Genetic Models Underestimate Rare Disease Prevalence

Hundreds of millions of people worldwide are affected by rare diseases. Behind every case is a family enduring a diagnostic odyssey that can last for years. To improve how public health systems help these patients, it is critically important that the number of rare disease cases be accurately counted.

In a new study, researchers at BioMarin Pharmaceutical, a biotechnology company that develops drugs for rare diseases, reported that genetics-based models of these diseases underestimate their prevalence. The team used real-world data taken from 23 million newborns as a comparison. The researchers showed that they could improve the models’ accuracy by incorporating other types of mutations and tweaking the models to better reflect the US population, but inclusion of more genetically diverse samples is still needed to make the models as accurate as possible. They published their findings in The American Journal of Human Genetics.¹

No Perfect Approach to Predict Rare Diseases

The most basic method for estimating how common a rare disease is involves counting diagnoses in hospital records. But many people living with a rare disease spend years without a diagnosis. Two people with the same condition might have very different symptoms.

Newborn screening programs (NBS), which prick the heels of newborns to measure disease-linked molecules in their blood, have very high uptake rates, but they aren’t perfect at identifying disease. The list of diseases tested by NBS changes very slowly, so there are many diseases for which these programs don’t currently test.

Genetic approaches have taken a different tack to the problem of estimating rare disease prevalence. Autosomal recessive rare diseases require both parents to pass on an affected allele. Many people will be unaffected carriers of a single allele. Genetic models can estimate how common these single alleles are in the wider population and then infer from that how many people will carry both alleles and be affected by the rare disease.

In the new study, Michael Sierant, a coauthor and bioinformatician at BioMarin Pharmaceutical, and his colleagues compared these genetic approaches with NBS data for the first time.

Systematic Underestimates in Rare Disease Models

The researchers calculated the prevalence of 28 autosomal recessive diseases using genetic databases such as gnomAD and ClinVar, before comparing them to the birth prevalence reported in NBS databases for 23 million newborns in the US, Guam, and Puerto Rico.

Genetic modeling techniques substantially underestimated the prevalence of nearly every rare disease tested. Half of the diseases were underestimated by at least 3.5 times in genetic modeling, and for 10 diseases, the genetic testing estimates were 10 times lower than the NBS estimates. Spinal muscular atrophy’s prevalence was 74 times lower in the genetic models. Only one disease, biotinidase deficiency, was overestimated. This was because an allele linked to the disease in genetic models didn’t actually make people sick.²

Making Genetic Tests More Accurate

To see if they could improve these systemic underestimates, the researchers made several tweaks to the models. Switching to a newer version of the gnomAD database enhanced accuracy. The team also weighed their genetic dataset, which was based on global allele frequencies, to match US census demographics. This improved concordance to the point where only four diseases were more than 10 times underestimated in the genetic models.

The team then refined the mutations included in the genetic analysis. The databases didn’t account for some protein-truncating and missense variants. Adding these mutations further improved concordance. Now, the median difference was just 1.4-fold lower than NBS.

Despite these improvements, limitations in the datasets that feed into genetic databases, such as low sample diversity, still hold back these models. “The underrepresentation of many ancestral groups will reduce the number of population-specific rare disease-causing variants,” wrote the authors in the paper.

But the team concluded that even rough genetic estimates can still be useful for public health efforts. Most rare diseases aren’t included on powerful NBS panels, so these models, which doctors can apply to many more rare diseases, can benefit patient groups that desperately want more information about their conditions.