Estimating the diagnostic yield of PGT-WGS

One of the most important questions we get at Orchid is:  “Whole-genome embryo screening sounds cool, but can your monogenic screening actually detect a significant amount of disease in the real world?”

A great recent preprint of young-adult ICU admissions from the PennMedicine Biobank (PMBB) gives us some compelling evidence that the answer is "yes".  By analyzing genes the authors found to be diagnostic for ICU admissions, we find that Orchid’s whole-genome embryo screening would have reported 52% of these variants — at the embryo stage, long before symptoms emerge — and that these variants in aggregate represent 13% of young-adult ICU admissions.

In Universal Exome Sequencing in Critically Ill Adults, Jessica Gold et al performed a retrospective analysis of 365 young adults (18-40) admitted to the ICU.  In this study, 25% were found to have a monogenic variant diagnostic of the disease responsible for the ICU admission (a monogenic diagnosis means that the researchers were able to identify well-characterized mutations on a single gene that have been confidently linked to the patient’s disease).

What caught my eye was that the methodology in this paper aligned closely with the reporting used at Orchid during preimplantation genetic screening with whole-embryo sequencing (PGT-WGS). It can be difficult to use external studies to estimate the real-life fraction of rare disease that PGT-WGS can detect: 

• The diagnostic yield computed in many studies include Variants of Uncertain Significance (VUS), which describe a class of genetic changes for which there is insufficient evidence to be deemed truly disease causing — Orchid does not report VUSs.  
• Similarly, many diagnostic studies perform a trio analysis against a patient’s family members to identify the variant which caused a disease.  However, a trio analysis isn’t applicable before the phenotypic presentation of disease, as is the case during embryo screening.

This study was different — when looking for a genetic diagnosis, this study did not rely on family members for a trio analysis, and the yield was explicitly computed for only pathogenic / likely pathogenic variants, all aligned with Orchid’s reporting.  Last, the supplemental data included the specific genes and variants identified, letting us directly compare against our PGT-WGS coverage.  

‍

So let’s check it out.

We can start with this great Sankey diagram from the preprint. 365 young-adult patients admitted to the ICU were screened for suspicious variants.  146 patients had at least one suspicious variant identified, for 166 total genes.  (One small discrepancy I noticed is that there are 365 patients discussed in the preprint, but 385 patients in the supplemental data. My understanding is that the authors replicated 20 records at this point to have a single suspicious variant per record.)

These variants were assessed for pathogenicity on ClinVar.  81 were determined to be pathogenic / likely pathogenic and autosomal dominant, and 9 were recessive and pathogenic.  My understanding is that the Penn Medicine BioBank uses short-read sequencing so it was usually ambiguous whether the compound heterozygous recessives were in cis (on the same chromosome) or trans (on opposite chromosomes), but given the observed disease, the authors considered them diagnostic.

So that gives us 90 clear path/lpath variants (“ClinVar Classification”). This is where PGT-WGS diverges from the clinical diagnosis in the preprint — the authors upgraded a number of VUS to path/lpath based on the observed disease phenotype (“Final Classification”).  Because we don’t have prior symptoms of disease on an embryo during PGT-WGS screening, we don’t report a VUS or attempt to reclassify it as pathogenic.

For technical and clinical reasons, Orchid’s standard screening panels do not cover every gene (our standard panels screen ~1,300 genes) or every class of variant.  So I wanted to see how many of those 90 variants would have been screened by Orchid’s embryo screening.

To be conservative, I filtered compound recessive variants with VUS or unclassified pathogenicity for either variant, leaving 84 records.  Filtering for Orchid’s standard screening panels left 68 records.  Two of the remaining records had larger deletions not covered by Orchid’s microduplication/microdeletion panels, so I’ve removed those as well, leaving 66 records.  Of these, 48 were determined to be diagnostic, plus 18 incidental findings (variants discovered during sequencing which are known to be pathogenic, but not the cause of this particular ICU admission).

Adjusting for Orchid’s > 98% validated sensitivity on trophectoderm biopsies, the conclusion from this data is that our PGT-WGS would have reported genetic variants diagnostic in ~47 admissions, representing 13% of young adult ICU admissions (47/365) and 52% of the retrospectively identified monogenic diagnoses (47/90).

‍

The ICU admission status corresponding to these 47 variants align with the expected hereditary cancer and cardiac birth defects included on Orchid’s standard screening panels.  Neurodevelopmental conditions are not heavily represented, likely because they manifest as chronic disability rather than acute ICU care.

‍

Last, but notably —  across the overall patient population, the average length of ICU stay was 2.3 days, with a mortality rate of 8.3%.  However, patients with monogenic diagnoses that would have been screened by PGT-WGS averaged an ICU stay of 2.94 days and a mortality rate of 10.4%.  Orchid’s panels are designed to screen for severe disease, and we see that reflected here.

It's estimated that 10% of Americans — 25 to 30 million people — are affected by a rare disease. 95% of those diseases have no treatment, and even accelerated pipelines for gene therapy will only make a dent. A finding that monogenic diagnoses represent 25% of ICU admissions even in young adults is a grim confirmation of those estimates.

This study though, serves as a great indirect validation that whole-genome embryo screening (PGT-WGS) can detect and prevent a large fraction of the burden of rare disease.  Based on this study, a conservative estimate is that Orchid’s whole-genome embryo screening would have reported on 52% of the variants determined to have caused ICU admissions, and that these variants in aggregate represent 13% of young-adult ICU admissions — admissions for severe disease which resulted in above-average mortality.  And as cheaper sequencing and larger biobanks allow medical genetics to link more real-world disease to causal variants, embryo screening will become an increasingly powerful tool for preventing genetic disease.

(This study of ICU admissions is a preprint.  We’ll update this post when the paper is published).