How to read a canine survival curve

Start with the event

To read a canine survival curve, identify the event being tracked and when the clock started, read the axes, follow the shape of the line, and check the number-at-risk table before trusting the tail. In veterinary and epidemiological research, “survival” doesn’t always mean death. It might mean time to relapse, time to a diagnosis, or time until a dog leaves a study for any reason.

Check the study’s methods for how the event is defined before looking at the curve itself. A curve tracking “death from any cause” tells a different story than one tracking “disease-specific death” or “time to first seizure.” Mixing these up leads to reading the chart wrong from the start.

Also note the starting point, sometimes called time zero. This could be the date of diagnosis, the date of enrollment in a study, or the date a treatment began. Two curves that look similar can represent very different journeys if their starting points differ.

With right censoring, a dog contributes its observed event-free follow-up up to the censoring time. What happens afterward is unknown to the analysis; censoring is not a death event and does not mean the dog survives indefinitely. This distinction matters for how the curve is built and how far you can trust it toward the right-hand side, where fewer dogs remain under observation.

A practical guide to Kaplan-Meier curves explains event times, censoring and risk sets in clinical research. These mathematical reading principles also apply when a study uses the method for dogs.

Before reading the curve itself, confirm three things: what counts as the event, when the clock started, and whether censoring is explained.

What do the axes and shape of the curve show?

Reading a survival curve well means working through it in a fixed order, starting from the beginning. A consistent sequence keeps you from over-reading a single point on the graph.

Start with the axes. The x-axis shows time (days, months, or years), and the y-axis shows the proportion of dogs who have not yet experienced the event, usually expressed as a percentage or a decimal between 0 and 1.

Next, focus on the overall shape of the curve rather than any single number on it:

• A steep early drop suggests many events happening soon after time zero.
• A flat stretch in an unrounded Kaplan-Meier curve has no recorded events during that interval; censoring can still occur.
• Small vertical tick marks on the line usually mark dogs who were censored at that point, separate from the events plotted along the curve.

Then check where multiple curves separate, if the study compares groups. The gap between two lines at a given time point reflects the difference measured at that moment, and it describes the group only, without predicting the outcome for any individual dog.

Finally, look for the number of dogs still being followed as time goes on. This number is often shown below the graph and shrinks the farther right you look on the timeline. A curve extending years beyond where meaningful numbers of dogs remain deserves more caution, not less.

Reading order matters because it stops you from anchoring on the wrong detail. A dramatic-looking dip late in a curve, based on two or three remaining dogs, carries far less information than the same size dip earlier when dozens of dogs were still being tracked.

Use a fictional reading example

A hypothetical curve helps illustrate the mechanics without claiming to represent any real dataset. Picture a chart labeled “Time to Event (Months)” on the x-axis and “Proportion Event-Free” on the y-axis, starting at 1.0.

Suppose the line stays flat near 1.0 for the first two months, then drops to roughly 0.6 by month six, then flattens again through month twelve. Reading this shape step by step:

• Months 0 to 2: before the first downward step, no events are recorded in this hypothetical group.
• Months 2 to 6: this is where most of the hypothetical events cluster.
• Months 6 to 12: no further events appear in this hypothetical interval. Check how many subjects remain at risk before interpreting that flat stretch.

None of these numbers come from a published dataset. This example exists purely to demonstrate how to move from raw shape to a plain-language description, without turning that description into a claim about any actual breed, disease, or treatment.

The same reading applies to two hypothetical curves compared side by side. If one imagined line sits above another for most of the timeline, that separation only describes the pattern within this fictional data. It says nothing on its own about why the separation exists or whether it would hold up in a different sample of dogs.

Treat any real curve you encounter with the same neutral, step-by-step approach: describe the shape first, and resist attaching meaning to it before checking the underlying numbers.

Check the risk table and uncertainty

The number-at-risk table underneath a survival curve tells you how much to trust each part of the line. This row of numbers, usually printed directly below the x-axis, shows how many dogs remain at risk just before the stated time points: still observed and not yet having experienced the defined event.

A curve can look dramatic in its tail while resting on very few remaining dogs. When the risk table drops to single digits, any bend or drop in the curve at that point reflects a small sample, not necessarily a stable pattern.

Confidence intervals, when shown as a shaded band around the line, describe a range of plausible values rather than a precise measurement. A wide band means the estimate carries more uncertainty at that time point; a narrow band means less. Bands often widen toward the right side of a curve as the number at risk shrinks.

Some studies report a median survival time, the point where the curve crosses 0.5 on the y-axis. This single number can be useful for a quick reference, but it summarizes an entire distribution into one point and can obscure how spread out individual outcomes were.

The KMunicate paper proposes clearer Kaplan-Meier displays and surveys stakeholder views, including displays with confidence intervals and expanded tables showing participants at risk, censored and with events. Its evidence comes from stakeholder surveys on reporting methods, and its scope stays outside any canine cohort.

Before drawing any conclusion from a curve’s tail, check the risk table first.

Avoid a causal leap

A survival curve describes what happened in a specific group of dogs. Explaining why it happened is a separate question, one the curve alone cannot answer. One error to avoid is treating a difference between two curves as proof that one treatment, diagnosis or condition caused the difference in outcomes.

Observational data, in particular, can reflect other differences between the groups being compared, separate from the factor a study is examining. Dogs in one group might differ in age, breed, disease stage, or access to care, and any of these could shape the curve independent of whatever variable the study set out to test.

The curve alone does not establish causation. Causal interpretation depends on the study design, including how groups were formed, and the assumptions and analyses used to address alternative explanations; statistical adjustment by itself does not prove causality.

A curve also can’t tell you what will happen to an individual dog. It summarizes a group. A curve showing 60% event-free at month six does not by itself establish a calibrated prognosis for a particular dog. Applying a group estimate to an individual requires the clinical context and an assessment of how well the study population and methods apply.

Questions about what a specific curve implies for a specific dog’s diagnosis, treatment, or prognosis belong with a veterinarian who has access to that dog’s full history and current condition. Reading a curve accurately is a research literacy skill, separate from clinical judgment about an individual case.