All articles

When two diagnoses explain one patient.

A patient can have two genetic conditions at once, and the blend fits neither. An analysis that stops at the first good answer misses the second.

A girl of seven has a hearing loss, a cardiac conduction defect, short stature and a learning difficulty. The exome finds a pathogenic variant in a well-known hearing loss gene, and the case is reported as solved. Her cardiologist, reading the report, notes that the gene has nothing to do with the heart, and writes to ask whether the laboratory looked for anything else. The honest answer is that the search stopped when the first gene fitted well enough, which is what most pipelines are built to do. The laboratory replies that the conduction defect is probably coincidental.

At reanalysis three years later, prompted by a new sibling, a second pathogenic variant is found in a gene for a syndrome that explains the heart, the stature and the learning difficulty. It had been in the first exome, in the candidate list, below the gene that had already answered the question the laboratory thought it was asking.

What the field knows

Dual molecular diagnoses are a recognised and measured part of clinical genomics. Posey and colleagues (2017), in a large clinical exome series, found that a small but consistent share of patients with a molecular diagnosis had two, in distinct genes, and that in many the phenotypes overlapped enough that the blend had been read as an unusual presentation of a single condition. Balci and colleagues (2017) reported similar findings in a research cohort and described the clinical reasoning that had missed them: a presentation that did not fit one syndrome had been explained as variable expressivity rather than as two syndromes.

Several things make a second diagnosis more likely. Consanguinity raises the chance of two recessive conditions in one child. A family with a known dominant condition can have a child with that and something new. Some phenotypes are frequent enough, hearing loss among them, that their co-occurrence with another condition by chance is expected in a sequencing service of any size. The phenotype that does not fit the first diagnosis is the signal, and the literature's consistent finding is that it was usually visible in retrospect.

The analytical consequence is that the ranking should not be read as a search for one answer. A phenotype-driven method scores each candidate against the whole phenotype, so a gene that explains four features scores above one that explains two, and that is right; the trap is in stopping when a gene explains some features well. Posey and colleagues observed that in most dual diagnoses each gene explained a distinct subset of the features, which is exactly the pattern a reviewer can look for once the first gene is in hand.

The cost of missing the second diagnosis is paid in the clinic. Surveillance for the unexplained condition is not arranged, recurrence counselling covers one condition when the family carries two, and a feature that was called coincidental is managed as if it had no cause. For the girl above, the conduction defect had a cause with implications for her and for her siblings that a hearing loss gene could never have carried.

Where practice breaks

The search stops at the first good answer. A pathogenic variant in a gene that fits part of the phenotype closes the case, because the workflow has a slot for one diagnosis and the slot is now full. The remaining features are explained away, as coincidence, as expressivity, or as a feature of the known condition that the literature has not yet described.

Nobody asks what the first diagnosis leaves unexplained. The report lists the features and the variant; it does not list the features the variant accounts for and the features it does not. A cardiologist can see the gap because the heart is their business, and a geneticist who reads the whole child can see it too, if the report is written to show it. The laboratory, holding the whole phenotype, could have seen it first.

The candidate list is not re-read with the first diagnosis removed. Once the hearing loss gene is accepted, the correct next step is to rank the remaining features alone, which moves the syndrome gene to the top. Instead the original ranking, dominated by the hearing loss, is the only one that was ever made.

Reanalysis inherits the closure. A solved case is not on any reanalysis list, and the second diagnosis is found, if at all, by accident: a sibling, a different clinician, a question from a family that never quite believed the first answer explained everything. The family is often right, and the system has no way to hear them.

How Tulparus handles it

A hypothesis in the workbench is scored against each finding, and the page shows which findings it explains and which it leaves. When a hypothesis is accepted, the unexplained findings are listed by name, with their specificity, and the case does not reach a closed state while specific findings remain unexplained unless a person records a reason for each. 'Coincidental' is an acceptable reason only when written down with a name beside it, and a specific finding with no reason keeps the case open.

Accepting a first hypothesis re-runs the ranking on the residual. The findings the accepted gene explains are set aside, the remaining findings are ranked as a phenotype in their own right, and the result is shown as a second list. For the girl above, the residual phenotype, conduction defect, short stature and learning difficulty, would have put the syndrome gene at the top of that list on the day the hearing loss gene was accepted.

The case model allows more than one accepted hypothesis, and the report, rendered from the case, describes each with the findings it accounts for, so that the family and every clinician involved can see what has and has not been explained. A case with one accepted hypothesis and specific residual findings is reported as partly explained, in those words.

Partly explained cases stay on the reanalysis list. New knowledge is matched against the residual phenotype as well as against the whole, and a new gene-disease association that fits the residual reopens the case with the reason stated.

What you can do tomorrow

The habit is a single question asked at the moment of closing.

  • Before reporting a diagnosis, write down which of the patient's features it explains and which it does not. If a specific feature is left over, say so in the report.
  • Re-rank the residual features as if they were the whole phenotype. It takes minutes with any phenotype-driven tool and it is where the second diagnosis sits.
  • Treat a feature that does not fit as a finding, not as noise. Ask the clinician whether it is real and how specific it is.
  • Keep partly explained cases on the reanalysis list, with the residual phenotype recorded, so that new knowledge is matched against what is still open.

Sources

  1. Posey JE, Harel T, Liu P, et al. Resolution of disease phenotypes resulting from multilocus genomic variation. New England Journal of Medicine, 2017.
  2. Balci TB, Hartley T, Xi Y, et al. Debunking Occam's razor: diagnosing multiple genetic diseases in families by whole-exome sequencing. Clinical Genetics, 2017.
  3. Smedley D, Jacobsen JOB, Jäger M, et al. Next-generation diagnostics and disease-gene discovery with the Exomiser. Nature Protocols, 2015.
  4. 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.

One answer, four features.See what the first diagnosis leaves.

A demo takes 20 minutes, screen shared, on synthetic cases. Bring the question from one of your own.