Using Capture-Recapture Methods to Estimate Animal Disease Prevalence

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

Key Takeaways

• Capture-recapture methods are essential for estimating true animal disease prevalence when surveillance systems are inherently incomplete, as they formally account for undetected cases missed by multiple diagnostic pathways (e.g., voluntary practitioner reporting, incomplete laboratory submissions, subclinical infections not presenting for clinical examination).
• The core principle relies on the overlap between independent lists of affected individuals; a smaller overlap between two or more surveillance sources (e.g., shelter ELISA testing and university PCR confirmation for FeLV) indicates a larger proportion of the diseased population remains undetected.
• Critical assumptions for valid estimation include a closed population (no births, deaths, immigration, or emigration during the study period) and source independence (detection by one system does not influence detection by another), with violations like positive dependence (e.g., severely ill animals being flagged by multiple systems) leading to underestimation of true prevalence.
• Individual capture heterogeneity, where some animals are inherently more likely to be detected (e.g., due to higher pathogen load or more frequent veterinary interaction), is a significant threat that can bias estimates downwards if not explicitly modeled, often requiring more complex statistical approaches like M(th) models.
• List linkage errors, such as incorrect or inconsistent individual identifiers (e.g., ear tag numbers, microchip IDs), are a major source of bias, typically leading to overestimation of disease prevalence due to artificially reduced overlap between surveillance lists.
• Log-linear models are preferred for three or more surveillance sources, allowing for the testing and adjustment of pairwise dependencies, while two-source estimates (e.g., Lincoln-Petersen estimator) are more susceptible to bias and should be interpreted cautiously, especially when source independence is questionable.

Animal disease surveillance systems rarely capture every affected individual. Clinical cases may go undetected, laboratory submissions may be incomplete, and reporting from private practitioners is often voluntary. When the goal is to estimate true disease prevalence instead of merely count reported cases, capture-recapture methods offer a formal statistical framework for estimating the size of the hidden population. Originally developed by animal ecologists to estimate wildlife abundance, these methods have been adapted for epidemiological use in human medicine and are increasingly applicable to veterinary populations Capture-recapture methodology: an option for surveillance of non-communicable diseases in the elderly.

This article explains how capture-recapture techniques can estimate disease prevalence and surveillance completeness in animal populations when multiple incomplete lists of affected individuals are available. It is written for veterinary researchers and epidemiologists designing prevalence studies or evaluating surveillance systems. The scope covers closed-population capture-recapture as applied to disease detection, including two-source and multiple-source approaches, the assumptions that underpin valid estimation, and the failure modes that produce biased results. Mark-recapture for estimating wildlife abundance is excluded, the focus is on estimating the number of diseased animals missed by one or more detection systems.

The core logic is straightforward. If two independent surveillance systems each detect a portion of the diseased population, the degree of overlap between the two lists reveals information about the total population size. A small overlap suggests both systems detect only a small fraction of cases, implying a large hidden population. A large overlap suggests each system captures most cases, implying few were missed. The same reasoning extends to three or more sources, where log-linear modeling can accommodate dependencies between sources Bayesian estimation of a cancer population by capture-recapture with individual capture heterogeneity and small sample.

At a Glance

Conceptual Foundations of Capture-Recapture

Capture-recapture estimation treats each surveillance system as a capture event. In the ecological origin of the method, animals are captured, marked, released, and later recaptured, the proportion of marked animals in the second capture sample estimates population size. In epidemiology, the "capture" is inclusion on a case list, and the "recapture" is the appearance of the same individual on a second independent list Capture and recapture method: a new methodology for epidemiological research. No physical marking is required, but the individual must be identifiable across lists through a unique identifier or a reliable matching algorithm.

The method estimates the number of cases missed by all sources. If two lists are available, the total population size N is estimated from the number captured by list A only, list B only, and both lists. The estimate of N equals the sum of observed cases plus the estimated number missed by both. With three or more lists, log-linear models fit the observed capture patterns and estimate the number of individuals with no captures, the zero-cell count. Model selection among possible source interactions determines which dependencies between sources are incorporated.

The Closed Population Assumption

All standard capture-recapture estimators assume a closed population during the study period. No animals enter the population through birth or immigration, and none leave through death or emigration. For disease prevalence estimation, this means the disease status of each animal must be stable across the study window, and the population at risk must not change composition. A chronic infection such as feline immunodeficiency virus suits this assumption well. An acute, rapidly fatal disease does not, because animals may die between list constructions, and the two lists may effectively sample different populations.

Violating the closure assumption biases estimates in predictable directions. If animals die between capture events, the second list samples a smaller population than the first, and the overlap is reduced, inflating the population estimate. If new infections occur during the study, the second list may capture animals that were not at risk when the first list was constructed, also distorting the overlap. Study design should therefore confine the capture period to a window short enough that population turnover is negligible.

Source Independence and Its Violations

The two-source estimator requires that the probability of appearing on list A is independent of the probability of appearing on list B. In practice, this assumption is rarely met. Animals with severe clinical signs are more likely to be examined, sampled, and reported by multiple systems simultaneously, producing positive dependence. Conversely, if two surveillance systems serve different geographic regions or different production sectors, an animal captured by one may be systematically less likely to appear on the other, producing negative dependence Capture-recapture, epidemiology, and list mismatches: two lists.

Positive dependence between two sources causes underestimation of the total population. The overlap is larger than independence would predict, so the estimated number missed is too small. Negative dependence causes overestimation. With three or more sources, log-linear models can test and adjust for pairwise dependencies, but they cannot fully resolve dependence structures that involve higher-order interactions. The practical consequence is that two-source estimates should be interpreted cautiously, and three or more sources are strongly preferred when the study design permits.

Individual Capture Heterogeneity

Equal catchability is another assumption inherited from ecology that frequently fails in disease surveillance. Individual animals differ in their probability of being detected. A severely affected animal may be more likely to be presented to a veterinarian, sampled, and reported. Age, sex, production class, and management system also influence detection probability. When heterogeneity is present and unmodelled, the population estimate is biased downward because the method assumes the observed capture probability applies to all individuals, including those never captured Bayesian estimation of a cancer population by capture-recapture with individual capture heterogeneity and small sample.

Model families that incorporate heterogeneity exist. The M(th) models allow capture probability to vary with time and with individual characteriztics. These models require more data and more complex computation than simple log-linear approaches, and they may perform poorly with small samples. In veterinary applications, where list sizes are often modest, the choice between bias and variance is a real trade-off. A parsimonious model that respects the data available is usually preferable to a complex model that cannot be estimated reliably.

List Construction and Linkage Errors

The quality of capture-recapture estimates depends entirely on the quality of list linkage. If identifiers are recorded incorrectly, or if matching algorithms fail to recognize the same animal across lists, true matches are missed. The observed overlap is then too small, and the population estimate is too large. This error is analogous to tag loss in ecological studies, where marked animals lose their tags and are misclassified as unmarked Capture-recapture, epidemiology, and list mismatches: two lists. In veterinary data, ear tags, microchip numbers, and laboratory accession numbers may be transcribed incorrectly, and farm records may use inconsistent animal identifiers.

Matching errors are also a nuisance, they can dominate the bias in the estimate. When error rates are high, ignoring them produces gross overestimates of the total population. Methods exist to estimate both the error probability and the corrected population size, but they require additional assumptions about the error process. In practice, the study design should invest in identifier quality, and the analysis should include a sensitivity assessment of how plausible error rates affect the final estimate.

Applications in Veterinary Surveillance

Capture-recapture methods have been applied to veterinary questions ranging from zoonotic pathogen carriage in urban wildlife to infectious disease dynamics in domestic species. In urban Norway rat populations, capture-mark-recapture sampling was used to estimate the prevalence of methicillin-resistant Staphylococcus aureus carriage and to evaluate whether kill-trapping interventions altered that prevalence Methicillin-resistant Staphylococcus aureus in urban Norway rat (Rattus norvegicus) populations: Epidemiology and the impacts of kill-trapping. Multi-event capture-recapture modeling has been used to estimate Toxoplasma gondii seroconversion rates in farm cats while explicitly accounting for uncertainty in age and serological status A multi-event capture-recapture analysis of Toxoplasma gondii seroconversion dynamics in farm cats. These applications demonstrate the adaptability of the framework across species and study designs.

The method also has a role in evaluating surveillance system completeness, a requirement of international animal health standards. The World Organization for Animal Health sets standards for surveillance and notification, and member countries must demonstrate that their surveillance systems detect disease at a defined level of confidence WOAH animal health surveillance standards. Capture-recapture provides one quantitative route to estimating the sensitivity of a surveillance system, complementing scenario-tree modeling and other approaches.

Worked Example: Two-Source Estimation of Feline Leukemia Virus Prevalence

The practical value of capture-recapture becomes clear when applied to a concrete surveillance problem. Consider a municipal shelter that tests all incoming cats for feline leukemia virus (FeLV) antigen using a point-of-care ELISA. The shelter also refers clinically suspicious cats to a university diagnostic laboratory for confirmatory PCR. Neither source alone captures every infected cat. The ELISA may miss early infections with low antigenaemia, while PCR testing is reserved for cats with clinical signs and therefore misses subclinical carriers.

The shelter records 42 ELISA-positive cats over a 12-month period. The diagnostic laboratory records 28 PCR-positive cats from the same catchment area during the same interval. Cross-referencing the two lists by microchip number and ear tattoo reveals 12 cats present on both lists. The two-source Lincoln-Petersen estimator gives:

N = (M × C) / R

where M is the number on the first list, C is the number on the second list, and R is the number of matches. Substituting the observed values:

N = (42 × 28) / 12 = 98

The estimated total number of FeLV-infected cats in the catchment population is therefore 98, of whom 42 + 28 - 12 = 58 were observed. The estimated number missed by both sources is 98 - 58 = 40. The estimated completeness of the combined surveillance system is 58 / 98, or 59.2 percent.

The variance of the estimator, using the Chapman modification recommended for small samples, is:

Var(N) = [(M + 1)(C + 1)(M - R)(C - R)] / [(R + 1)²(R + 2)]

This yields a standard error of approximately 15.4, giving a 95 percent confidence interval of roughly 68 to 128 infected cats. The width of this interval reflects the modest overlap between the two lists. When R is small relative to M and C, the estimate becomes unstable and the confidence interval widens substantially.

The critical assumption here is that the two sources are independent. That assumption fails if ELISA-positive cats are more likely to be referred for PCR confirmation, which is plausible in a shelter that follows a positive antigen test with confirmatory testing before euthanasia decisions. Under positive dependence, where capture on one list increases the probability of capture on the other, the Lincoln-Petersen estimator underestimates the true population size. The direction of bias reverses under negative dependence. Seber and colleagues describe methods for adjusting two-list estimates when list matching is imperfect, using the concept of tag loss borrowed from animal population studies to model missed matches Seber, Huakau, and Simmons describe list mismatch corrections for two-source estimates.

Selecting an Estimator for Three or More Sources

When three or more lists are available, log-linear modeling provides a more flexible framework. The observed counts in the seven cells of the three-way contingency table are fitted to a Poisson model with main effects for each source and interaction terms representing dependence between sources. The model estimates the count in the missing eighth cell, which represents animals missed by all sources.

The model selection sequence follows a defined logic. Start with the saturated model containing all two-way interactions. Remove interaction terms that do not improve fit, using the Akaike information criterion or likelihood ratio tests. The preferred model is the simplest one that adequately reproduces the observed table. Models with a single two-way interaction assume that the corresponding pair of sources is dependent while all others are independent. Models with all three two-way interactions assume pairwise dependence throughout.

Individual capture heterogeneity, where some animals are inherently more likely to be detected than others, violates the equal catchability assumption of standard log-linear models. This is a particular concern in disease surveillance because severely affected animals are more likely to present to multiple diagnostic pathways. Bailly and colleagues applied models incorporating time and individual heterogeneity parameters to cancer registry data and found that classical log-linear models produced different estimates than heterogeneity-adjusted models Bailly et al. compare log-linear and heterogeneity-adjusted capture-recapture models. In veterinary practice, heterogeneity arises when animals with high pathogen loads are more likely to be tested, more likely to test positive, and more likely to be reported to multiple surveillance streams.

Diagnostic Testing and Sampling Protocols

Capture-recapture analysis requires that the diagnostic test used by each source has known performance characteriztics. A test with imperfect sensitivity will miss infected animals even when they are sampled, and the capture-recapture estimate will then underestimate the true prevalence. The magnitude of this bias depends on the sensitivity of each source's test and the degree of overlap between sources. Where possible, use the same diagnostic platform across sources, or adjust the analysis for known sensitivity differences.

Sampling design determines whether the closed population assumption holds. For chronic infections such as FeLV, bovine viral diarrhea virus persistence, or tuberculosis, a defined study period with a closed cohort is appropriate. For acute infections with rapid turnover, the population is open and standard closed-population estimators are biased. Multi-event capture-recapture models extend the framework to accommodate transitions between disease states. Simon and colleagues used a multi-event approach to model Toxoplasma gondii seroconversion in farm cats, explicitly accounting for uncertainty in age and serological status while estimating seroconversion rates across seasons Simon et al. apply multi-event capture-recapture to seroconversion dynamics in farm cats.

Documentation and Reporting Standards

Surveillance outputs should report the estimator used, the assumptions made, the number of sources, the overlap counts, and the confidence interval around the final estimate. The World Organization for Animal Health publishes surveillance standards that specify the information expected in national disease reports, including the methods used to estimate prevalence and the completeness of detection WOAH animal health surveillance standards. Reports intended for trade partners should follow the Terrestrial Animal Health Code provisions on surveillance and notification WOAH terrestrial animal health code.

Documentation should include the case definition used by each source, the linkage method, the number of records excluded for missing identifiers, and the results of any sensitivity analysis. A sensitivity analysis that varies the assumed match error rate or the degree of source dependence provides a range of plausible estimates instead of a single point value.

Species and Production System Considerations

The correct estimator and sampling design vary by species and production context. In companion animal populations, microchip databases and veterinary practice records provide reliable individual identifiers, making list linkage straightforward. In production animal systems, individual identification may be absent in small ruminants or poultry, forcing reliance on group-level data or temporary marks. Wildlife populations present additional challenges, although the same statistical framework applies to pathogen detection in free-ranging species. Lee and colleagues used capture-mark-recapture to estimate MRSA carriage prevalence in urban Norway rats, sampling both oropharyngeal and rectal sites before and after a kill-trapping intervention Lee et al. assess MRSA prevalence in urban rats using capture-mark-recapture.

Herd-level applications differ from individual-level applications. When the unit of interest is the herd instead of the animal, the lists consist of herds detected by different surveillance streams, and the estimator produces an estimate of the total number of infected herds. This approach is particularly useful for diseases with statutory reporting requirements, where underreporting is a known problem.

Model Selection Decision Framework

The choice between these approaches depends on the number of available sources, the suspected dependence structure, and the quality of individual identifiers. When in doubt, report estimates from multiple models and present the range as the uncertainty envelope.

Recognized Complications and Early Detection

Capture-recapture estimates fail silently when their structural assumptions are violated. The most consequential failure modes are source dependence, individual heterogeneity, and list linkage errors, each of which biases the population estimate in a characteriztic direction.

Positive dependence between sources, where capture on one list increases the probability of capture on another, produces underestimation of the hidden population. Negative dependence produces overestimation. The two-source Lincoln-Petersen estimator cannot detect dependence at all, because with only two lists the overlap count is the sole information available. With three or more sources, log-linear models provide a formal test: the fit of the model that includes all two-way interaction terms is compared against the saturated model, and a significant lack of fit signals dependence that the model cannot accommodate. A simpler diagnostic is to examine the pattern of pairwise overlaps. If the observed overlap between two sources greatly exceeds the product of their marginal capture probabilities, positive dependence is likely.

Individual heterogeneity, where some animals are intrinsically more capturable than others, biases two-source estimates downward. In three-source models, heterogeneity can masquerade as a three-way interaction term. The sample coverage approach and the M(th) models described in the cancer epidemiology literature offer an alternative that explicitly parameterises heterogeneity, though these require more data and more computational effort than classical log-linear models Bayesian estimation of a cancer population by capture-recapture with individual capture heterogeneity and small sample. Early detection relies on comparing estimates from models with and without heterogeneity terms, a wide divergence indicates that the simpler model is misspecified.

Linkage errors are the most insidious failure mode because they are invisible in the final estimate. When records for the same animal are not matched across lists, the overlap count is artificially low and the population is overestimated. When distinct animals are incorrectly merged, the opposite occurs. The tag-loss framework borrowed from animal ecology provides a correction for two-source designs, but it requires validation data on the error rate Capture-recapture, epidemiology, and list mismatches: two lists. In practice, a random sample of apparent matches should be verified against original records, and a random sample of non-matches should be checked for missed matches. Error rates above 5% warrant correction or abandonment of the analysis.

Common Errors and Corrective Actions

The most frequent error in applied capture-recapture is treating the closed population assumption as a formality. In veterinary data, the population is rarely closed: animals die, are culled, move between premises, or recover from the condition under study. The correction is to restrict the observation window to a period short relative to the population turnover rate, or to use open-population models that estimate entry and exit parameters.

A second common error is pooling sources that are not independent by construction. Two laboratory databases that receive samples from the same referral clinic are not independent sources, even if the databases are maintained by different organizations. The corrective action is to trace the referral pathway for each case and treat sources as independent only when their case ascertainment mechanisms are genuinely distinct.

A third error is over-interpreting the completeness estimate. The estimated total population size is only as good as the model that produced it. Reporting a single point estimate without confidence intervals, or without a sensitivity analysis across plausible models, invites false precision. The corrective action is to report the range of estimates across candidate models and to state the model selection criteria explicitly.

Limitations of the Current Evidence

The veterinary literature on capture-recapture for disease prevalence is thinner than the human epidemiology literature. Most published veterinary applications concern wildlife or peri-domestic species, such as the use of capture-mark-recapture to assess methicillin-resistant Staphylococcus aureus carriage in urban Norway rats Methicillin-resistant Staphylococcus aureus in urban Norway rat populations and multi-event modeling of Toxoplasma gondii seroconversion in farm cats A multi-event capture-recapture analysis of Toxoplasma gondii seroconversion dynamics in farm cats. These studies demonstrate feasibility, but they do not establish generalizable performance characteriztics for livestock or companion animal surveillance.

Expert opinion still differs on the minimum number of sources required for reliable estimation. Some authorities accept two-source estimates when the sources are believed to be independent, while others insist on three or more sources because only then can dependence be tested. The conservative position is that two-source estimates should be treated as exploratory and corroborated by an independent method where possible.

Referral, Consultation, and Reporting Triggers

Capture-recapture analysis is not a routine clinical procedure, and most cases do not require escalation. Referral to a veterinary epidemiologist or biostatistician is warranted when the analysis will inform a regulatory decision, when the study population is small and heterogeneity is suspected, or when model diagnostics indicate dependence that cannot be resolved by source redefinition. Laboratory involvement is indicated when diagnostic test sensitivity or specificity is uncertain, because test error propagates directly into capture probabilities.

Regulatory reporting obligations are governed by the relevant animal health authority. The World Organization for Animal Health maintains international standards for surveillance and notification, and these standards specify which diseases must be reported and what surveillance data are expected WOAH animal health surveillance standards. Where a capture-recapture analysis suggests that the true prevalence of a notifiable disease exceeds the reported figure, the responsible clinician or laboratory should contact the competent authority instead of publish the estimate independently. The same principle applies to trade-sensitive findings under the terrestrial animal health code WOAH terrestrial animal health code.

Troubleshooting Table

Frequently Asked Questions

How much does a capture-recapture study cost compared with a traditional prevalence survey?

Capture-recapture methods are often less expensive than exhaustive surveillance because they exploit existing data sources instead of requiring de novo sampling. The main costs are data cleaning, record linkage, and statistical modeling time, not field work. When suitable lists already exist, such as diagnostic laboratory submissions, treatment records, and slaughterhouse reports, the marginal cost is modest. Building new lists from active surveillance is more expensive. The method is particularly cost-effective for rare or elusive conditions, where traditional sampling would require impractically large sample sizes to detect enough cases. As noted in early epidemiological applications, capture-recapture allows more accurate estimates than traditional methods and is more cost-effective for studying rare populations.

What can I do when the available lists are clearly dependent, for example when one laboratory serves both referral and primary care clinics?

Dependence between sources is the most common practical obstacle. With three or more lists, log-linear models can explicitly estimate pairwise interaction terms, so dependence becomes a modelled parameter instead of a fatal flaw. With only two lists, dependence cannot be estimated from the data alone, and you must either accept a biased estimate or seek a third source. One practical strategy is to split a single heterogeneous source into two sub-sources, such as separating primary care and referral submissions from the same laboratory, provided the case definitions and capture periods remain identical. Another option is to adjust for known dependence using external information, though this requires defensible assumptions. When dependence is suspected but unquantifiable, report a range of estimates under different dependence scenarios instead of a single figure.

How does capture-recapture perform in wildlife or free-roaming populations where individuals cannot be physically marked?

The method does not require physical marking at all. What matters is the ability to recognize the same individual across lists. In wildlife studies, physical marks are common, but veterinary applications often rely on naturally occurring identifiers. Microchip numbers, ear tags, tattoo codes, and even DNA profiles from fecal or hair samples serve the same function. For domestic species, registration databases and microchip records are often the backbone of multi-source studies. For truly unmarked populations, such as certain wildlife reservoirs, DNA-based individual identification is increasingly feasible but adds laboratory costs. The statistical machinery is identical once individual identity is established. The key distinction is that the method estimates the number of affected animals, not the total population size, so the sampling frame must be defined around disease status instead of geographic presence.

What record-keeping practices improve the accuracy of future capture-recapture analyzes?

Standardized unique identifiers are the single most valuable practice. Microchip numbers, ear tags, and national registration identifiers should be recorded consistently across all clinical and laboratory systems. Dates of birth or age estimates, species, breed, sex, and location should follow a shared data dictionary. Diagnostic results should include the test method, laboratory, and date of sampling, also a positive or negative outcome. When records are merged across institutions, preserve the original identifiers and document the linkage algorithm. As demonstrated in studies of list mismatches, errors in identifying information can cause gross overestimates of disease frequency, so invest in validation steps that check for inconsistent dates, implausible ages, and duplicate entries before analysis.

How should I explain capture-recapture findings to a producer or practice owner who expects a simple prevalence percentage?

Frame the estimate as a correction for undercounting. Explain that routine records capture only a fraction of affected animals, and capture-recapture uses the overlap between different record systems to estimate how many were missed. Use a concrete analogy: if two independent observers each count a proportion of a flock and you know how many birds both saw, you can estimate the total flock size. Emphasize that the method produces a range, not a single number, and that the range reflects real uncertainty about hidden cases. Avoid presenting the estimate as definitive. Tie the result to a practical decision, such as whether herd-level prevalence exceeds a threshold that justifies vaccination or culling. Reference the relevant international surveillance standards when the finding affects trade or reporting obligations.

When is capture-recapture inappropriate, and what alternatives should I consider?

Avoid capture-recapture when the population is open, meaning animals enter and leave through birth, death, or movement during the study period, unless you use open-population models that require substantially more data. Avoid it when case definition is inconsistent across sources, because list mismatches will reflect definitional differences instead of true overlap. Avoid it when fewer than two independent sources exist. Alternatives include stratified sampling with laboratory confirmation, which provides unbiased prevalence estimates when a sampling frame exists, and participatory epidemiology methods, which work well in low-resource settings where formal records are sparse. For conditions with strong clinical signs, sentinel surveillance may be more practical. The choice depends on the question, the available data, and the consequences of underestimation, particularly for zoonotic or trade-regulated diseases where underreporting has public health implications.

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

• [[Capture-recapture methodology: an option for surveillance of non-communicable diseases in the elderly].](https://pubmed.ncbi.nlm.nih.gov/11175530/). 2000.
• [[Capture and recapture method: a new methodology for epidemiological research].](https://pubmed.ncbi.nlm.nih.gov/7660051/). 1994.
• Bayesian estimation of a cancer population by capture-recapture with individual capture heterogeneity and small sample.. 2015.
• Capture-recapture, epidemiology, and list mismatches: two lists.. 2000.
• Methicillin-resistant Staphylococcus aureus in urban Norway rat (Rattus norvegicus) populations: Epidemiology and the impacts of kill-trapping.. 2019.
• A multi-event capture-recapture analysis of Toxoplasma gondii seroconversion dynamics in farm cats.. 2018.
• 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

• Stratified Sampling for Disease Prevalence Estimation
• Measuring Disease Frequency: Incidence and Prevalence in Animal Populations
• Participatory Epidemiology Methods for Livestock Disease Surveillance
• Compartmental Models in Veterinary Disease Dynamics
• Sensitivity Analysis in Veterinary Disease Models

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.