# Survival Analysis in Veterinary Medicine: Kaplan-Meier and Cox Regression


## Key Takeaways

- Survival analysis methods, specifically Kaplan-Meier estimation and Cox proportional hazards regression, are critical for analyzing time-to-event data in veterinary medicine, such as time to death, tumor recurrence, or recovery from infection. These techniques inherently accommodate censoring, a common occurrence where an event has not yet happened by the end of observation or due to loss to follow-up, preventing bias that would arise from excluding such data.

- The Kaplan-Meier estimator provides a non-parametric method to describe survival probabilities over time, yielding a step-function curve and enabling the calculation of median survival time. Comparisons between groups are typically performed using the log-rank test, which assesses differences in survival distributions, though it can be less powerful if survival curves cross.

- The Cox proportional hazards model offers a semiparametric approach for multivariable analysis, allowing the assessment of multiple covariates (e.g., breed, age, treatment) on the hazard rate of an event. Its core assumption is proportional hazards, meaning the hazard ratio between any two individuals remains constant over time, which must be rigorously checked using methods like Schoenfeld residual plots.

- Key practical steps in veterinary survival analysis include defining the cohort and index date, clearly specifying the event and censoring rules (especially for euthanasia), extracting and cleaning data, describing survival with Kaplan-Meier curves, comparing groups with the log-rank test, and fitting a Cox model with thorough assumption checking and influential observation assessment.

- Common pitfalls include misinterpreting censoring (e.g., treating euthanasia for unrelated causes as an event), violating the proportional hazards assumption, reporting median survival when the 50% threshold is not reached, and over-adjusting for variables on the causal pathway. The number of events, not just the total sample size, is crucial for statistical power, and power calculations should be based on the expected event rate.

---

Survival analysis comprises a family of statistical methods for analyzing time-to-event data, where the outcome of interest is the time until a defined endpoint such as death, tumor recurrence, or recovery from infection. These methods are distinguished from ordinary regression approaches by their capacity to accommodate censoring, the situation in which the event has not occurred by the end of the observation period or the subject is lost to follow-up. For veterinary researchers, survival analysis provides the quantitative framework for comparing treatment protocols, identifying prognostic factors, and evaluating disease outcomes across species and clinical settings.

This article addresses the conceptual foundations and practical application of the two most widely used survival analysis techniques in veterinary research: the Kaplan-Meier estimator for descriptive survival curves and the Cox proportional hazards model for multivariable analysis. The target reader is a veterinary researcher or graduate student who needs to interpret survival analyzes in the literature, design studies with time-to-event outcomes, or perform such analyzes on clinical data. The article covers censoring mechanisms, the assumptions underlying each method, model building and diagnostics, and the reporting standards expected in peer-reviewed veterinary journals. Other regression models, including parametric survival models and competing risk approaches, are outside the scope of this reference.

## At a Glance

| Parameter | Kaplan-Meier | Cox Proportional Hazards |
| --- | --- | --- |
| Primary purpose | Descriptive estimation of survival probability over time | Multivariable modeling of factors associated with event risk |
| Censoring handling | Yes, right censoring accommodated | Yes, right censoring accommodated |
| Covariates | Not accommodated directly, subgroup comparisons only | Multiple continuous and categorical covariates |
| Key assumption | None beyond independent censoring | Proportional hazards over time |
| Output | Step function with survival probabilities and median survival time | Hazard ratios with confidence intervals for each covariate |
| Typical veterinary application | Comparing survival between treatment groups in a clinical trial | Identifying prognostic factors in a retrospective cohort |
| Software implementation | Base functions in R, Stata, SAS, and commercial statistics packages | Same packages, with additional diagnostic tools |

## The Structure of Time-to-Event Data

Time-to-event data in veterinary medicine arise whenever the research question concerns the interval between a starting point and an endpoint. The starting point may be diagnosis, treatment initiation, or study enrollment. The endpoint may be death, euthanasia, metastasis, or resolution of clinical signs. The defining feature of such data is that some subjects will not experience the event during the observation window. These subjects are said to be censored.

Right censoring, the most common form, occurs when the event has not happened by the time the study ends or the subject is lost to follow-up. A dog with oral melanoma that is alive at the conclusion of a vaccine trial contributes a censored observation, as does a cat that moves to another practice and cannot be traced. Left censoring, where the event occurred before the subject entered the study, and interval censoring, where the event is known to have occurred within a window but the exact time is unknown, occur less frequently in veterinary data but require different analytical approaches. The distinction between censoring and a true event is critical: treating censored observations as events underestimates survival, while excluding them biases the sample toward shorter survival times.

The veterinary literature contains numerous examples of survival analysis applied to clinical questions. A study of chronic kidney disease in dogs attending UK veterinary practices used Kaplan-Meier survival curves and multivariable Cox regression to estimate survival after diagnosis and to identify risk factors associated with poorer outcomes [O'Neill DG, Elliott J, Church DB, et al. Chronic kidney disease in dogs in UK veterinary practices: prevalence, risk factors, and survival](https://pubmed.ncbi.nlm.nih.gov/23647231/). Similarly, an evaluation of a xenogeneic DNA vaccine for oral malignant melanoma in dogs compared survival of vaccinated dogs with historical controls using Kaplan-Meier analysis, with death attributable to melanoma as the event of interest [Grosenbaugh DA, Leard AT, Bergman PJ, et al. Safety and efficacy of a xenogeneic DNA vaccine encoding for human tyrosinase as adjunctive treatment for oral malignant melanoma in dogs following surgical excision of the primary tumor](https://pubmed.ncbi.nlm.nih.gov/22126691/).

## The Survival Function and Hazard Function

The survival function, denoted S(t), represents the probability that a subject survives beyond time t. It is a non-increasing function that starts at 1.0 at time zero and approaches zero as time increases. The hazard function, denoted h(t), represents the instantaneous rate of the event at time t conditional on having survived to that time. The hazard is not a probability and can exceed 1.0, it is better understood as a rate per unit time.

These two functions are mathematically linked. The cumulative hazard, H(t), is the integral of the hazard function from zero to t, and the survival function equals exp[-H(t)]. This relationship matters for practical reasons. When the proportional hazards assumption holds, the ratio of hazards between two groups remains constant over time, which permits the Cox model to estimate hazard ratios without specifying the baseline hazard. When the assumption fails, the hazard ratio becomes a time-dependent quantity and the Cox model produces misleading estimates.

## The Kaplan-Meier Estimator

The Kaplan-Meier method estimates the survival function non-parametrically from observed event and censoring times. At each distinct event time, the survival probability is multiplied by the proportion of subjects at risk who experience the event. Subjects censored at a given time are removed from the risk set at that point but contribute to the survival estimate up to the time of censoring.

The resulting curve is a step function that drops at each event time. The median survival time, the time at which survival probability reaches 0.5, is the most commonly reported summary statistic. It can only be estimated when at least half of the subjects have experienced the event, otherwise the median is reported as not reached. In the canine melanoma vaccine trial, median survival time could not be determined for vaccinated dogs because fewer than half had died from melanoma by the end of the study, a finding that itself indicated improved survival relative to controls [Grosenbaugh DA, et al.](https://pubmed.ncbi.nlm.nih.gov/22126691/).

Comparison of two or more Kaplan-Meier curves is performed with the log-rank test, which evaluates whether the survival distributions differ significantly. The log-rank test gives equal weight to all event times and is most powerful when the hazard ratio is constant over time. When survival curves cross, indicating that the relative risk changes over time, the log-rank test loses power and alternative tests such as the Wilcoxon test, which weights earlier events more heavily, may be more appropriate.

## The Log-Rank Test and Group Comparisons

The Kaplan-Meier estimator describes survival within a single group. Clinical questions, however, usually compare two or more groups: treated versus untreated, breed A versus breed B, or one staging category against another. The log-rank test is the standard nonparametric procedure for this comparison. It tests the null hypothesis that the survival functions of the groups are identical across the entire follow-up period.

The test statistic is computed by pooling all observed event times across groups. At each distinct event time, the observed number of events in each group is compared with the number expected under the null hypothesis, given the number at risk in each group at that moment. These deviations are summed over all event times, and the resulting statistic follows a chi-square distribution with degrees of freedom equal to one less than the number of groups.

The log-rank test gives equal weight to every event time. This property is a strength when the hazard ratio is roughly constant over time, but it becomes a limitation when survival curves cross. If two curves cross, the log-rank test may report no significant difference even when early survival strongly favours one group and late survival favours the other. In that situation, examine the plotted curves before interpreting the test result. The Wilcoxon or Breslow variant, which weights early events more heavily, can serve as a supplementary check.

The log-rank test is an omnibus test. It detects any difference between survival functions but does not quantify the size of that difference. For effect size, report the hazard ratio obtained from a Cox model, as described below. The test also assumes that censoring is independent of the event of interest. If animals lost to follow-up differ systematically from those retained, the comparison is biased.

## The Cox Proportional Hazards Model

The Cox proportional hazards model extends survival analysis to accommodate multiple covariates simultaneously. It models the hazard function as a baseline hazard multiplied by an exponential function of the covariates:

h(t|X) = h₀(t) × exp(β₁X₁ + β₂X₂ +... + βₚXₚ)

The baseline hazard h₀(t) is left unspecified. The model estimates the regression coefficients β, and exp(β) is the hazard ratio for a one-unit increase in the corresponding covariate, holding all other covariates constant. Because the baseline hazard is not parameterised, the Cox model is called semiparametric. This flexibility makes it the default choice for multivariable survival analysis in veterinary research.

The proportional hazards assumption requires that the hazard ratio between any two individuals remains constant over time. If the assumption is violated, the model produces biased estimates. Assess the assumption graphically by plotting log-minus-log survival curves for categorical covariates, parallel curves support proportionality. For continuous covariates, examine Schoenfeld residuals against time. A formal test of the scaled Schoenfeld residuals is available in most statistical software. When proportionality fails, options include stratifying on the offending variable, using time-dependent covariates, or fitting a parametric model with a different structure.

The UK veterinary practice study of chronic kidney disease in dogs illustrates the standard workflow. The investigators used Kaplan-Meier curves to describe survival and multivariable Cox regression to identify independent risk factors, including breed, age, and comorbid cardiac disease [O'Neill et al., 2013](https://pubmed.ncbi.nlm.nih.gov/23647231/). The Cox model allowed them to separate the effect of breed from the effect of age, which would be confounded in a crude comparison.

## Building and Checking the Model

Model building in survival analysis follows the same principles as other multivariable regression. Start with a causal model based on the literature and biological plausibility, not with automated stepwise selection alone. Include variables that are known confounders even if their p-values are not significant. Exclude variables that lie on the causal pathway between the exposure of interest and the outcome, because adjusting for them removes part of the exposure effect.

Check for collinearity before fitting. Two highly correlated covariates, such as body weight and body condition score, produce unstable coefficient estimates with inflated standard errors. Examine the correlation matrix and variance inflation factors before finalising the model.

After fitting, assess influential observations. A single animal with an unusual combination of covariates and a short survival time can dominate the results. Plot dfbeta statistics against time and inspect any observation with a large influence. In small veterinary datasets, one or two influential cases can change the conclusions, so this step is not optional.

The proportional hazards assumption must be checked for every covariate in the final model. If a treatment effect diminishes over time, as seen in some oncology studies, the Cox model with a constant hazard ratio will underestimate early benefit and overestimate late benefit. The DNA vaccine study for canine oral melanoma reported improved survival for vaccinates, but the median survival time for the vaccine group could not be determined because too few dogs had died by the end of follow-up [Grosenbaugh et al., 2011](https://pubmed.ncbi.nlm.nih.gov/22126691/). This pattern, where the survival curve for one group never reaches 50%, is common in veterinary oncology and should be reported explicitly instead of obscured.

## Practical Workflow for a Veterinary Survival Analysis

The following sequence applies to a typical retrospective cohort study using medical records. Adjust the details for prospective studies or for data extracted from clinical trials.

**Step 1: Define the cohort and the index date.** The index date is the moment from which survival is measured. Common choices are diagnosis date, treatment start date, or surgery date. The choice must be clinically meaningful and must be applied identically to every animal. For a study of surgical outcomes, the index date is the surgery date. For a study of disease progression, it is the date of definitive diagnosis.

**Step 2: Define the event and the censoring rules.** The event must be unambiguous. All-cause mortality is the most objective outcome but mixes deaths from unrelated causes with deaths from the disease of interest. Disease-specific mortality requires accurate cause-of-death determination, which is often difficult in veterinary medicine. Euthanasia is a particular problem. Euthanasia for progression of the disease under study is often treated as an event, while euthanasia for unrelated causes is censored. The distinction requires careful record review and explicit criteria. State these criteria in the methods section.

**Step 3: Extract and clean the data.** Record the index date, the event date or last known alive date, the event status, and all covariates. Verify date consistency: the event date must not precede the index date, and the censoring date must not exceed the end of the study period. Check for duplicate records and for animals with missing covariate data. Decide a priori how missing data will be handled. Complete-case analysis is simple but can introduce bias if missingness is related to outcome.

**Step 4: Describe survival with Kaplan-Meier curves.** Plot the overall survival curve and curves for each categorical covariate of interest. Report median survival times with confidence intervals. If the median is not reached, report the survival proportion at a fixed time point, such as the one-year survival proportion, with its confidence interval.

**Step 5: Compare groups with the log-rank test.** Use the log-rank test for the primary comparison. Examine the curves for crossing before interpreting the result. If curves cross, consider whether the proportional hazards assumption is plausible and whether a different analysis is needed.

**Step 6: Fit the Cox model.** Include covariates based on the causal model. Check the proportional hazards assumption for each covariate. Examine influential observations. Report hazard ratios with 95% confidence intervals and p-values. Present the final model in a table with the number of events, also the number of animals, because the effective sample size in survival analysis is the number of events.

**Step 7: Document the analysis.** Record the software and version, the functions used, the model specification, and the diagnostic checks performed. This documentation allows another researcher to reproduce the analysis.

## Reporting Standards and Common Pitfalls

| Pitfall | Consequence | Prevention |
| --- | --- | --- |
| Censoring animals at last visit instead of last known alive date | Underestimates survival time | Use the later of last examination or last owner contact |
| Treating euthanasia for unrelated causes as disease events | Overestimates disease-specific mortality | Apply explicit euthanasia classification criteria |
| Ignoring the proportional hazards assumption | Biased hazard ratios | Test the assumption and stratify or use time-dependent terms |
| Reporting median survival when the curve does not reach 50% | Misleading summary statistic | Report survival proportion at a fixed time point instead |
| Adjusting for variables on the causal pathway | Removes part of the treatment effect | Build the causal model before analysis |
| Failing to report the number of events | Overstates precision | Report events alongside animals |

The number of events deserves emphasis. A study with 200 animals but only 20 events has the effective power of a much smaller study. Confidence intervals will be wide, and the analysis may fail to detect clinically meaningful differences. Power calculations for survival studies should be based on the expected event rate, not the expected sample size.

Species and clinical context change the analysis in specific ways. In production animal medicine, the unit of analysis may be the herd instead of the individual animal, and culling decisions create informative censoring that is difficult to handle. In wildlife studies, the index date is often unknown and left-truncation is common. In companion animal oncology, euthanasia decisions are influenced by owner finances and clinician recommendation, creating a competing risk that the standard Cox model does not address. The [WOAH terrestrial animal health standards](https://www.woah.org/en/what-we-do/standards/codes-and-manuals/terrestrial-code-online-access/) provide context for surveillance definitions, but the statistical methods described here apply across species with these caveats in mind.

Competing risks analysis, where death from other causes precludes the event of interest, is an extension beyond the Cox model that should be considered when euthanasia or unrelated mortality is common. The standard Kaplan-Meier approach treats competing events as censoring, which overestimates the cumulative incidence of the event of interest. For veterinary oncology studies where euthanasia is frequent, this bias can be substantial.

## Recognized Complications and Early Detection

Survival analyzes fail in characteriztic ways, and most failures become apparent before formal testing if the data structure is examined critically. The most consequential failure is informative censoring, where the reason an animal leaves follow-up is related to its prognosis. A dog withdrawn from a oncology trial because of progressive disease, a cow culled for declining production that reflects subclinical illness, and a horse euthanased for financial reasons when treatment options are exhausted all violate the assumption that censoring is independent of the event of interest. Detection requires a tabulation of censoring reasons by treatment group and by time interval. If censoring proportions differ markedly between groups, or if censoring clusters in the period immediately preceding expected events, the assumption is suspect and a sensitivity analysis treating those cases as events should be run.

A second failure mode is left truncation, where animals that die before enrollment are invisible to the analysis. This arises in practice when electronic records are searched retrospectively and only animals with a confirmed diagnosis are included. Animals that died before diagnosis are excluded, and survival times are overestimated. The same problem occurs when the start time is mis-specified. Using the date of diagnosis instead of the date of onset of clinical signs is standard, but if the two are conflated, the survival clock starts late and the analysis is biased. Early detection requires a clear written definition of time zero before data extraction begins, and a check that no animal has a recorded event before its stated entry time.

Proportional hazards violations are the third major complication. The Cox model assumes the hazard ratio between groups is constant over time. When treatment benefit wanes, or when an intervention is more effective in the first year than later, the assumption fails. Schoenfeld residuals plotted against time will show a trend, and a formal test of their correlation with time will be significant. A stratified Cox model, or a model with a time-dependent coefficient, is the corrective step. The [CDC principles of epidemiology](https://www.cdc.gov/csels/dsepd/ss1978/index.html) describe the underlying logic of time-based measures that these diagnostics rely on.

## Common Errors and Corrective Action

Less experienced analysts frequently treat the median survival time as the only result worth reporting. The median is uninformative when the survival curve does not reach 50%, as occurred in the canine melanoma vaccine trial where median survival could not be determined for vaccinates because the event rate was too low [safety and efficacy of a xenogeneic DNA vaccine for oral malignant melanoma in dogs](https://pubmed.ncbi.nlm.nih.gov/22126691/). Reporting the survival proportion at fixed time points, such as one and two years, together with confidence intervals, is the corrective action.

A second recurring error is the analysis of survival time as a continuous outcome using linear regression. This ignores censoring entirely and produces biased estimates whenever any animal has not reached the event. The error is detectable when the dataset contains animals with follow-up times but no event, and the analyst has excluded them or assigned their follow-up time as the outcome. The corrective action is to use a method that accommodates censoring, such as the Kaplan-Meier estimator or Cox regression, as applied in the canine chronic kidney disease study that used both approaches to estimate survival [chronic kidney disease in dogs in UK veterinary practices](https://pubmed.ncbi.nlm.nih.gov/23647231/).

A third error is over-adjustment. Including variables on the causal pathway between exposure and outcome, such as adjusting for disease stage when evaluating a treatment that prevents progression, attenuates or removes the effect of interest. The corrective action is to specify the causal model before analysis and to distinguish confounders from mediators.

## Limitations of the Current Evidence

Veterinary survival data are frequently drawn from referral populations, which over-represent severe disease and owners with greater financial capacity. Primary care data, such as the UK practice database used in the CKD study, are more representative but often lack standardized staging and follow-up protocols [chronic kidney disease in dogs in UK veterinary practices](https://pubmed.ncbi.nlm.nih.gov/23647231/). The evidence base for many species is thin. Survival estimates for exotic pets, production animals outside commercial settings, and wildlife are often derived from small convenience samples with heterogeneous management.

Expert opinion differs on the handling of euthanasia as an event. Some argue euthanasia should be treated as the event of interest because it reflects disease progression in most companion animal settings. Others contend that euthanasia for financial or behavioral reasons is informative censoring and should be modelled separately. A competing risks framework, where euthanasia and natural death are treated as distinct outcomes, is increasingly recommended but is not yet standard in the veterinary literature. The [WOAH animal health surveillance standards](https://www.woah.org/en/what-we-do/animal-health-and-welfare/disease-data-collection/) emphasize that outcome definitions must be harmonised across sites, a principle that applies equally to survival endpoints in multi-center studies.

## Escalation and Referral

Statistical consultation is warranted when the dataset has complex censoring patterns, when competing risks are present, or when the proportional hazards assumption cannot be satisfied by stratification. A veterinary epidemiologist or biostatistician should be involved before data collection if the study design involves multiple sites, variable follow-up intervals, or cluster-randomised interventions. Laboratory involvement is required when outcome ascertainment depends on diagnostic tests with imperfect sensitivity or specificity, because misclassification of the event biases survival estimates in ways that routine software output will not reveal.

Regulatory reporting obligations apply when survival data are generated for product licensure, vaccine efficacy, or notifiable disease surveillance. The [WOAH terrestrial animal health code](https://www.woah.org/en/what-we-do/standards/codes-and-manuals/terrestrial-code-online-access/) sets out reporting requirements for listed diseases, and national authorities may impose additional obligations. When survival analysis informs a clinical trial of a biologic or pharmaceutical, the study protocol and statistical analysis plan should be reviewed by the relevant regulatory body before enrollment begins. The [AVMA practice resources](https://www.avma.org/resources-tools) provide guidance on professional obligations in clinical research settings.

## Troubleshooting Table

| Observation | Likely Cause | Discriminating Check |
|---|---|---|
| Survival curve drops sharply at study end | Administrative censoring mis-coded as events | Inspect event dates against last contact dates |
| Hazard ratio changes sign over follow-up | Non-proportional hazards | Schoenfeld residual plot and test |
| Median survival missing despite adequate follow-up | Event rate below 50% | Report survival proportions at fixed times instead |
| Censoring proportion much higher in one group | Informative censoring | Tabulate censoring reasons by group |
| Survival time exceeds study duration for many animals | Left truncation or late entry | Verify time zero definition against records |
| Very wide confidence intervals | Small sample or sparse events | Report events per variable and consider penalised methods |

## Frequently Asked Questions

### How Many Animals Do I Need for a Reliable Kaplan-Meier or Cox Analysis?

There is no universal minimum, but the number of events, not the total sample size, drives statistical power. A common rule of thumb is at least 10 events per predictor variable in a Cox model. For Kaplan-Meier curves, fewer than 10 events in a group makes the confidence intervals so wide that clinically meaningful comparisons become unreliable. Veterinary datasets from practice databases can be large, as in a study of 107,214 dogs attending UK practices, yet the chronic kidney disease case count still limited precision [O'Neill et al., 2013](https://pubmed.ncbi.nlm.nih.gov/23647231/). If events are scarce, consider a simpler model, combine categories, or report the analysis as descriptive only.

### What Should I Do When My Clinic Lacks Statistical Software?

Spreadsheet programs can produce Kaplan-Meier tables and simple log-rank tests manually. For Cox regression, free statistical environments such as R handle survival packages competently. Many veterinary schools offer collaborative statistical support or short courses. The [CDC principles of epidemiology](https://www.cdc.gov/csels/dsepd/ss1978/index.html) provide the underlying methods for calculating person-time and event rates by hand, which is useful for small datasets. When software access is limited, prioritize a clean, structured spreadsheet with one row per animal, a time variable, an event indicator, and a censoring indicator. This format transfers directly into any analysis package later.

### How Do Survival Results Differ Between Dogs and Cats or Production Animals?

The survival analysis framework is species-neutral, but the practical constraints differ. Dogs and cats in companion animal practice often have irregular follow-up because owners move or change clinics, producing heavy right censoring. Production animals are usually observed until a fixed culling or slaughter point, so interval censoring and competing risks such as culling for production failure become central. The [MSD Veterinary Manual](https://www.msdvetmanual.com/) notes that disease definitions and staging systems vary by species, which directly affects how you define the event. A death event in a dog is unambiguous, whereas in a dairy cow the event may be culling, and the reason for culling must be recorded carefully.

### How Should I Handle Animals Lost to Follow-Up or Euthanized for Unrelated Causes?

Record the date and reason for every exit from the study. Euthanasia for the disease of interest counts as an event. Euthanasia for an unrelated condition, such as trauma, should be censored at the date of euthanasia. Animals lost to follow-up are censored at their last known alive date. The key assumption is that censoring is independent of the outcome, meaning censored animals would have had the same future risk as those remaining. This assumption is violated if owners elect euthanasia for financial reasons in a study of treatment cost. Sensitivity analysis, treating uncertain euthanasia cases first as events and then as censored, shows how robust your conclusions are.

### What Are the Minimum Data Fields I Must Record for a Future Survival Study?

Record a unique animal identifier, species, breed, age at entry, sex and neuter status, date of entry into the study, the primary diagnosis and staging information, the date of the event or censoring, and the event type. Also record the date of last contact for censored animals. For multicenter studies, include the clinic identifier because practice-level variation in diagnostic workup and treatment protocols can confound results. The [WOAH animal health surveillance standards](https://www.woah.org/en/what-we-do/animal-health-and-welfare/disease-data-collection/) emphasize consistent case definitions and data quality, which apply equally to clinical research databases. A standardized record form completed at enrollment prevents missing baseline covariates.

### How Do I Explain Survival Results to an Owner or a Referring Veterinarian?

Use absolute terms instead of hazard ratios. A hazard ratio of 2.0 does not mean the animal will live half as long. Instead, state the median survival time from the Kaplan-Meier curve, for example, "dogs in this group lived a median of 18 months." The [AVMA practice resources](https://www.avma.org/resources-tools) advise communicating prognosis with clear caveats about individual variation. Explain that the median means half of similar animals lived longer and half lived shorter. Acknowledge that published survival times come from specific populations and may not match your patient exactly. Offer the range of survival times, also the median, and frame the discussion around quality of life milestones.

## Related Clinical & Scientific Guides

* [Evaluating Veterinary Surveillance System Attributes](/knowledge/veterinary-medicine/veterinary-epidemiology/evaluating-veterinary-surveillance-system-attributes)
* [Network Analysis for Infectious Disease Spread in Animal Populations](/knowledge/veterinary-medicine/veterinary-epidemiology/network-analysis-infectious-disease-spread-animal-populations)
* [Randomized Controlled Trials in Veterinary Field Settings](/knowledge/veterinary-medicine/veterinary-epidemiology/randomized-controlled-trials-veterinary-field-settings)


## References and Further Reading

- [Chronic kidney disease in dogs in UK veterinary practices: prevalence, risk factors, and survival.](https://pubmed.ncbi.nlm.nih.gov/23647231/). 2013.
- [A comparison of T-wave alternans, signal averaged electrocardiography and programd ventricular stimulation for arrhythmia risk stratification.](https://pubmed.ncbi.nlm.nih.gov/11127468/). 2000.
- [Anesthetic technique may affect prognosis for ovarian serous adenocarcinoma: a retrospective analysis.](https://pubmed.ncbi.nlm.nih.gov/21436156/). 2011.
- [Ultraviolet B but not ultraviolet A radiation initiates melanoma.](https://pubmed.ncbi.nlm.nih.gov/15374941/). 2004.
- [Safety and efficacy of a xenogeneic DNA vaccine encoding for human tyrosinase as adjunctive treatment for oral malignant melanoma in dogs following surgical excision of the primary tumor.](https://pubmed.ncbi.nlm.nih.gov/22126691/). 2011.
- [Use of dietary linoleic acid for secondary prevention of coronary heart disease and death: evaluation of recovered data from the Sydney Diet Heart Study and updated meta-analysis.](https://pubmed.ncbi.nlm.nih.gov/23386268/). 2013.
- [WOAH Animal Health Surveillance Standards](https://www.woah.org/en/what-we-do/animal-health-and-welfare/disease-data-collection/). WOAH.
- [CDC Principles of Epidemiology in Public Health Practice](https://www.cdc.gov/csels/dsepd/ss1978/index.html). CDC.
- [MSD Veterinary Manual, Professional Edition](https://www.msdvetmanual.com/). MSD Veterinary Manual.

## Related Articles

- [Cohort Studies in Veterinary Medicine: Design and Analysis](/knowledge/veterinary-medicine/veterinary-epidemiology/cohort-studies-veterinary-medicine-design-analysis)
- [Regression Analysis in Veterinary Epidemiology: Logistic and Poisson Models](/knowledge/veterinary-medicine/veterinary-epidemiology/regression-analysis-veterinary-epidemiology-logistic-poisson-models)
- [Sensitivity Analysis in Veterinary Disease Models](/knowledge/veterinary-medicine/veterinary-epidemiology/sensitivity-analysis-veterinary-disease-models)
- [Outbreak Investigation in Veterinary Medicine: A Step-by-Step Guide](/knowledge/veterinary-medicine/veterinary-epidemiology/outbreak-investigation-veterinary-medicine-step-by-step-guide)
- [Risk Factor Analysis for Disease in Animal Populations](/knowledge/veterinary-medicine/veterinary-epidemiology/risk-factor-analysis-disease-animal-populations)

> 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.