Empowering Geneticists to Do Genomic Reanalysis Routinely
Laboratory Genetics and Genomics
-
Primary Categories:
- Genomic Medicine
-
Secondary Categories:
- Genomic Medicine
Introduction:
Patients not diagnosed on an initial genomic sequencing analysis can benefit from reanalysis that adds 3 types of information:1) New associations of a gene with disease, 2) New findings in a disease, or 3) New clinical and non-genetic lab findings in the patient.
Historically such a reanalysis has been costly and unreimbursed. However, with advances in software for clinical correlation, reanalysis could incorporate all this information. We studied the practicality of making such reanalysis part of routine follow-up care by the clinician best positioned to do the clinical correlation: the patient’s geneticist.
Methods:
We evaluated two patient cohorts: 1) those who had clinical exome sequencing at St. George’s Hospital in London UK, and 2) those who had genome sequencing from the Undiagnosed Diseases Program at Vanderbilt University Medical Center (VUMC).
Annotated variant tables were prepared by Congenica, using BAM files from St. George’s and unannotated VCFs from VUMC generated by Baylor. Reanalysis of SNVs was done using the SimulConsult diagnostic decision support software. This software ranks plausible gene zygosities using a metric of pertinence to the patient’s clinical and lab findings. Loading the files to do the reanalysis takes ~10 seconds and a clinician can review the case in minutes.
Results:
De novo variants averaged 71 for St. George’s, done from BAM files, and 406 for VUMC cases, done from VCF files. Plausible zygosities with known clinical phenotypes averaged 18.6 for St. George’s and 31.4 for VUMC.
A diagnosis was made by St. George’s and Congenica initially in 8/20 (40%) of cases. A ninth (TAB2) was found on reanalysis by Congenica. For all 9/20 (45%), the diagnostic software ranked the correct diagnosis as #1 in pertinence of the detected gene zygosities from the original files. The pertinence metric for the correct gene in these 9 cases averaged 62.5%, and was 100% in 5 cases. A tenth diagnosis (CAPN3) was later made based on deletion analysis, and although no CAPN3 variants were in the variant table, the software ranked CAPN3 #1 in genes recommended for further testing. In 7 cases, additional findings were added based on chart review, improving the gene pertinence.
A diagnosis was made in 5/7 (71%) of the diagnosed and none of the 10 undiagnosed VUMC cases. For all 5, the diagnosis was ranked #1, with average pertinence 60.5%. For 1 (SSR4) the variant was from a deletion not in the variant table, and for another (NR2F2) the gene was picked because a parent with shared findings was mosaic. One of 10 undiagnosed cases had a recently discovered (RNU4-2) non-coding variant; in the software RNU4-2 was ranked #1 in genes for further testing. Three more cases are under review.
Conclusion:
Our data show that rapid reanalysis by clinicians can increase the yield of genetic diagnosis with minimal effort and no new unreimbursed lab expenses. In 1/40 cases this was through recent database information and in 2 it was through suggestions of other genes to check.
The study also demonstrates how to further increase the yield:
We conclude that reanalysis using diagnostic software during clinical follow-up is practical and efficient. This could be improved further by joint variant calling and by including deletion analysis.
Patients not diagnosed on an initial genomic sequencing analysis can benefit from reanalysis that adds 3 types of information:1) New associations of a gene with disease, 2) New findings in a disease, or 3) New clinical and non-genetic lab findings in the patient.
Historically such a reanalysis has been costly and unreimbursed. However, with advances in software for clinical correlation, reanalysis could incorporate all this information. We studied the practicality of making such reanalysis part of routine follow-up care by the clinician best positioned to do the clinical correlation: the patient’s geneticist.
Methods:
We evaluated two patient cohorts: 1) those who had clinical exome sequencing at St. George’s Hospital in London UK, and 2) those who had genome sequencing from the Undiagnosed Diseases Program at Vanderbilt University Medical Center (VUMC).
Annotated variant tables were prepared by Congenica, using BAM files from St. George’s and unannotated VCFs from VUMC generated by Baylor. Reanalysis of SNVs was done using the SimulConsult diagnostic decision support software. This software ranks plausible gene zygosities using a metric of pertinence to the patient’s clinical and lab findings. Loading the files to do the reanalysis takes ~10 seconds and a clinician can review the case in minutes.
Results:
De novo variants averaged 71 for St. George’s, done from BAM files, and 406 for VUMC cases, done from VCF files. Plausible zygosities with known clinical phenotypes averaged 18.6 for St. George’s and 31.4 for VUMC.
A diagnosis was made by St. George’s and Congenica initially in 8/20 (40%) of cases. A ninth (TAB2) was found on reanalysis by Congenica. For all 9/20 (45%), the diagnostic software ranked the correct diagnosis as #1 in pertinence of the detected gene zygosities from the original files. The pertinence metric for the correct gene in these 9 cases averaged 62.5%, and was 100% in 5 cases. A tenth diagnosis (CAPN3) was later made based on deletion analysis, and although no CAPN3 variants were in the variant table, the software ranked CAPN3 #1 in genes recommended for further testing. In 7 cases, additional findings were added based on chart review, improving the gene pertinence.
A diagnosis was made in 5/7 (71%) of the diagnosed and none of the 10 undiagnosed VUMC cases. For all 5, the diagnosis was ranked #1, with average pertinence 60.5%. For 1 (SSR4) the variant was from a deletion not in the variant table, and for another (NR2F2) the gene was picked because a parent with shared findings was mosaic. One of 10 undiagnosed cases had a recently discovered (RNU4-2) non-coding variant; in the software RNU4-2 was ranked #1 in genes for further testing. Three more cases are under review.
Conclusion:
Our data show that rapid reanalysis by clinicians can increase the yield of genetic diagnosis with minimal effort and no new unreimbursed lab expenses. In 1/40 cases this was through recent database information and in 2 it was through suggestions of other genes to check.
The study also demonstrates how to further increase the yield:
- Clinicians adding useful findings: updated patient information can be added during follow-up visits, guided by the software’s usefulness rankings, which take into account the patient’s gene variants.
- Doing joint variant calling: by reducing false de novo calls the signal to noise ratio in the reanalysis would be increased substantially.
- Inclusion of deletion analysis: this can included in the variant table read by the software and analyzed together with small variants.
We conclude that reanalysis using diagnostic software during clinical follow-up is practical and efficient. This could be improved further by joint variant calling and by including deletion analysis.