A girl of five with a progressive neurological disorder has had a panel, an exome and a chromosomal microarray, all negative. The multidisciplinary meeting proposes a genome. The genome will cost more than everything so far, take four months, and, for the two hypotheses the team actually holds, answer neither: one needs a repeat examination to establish whether a hedged early finding is real, the other needs a metabolic test that costs a tube of blood.
Both were discussed and neither was written down as a test. The genome was ordered because it was the next thing on the list.
What the field knows
The idea that the best next test is the one that most reduces uncertainty is older than genomics. Decision analysis formalised it as the value of information: a test is worth doing in proportion to how much its result would change what you would do next, weighted by how likely each result is. Clinical reasoning has applied the principle informally for a century, choosing the investigation that best separates the leading explanations rather than the one that is most thorough.
In genomics the principle has been quantified for the big decisions. Studies comparing singleton with trio sequencing (Farwell et al., 2015; Retterer et al., 2016) and exome with genome (Clark et al., 2018, in a meta-analysis of diagnostic yields) measured how much each step added, and the rapid-genome programmes (Kingsmore et al., 2019) showed that for critically ill infants the information per day was high enough to justify the cost. What the literature rarely addresses is the smaller decision inside a single case: given these particular hypotheses, what should this patient have next?
That decision depends on the hypotheses. A case with two strong candidates separated by one phenotypic feature needs an examination. A case with one pathogenic allele in a recessive gene needs a search for the second, by copy-number analysis or RNA. A case with no candidates at all needs more data, and a genome may be the right call. The same negative exome can call for any of these, and the choice is only visible if the hypotheses have been written down with what each one lacks.
Diagnostic yield is a population statistic and a poor guide for the individual case. A genome adds diagnoses across a cohort because it finds variant classes the exome missed; whether it will add one for this patient depends on whether this patient's hypotheses point at those classes. The literature on yield says what works on average. The case says what is missing here. The two can point in opposite directions, and when they do, the case should win, because the cohort is not the patient in the room.
Where practice breaks
The next step is chosen from a menu rather than from the case. Panel, exome, genome, in that order, with reanalysis between. The sequence is sensible as a service policy and blind to the individual case, in which the most informative next action may be a phone call to the referring clinician.
Cheap tests are undervalued because they are cheap. A repeat examination, a metabolic screen, a family member's blood sample or a review of the original imaging each cost little and can settle a hypothesis outright. They are rarely proposed with the seriousness of a sequencing test, because they do not feel like progress, and so they are discussed at the meeting and not ordered.
The hypotheses are not written down, so nobody can say what each needs. The team holds them in conversation; the meeting minute records a decision; the specific gap that the genome would or would not fill is never stated, and the genome is ordered on the general ground that more data are better.
The cost of the wrong next step is paid in time. A four-month genome that answers neither live hypothesis delays the metabolic test that would have answered one of them, and the child is four months further into a progressive disorder when the simpler test is finally done. Budgets pay too: a service that orders the broadest test by default spends its sequencing allocation on cases a cheaper step would have advanced, and the cases that genuinely need a genome wait behind them.
How Tulparus handles it
The workbench keeps each hypothesis with its missing evidence as structured items, and the next-action engine works from those items. For every gap it knows the actions that could close it, from the case model and the knowledge store: an examination for an unestablished feature, a specific assay for an unresolved variant, a sample from a named relative for an inheritance question, a broader sequencing test where the hypotheses point at classes the current data cannot see.
Each proposed action is shown with what it would do. For each live hypothesis the page states how each possible result of the action would move its standing, together with the action's cost, turnaround and whether the material already exists. The proposals are ordered by how much they would change the picture across all the hypotheses, so that a tube of blood that would settle one of two live hypotheses outranks a genome that would touch neither.
Nothing is ordered by the system. The list is a proposal for the clinician and the laboratory, and each action taken or declined is recorded on the case with a reason, so that the next reviewer sees why the genome was deferred and the metabolic test was done first. If the hypotheses change, the proposals are recomputed, and the record of earlier proposals stays.
For the girl above, the page would have put the repeat examination and the metabolic test at the top, with the two hypotheses they bear on and the result each would need, and the genome below them with the honest note that neither live hypothesis depends on it. The meeting could still have chosen the genome. It would have chosen it knowing what it would and would not answer.
What you can do tomorrow
The question can be asked at any multidisciplinary meeting without a product in the room.
- Before proposing the next test, write the live hypotheses on the board, and under each one the single piece of evidence it most lacks.
- For each candidate action, say which hypothesis it bears on and what each possible result would do to it. If the answer is none, say so out loud.
- Give cheap actions the same line in the plan as expensive ones: a repeat examination is a test, and it gets an owner and a date.
- Record the reasoning in the case note, in one sentence per action: what it is for and what would change if it came back either way.
Sources
- Farwell KD, Shahmirzadi L, El-Khechen D, et al. Enhanced utility of family-centered diagnostic exome sequencing with inheritance model-based analysis: results from 500 unselected families with undiagnosed genetic conditions. Genetics in Medicine, 2015.
- Retterer K, Juusola J, Cho MT, et al. Clinical application of whole-exome sequencing across clinical indications. Genetics in Medicine, 2016.
- Clark MM, Stark Z, Farnaes L, et al. Meta-analysis of the diagnostic and clinical utility of genome and exome sequencing and chromosomal microarray in children with suspected genetic diseases. npj Genomic Medicine, 2018.
- Kingsmore SF, Cakici JA, Clark MM, et al. A randomized, controlled trial of the analytic and diagnostic performance of singleton and trio, rapid genome and exome sequencing in ill infants. American Journal of Human Genetics, 2019.
- 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.