Cluster Randomized Trials in Veterinary Research: Design and Analysis

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

Key Takeaways

• Cluster randomized trials (CRTs) are essential in veterinary research when interventions are applied at the group level (e.g., herd health programs, biosecurity protocols) making individual randomization infeasible or scientifically invalid due to shared environments or social structures.
• The Intracluster Correlation Coefficient (ICC) quantifies the correlation of outcomes within clusters (e.g., herds, pens) and is a critical driver for sample size calculations, as it inflates the required number of animals compared to individually randomized trials.
• Analysis must explicitly account for clustering to avoid underestimating standard errors and inflating Type I error rates; preferred methods include mixed-effects models, generalized estimating equations, or cluster-level summary analyses.
• Common design errors include randomizing too few clusters, ignoring clustering in analysis, contamination between control and intervention groups, and a mismatch between the unit of randomization and analysis, all of which compromise the validity and generalizability of findings.
• Reporting standards, such as the CONSORT extension for CRTs and the ARRIVE guidelines, are crucial for transparency, requiring detailed reporting of the randomization unit, cluster sizes, allocation concealment, ICC estimation, and analytical methods used to account for clustering.
• Species-specific considerations are vital, with companion animal trials often having many small clusters (households, kennels) while production animal trials may have fewer, larger clusters (herds, farms), impacting design choices and statistical power.

Cluster randomized trials occupy a distinct position in the veterinary evidence hierarchy. When interventions are delivered at the group level, such as a herd health program, a biosecurity protocol applied to an entire production facility, or a community-level zoonosis control campaign, individual randomization is either logistically impossible or scientifically invalid because the intervention inherently operates on the social or physical environment shared by animals within a unit. This article provides a structured account of the design decisions, analytical methods, and reporting standards that govern cluster randomized trials in veterinary research. It serves the veterinary researcher who must appraise such studies critically or design one that will withstand peer review and regulatory scrutiny.

The central question this article answers is how to obtain an unbiased estimate of an intervention effect when the unit of assignment differs from the unit of analysis. That distinction carries consequences for sample size, inference, and interpretation that are frequently underestimated. The article covers the logic of cluster randomization, the intracluster correlation coefficient and its estimation, sample size calculation under clustering, analytical approaches from simple adjustment to mixed-effects modeling, and the reporting standards that apply. Cross-species examples are used throughout because the principles transfer across companion animal, livestock, and wildlife research contexts.

At a Glance

The Rationale for Cluster Randomization

Cluster randomized trials assign intact groups of animals to intervention or control conditions. The justification for this design is usually one of three forms. First, the intervention may be delivered at the group level by necessity. A ventilation modification, a feed additive incorporated at the mill, or a vaccination program administered by a single veterinary team cannot be applied to some animals within a pen while withholding it from others without contamination. Second, the intervention may target the environment or the social structure of the group, so that individual-level assignment would dilute or distort the mechanism of action. Third, practical constraints such as cost, farmer compliance, or the risk of behavioral contamination between animals in close contact may make individual randomization infeasible.

The BOHEMIA trial in Mozambique illustrates the first rationale in a One Health context. That study assigned entire geographic areas to receive ivermectin mass drug administration to humans, livestock, or both, with the outcome being malaria transmission. The intervention could not plausibly be delivered to selected households within a village while excluding neighboring households, because mosquito movement and livestock management practices cross property boundaries. The demographic survey conducted before randomization mapped over 25,000 households and registered more than 131,000 individuals, which provided the sampling frame for cluster identification and baseline characterization Ruiz-Castillo et al., institutional publication.

Veterinary examples follow the same logic. A study of sanitation interventions in rural Bangladesh assigned entire compounds to receive improved toilets and child feces management tools, then measured fecal contamination markers on hands, soil, and drinking water. The investigators detected ruminant and avian fecal markers that scaled with animal ownership, demonstrating that the compound environment was shared across species and that environmental contamination could not be attributed to individual households or individual animals within a compound Boehm et al., institutional publication. Randomizing individual animals within such an environment would have been meaningless because the intervention altered the shared environment.

The Intracluster Correlation Coefficient

The statistical consequence of cluster randomization is that outcomes for animals within the same cluster are correlated. Animals in the same pen share pathogens, management, genetics, and environmental exposures. This correlation is quantified by the intracluster correlation coefficient, denoted ρ or ICC. The ICC is defined as the proportion of total outcome variance that lies between clusters. An ICC of zero means no clustering effect and outcomes within a cluster are as variable as outcomes across the whole population. An ICC approaching one means all animals within a cluster respond identically and all variability is between clusters.

The ICC is not a nuisance parameter to be estimated and discarded. It determines the design effect, also called the variance inflation factor, which is calculated as 1 + (m, 1)ρ, where m is the average cluster size. A trial with an average cluster size of 20 animals and an ICC of 0.05 has a design effect of 1.95, meaning the required sample size nearly doubles compared with an individually randomized trial. At an ICC of 0.10 with the same cluster size, the design effect is 2.9. The design effect multiplies the sample size that would be required under individual randomization.

Published ICC values for veterinary outcomes vary widely by species, outcome type, and cluster definition. Growth outcomes in livestock production systems often show ICCs between 0.05 and 0.15. Infectious disease outcomes can show substantially higher values because transmission creates strong within-cluster dependence. When no prior estimate exists, the researcher should report the assumed ICC explicitly and conduct sensitivity analyzes across a plausible range. The cluster randomized trial of caterpillar cereal as a complementary food in the Democratic Republic of Congo illustrates the practical consequence of clustering: 175 infants were enrolled across clusters, and the analysis detected a significant effect on hemoglobin concentration but no effect on stunting, a pattern that reflects both the intervention's true effects and the precision available under cluster assignment Bauserman et al., institutional publication.

Sample Size Determination

Sample size calculation for a cluster randomized trial requires four inputs: the expected effect size, the within-cluster variance, the ICC, and the average cluster size. The number of clusters is often the binding constraint. Trials with few clusters per arm have limited degrees of freedom for the cluster-level comparison, and increasing the number of animals within existing clusters yields diminishing returns once the design effect has been accounted for.

The Kenyan school snack trial randomized only twelve schools, three per arm, and followed 902 children across two cohorts Neumann et al., institutional publication. With three clusters per arm, the trial had limited power to detect modest differences, and the authors appropriately emphasized the descriptive pattern of morbidity declines across feeding groups instead of relying solely on pairwise significance tests. This trial demonstrates a critical design principle: the number of clusters, not the number of animals, usually determines statistical power. Researchers planning cluster trials should aim for at least four to six clusters per arm when feasible, and should recognize that trials with fewer clusters require analytical methods that accommodate the small number of independent units.

Analytical Approaches

Three analytical strategies are available for cluster randomized trials, each with distinct assumptions and use cases. The cluster-level analysis treats the cluster summary statistic, such as the mean outcome or the proportion of affected animals, as the unit of analysis. This approach is simple, robust, and appropriate when the number of clusters is small, but it loses power when cluster sizes vary substantially and cannot easily adjust for individual-level covariates.

The mixed-effects model, also called the hierarchical or multilevel model, includes cluster as a random effect and individual animals as fixed-effect units. This approach retains individual-level data, allows adjustment for covariates at both the animal and cluster levels, and provides valid inference when the number of clusters is adequate, generally considered at least ten to fifteen per arm. The generalized estimating equation approach provides population-averaged estimates and is robust to misspecification of the correlation structure, making it attractive for binary outcomes.

The choice among these methods should be made at the design stage, not after data collection. The analysis plan must specify the primary outcome, the method of estimation, and the handling of missing data. The reporting of the trial must follow the CONSORT extension for cluster randomized trials, which requires specification of the randomization unit, the methods used to account for clustering in the analysis, and the ICC estimate. The EQUATOR Network maintains the library of reporting guidelines that includes this extension and related standards for observational and diagnostic studies EQUATOR Network reporting guidelines.

Reporting and Transparency Standards

Cluster randomized trials in veterinary research face a persistent problem: the design features that distinguish them from individually randomized trials are often underreported. Readers cannot assess the validity of a trial if the number of clusters, the cluster sizes, the method of allocation concealment, and the intracluster correlation coefficient used in the sample size calculation are missing from the manuscript. The EQUATOR Network reporting guidelines maintain a library of reporting standards that includes the CONSORT extension for cluster randomized trials, and veterinary journals increasingly require adherence to these standards. The ARRIVE guidelines 2.0 for reporting animal research specify the minimum information required for transparent and reproducible publications, and their checklist items apply directly to cluster trials in veterinary settings.

For veterinary cluster trials specifically, the REFLECT statement (Reporting guidelines For randomized controlled trials in livestock and food safety) extends CONSORT to accommodate the realities of production animal research, including the frequent need to randomize at the herd or pen level while measuring outcomes at the individual animal level. Authors should report the number of clusters allocated to each arm, the number lost to follow-up at both the cluster and individual levels, and the analytic method used to account for clustering. A table showing baseline characteriztics by trial arm should present cluster-level summaries, not individual-level summaries, because individual-level baseline tables can mislead when clustering is present.

Common Design Errors and Their Consequences

Several recurring errors undermine the validity of cluster randomized trials in veterinary research. The first is randomizing too few clusters. A trial with two herds per arm provides no reliable estimate of the between-cluster variance, and the trial's results cannot be generalized beyond the specific herds studied. The BOHEMIA cluster randomized trial in Mozambique illustrates the scale of infrastructure required for a rigorous cluster trial, with 25,550 households mapped and 131,818 individuals registered before randomization. Veterinary trials rarely reach this scale, but the principle holds: the number of clusters, not the number of animals, drives statistical power.

The second common error is ignoring clustering in the analysis. Analyzing individual animals as independent observations when they are grouped within herds, pens, or kennels produces artificially narrow confidence intervals and inflated type I error rates. A trial that reports a significant treatment effect without accounting for clustering should be viewed with suspicion, particularly when the intracluster correlation coefficient is likely to be substantial.

The third error is contamination between clusters. In veterinary field trials, herds on adjacent premises may share personnel, equipment, or grazing land, and a control herd may inadvertently receive the intervention. The cluster randomized trial of caterpillar cereal as a complementary food in the Democratic Republic of Congo randomized villages instead of individual infants specifically to avoid contamination between intervention and control groups, a design choice that reflects the practical reality that feeding practices are household and community behaviors. Veterinary researchers should consider the biologic and behavioral pathways by which an intervention could spread to control clusters and choose the cluster definition accordingly.

The fourth error is a mismatch between the unit of randomization and the unit of analysis. When herds are randomized but animals are the unit of analysis, the analysis must account for the hierarchical structure. Failure to do so is the most common analytical error in published cluster trials.

Worked Example: A Herd-Level Mastitis Intervention

Consider a researcher planning a trial of a novel dry cow therapy protocol in dairy herds. The intervention is applied at the herd level because the treatment protocol is implemented by farm staff, and the outcome is the incidence of clinical mastitis in the first 30 days of the next lactation, measured in individual cows.

The researcher identifies 40 eligible herds and randomizes 20 to the intervention protocol and 20 to the standard protocol. The average herd size is 120 lactating cows, and the researcher expects a baseline mastitis incidence of 15% in the control herds. From previous studies, the intracluster correlation coefficient for mastitis within herds is estimated at 0.05.

The design effect is calculated as 1 + (m - 1) × ICC, where m is the average cluster size. With m = 120 and ICC = 0.05, the design effect is 1 + (119 × 0.05) = 6.95. The required sample size under individual randomization is multiplied by this design effect. If individual randomization would require 200 cows per arm, the cluster trial requires approximately 1,390 cows per arm, or about 12 herds per arm. The planned 20 herds per arm provides a margin for herd-level loss to follow-up.

The analysis plan should specify a mixed-effects logistic regression model with herd as a random effect, or a generalized estimating equation with an exchangeable correlation structure. The primary analysis should be by intention to treat, with herds analyzed according to their allocated protocol regardless of compliance. A secondary per-protocol analysis may be informative, but it should be clearly labeled as such and interpreted with caution because noncompliance is rarely random.

The researcher should also plan a sensitivity analysis using different values of the ICC. If the true ICC is 0.10 instead of 0.05, the design effect rises to 12.9 and the required number of herds per arm increases to approximately 22. Reporting the power of the trial across a plausible range of ICC values gives readers a more honest picture of the trial's ability to detect a treatment effect than a single point estimate.

Species-Specific and Setting-Specific Considerations

The correct choices in cluster trial design depend heavily on the species, production system, and research setting. In companion animal practice, clusters are typically households or kennels, and cluster sizes are small, often one to three animals. The design effect is correspondingly modest, but the number of available clusters may be large, which favors trials with many small clusters. In contrast, production animal trials often have large clusters with hundreds or thousands of animals, and the design effect can be substantial even when the ICC is small.

The school snack trial in Kenyan schoolchildren randomized twelve schools to four arms, with three schools per arm, and followed two cohorts of children within those schools over two years. The small number of clusters per arm limited the trial's statistical power, and the authors acknowledged this constraint. Veterinary researchers working with a limited number of large herds face the same constraint and should consider whether a crossover design, a stepped-wedge design, or an individually randomized design is feasible before committing to a cluster trial with an inadequate number of clusters.

The sanitation trial in rural Bangladesh demonstrates the importance of measuring outcomes at the correct level. The trial randomized compounds to improved sanitation and measured fecal contamination in drinking water, child hands, and soil, with host-associated genetic markers distinguishing human from ruminant and avian sources. The finding that non-human fecal contamination scaled with animal ownership illustrates a broader lesson for veterinary cluster trials: outcomes may be influenced by factors outside the cluster definition, and the trial should measure and account for these factors.

Aquatic species present particular challenges. Fish are housed in tanks or ponds that function as clusters, but water quality parameters such as temperature, oxygen, and ammonia can vary substantially between tanks and over time. The ICC for growth and mortality outcomes in fish is often high because fish within a tank share a common environment and genetic background. Researchers should measure water quality parameters at the tank level and consider them as covariates in the analysis.

The WOAH terrestrial animal health standards are relevant when cluster trials are conducted to support disease control programs or trade decisions. Trials evaluating vaccination strategies, biosecurity protocols, or diagnostic surveillance programs may need to align with international standards for disease freedom and surveillance, and the cluster design must accommodate the regulatory requirements of the relevant jurisdictions.

Documentation and Data Management

Cluster trials generate data at multiple levels, and the data management plan must preserve the hierarchical structure throughout the study. Each animal should have a unique identifier that links to its cluster identifier, and the cluster identifier must be recorded in every data file. The demographic survey conducted for the BOHEMIA trial assigned each household a unique identification number and geolocated every household, creating a permanent sampling frame for the trial. Veterinary trials should adopt a similar approach, with herd or pen identifiers recorded at enrollment and maintained through analysis.

The trial protocol should specify the stopping rules for cluster-level adverse events, the procedures for handling herds that withdraw from the trial, and the methods for monitoring data quality at both the cluster and individual levels. A data monitoring committee may be appropriate for trials with high morbidity or mortality outcomes, particularly in production animal settings where the intervention could affect animal welfare or food safety.

The AVMA practice resources provide general guidance on professional standards for veterinary research and practice, and the MSD Veterinary Manual offers species-specific clinical reference material that can inform outcome definitions and measurement protocols. Neither source provides specific guidance on cluster trial design, but both are useful for ensuring that the clinical outcomes measured in a cluster trial are meaningful and standardized across participating sites.

Recognized Complications and Early Detection

Cluster randomized trials in veterinary settings fail most often through contamination, cluster attrition, and secular drift. Contamination occurs when control clusters receive the intervention through animal movement, shared personnel, or owner crossover. In production animal studies, neighbouring herds sharing grazing land, water sources, or veterinary staff can transmit treatment effects across allocation boundaries. Detect contamination early by recording proximity between clusters, documenting movement events, and monitoring a sham or placebo outcome that should remain unaffected if allocation integrity holds.

Cluster attrition differs from individual dropout because losing a cluster removes many observations at once and often correlates with the outcome of interest. A herd that leaves because of an outbreak, a financial collapse, or a change in ownership contributes missing data that are not missing at random. Track retention at each follow-up wave, compare baseline characteriztics of retained and lost clusters, and pre-specify the threshold of cluster loss that would compromise the trial's interpretability.

Secular drift arises when external conditions change during the trial period. Feed prices, weather patterns, disease circulation, or regulatory shifts can alter the outcome in ways unrelated to the intervention. A concurrent control group protects against drift only if both arms experience the same external forces. Detect drift by plotting outcome trends over calendar time within each arm and by recording major external events in a trial log.

Common Errors and Corrective Action

Less experienced investigators frequently analyze cluster trials as though observations were independent. This error inflates precision, narrows confidence intervals, and produces false-positive conclusions. The corrective action is to fit a mixed-effects model with a random intercept for cluster or to use generalized estimating equations with a cluster-robust variance estimator. Report the intracluster correlation coefficient so readers can assess the magnitude of the clustering effect.

A second error is randomising too few clusters. Trials with fewer than four clusters per arm cannot support reliable variance estimation, and the analysis will be sensitive to the choice of statistical method. The corrective action is to plan for at least four clusters per arm and to consider whether the research question can be answered with a different design if cluster numbers are constrained.

A third error is ignoring the distinction between the unit of allocation and the unit of analysis. Some investigators randomise herds but analyze animals without accounting for the herd effect, while others randomise animals within herds but analyze at the herd level, losing power unnecessarily. The corrective action is to specify both units explicitly in the protocol and to ensure the analysis model reflects the hierarchy.

Limitations of the Current Evidence

The veterinary cluster trial literature remains sparse compared with human medicine. Many published trials in livestock settings use small numbers of clusters, short follow-up periods, and outcome measures that are surrogate instead of clinical. The reporting standards catalogued by the EQUATOR Network include extensions for cluster trials, but uptake in veterinary journals has been uneven. The ARRIVE guidelines for animal research reporting address transparency in animal studies generally but do not provide cluster-specific guidance.

Expert opinion still differs on several points. There is no consensus on the minimum number of clusters required for reliable inference when using generalized estimating equations, with recommendations ranging from ten to forty depending on the outcome distribution and the strength of clustering. Analysts also disagree on whether to use cluster-level summaries or individual-level models when the number of clusters is small. The cluster-level approach is more conservative and more robust, but it discards information and can be underpowered. Individual-level models are more efficient but rely on asymptotic approximations that may fail with few clusters.

Referral, Consultation, and Regulatory Reporting

Veterinary researchers should seek specialist consultation when the trial design involves complex hierarchies, such as animals nested within pens within barns within sites, or when the outcome requires sophisticated modeling such as competing risks or survival analysis. A statistician with experience in clustered data should be engaged before randomisation, not after data collection.

Laboratory involvement is warranted when outcome ascertainment depends on diagnostic testing. The MSD Veterinary Manual professional edition provides species-specific guidance on sample collection and interpretation, but the trial protocol must specify laboratory procedures, quality control, and blinding of laboratory staff to allocation status.

Regulatory reporting obligations vary by jurisdiction and species. Trials involving investigational veterinary products, food-producing animals, or notifiable diseases may require notification to competent authorities. The World Organization for Animal Health terrestrial animal health standards define reporting obligations for listed diseases, and the American Veterinary Medical Association practice resources summarize professional obligations in the United States. Researchers must determine the applicable requirements in their own jurisdiction before the trial begins.

Troubleshooting Table

Frequently Asked Questions

How Many Clusters Do I Need When the Number of Animals per Cluster Is Fixed by the Practice or Herd Structure?

When cluster size is fixed, you must recruit more clusters to compensate for a high intracluster correlation coefficient (ICC). A small number of large clusters can leave you underpowered because the effective sample size is the number of clusters divided by the design effect, not the total animal count. Use a sample size calculator that accepts a range of plausible ICC values and run a sensitivity analysis. If the number of available clusters is genuinely limited, consider a stepped-wedge design, which assigns clusters to the intervention at different time points and can improve power relative to a parallel design with the same cluster count. Consult a veterinary epidemiologist or statistician before finalising the protocol.

What Can I Do When Randomization at the Intended Cluster Level Is Not Feasible in the Field?

When true cluster randomization is impossible, the next best option is a quasi-experimental design with carefully matched control clusters. Document every reason for the deviation and the steps taken to reduce selection bias. Use baseline covariates in the analysis to adjust for measured differences between clusters. Report the design transparently so readers can judge the risk of bias. The ARRIVE guidelines require clear descriptions of allocation methods and any departures from the planned design. If the intervention is delivered at the individual level but contamination between animals is the concern, consider a cluster crossover design where each cluster receives both treatments in separate periods, provided the intervention has no carryover effect.

How Do I Handle Missing Outcome Data When Entire Clusters Are Lost to Follow-Up?

Whole-cluster loss is a serious threat to validity because it is rarely random. First, compare the baseline characteriztics of lost and retained clusters. If they differ, the missingness is informative and a complete-case analysis will be biased. Use multiple imputation at the cluster level, or fit a mixed-effects model that can accommodate missing data under a missing-at-random assumption. Report the number of clusters lost and the reasons in the participant flow diagram. Sensitivity analyzes should include a worst-case scenario where all missing clusters are assumed to have failed. The EQUATOR Network hosts reporting checklists that specify how attrition must be documented, and reviewers will expect this level of detail.

Does the Analysis Differ When the Intervention Is Applied to Animals but the Outcome Is Measured in Humans or the Environment?

Yes. The unit of randomization, the unit of intervention delivery, and the unit of outcome measurement can all differ. In a trial where livestock receive an endectocide and the outcome is malaria in nearby humans, the analysis must account for clustering at the household or village level, as demonstrated in the BOHEMIA trial baseline demographic survey. Similarly, a sanitation intervention randomised at the compound level may measure fecal contamination in water, soil, and on child hands, as reported in a cluster-randomized sanitation trial in rural Bangladesh. The analytical model must include random effects for every level at which outcomes are correlated, and the interpretation should state clearly which population the inference applies to.

How Should I Budget for the Additional Costs of a Cluster Randomized Trial?

Cluster trials are often more expensive per animal than individually randomized trials because you must enrol entire herds or flocks, which increases monitoring and data management costs. Budget for a dedicated data manager, because the hierarchical data structure creates complex tracking requirements. Travel costs are higher when clusters are geographically dispersed. Laboratory costs may rise if you need to collect samples from all animals in a cluster to estimate the ICC accurately. Consider a pilot phase to measure recruitment rates and the ICC before committing to the full trial. The AVMA practice resources provide general guidance on research budgeting and practice-based study logistics, though specific cost estimates will depend on your species and setting.

How Do I Explain the Results of a Cluster Trial to a Producer or Practice Owner?

Translate the effect estimate into the metric that matters to the decision maker. An odds ratio or adjusted mean difference is less useful than the number of herds needed to treat or the expected reduction in disease incidence per herd per year. Explain that the confidence interval reflects uncertainty at the herd level, also the animal level, and that results may not apply to their herd if their management differs from the trial population. Be explicit about the intervention's cost per animal and the expected benefit. The MSD Veterinary Manual offers species-specific background on disease prevalence and production impacts that can help contextualise the findings for a particular operation.

Related Clinical & Scientific Guides

• Conducting Systematic Reviews of Veterinary Diagnostic Test Accuracy
• Bias in Veterinary Research: Types, Sources, and Mitigation
• Performing Economic Evaluations of Veterinary Interventions

References and Further Reading

• BOHEMIA a cluster randomized trial to assess the impact of an endectocide-based one health approach to malaria in Mozambique: baseline demographics and key malaria indicators.. 2023.
• Occurrence of Host-Associated Fecal Markers on Child Hands, Household Soil, and Drinking Water in Rural Bangladeshi Households.. 2016.
• A cluster-randomized trial determining the efficacy of caterpillar cereal as a locally available and sustainable complementary food to prevent stunting and anemia.. 2015.
• School snacks decrease morbidity in Kenyan schoolchildren: a cluster randomized, controlled feeding intervention trial.. 2013.
• Using virtual agents to increase physical activity in young children with the virtual fitness buddy ecosystem: Study protocol for a cluster randomized trial.. 2020.
• Molecular Identification of Hookworm Isolates in Humans, Dogs and Soil in a Tribal Area in Tamil Nadu, India.. 2016.
• ARRIVE Guidelines 2.0 for Reporting Animal Research. PLOS Biology, 2020.
• EQUATOR Network Reporting Guidelines. EQUATOR Network.
• MSD Veterinary Manual, Professional Edition. MSD Veterinary Manual.

Related Articles

• Designing and Analyzing Factorial Trials in Veterinary Research
• Applying Competing Risks Analysis in Veterinary Research
• Assessing Risk of Bias in Veterinary Randomized Trials
• Critical Appraisal of Randomized Controlled Trials in Veterinary Medicine
• Designing Crossover Trials for Veterinary Therapeutics

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.