A laboratory's policy is to reanalyse undiagnosed exomes at two years. A boy tested in January 2025 is due in January 2027. In March 2026 a paper reports a new gene for his phenotype, with de novo missense variants clustered in one domain. He has one. It was in his original variant file and was filtered out because the gene had no disease association at the time.
Nobody connected the paper with his file. He will be reanalysed as scheduled, ten months after the answer existed, if the laboratory still has the capacity that month and he is still under the same clinician.
What the field knows
Reanalysis works because knowledge changes. Wenger and colleagues (2017) reanalysed clinical exomes after an interval, found new diagnoses, and attributed most of them to gene-disease associations published after the original analysis. Wright and colleagues (2018), reanalysing the Deciphering Developmental Disorders cohort, reported the same ordering: the largest contributor to new diagnoses was newly described disease genes, followed by improved variant calling and better phenotype data. Liu and colleagues (2019) reanalysed a large clinical series and again found that recently reported genes accounted for much of the additional yield.
The timing recommendations that followed were calendar-based because the evidence was. The studies compared a first analysis with a second at a fixed interval, so the natural recommendation was an interval. Ewans and colleagues (2018) made the case for reanalysis at around twelve months. The ACMG points to consider (Deignan et al., 2019) list the events that should prompt a re-evaluation, among them new clinical information, new evidence about a reported variant and a new gene-disease association, and discuss who carries the responsibility for starting one.
The resources that track the knowledge are updated continuously. Gene-disease associations are curated as they are published, classification archives are released on a schedule of weeks, and expert panels re-evaluate existing associations throughout the year. Any given case is touched by a small number of these changes in a year, and in most years by none. Calendar reanalysis spends the same effort on every case whether or not anything relevant has changed.
Who should start a reanalysis is unsettled. Deignan and colleagues note that laboratories, ordering clinicians and patients each have a part, and that no one of them has been given the duty. In practice the duty falls to whoever notices, and whoever notices is often the family, when they read about a new gene and ring the clinic.
Where practice breaks
Capacity goes first. A laboratory with a few thousand undiagnosed cases and a two-year policy must reanalyse dozens a week to keep pace, and most of those reanalyses find nothing because nothing relevant changed. The queue grows, the policy slips to three years, and the backlog becomes the thing that is managed.
The case that should be reopened waits with the rest. The boy above is not due. The paper that would solve his case is read, perhaps, by someone in the laboratory who does not have his file in mind. The connection between a new association and the cases it touches is made by human memory, and human memory does not keep a backlog.
When the scheduled reanalysis does happen, it is usually a re-run: the same pipeline on the same variant file with updated annotations, read by a scientist who starts from the top of a new list. What changed since the first analysis, which is the only reason to look again, is not shown. The scientist re-reads candidates that were dismissed for good reasons the first time and has to rediscover the reasons.
The reverse failure is quieter. A case closed on a likely pathogenic variant is on no reanalysis schedule. If that variant is later downgraded, nobody is prompted. Closed cases are assumed to stay closed.
The policy also treats the cases as alike. A case with a strong candidate set aside only because the gene had no disease association is one paper away from an answer. A case with no candidate at all needs a new pipeline or a new phenotype, and no amount of re-annotation will help it. Both wait the same two years.
How Tulparus handles it
The workbench records the knowledge release each investigation was performed against: the gene-disease association set, the classification archive, the ontology and the annotation versions. When any of these is updated, the new release is loaded as a new release, and the difference from each case's recorded release is computed as a list of specific changes: a gene newly associated with a disease, a variant whose classification moved, a disease whose phenotype annotations changed.
Each change is matched against the cases it could touch. A new association is matched against the cases carrying a qualifying variant in that gene, with the phenotype compared against the disease's annotations. A reclassification is matched against the cases in which that variant was considered, whether the case was closed on it or set it aside. The result is a queue in which every entry has a reason: this case, this change, this variant, this release. Entries are ordered by how specific the match is, so that a de novo missense variant in a newly associated gene with a close phenotype match is read before a common variant in a gene whose association was merely strengthened. The queue is a work order. No entry states a chance that the case is solved.
The queue is offered and never acted on alone. The workbench proposes a re-draft of the investigation, and a clinician or scientist decides whether to open it. When they do, the reanalysis starts from the previous investigation's record, with the earlier decisions and their reasons visible, and shows what is different now, so the scientist reads the change rather than the whole list again.
Calendar review keeps a place, for the changes no diff can see: an improved pipeline, a new phenotype from the clinic. The workbench supports it as a second trigger. The boy above would have appeared in the queue in the week the association was loaded, with the gene, his variant and the phenotype match shown beside his case.
What you can do tomorrow
The first two steps need no new software.
- Record, for each case, the versions of the gene-disease and classification resources used. Without them there is nothing to compare later.
- When a relevant new gene is published, search your stored variant files for qualifying variants in it. Most laboratories can do this today; few do it routinely.
- Keep closed cases in the loop. A downgrade of a reported variant should prompt a review, and the simplest way is to check the reported variants against each new classification release.
- Keep the scheduled review for pipeline and phenotype changes, and stop expecting it to catch the knowledge changes on time.
Sources
- Wenger AM, Guturu H, Bernstein JA, Bejerano G. Systematic reanalysis of clinical exome data yields additional diagnoses: implications for providers. Genetics in Medicine, 2017.
- Wright CF, McRae JF, Clayton S, et al. Making new genetic diagnoses with old data: iterative reanalysis and reporting from genome-wide data in 1,133 families with developmental disorders. Genetics in Medicine, 2018.
- Liu P, Meng L, Normand EA, et al. Reanalysis of clinical exome sequencing data. New England Journal of Medicine, 2019.
- Ewans LJ, Schofield D, Shrestha R, et al. Whole-exome sequencing reanalysis at 12 months boosts diagnosis and is cost-effective when applied early in Mendelian disorders. Genetics in Medicine, 2018.
- Deignan JL, Chung WK, Kearney HM, Monaghan KG, Rehder CW, Chao EC. Points to consider in the reevaluation and reanalysis of genomic test results: a statement of the American College of Medical Genetics and Genomics (ACMG). Genetics in Medicine, 2019.