Where do these figures come from, and what did you not do?

Every number on this site comes from two published sources joined by hand: a peer-reviewed study or screening registry that says how often a breed gets a condition, and the insurer's own policy document that says what it does with that condition. We add the join and the arithmetic. We did not test anything, buy any policy, or file any claim.

What exactly is the number under each insurer?

For one breed we count the conditions that a published source says the breed is at raised risk of. That is N. Then, for each insurer, we count how many of those carry a documented restriction in the first twelve months. A condition counts as restricted when any of these is true on day 365: the waiting period for its class has not passed, the policy excludes it by breed or by age, the policy does not cover its category at all, or the condition usually affects both sides of the body and this insurer's own wording reaches it.

That last test has two halves, and the second one is easy to get wrong. A policy writes its second-side clause in one of two ways. Some define it by class, so the rule covers every condition in that class. Others name the conditions it applies to, and one policy in this register does exactly that: it names ligament, patella, meniscus and soft tissue disorders of the knee. An elbow is not in that list. Where a policy names its conditions, we apply the rule only to the ones it names, and the row says so instead of stretching the clause to fit the class.

The waiting-period test and the second-side test are shown apart from each other on every page, because they are not the same kind of problem. A waiting period that has not passed blocks a claim outright. The second-side rule only bites if one side was already known before you enrolled. Folding them into one figure would make the number look worse than it is, and a number that overstates is as useless as one that flatters.

Which sources are allowed in the prevalence layer?

Three, and every one of them has to state its own sample size. Peer-reviewed VetCompass studies from the Royal Veterinary College, which report how much more often a breed is diagnosed with a disorder than other dogs. Screening registry statistics, which report what share of examined animals were affected. And peer-reviewed veterinary literature that gives a prevalence or a relative risk.

What is not allowed: breed descriptions from breeders or kennel clubs, because they are an interested party; and insurers' own lists of "conditions common in this breed", because the insurer's rules are what the second layer measures, so using their risk list as input would make the whole thing circular.

Why does a breed sometimes have no page at all?

Because it has fewer than three conditions from a source that states its sample size. That threshold is the reason this register covers a small number of breeds well rather than two hundred badly. A page built on one screening figure would be the kind of template page that says nothing, and we would rather have a gap you can see than a page you cannot rely on.

How is the insurer layer read?

From the specimen policy each company publishes, and from nothing else. Not the marketing page, not the FAQ, not a comparison site. For each insurer we record four things per condition class: how long the waiting period runs, what happens to the second side of the body, whether hereditary and congenital conditions are covered, and when a cured condition becomes insurable again. Each of those carries the policy form code and the section number, so you can open the document and land on the sentence.

Policies rarely name every condition class. Most set one general illness waiting period and then carve out the expensive classes with a longer one. Where an insurer states a rule for a class, that rule wins. Where it does not, the general illness rule governs, and the page says so on the row rather than leaving it blank.

What do you not do?

We do not request quotes and we do not publish premiums per breed. The original plan for this site was quarterly quote collection under fixed assumptions; it was dropped because getting a quote means entering personal details into someone else's form, and a quarterly promise we cannot keep is worse than no promise. Market-wide averages we do cite, with their source: for 2025 the North American Pet Health Insurance Association reports $836 a year for dogs, up 11.5%, and $435 for cats, up 12.6%. That is a national average and it is never a price for your breed.

We have not bought any of these policies and have never filed a claim under one. We do not examine animals and we give no veterinary advice. We do not recommend an insurer, and the register sorts on one criterion only, which is stated above every table together with what that criterion ignores: premium, payout speed, annual limit and customer service.

We also have no licensed insurance professional on the desk. A competitor in this field puts a licensed agent's name above its pages. That is a real credential and we do not have it, so we do not imitate it in any form. What we offer instead is that you can check every cell against the document it came from.

What are the known weaknesses?

Screening registry figures come from radiographs that breeders chose to submit, so the animals are pre-selected and the numbers read better than the breed really is. Treat them as a floor. VetCompass measures UK first-opinion practices, and our market is the United States; the biology carries over, referral patterns and recording habits do not. An odds ratio compares a breed against other dogs and is not the chance that your dog will be affected.

Beyond the sources, two limits are ours. Because N depends on which studies exist rather than on which conditions a breed actually has, a well-researched breed gets a higher N than a poorly-researched one of identical health. So we compare insurers within one breed and never rank breeds against each other. And insurers do not use the same condition classes: what one calls orthopedic another calls a knee condition. Where that mapping is arguable we mark the row instead of choosing quietly.

How often is this rechecked, and what happens if a check is missed?

Policy wording every quarter, by pulling each document again and comparing the form code; a changed code means the document is read from scratch. Prevalence sources every six months. State legislation every quarter. Market averages once a year when the new report appears.

The visible "last checked" date on a page is the oldest check date of the rows that page uses, not the newest, so one fresh row cannot make a stale page look current. A date only moves when something was actually rechecked. If a quarter is missed, the row keeps its old date and the page shows how old the check is. Nothing is quietly presented as fresh.