A multidisciplinary meeting discusses a young man with progressive ataxia and a normal repeat-expansion panel. The exome shows a heterozygous missense variant in a recessive ataxia gene, a second heterozygous variant in the same gene of unknown phase, and a de novo variant in a gene with one published family. After twenty minutes the room agrees to look into the recessive gene. Nobody writes down what the de novo variant would need to become the leading explanation, or what would rule the recessive gene out.
Six weeks later the parental samples arrive, and the question has to be reconstructed from memory.
What the field knows
The idea that an investigator should hold several explanations at once, and work to discriminate between them, is older than genetics. Chamberlin (1890) called it the method of multiple working hypotheses and argued that a single favoured explanation bends the collection of evidence towards itself. Platt (1964) made the same point for experimental science under the name strong inference: state the alternatives, devise the observation that excludes one or more of them, make it, and repeat. The method is not a philosophy of science. It is a procedure for deciding which test to order next.
Clinical reasoning research describes what happens without it. Croskerry (2003) catalogued the cognitive errors behind diagnostic mistakes, and anchoring and premature closure sit near the top: the first plausible explanation is adopted, the search narrows to evidence that fits it, and the alternatives are not so much rejected as forgotten. A multidisciplinary meeting that ends with a single gene to look into has, in that vocabulary, closed early.
Variant interpretation already works as evidence accumulation. The ACMG and AMP standards (Richards et al., 2015) classify a variant by combining criteria for and against pathogenicity, and Tavtigian and colleagues (2018) showed the framework behaves like a Bayesian one in which each criterion moves the odds. Several of those criteria are discriminating observations in Platt's sense: segregation in affected relatives, the phase of two variants in a recessive gene, the presence of a variant in an unaffected parent. Which of them to obtain next depends on which hypotheses are in play, and that is the part the framework leaves to the room.
Where practice breaks
Meetings record actions, not hypotheses. The minute says request parental samples, or look into the recessive gene. It does not say that the recessive gene is the leading explanation because of the phenotype match and the two variants, that it would be excluded if both variants prove to be on the same allele, and that the de novo variant becomes the leading explanation in that case if the one published family's phenotype overlaps on review. All of that was in the room. None of it is in the record.
So when the parental results arrive and show the two variants in cis, the result has to be interpreted by whoever opens it, with whatever they remember. If the person who argued for the de novo variant has rotated away, that argument may not be made again. If the meeting had written down the condition under which the recessive gene fell, the result would have carried its own interpretation.
Evidence against is the first casualty. Teams record what supports a candidate because that is what they will have to defend. What argues against it is said aloud and lost. A year later, at reanalysis, the candidate is dismissed again for a reason that was good once and may no longer hold, or is kept for a reason that was weak once and nobody wrote down.
The last failure is in choosing the next test. Without the alternatives written side by side, the test that gets ordered is the one that confirms the favourite rather than the one that separates the favourite from the second. Segregation studies are requested when phasing would have settled it, functional work is commissioned on a variant that a frequency check would have retired, and the case takes a year to reach the question a sentence could have asked.
The cost is measured in months. Each request to a family for another sample, each assay commissioned on the wrong variant and each meeting that re-discusses a candidate nobody wrote down adds a cycle of weeks to a case whose question has not changed since the first discussion. Writing the frame down takes one person ten minutes at the end of the meeting.
How Tulparus handles it
The workbench holds the case as a set of competing hypotheses rather than a ranked list with one winner. Each hypothesis names its candidate and its mechanism, the recessive gene with two variants in trans, the de novo variant acting dominantly, and carries the evidence for it and the evidence against it side by side, with every item traced to its source: a phenotype term from a letter, a frequency from a population release, a classification from a database with its date. What a hypothesis would need and does not have is listed as missing evidence.
From that frame the workbench proposes next steps and states what each outcome would change. Phasing the two variants is proposed because the recessive hypothesis stands or falls on it; the proposal says that a result in cis retires that hypothesis and promotes the de novo one, and that a result in trans adds a specific ACMG criterion to the recessive case. A clinician can ask what would rule a hypothesis out and get an answer drawn from the case's own evidence, with its sources shown and anything unverifiable marked.
When the result arrives, it lands on the frame the team wrote. The evidence against the recessive gene now includes the phasing result and its date; the de novo hypothesis moves, and the record shows who reviewed the change and what they decided. A reanalysis a year later starts from the same frame, so a candidate dismissed for a reason that no longer holds is visible as exactly that. Scores on the page rank the hypotheses against each other and never state the chance that any one of them is this patient's diagnosis. The decision at each step is a person's, recorded with their name.
What you can do tomorrow
The method fits on one side of a page per case.
- At the end of each case discussion, write each hypothesis as a sentence: the gene, the mechanism, the inheritance. Two or three is normal. One is a warning sign.
- Under each, write what supports it and what argues against it, with a source for every item, even if the source is a letter in the notes.
- For each, write the observation that would exclude it. Then order the test that excludes the most, not the one that confirms the favourite.
- When results arrive, read them against the written frame before discussing them, and record what moved. Keep the page with the case; it is the first thing the next reanalysis should read.
Sources
- Chamberlin TC. The method of multiple working hypotheses. Science, 1890.
- Platt JR. Strong inference: certain systematic methods of scientific thinking may produce much more rapid progress than others. Science, 1964.
- Croskerry P. The importance of cognitive errors in diagnosis and strategies to minimize them. Academic Medicine, 2003.
- Richards S, Aziz N, Bale S, et al. Standards and guidelines for the interpretation of sequence variants: a joint consensus recommendation of the ACMG and the Association for Molecular Pathology. Genetics in Medicine, 2015.
- Tavtigian SV, Greenblatt MS, Harrison SM, et al. Modeling the ACMG/AMP variant classification guidelines as a Bayesian classification framework. Genetics in Medicine, 2018.