What’s the point of a Breed Health Survey? – my April 2016 “Best of Health” article
A few weeks ahead of Crufts, the KC published an overall pedigree breed summary report and breed-specific reports from its health survey which was originally conducted at the end of 2014.
The 2014 survey’s stated aim was to find the most prevalent conditions currently affecting live dogs and the most common causes of death and reasons for euthanasia.
The KC’s press release also identified other benefits arising from the survey: “These results give an idea of where progress has been made, where it still needs to be made, and how it is best to make appropriate changes”.
This was the largest survey of its kind, promoted to 385,000 owners of Kennel Club registered dogs, from 215 Kennel Club recognised dog breeds. It resulted in responses covering 43,207 live dogs and deaths of 5684 dogs, compared with 2004 survey responses for 36,006 live dogs and 15,881 deaths.
Why sampling really matters, or does sampling matter?
The sampling approaches were different in 2004 and 2014, so it’s not surprising that the response rates were also different: 24% in 2004 and 11% in 2014. I know some Breeds were disappointed with their response rates in 2004, but actually, for a paper-based survey issued via Breed Clubs, the overall response was pretty good. According to the 2004 survey report, approximately 57,000 forms were posted out (although the KC website says 70,000 were sent out to Breed Clubs for distribution). Herein lies another key difference between the 2004 and 2014 samples; pedigree dog owners who were members of Breed Clubs vs. pedigree dog owners with a dog registered at the KC. Those different owners will mean there are slightly different biases in the reported results and I mean “bias” from a statistical point of view when it comes to looking at the results.
Working out whether or not these response rates are “good” or “bad” really boils down to tests of statistical significance (where terms like “good” and “bad” are meaningless). People designing surveys do this sort of calculation all the time. You can even work it out for yourself with many of the free online calculators (e.g. http://www.surveysystem.com/sscalc.htm). Estimate the UK population of your chosen breed by multiplying its average age of death by your average number of annual KC registrations. For Dachshunds, that gives us a current population of about 65,000 UK registered Dachshunds. With 789 reports on live Dachshunds in 2014 we can calculate how confident we can be in the results and a margin of error. Overall, I’m pretty happy to use the results from both 2004 and 2014 to help us make informed decisions about Dachshund health. Other breeds should look at their response rates and make their own, informed, decisions.
Does the person on the Clapham Omnibus care about sampling?
The simple answer is no. He, or she, if they are in the slightest bit interested in the health of pedigree dogs, would probably only want to know the answers to three questions:
- what are the biggest health issues in any particular breed (because I own one or might want to own one)?
- how long can I expect my pedigree dog to live?
- have things got better, or worse, in the 10 years between the two KC surveys?
Every breed that has a KC report can look at the data and clearly see what the most frequently reported health issues are. If you’re unlucky (or lucky, depending on your perspective!), there will be so few reports that you can’t find priority issues. For example, in Smooth Dachshunds our report shows 18 different conditions, each of which had just one report. In Long-haired Dachshunds, the KC report tells us there were 19 conditions reported, but not how many dogs affected by each. In these cases, there are simply too few reports to be of much use and our own breed surveys from 2012 and 2015 had more responses and gave us more insights.
You do, however, have to look a bit more carefully because of the way some responses have been categorised. For example, in Mini Smooth Dachshunds there are reports of “Hypersensitivity (allergic) skin disorder”, “Dermatitis”, “Pyoderma”, “Chronic Itching” and “Unspecified Skin, Ear or Coat”. Individually, the highest has a prevalence of just under 4%, but cumulatively, they account for nearly 13% of disease prevalence and this is actually higher than the reports of IVDD (back disease). Skin problems are notoriously difficult to diagnose and resolve and, in some cases, may be due to underlying issues such as Autoimmune disease which may be a consequence of inbreeding, or Immunosuppression caused by certain medications or other illnesses.
So, for us in the Dachshund Health team, the survey has highlighted an area for further investigation (“skin problems”) and we have immediately reviewed and updated our online health reporting categories to help us look at more and better data.
It’s not just disease frequency; impact matters too
Just because a particular health condition comes out as being most prevalent, doesn’t mean it’s the one that has the highest welfare impact on the dogs at the individual and the breed level. A condition that has a low prevalence currently, might have the potential to have a serious impact on the breed in the longer term. For example, if it is known to have a recessive mode of inheritance, there may already be a high Carrier frequency in the breed.
We also need to consider age of onset. Like people, all dogs are going to be ill with something, at some time. Many breeds will have reports showing cardiac conditions; Heart Murmurs had an overall prevalence of 0.9% in the 2014 survey, but without knowing their age of onset, progression and prognosis we are missing lots of useful information. Any early onset condition that results in “early death” is clearly of more significance to owners than a late onset condition where death “due to old age” might be reported.
Age of diagnosis (which may not be the same as onset) is reported in the KC’s overall and breed reports. It’s presented in the form of the visually confusing “Box and Whisker” plots, but once again, breeds do have the opportunity to look more deeply into the health conditions being reported. In the overall data there aren’t many surprises: food allergies, itchy skin, elbow dysplasia all show up early (on average). Cancers and tumours show up later in life (on average). Since skin conditions do not appear as a high frequency cause of death, unlike cancers, we might conclude that many dogs may be suffering these conditions for most of their lives, with or without effective treatment.
The person on the Clapham Omnibus may also be wondering if his/her chosen pedigree breed is more likely to suffer from a particular condition than a different breed, or from pedigree dogs in general.
Breed Clubs can answer that question quite easily using a statistical technique called the Odds Ratio. This has been used extensively to report data from the VetCompass studies. It lets you test, for example, whether the 3% prevalence of Heart Murmurs in breed X is statistically significantly different from the 0.9% prevalence across the whole 2014 survey sample for all breeds. We’ve been able to use the Odds Ratio to confirm the higher rate of epilepsy in Mini Long Dachshunds (which we already knew about) as well as skin conditions in several of the Dachshund varieties (which we didn’t previously know about).
What exactly is “old age”?
Even with the smaller number of deaths reported in 2014 compared with 2004, the KC is able to report the Median age of death overall for pedigree dogs (10 vs. 11.25) and the numbers are also presented within the Breed Reports.
They also report age of death and cause of death using more Box and Whisker charts. There is a wide range of “old age” age of death reports, with 95% ranging from 8-19 years old. This is not surprising because we know that “old age” is different in big dogs and small dogs. There’s plenty of research being done on this, including an epigenetics project which the AHT is participating in and using Miniature Longhaired Dachshunds as their model for a breed with high longevity.
If you’re lucky, your Breed Report will also have an Age of Death Histogram. This shows how many dogs were reported to have died at each age from under one, in one year increments, to whatever the oldest dog was. Looking at the shape of this histogram can be useful. The highest point tells you the age when most of the dogs died; it’s not necessarily the average (Mean) age unless the distribution is perfectly symmetrical.
You might also get a surprise, like in this example which shows a cluster of deaths aged 4-6 and then again at 14-17. Anyone who knows Dachshunds will be able to explain this immediately!
So, are things better or worse after 10 years?
This is the question Jemima Harrison leapt to answer on her blog shortly after the KC published its results, proclaiming an “apocalyptic drop in purebred dog longevity”. It’s an entirely predictable question given the “10 years on” dimension of the 2014 survey. However, on the basis of the results reported, it can’t be answered; at least not with any statistical confidence. So, we’re left with “apocalypse now” headlines, cherry-picked to make a point.
Not only can’t you tell if average age of death has reduced, but it’s hard to make comparisons at the level of individual health conditions. So, in Dachshunds, in the 2004 survey, IVDD was categorised in the same group as Epilepsy as a “Neurological condition”. We have nothing we can directly compare between 2004 and 2014 unless there is some way of revisiting the source data for conditions reported in 2004.
In Dachshunds, we now have three sets of data to help us make comparisons at the level of individual health conditions; the KC’s 2014 results and our own surveys from 2012 and 2015.
However, 3 years isn’t enough for us to expect to see any meaningful changes in prevalence.
What exactly is the point of Breed Health Surveys?
The “point” of Breed Health Surveys depends totally on “the exam question” you are setting out to answer. If the exam question is “what’s the biggest issue?”, most Breed Clubs’ surveys have been able to answer that. The KC’s latest survey does too.
If you want to know where to focus effort on improvement, asking about prevalence and age of diagnosis isn’t going to get you very far. You need to know about severity and welfare impact as well.
Finally, if you want to know if there’s been any change (for the better or worse) over a period of time, you’d better make sure you’re measuring the same thing, in the same way, at the start and the end. Otherwise, you’ll end up with “Garbage in, Gospel out” with someone cherry-picking the results to suit their own argument!
Our approach in Dachshunds has been to answer very specific “exam questions” in each of our surveys and I am convinced that Breed Clubs are best placed to do this sort of work. The KC’s 2004 survey results provided a fantastic platform for breeds to use to shape their own future work. The challenge is that far too many of them don’t have the capability or the capacity to do what’s needed, even if they have the motivation and desire.
I’ve written before about the value of large-scale epidemiological studies such as the VetCompass Project. Yes, there are coding and classification challenges with “dirty data”, but the data set is already massive and continues to grow. Last year it had around 2 million unique dogs in the database, from over 450 veterinary practices. This resource is only going to get more useful, but I suspect the biggest constraint to its value will be the available resources to mine and analyse the data.
One other thing VetCompass does really well is communications; it publishes open-access peer-reviewed papers that meet the needs of a technical audience, with supporting data available. It also publishes infographics; easily digestible summaries that are visually appealing and comprehensible to the “person on the Clapham Omnibus”. VetCompass seems to practice a very “agile” approach to the use of its data assets. It publishes useful chunks of information at frequent intervals, that answer specific questions. We’ve tried a similar agile approach with our Dachshund Breed Surveys; feeding people results and insights as soon as we can. This helps people see the value of their inputs quickly and repeatedly, so they are more likely to support future initiatives.
VetCompass gives us “big data”; breed health surveys give us “small data”. We need both, but most importantly, we need to define the “exam questions” before rushing off to design surveys, or to see if we have data already available.