The Problem of Rarity: Diagnosis Rates are More Than Just a Formula

How many people live with a given rare disease? How many more remain undiagnosed? Can we develop a formula to predict improvement in diagnosis rates across diseases? These questions fill many conference rooms as researchers, policy makers, and business executives search for accurate preclinical data and strive to build accurate market forecasts. A recent article by the International Society for Pharmacoeconomics and Outcomes Research attempts to answer these riddles with impressive, yet imperfect results. “The problem of rarity: Estimation of prevalence in rare disease” quantifies the factors that cause underdiagnosis of rare diseases, specifically by examining the case of Dravet syndrome (a severe form of epilepsy in children). This is a tantalizing idea for orphan drug companies: what if a formula could predict the improvement of diagnosis rates in other diseases? We could eliminate a huge source of uncertainty in our market forecasts.

Sadly, for our purposes, this formula only works with hindsight. As the article’s title suggests, rarity is a problem: one effect of rarity is that seemingly-small differences between diseases can make it impossible to use the history of one disease to predict the future of another.

Incident cases are hard to find

Dravet syndrome (DS) is unusual among rare diseases in several ways. For example, the symptoms of DS are dramatic and easily recognizable: infants with repeated, severe epileptic seizures are quickly referred to a specialist. A strong advocacy organization (the International League Against Epilepsy) has ensured that neurologists know to test such children for DS. Because of these two factors (the disease’s clear presentation and physicians’ vigilance), the true incidence of DS can be reliably measured among infants. By contrast, the true incidence of many (or most) ultra-rare diseases is much harder to discover. Without this fundamental piece of information, a key element of this paper’s calculations would be missing.

Even for ultra-rare diseases where physicians succeed in diagnosing nearly all incident cases, epidemiologists may not be able to count these patients. Dravet syndrome is easy to identify in large populations because patients generate a distinctive set of medical claims: a simple query of healthcare databases can find the vast majority of affected children.

Incidence may differ across countries

Even if these barriers can be overcome for other ultra-rare diseases, more challenges remain: the healthcare databases may not cover the right populations. Nearly all cases of Dravet syndrome are caused by de novo mutations, meaning that the incidence should be the same among all populations and markets. This paper was able to use databases that covered the entire populations of Sweden and Denmark. But for most genetic diseases (which are not caused by de novo mutations), it would be risky to assume that other countries have the same incidence as these Scandinavian countries—their population genetics differ from other Europeans. In many cases, a country-by-country analysis of databases would be needed. Frustratingly, the largest pharmaceutical market presents the greatest challenges: US databases offer limited population coverage and are expensive to access.

Diagnosis rates are influenced by disease-specific factors

The problems discussed above are just part of the story because they only pertain to the absolute number of patients today. This paper’s approach also depends on understanding how the diagnostic process has changed over time. For Dravet syndrome, the diagnosis rate has been boosted by specific events, such as identification of the disease gene and the approval of a DS-labeled drug. In other ultra-rare diseases, the equivalent factors might be more subtle or harder to predict: for example, it is unclear how another DS diagnosis factor identified by the authors—the development of clinical trial classification—should be translated to other diseases. Clinical trial design is an iterative process for most diseases rather than a single “before/after” event; for a different disease, when can we be confident that it has achieved this milestone?

If a formula is the wrong way to predict diagnosis rates, what is the right one?

It would be ideal if a simple formula could predict future diagnosis rates. Without this magic bullet, what other tools are available? A two-pronged strategy is needed:

1. Use high-quality epidemiology data to understand today’s baseline. For some diseases, this may be simple like Dravet syndrome, where incidence, prevalence, and diagnosis rates are easily found. But most rare diseases require careful analysis and interpretation of data from flawed sources. It’s important to identify the limits of the data. Consider whether you will have to do new research to generate the needed data.
2. Learn how the disease is treated by the healthcare system and build an appropriate commercial strategy. Consult experts who have experience with similar markets to understand how you can improve the diagnosis rate. Examine physician awareness, referral patterns, diagnostics, the patient journey… and much more. These factors can rarely be reduced to a simple formula, so it is essential to compare your target disease to as many analogs as possible.