# Meta-Analysis of Veterinary Diagnostic Test Accuracy


## Key Takeaways

- Diagnostic test accuracy (DTA) meta-analyses in veterinary medicine require bivariate random-effects or Hierarchical Summary Receiver Operating Characteristic (HSROC) models to account for the inherent correlation between sensitivity and specificity, which trade off with changing test thresholds.
- Each study contributes a 2x2 table (true positives, false positives, false negatives, true negatives) per threshold, and these must be extracted and analyzed to preserve the joint accuracy structure, avoiding independent pooling of sensitivity and specificity.
- Imperfect reference standards (e.g., fecal flotation for parasites, clinical signs for some viral infections) necessitate acknowledgment of potential bias, consideration of latent class methods, or sensitivity analyses restricted to studies with more robust reference standards like PCR or histopathology.
- Heterogeneity in DTA meta-analyses is common due to variations in species (e.g., dairy cattle vs. companion dogs), sample type (e.g., serum vs. feces), disease prevalence, and reference standard quality, requiring exploration via meta-regression with pre-specified covariates.
- Reporting must adhere to PRISMA-DTA guidelines, detailing search strategies, inclusion criteria, data extraction, and statistical models, with software like R's `mada` package or Stata's `metandi` command being essential for implementing bivariate or HSROC frameworks.

---

Meta-analysis of diagnostic test accuracy (DTA) studies synthesizes sensitivity and specificity estimates across multiple primary evaluations of a test or index modality. In veterinary medicine, such syntheses inform test selection for surveillance, herd health programs, and clinical diagnosis across species. This article covers the statistical foundations, model choices, and reporting standards for DTA meta-analysis in veterinary research. It serves veterinary researchers, residents, and clinicians who appraise or conduct quantitative syntheses of diagnostic studies. The methods described apply across species and production systems, from companion animal point-of-care assays to livestock pathogen detection.

A DTA meta-analysis differs fundamentally from a therapeutic meta-analysis. The outcome is bivariate: sensitivity and specificity are jointly estimated because they trade off against each other as test thresholds change. Pooling them independently produces biased summary estimates. The analytic framework must therefore preserve the correlation structure between the two accuracy indices. This article explains the bivariate random-effects model and the hierarchical summary receiver operating characteriztic (HSROC) model, their equivalence under common conditions, and practical decisions about data extraction, threshold handling, and heterogeneity assessment. It excludes therapeutic meta-analysis and focuses on procedural guidance for veterinary applications.

## At a Glance

| Parameter | Decision or Fact |
|---|---|
| Primary analysis model | Bivariate random-effects model or HSROC, both account for correlation between sensitivity and specificity |
| Unit of analysis | A 2x2 table (true positives, false positives, false negatives, true negatives) per study, per threshold |
| Threshold handling | If studies report different thresholds, use HSROC or bivariate model with threshold as covariate, do not pool across thresholds without adjustment |
| Reference standard | Must be defined and consistent, imperfect reference standards require latent class methods or explicit acknowledgment of bias |
| Heterogeneity | Report I² analogues, between-study variance components, and explore covariates (species, sample type, disease prevalence, test format) |
| Minimum studies | No universal minimum, but models with fewer than four studies often fail to converge, consider fixed-effects or descriptive synthesis |
| Reporting standard | Follow PRISMA-DTA, available through the EQUATOR Network reporting guidelines library |
| Software | R packages (mada, lme4, rjags), SAS PROC NLMIXED, or Stata metandi, all implement bivariate or HSROC frameworks |

## Conceptual Foundations of Diagnostic Accuracy Synthesis

### The Bivariate Nature of Test Accuracy

Sensitivity and specificity are not independent properties of a test. They vary with the threshold used to define a positive result, with disease spectrum, and with population characteriztics. A meta-analysis that pools sensitivity across studies while separately pooling specificity discards the within-study correlation between these measures. The bivariate model treats the logit-transformed sensitivity and specificity from each study as a pair of correlated random effects. This preserves the trade-off structure and yields a summary operating point with a confidence ellipse that reflects joint uncertainty.

The bivariate model is parameterized as a random-effects logistic regression with two outcomes per study. Between-study variance is captured by two variance components and one covariance term. The model produces a summary sensitivity and summary specificity, along with a 95% confidence region. When studies report multiple thresholds, the model can be extended to include threshold as a covariate, though the HSROC framework is often more natural for that setting.

### The HSROC Model

The HSROC model, developed by Rutter and Gatsonis, parameterizes the summary ROC curve directly. It includes three parameters: a threshold parameter, an accuracy parameter, and a shape parameter. The model accommodates studies that report different thresholds by modeling the implicit threshold in each study. Under the assumption of a symmetric ROC curve, the HSROC model and the bivariate model are mathematically equivalent. This equivalence means that the choice between them is often one of software convenience or reporting preference instead of statistical necessity.

For veterinary datasets with few studies, the HSROC model may be overparameterized. In such cases, a simplified version that fixes the shape parameter or assumes symmetry can improve convergence. Researchers should report which parameterization was used and justify simplifications.

## Study Selection and Data Extraction

### Defining the Index Test and Reference Standard

The research question must specify the index test, the target condition, and the reference standard before literature searching begins. In veterinary DTA studies, the reference standard may be culture, histopathology, PCR, or a composite of clinical and laboratory findings. When the reference standard is imperfect, as with many parasitic diagnoses where no gold standard exists, the meta-analysis must acknowledge this limitation. The FLOTAC techniques, for example, were developed to improve copromicroscopic diagnosis of parasites in animals and humans, yet even improved fecal flotation methods have imperfect sensitivity compared with true infection status. A meta-analysis of such tests should consider latent class approaches or at minimum discuss the direction of expected bias.

### Extracting 2x2 Tables

Each included study contributes at least one 2x2 table. When a study reports accuracy at multiple thresholds, extract each threshold as a separate data row but record the threshold value. When a study reports only summary statistics such as sensitivity and specificity with confidence intervals, reconstruct the 2x2 table from the reported totals if possible. If reconstruction is impossible, the study may need to be excluded or handled in sensitivity analysis. Data extraction should be performed in duplicate, with disagreements resolved by consensus or a third reviewer.

## Statistical Models in Practice

### Model Fitting and Convergence

The bivariate model is typically fitted by restricted maximum likelihood or Markov chain Monte Carlo. Convergence problems arise with sparse data, zero cells, or few studies. A zero cell in a 2x2 table can be handled by adding 0.5 to all cells, though this adjustment can bias estimates when applied uniformly. Alternative approaches include exact methods or penalized likelihood. With fewer than four studies, random-effects variance components are estimated imprecisely, and a fixed-effects model or a purely descriptive synthesis may be more appropriate.

### Exploring Heterogeneity

Between-study heterogeneity in DTA meta-analyzes is often substantial. Sources include differences in disease prevalence, spectrum of disease severity, sample handling, test operator skill, and reference standard quality. Covariates can be added to the bivariate model to explain heterogeneity. In veterinary applications, species and production system are natural covariates. A test validated in dairy cattle may perform differently in beef cattle or small ruminants, and pooling across species without adjustment can produce misleading summary estimates. The methodological quality of primary studies also contributes to heterogeneity, and tools for assessing study quality in diagnostic accuracy research differ from those used for intervention studies. The systematic review of methodological assessment tools by Zeng and colleagues identifies instruments appropriate for different study designs, and researchers should select a quality tool validated for diagnostic accuracy studies instead of applying a generic risk-of-bias instrument.

## Reporting and Interpretation

### Reporting Standards

DTA meta-analyzes should be reported according to PRISMA-DTA, the extension of the PRISMA statement for diagnostic test accuracy reviews. The EQUATOR Network maintains a comprehensive library of reporting guidelines, including PRISMA-DTA, and authors should consult this resource before submission. Transparent reporting of search strategies, inclusion criteria, data extraction methods, and statistical models is essential for reproducibility. The ARRIVE guidelines, while developed for primary animal research, also inform the reporting of the primary studies that feed into a DTA meta-analysis, and reviewers should consider whether included studies meet basic reporting standards.

### Interpreting Summary Estimates

A summary sensitivity and specificity from a bivariate model should be interpreted as the expected accuracy of the test in a population similar to those represented by the included studies. The confidence ellipse around the summary point reflects uncertainty in both dimensions simultaneously. When the ellipse is wide, the evidence base is insufficient to support strong clinical recommendations. Researchers should present the summary ROC curve graphically when studies report multiple thresholds, and they should report the range of thresholds across studies to help readers judge clinical applicability.

## Practical Workflow for a Diagnostic Accuracy Meta-Analysis

### Step 1: Define the Clinical Question and Scope

The clinical question must specify the target condition, the index test, the reference standard, the population, and the intended clinical use. A question such as "How accurately does thoracic radiography detect pulmonary metastases in dogs with appendicular osteosarcoma?" is answerable. A question such as "How accurate is imaging in veterinary oncology?" is not.

Decide whether the meta-analysis will estimate accuracy for a single test, compare multiple index tests, or evaluate a test across different populations. This decision determines the statistical model and the data requirements. The intended clinical role of the test, replacement of an existing test, triage before a more expensive test, or add-on testing, changes the relevant accuracy parameters and the acceptable trade-offs between sensitivity and specificity.

### Step 2: Assemble the Analytic Dataset

Each included study contributes one or more 2x2 tables. A study may contribute multiple tables when it evaluates multiple thresholds, multiple index tests, or multiple subgroups. These tables are not independent observations. Failing to account for this clustering produces artificially narrow confidence intervals.

Before fitting any model, construct a dataset with one row per 2x2 table and columns for study identifier, index test, threshold, reference standard, true positives, false positives, false negatives, true negatives, and any covariates planned for heterogeneity exploration. Verify that the sum of the four cells equals the total number of animals or samples tested. Discrepancies indicate extraction errors or unreported missing data.

### Step 3: Choose the Statistical Model

The bivariate model and the hierarchical summary receiver operating characteriztic (HSROC) model are mathematically equivalent when no covariates are included. They differ in parameterisation. The bivariate model estimates mean logit sensitivity and mean logit specificity plus their variances and covariance. The HSROC model estimates a summary curve with parameters for accuracy, threshold, and shape. The choice between them is largely one of software convenience and reporting preference.

When covariates are added, the two models diverge. The bivariate model allows covariates to affect sensitivity and specificity separately. The HSROC model allows covariates to affect accuracy and threshold. The bivariate parameterisation is often easier to interpret clinically because it directly answers the question of whether a covariate changes sensitivity, specificity, or both.

### Step 4: Fit the Model and Assess Convergence

Fit the chosen model using restricted maximum likelihood or a Bayesian approach with weakly informative priors. Common software options include the `mada` and `lme4` packages in R, the `metandi` command in Stata, and the `metaDTA` web application. Bayesian fitting via Stan or JAGS is useful when studies are few or cell counts are zero.

Convergence problems are common with sparse data. Symptoms include failure to estimate the covariance parameter, extreme standard errors, or estimates at the boundary of the parameter space. Remedies include simplifying the model, removing the covariance term, using a fixed-effects model, or adding a continuity correction of 0.5 to zero cells. A continuity correction biases estimates toward the null and should be reported when used.

### Step 5: Explore Heterogeneity

Heterogeneity in diagnostic accuracy meta-analyzes is the rule instead of the exception. The I-squared statistic is less informative here than in therapeutic meta-analysis because sensitivity and specificity are correlated and both contribute to total variation. Report the variances of the random effects and the covariance between logit sensitivity and logit specificity. A large negative covariance indicates a threshold effect, meaning studies used different positivity criteria.

Pre-specify covariates that might explain heterogeneity. Candidate covariates include species, breed, disease prevalence, disease severity, sample type, laboratory, operator experience, and study design features such as blinding or consecutive enrollment. Examine each covariate in a single-covariate model before building multivariable models. With fewer than ten studies, multivariable modeling is unreliable and single-covariate analyzes should be interpreted cautiously.

### Step 6: Sensitivity Analyzes

Perform sensitivity analyzes to test the robustness of the summary estimates. Exclude studies at high risk of bias and refit the model. Exclude studies with outlying results and refit. Restrict the analysis to studies using the same reference standard. If the conclusions change materially, report this and discuss the implications.

### Step 7: Present Results

Report the summary sensitivity and specificity with 95% confidence intervals, the summary ROC curve when thresholds vary, and a forest plot of study-level estimates. Present a coupled forest plot showing sensitivity and specificity for each study side by side. This display reveals trade-offs between the two parameters that a conventional forest plot obscures.

## Worked Example: Point-of-Care Tests for Canine Parvovirus

Consider a meta-analysis of point-of-care antigen tests for canine parvovirus compared with PCR as the reference standard. The search identifies 12 eligible studies reporting 14 2x2 tables. Two studies evaluated two different commercial tests.

The bivariate model yields summary sensitivity of 0.82 (95% CI 0.74 to 0.88) and summary specificity of 0.97 (95% CI 0.94 to 0.99). The covariance is negative, suggesting threshold variation. Adding a covariate for sample type, fecal versus vomitus, shows that sensitivity is higher for fecal samples, with a difference of 0.09 (95% CI 0.01 to 0.17), while specificity is unchanged.

A sensitivity analysis excluding three studies that did not report blinding of the index test interpreter raises summary sensitivity to 0.85 but leaves specificity unchanged. The conclusion is that these tests are highly specific but miss roughly one in five infections, and that sample type materially affects sensitivity.

## Software Recommendations

| Software | Model support | Strengths | Limitations | Best for |
|---|---|---|---|---|
| R package `mada` | Bivariate, HSROC | Free, flexible, well documented | Requires R proficiency | Most analyzes |
| Stata `metandi` | Bivariate, HSROC | Command-line simplicity, integrated with `midas` plots | Licensed, less flexible for complex models | Rapid analysis |
| `metaDTA` web app | Bivariate, HSROC | No programming required, generates plots automatically | Limited covariate handling | Teaching, quick analyzes |
| Stan or JAGS | Custom Bayesian models | Handles sparse data, complex hierarchies | Steep learning curve, requires prior specification | Research-grade analyzes with few studies |

Species and production system alter the practical choices. In food animals, the cost of the index test and the consequences of false positives for trade or culling decisions may dominate the analysis. The [WOAH terrestrial animal health standards](https://www.woah.org/en/what-we-do/standards/codes-and-manuals/terrestrial-code-online-access/) specify accuracy requirements for tests used in international trade, and these requirements may differ from those appropriate for clinical decision-making. In companion animals, the reference standard may be more invasive or expensive, and studies may be smaller. In wildlife or exotic species, sample sizes are often very small and Bayesian methods with informative priors may be the only feasible approach.

## Common Failure Modes and Their Remedies

The most frequent error is treating multiple 2x2 tables from one study as independent. This artificially inflates precision. The remedy is to include study as a random effect or to fit a model that accounts for within-study correlation.

A second failure is ignoring the reference standard. If the reference standard is imperfect, the summary estimates are biased. Latent class models can adjust for imperfect reference standards but require additional assumptions and often more data than are available. Report the reference standard for each study and consider a sensitivity analysis restricted to studies using the same reference standard.

A third failure is over-interpreting summary estimates when the included studies are clinically heterogeneous. A summary sensitivity of 0.82 is meaningless if the studies enrolled animals with different disease prevalence, severity, and vaccination status. The [EQUATOR Network reporting guidelines](https://www.equator-network.org/) and the [ARRIVE guidelines for animal research](https://arriveguidelines.org/) both emphasize transparent reporting of study populations, and the meta-analyst should demand the same transparency from primary studies before pooling their results.

A fourth failure is inadequate reporting of the search and selection process. The meta-analysis must be reproducible. Document the databases searched, the search dates, the inclusion and exclusion criteria, and the number of studies screened, assessed, and included at each stage.

## Recognized Complications and Early Detection

Meta-analyzes of diagnostic test accuracy fail in characteriztic ways. The most consequential failure is spectrum bias introduced through selective inclusion. Studies that enrol severely affected animals and obviously healthy controls inflate sensitivity and specificity estimates, and the summary receiver operating characteriztic curve shifts toward the upper-left corner. Detect this early by comparing the distribution of disease severity and signalment across included studies, and by inspecting the spread of observed sensitivities and specificities for implausible clustering.

Partial verification bias occurs when only a subset of animals receives the reference standard. If the decision to verify depends on index test results, the 2x2 table is systematically distorted. The discriminating check is a cross-tabulation of index test result against reference standard completion. Differential verification, where different reference standards are applied to different animals, produces similar distortion and should be examined during data extraction.

Publication bias remains difficult to detect in diagnostic accuracy syntheses. Funnel plot asymmetry testing has poor statistical properties when applied to sensitivity and specificity pairs, and the bivariate model does not provide a simple analogue to Egger's regression. A pragmatic approach is to compare the number of small studies with high accuracy against the number with low accuracy, and to search trial registries and conference proceedings for unpublished work. The [methodological quality assessment tools catalogued by Zeng and colleagues](https://pubmed.ncbi.nlm.nih.gov/25594108/) include items for verifying complete reporting, which helps identify studies whose missing data may signal selective publication.

Convergence failure in the bivariate or hierarchical summary receiver operating characteriztic model is common with sparse data. The model may report implausible variance estimates or fail to produce a valid covariance matrix. Detect this by examining the correlation parameter between sensitivity and specificity, which should fall between negative one and one, and by running the model from multiple starting values.

## Common Errors and Corrective Action

Less experienced analysts often meta-analyze sensitivity and specificity separately. This produces two independent pooled estimates and ignores the negative correlation that typically exists between them. The corrective action is to fit a bivariate model that estimates the covariance structure explicitly.

A second error is treating the summary operating point as a universal threshold. The bivariate model produces a summary sensitivity and specificity at the average threshold across studies, but this does not describe test performance at any specific cut-off. The hierarchical summary receiver operating characteriztic curve should be presented when studies use different thresholds.

A third error is failing to account for the ordinal or continuous nature of the index test. Dichotomizing at a threshold chosen after seeing the data overestimates accuracy. The corrective action is to extract data at prespecified thresholds or to use the hierarchical summary receiver operating characteriztic model, which accommodates threshold variation.

A fourth error is pooling studies with different reference standards. A molecular assay and a culture-based method do not measure the same latent condition, and pooling them produces an uninterpretable summary. The corrective action is to stratify by reference standard or to restrict inclusion to one reference standard.

| Observation | Likely cause | Discriminating check |
|---|---|---|
| Summary sensitivity near 1.00 with wide confidence interval | Sparse data or perfect test results in small studies | Examine individual study 2x2 tables for zero cells |
| Model fails to converge | Insufficient studies or extreme correlation | Reduce model complexity, check starting values |
| Summary point lies outside the observed data cloud | Threshold effects across studies | Plot study-level points against the summary curve |
| Heterogeneity variance estimate at boundary | Overdispersion or outlying study | Run leave-one-out analysis |
| Sensitivity and specificity both decline across studies | Threshold variation or spectrum shift | Inspect threshold values and disease severity per study |

## Limitations of Current Evidence

The veterinary diagnostic accuracy literature remains sparse relative to human medicine. Many evaluations are single-center studies with small sample sizes, and few are prospectively registered. The [reporting standards promoted by the EQUATOR Network](https://www.equator-network.org/) include STARD for diagnostic accuracy studies, but uptake in veterinary journals has been uneven. This limits the reliability of any synthesis built on poorly reported primary studies.

Expert opinion differs on whether the bivariate model or the hierarchical summary receiver operating characteriztic model should be preferred. Both are mathematically related and often produce similar results, but the hierarchical summary receiver operating characteriztic model is more flexible when thresholds vary. Some methodologists argue that the bivariate model is easier to interpret for clinicians, while others prefer the hierarchical model for its explicit threshold parameter. The choice matters less than the quality of the underlying data.

Reference standard problems are pervasive. For many veterinary infections, no true gold standard exists, and latent class analysis is required to estimate accuracy without assuming a perfect reference. This approach is underused in veterinary medicine, partly because it requires larger sample sizes and more complex modeling.

## Referral and Escalation

Statistical consultation is warranted when the analytic dataset contains fewer than five studies, when the bivariate model fails to converge after reasonable attempts, or when latent class methods are needed. A veterinary epidemiologist or biostatistician with experience in diagnostic accuracy synthesis should be engaged before the analysis is abandoned or simplified.

Laboratory involvement is appropriate when the reference standard itself is in question. Clinical pathologists and diagnostic laboratory directors can clarify whether the reference assay was performed under validated conditions and whether sample handling varied across sites. This information is essential for interpreting heterogeneity.

Regulatory reporting obligations arise when a diagnostic test is used in the context of notifiable disease surveillance. The [World Organization for Animal Health terrestrial standards](https://www.woah.org/en/what-we-do/standards/codes-and-manuals/terrestrial-code-online-access/) specify requirements for test validation and reporting in trade-related contexts. A meta-analysis that synthesises accuracy data for such a test should confirm that the included primary studies meet these validation standards, and any discrepancy should be reported to the relevant veterinary authority.

## Frequently Asked Questions

### How Many Studies Are Needed Before a Bivariate or HSROC Meta-Analysis Is Feasible?

There is no fixed minimum, but fewer than four studies typically produce unstable estimates and convergence failures. With four or more studies, a bivariate model can usually be fitted, although confidence intervals will be wide. The HSROC model requires more data to estimate its shape and threshold parameters reliably, so it often needs at least five to six studies. If fewer studies are available, consider a simpler approach, such as pooling sensitivity and specificity separately with random-effects models, and present the results with explicit caveats about imprecision. Report the number of studies and the model used transparently so readers can judge the stability of the summary estimates.

### What Should I Do When Primary Studies Report Multiple Index Test Thresholds or Subgroups?

Extract data for each threshold or subgroup as a separate 2x2 table, but do not treat them as independent studies in the same model. For a bivariate analysis, select one threshold per study, preferably the one prespecified in the study protocol or the clinically recommended cut-off. If thresholds vary widely across studies, the HSROC model is more appropriate because it can accommodate threshold variation. Alternatively, perform a separate analysis for each clinically meaningful threshold if at least four studies report that threshold. Document all threshold selections in the supplementary material and test their influence in sensitivity analyzes.

### How Do I Handle Studies With Different Reference Standards or Imperfect Reference Tests?

In veterinary diagnostics, a perfect reference standard is rare, and studies may use bacterial culture, histopathology, PCR, or a composite of clinical and laboratory findings. Include studies with different reference standards only if each reference standard is acceptable for the target condition, and record the reference standard used in each study. If one reference standard is clearly superior, restrict the primary analysis to studies using it and use the broader set in a sensitivity analysis. When the reference standard is imperfect, summary sensitivity and specificity will be biased toward the reference standard's performance. Discuss this limitation explicitly and consider latent class analysis if the data permit, although such models require strong assumptions and often more studies than are available.

### Can I Combine Studies From Different Species or Production Systems in One Meta-Analysis?

Combining across species is statistically permissible but clinically meaningful only when the target condition, index test mechanism, and reference standard are equivalent. For example, a point-of-care antigen test for a viral pathogen may behave similarly across dogs and cats, but a test for a production-limiting disease in cattle may have different operating characteriztics in dairy versus beef systems. Decide the scope before analysis and justify it in the protocol. If you combine across species, include species as a covariate in the meta-regression to test for differences in accuracy. If heterogeneity is substantial, report separate summary estimates for each species or production system instead of a single pooled value.

### What Resources and Statistical Expertise Are Required to Perform This Analysis?

A diagnostic accuracy meta-analysis requires statistical software capable of fitting nonlinear mixed models, such as R with the mada or lme4 packages, SAS PROC NLMIXED, or Stata with the metandi command. You also need a systematic review team that includes a librarian or information specialist for the search, two reviewers for independent screening and data extraction, and a statistician familiar with bivariate models. Time and funding should cover database access, full-text retrieval, and at least two rounds of protocol revision. If this expertise is unavailable locally, consider collaborating with a veterinary epidemiologist or biostatistician at another institution. The [EQUATOR Network reporting guidelines](https://www.equator-network.org/) provide checklists that can help structure the project and identify missing elements early.

### How Should I Document the Analysis for Regulatory or Funding Bodies?

Maintain a complete audit trail from protocol to final report. Keep the registered protocol, search strategies with dates, screening decisions, full data extraction forms, and all analysis scripts or command files. Record version numbers of software packages and any manual data corrections. For each analysis step, note the date, the analyst, and the rationale for decisions such as excluding a study or changing a covariate. The [ARRIVE guidelines for reporting animal research](https://arriveguidelines.org/) and the [AVMA practice resources](https://www.avma.org/resources-tools) offer frameworks for transparent reporting that satisfy most institutional and funding requirements. Store the final dataset in a machine-readable format with a data dictionary so that reviewers can reproduce the analysis independently.

## Related Clinical & Scientific Guides

* [Conducting Systematic Reviews of Veterinary Diagnostic Test Accuracy](/knowledge/veterinary-medicine/veterinary-research-methods/conducting-systematic-reviews-veterinary-diagnostic-test-accuracy)
* [Bias in Veterinary Research: Types, Sources, and Mitigation](/knowledge/veterinary-medicine/veterinary-research-methods/bias-veterinary-research-types-sources-mitigation)
* [Cluster Randomized Trials in Veterinary Research: Design and Analysis](/knowledge/veterinary-medicine/veterinary-research-methods/cluster-randomized-trials-veterinary-research-design-analysis)


## References and Further Reading

- [The Impact of Mild Cognitive Impairment on Gait and Balance: A Systematic Review and Meta-Analysis of Studies Using Instrumented Assessment.](https://pubmed.ncbi.nlm.nih.gov/27172932/). 2017.
- [Cerebral blood flow threshold of ischemic penumbra and infarct core in acute ischemic stroke: a systematic review.](https://pubmed.ncbi.nlm.nih.gov/16574919/). 2006.
- [Accuracy of linear measurements on CBCT images related to presurgical implant treatment planning: A systematic review.](https://pubmed.ncbi.nlm.nih.gov/30328204/). 2018.
- [The methodological quality assessment tools for preclinical and clinical studies, systematic review and meta-analysis, and clinical practice guideline: a systematic review.](https://pubmed.ncbi.nlm.nih.gov/25594108/). 2015.
- [FLOTAC: new multivalent techniques for qualitative and quantitative copromicroscopic diagnosis of parasites in animals and humans.](https://pubmed.ncbi.nlm.nih.gov/20203667/). 2010.
- [Intranasal oxytocin versus placebo in the treatment of adults with autism spectrum disorders: a randomized controlled trial.](https://pubmed.ncbi.nlm.nih.gov/23216716/). 2012.
- [ARRIVE Guidelines 2.0 for Reporting Animal Research](https://arriveguidelines.org/). PLOS Biology, 2020.
- [EQUATOR Network Reporting Guidelines](https://www.equator-network.org/). EQUATOR Network.
- [MSD Veterinary Manual, Professional Edition](https://www.msdvetmanual.com/). MSD Veterinary Manual.

## Related Articles

- [Diagnostic Test Accuracy Studies in Veterinary Medicine: Design and Reporting](/knowledge/veterinary-medicine/veterinary-research-methods/diagnostic-test-accuracy-studies-veterinary-design-reporting)
- [Conducting Systematic Reviews of Veterinary Diagnostic Test Accuracy](/knowledge/veterinary-medicine/veterinary-research-methods/conducting-systematic-reviews-veterinary-diagnostic-test-accuracy)
- [Network Meta-Analysis for Veterinary Treatment Comparisons](/knowledge/veterinary-medicine/veterinary-research-methods/network-meta-analysis-veterinary-treatment-comparisons)
- [Appraising Diagnostic Accuracy Studies in Veterinary Medicine](/knowledge/veterinary-medicine/veterinary-research-methods/appraising-diagnostic-accuracy-studies-veterinary-medicine)
- [Measuring Agreement in Veterinary Diagnostic Tests](/knowledge/veterinary-medicine/veterinary-research-methods/measuring-agreement-veterinary-diagnostic-tests)

> 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.