earlier than the date of onset. Cross-record checks identify duplicates by matching on premises identifier, species, and date. The MSD Veterinary Manual provides species-specific reference ranges that can inform the upper and lower bounds used in range checks, though the manual is a clinical reference and not a surveillance standard, so limits should be derived from the surveillance population itself where possible.

Data Quality Audit Checklist

A structured audit checklist provides a repeatable method for assessing data quality at regular intervals. The checklist below is organized by the stages of the data lifecycle and can be adapted to the scale of the surveillance system, from a single veterinary practice to a national program.

The thresholds in the table are starting points for discussion instead of universal standards. A research database supporting a clinical trial may require 100% completeness for primary outcome fields, while a passive surveillance system relying on voluntary reporting may accept lower completeness for non-mandatory fields. The WOAH terrestrial animal health code sets out reporting obligations for listed diseases, and those obligations define the minimum completeness expectations for official reporting. Systems that feed international trade decisions must meet the standards of the importing country, and the audit should verify that the data would withstand scrutiny by an external assessor.

The audit should be conducted at a frequency proportional to the risk of data degradation. Systems with manual data entry and multiple staff members warrant monthly audits. Fully electronic systems with automated validation may be audited quarterly. After any major system change, such as a new laboratory information system or a change in case definitions, an immediate audit is warranted regardless of the routine schedule.

Documentation and Audit Trails

Every data quality activity should leave a trace that allows an external reviewer to reconstruct what was done. The audit trail records who entered each record, when it was entered, when it was modified, and what the modification was. This is also a technical requirement. The quality assurance framework for animal disease surveillance described by Salman and colleagues notes that well-documented systems with specified objectives and integrated quality assurance mechanisms are easier to evaluate and more likely to be accepted by trading partners. An undocumented correction is indistinguishable from a fabrication.

The documentation set for a surveillance database should include a data dictionary, a validation rule specification, a completeness assessment protocol, and a correction log. The data dictionary defines every field, its type, its permitted values, and its relationship to other fields. The validation rule specification lists every automated check and the action taken when the check fails. The completeness assessment protocol describes the three-pass audit described above and records the results of each audit. The correction log records every change to existing records, including the original value, the corrected value, the reason for the change, and the person who authorised it.

Species and production system differences affect documentation requirements. A surveillance system covering wildlife relies on opportunistic sampling where completeness may be inherently low, and the documentation must distinguish between "not present" and "not looked for". A system covering intensively managed poultry may have complete denominator data from hatchery records, while a system covering backyard poultry in a low-resource setting may have no reliable denominator at all. The assessment of pathogen genomic surveillance capacity in lower resource settings in Asia identified limited quality assurance mechanisms as a barrier to effective surveillance, and the same limitation applies to conventional surveillance data. In settings where laboratory information systems are unreliable, paper-based logs with periodic electronic entry may provide a more honest audit trail than a partially automated system that silently drops records.

The level of documentation should match the intended use of the data. Data used only for internal monitoring may require a lighter documentation burden than data submitted to an international organization or used in a regulatory decision. However, the cost of documentation is small compared to the cost of discovering, after an outbreak investigation or a trade dispute, that the data cannot be defended.

Recognized Complications and Failure Modes

Surveillance data quality degrades through identifiable failure modes that often interact. The most consequential is systematic under-reporting, where submitting practices or laboratories omit cases because reporting competes with clinical workload. This produces incidence estimates that are biased downward, and the bias is rarely constant across regions or time periods. Early detection relies on comparing reporting rates between similar administrative units and monitoring submission volumes against expected baselines derived from historical data.

A second failure mode is diagnostic misclassification, which arises when case definitions are applied inconsistently. This is particularly problematic in syndromic surveillance, where clinical signs overlap across aetiologies. The discriminating check is to compare the distribution of final diagnoses against the distribution of initial syndromic classifications, large discrepancies indicate that the case definition or its application requires revision.

Data entry errors constitute a third category. Transcription errors in animal identification numbers, dates, or laboratory results propagate silently into downstream analyzes. Detection requires systematic cross-validation between the original submission form and the electronic record, ideally performed on a sampled basis when full verification is impractical.

The table below summarizes common observations, their likely causes, and the discriminating checks that distinguish between them.

Common Errors and Corrective Actions

Less experienced personnel tend to confuse data validation with data cleaning. Validation is the prospective process of checking data against defined rules at the point of entry. Cleaning is the retrospective correction of errors already present. Performing cleaning without validation allows errors to accumulate and forces repeated correction cycles. The corrective action is to implement validation rules at the earliest point of data capture, whether that is a web form, a mobile application, or a laboratory information system.

A second recurring error is the treatment of missing data as equivalent to absence of disease. In surveillance, a missing record for a premises does not mean the premises had no cases. It means the surveillance system did not observe that premises. Analysts and clinicians alike must distinguish between "not reported" and "not present". The corrective action is to document the reason for missingness wherever possible, using codes that distinguish between "no case occurred", "case occurred but not reported", and "reporting status unknown".

A third error involves the inappropriate reuse of reference standards. Assay validation parameters, including diagnostic sensitivity and specificity, are specific to the population and laboratory conditions under which they were established. Applying validation data from one production system to another without re-evaluation introduces systematic misclassification. The corrective action is to revalidate or at least verify assay performance when the target population changes materially, following the standardization principles described in the FAO and IAEA guidance on ELISA validation standardization and validation of ELISA techniques for antibody detection.

Limitations of Current Evidence

The evidence base for veterinary surveillance data quality is uneven. Published guidance on quality assurance frameworks exists, particularly from the World Organization for Animal Health, which sets international standards for surveillance and reporting WOAH animal health surveillance standards. However, empirical studies quantifying the impact of specific quality interventions on surveillance outcomes are scarce. Most published work describes methods and frameworks instead of controlled evaluations.

Expert opinion differs on the appropriate rigour of validation for different surveillance purposes. For early warning systems, speed of reporting may justify accepting lower per-record completeness, provided that critical fields such as location and case status are complete. For trade-related surveillance, where the absence of disease must be demonstrated, the standard of evidence is necessarily higher, and the requirements are set out in the terrestrial animal health code WOAH terrestrial animal health code. These differing purposes imply different quality thresholds, and applying a single standard across all surveillance activities is neither feasible nor appropriate.

Genomic surveillance introduces additional quality considerations. In lower-resource settings, limited quality assurance mechanisms have been identified as a barrier to the effective use of pathogen genomics in surveillance pathogen genomic surveillance status among lower resource settings in Asia. The evidence base for genomic data quality standards is still developing, and consensus on minimum quality metrics for sequence data used in veterinary surveillance has not yet been reached.

Escalation and Referral Criteria

Certain observations warrant escalation beyond routine quality assurance procedures. When validation checks reveal systematic errors affecting more than a small proportion of records from a single source, the submitting organization should be contacted directly. If the pattern persists after feedback, the issue should be referred to the program coordinator, who may decide to suspend data submissions from that source until corrective action is verified.

Laboratory involvement is indicated when discrepancies arise between field observations and laboratory results. This may reflect sample degradation, transport delays, or assay problems. The laboratory should be consulted to review its internal quality control data before field-side explanations are accepted.

Regulatory reporting obligations are triggered by specific findings, not by data quality problems themselves. However, data quality failures can compromise the ability to meet reporting obligations. If records are incomplete or unreliable, the responsible authority must be informed that the surveillance system cannot currently support a statement of freedom from disease or a reliable estimate of disease prevalence. This notification should occur before any official declaration is made, not after.

Referral to a specialist epidemiologist is warranted when missing data mechanisms are complex, when validation results suggest bias that cannot be corrected by simple cleaning, or when the surveillance system must support trade-related claims. The epidemiologist can advise on appropriate analytical methods and on the documentation required to support the surveillance conclusions.

Frequently Asked Questions

How much does a formal data quality assurance program cost, and how do I justify it to my institution?

Costs scale with system complexity, but the largest expenditures are usually personnel time for validation and correction, not software. A program built on existing spreadsheets and laboratory information systems can start with a defined validation protocol and a designated data steward. Justification should focus on the cost of poor data: false negative surveillance results delay detection, while false positives trigger unnecessary regulatory action and trade restrictions. Reliable surveillance data also supports claims of disease freedom, which is a requirement for international reporting under WOAH animal health surveillance standards. Frame the program as insurance against these downstream costs instead of as an administrative burden.

What should I do when the ideal validation software or automated checks are not available?

Manual validation is slower but can achieve comparable rigour if structured systematically. Use paper-based or spreadsheet checklists that mirror the automated rules you would implement: range checks, mandatory field lists, and cross-field consistency checks. Assign one person to perform entry and a second to verify a sample of records, ideally ten percent or more when error rates are high. Document every correction on a paper log with date, initials, and reason. The CDC principles of epidemiology in public health practice describe systematic data handling procedures that work without specialised software. Prioritize validation effort on fields that drive case classification and reporting decisions, since errors there have the greatest consequences.

Does the approach to data quality assurance differ between companion animal and livestock surveillance systems?

The core principles are identical, but the practical constraints differ. Livestock systems usually aggregate data at herd or flock level, so completeness checks must account for denominator uncertainty: you may know how many animals were tested but not the true population at risk. Companion animal systems typically record individual patient data, which makes record-level validation more straightforward but introduces owner-identifiable information and corresponding privacy obligations. Production animal surveillance often feeds directly into trade certification, so documentation standards must satisfy importing country requirements under WOAH terrestrial animal health standards. Companion animal data more frequently supports zoonotic disease detection, where timeliness of reporting may outweigh exhaustive validation.

How should I handle data quality problems that I cannot fix at my level?

Escalate when you detect systematic errors that suggest a process failure instead of isolated mistakes. This includes recurring misclassification of a common condition, persistent missing data from one submitting clinic, or laboratory results that consistently fail range checks. Document the pattern with specific examples and the time period affected, then report through your institutional chain to the person responsible for the surveillance system design. Do not silently correct recurring errors, since masking the pattern prevents identification of the underlying cause. If the issue affects official reporting, inform the relevant veterinary authority promptly. The quality assurance methods for animal disease surveillance systems described by Salman and colleagues emphasize that unresolved quality problems must be visible to system management, not absorbed by individual operators.

What records should I keep to demonstrate that my surveillance data are trustworthy?

Maintain a validation log that records every check performed, the date, the operator, and the outcome. Keep a separate correction log that documents each change made to a record, including the original value, the corrected value, the reason, and the authorisation. Retain case definitions and standard operating procedures in versioned documents so that classification changes over time are traceable. Store raw laboratory outputs separately from the cleaned dataset so that any correction can be audited against the original result. This documentation supports external audit and provides the transparency required for international reporting under WOAH animal health surveillance standards. Records should be retained for at least the period required by your jurisdiction or funding agreement, whichever is longer.

How do I explain data quality limitations to a client, producer, or non-technical supervisor without undermining confidence in the surveillance system?

Be direct about what is known and what is uncertain. State that the system detected a specific problem, describe the affected data fields and time period, and explain the impact on interpretation in concrete terms. For example, say that missing treatment records for three months mean the farm's antimicrobial use figures are underestimates, not that the data are unusable. Distinguish between data that are incomplete and data that are wrong, since the corrective actions differ. Offer a clear plan for correction and a timeline. The application of quality assurance principles at farm level described by Noordhuizen and Frankena shows that producers respond well to structured quality control approaches such as HACCP, which frame data problems as process issues to be managed instead of failures to be hidden.

Related Clinical & Scientific Guides

• Evaluating Veterinary Surveillance System Attributes
• Network Analysis for Infectious Disease Spread in Animal Populations
• Randomized Controlled Trials in Veterinary Field Settings

References and Further Reading

• Standardization and validation of enzyme-linked immunosorbent assay techniques for the detection of antibody in infectious disease diagnosis.. 1993.
• Quality assurance applied to animal disease surveillance systems.. 2003.
• Surveillance and diagnosis of zoonotic foodborne parasites.. 2018.
• Epidemiology and quality assurance: applications at farm level.. 1999.
• Antibiotic drug usage in pigs in Germany-Are the class profiles changing?. 2017.
• Pathogen genomic surveillance status among lower resource settings in Asia.. 2024.
• WOAH Animal Health Surveillance Standards. WOAH.
• CDC Principles of Epidemiology in Public Health Practice. CDC.
• MSD Veterinary Manual, Professional Edition. MSD Veterinary Manual.

Related Articles

• Designing Case Definitions for Veterinary Surveillance
• Designing and Implementing Animal Disease Surveillance Systems
• Designing Participatory Disease Surveillance in Livestock Systems
• One Health Surveillance: Integrating Human, Animal, and Environmental Data
• Syndromic Surveillance in Veterinary Practice

This article is educational professional reference material for veterinary audiences. It is not a substitute for veterinary diagnosis, individual clinical judgment, current product labeling, or applicable regulatory requirements.