All articles

Age of onset narrows the search.

When a feature first appeared separates conditions that share the same terms. Forms keep the feature and lose the when, and the ranking cannot use it.

Two patients are referred in the same month with the same three terms: ataxia, peripheral neuropathy, cardiomyopathy. One is a girl of eleven whose gait changed at eight. One is a man of fifty-four whose balance went in his forties and whose heart was found enlarged at a routine scan. The laboratory's request form has a field for each term and no field for when it began. Both exomes run through the same ranking, against the same disease annotations, and produce nearly the same candidate list.

The conditions that explain an eleven-year-old and a fifty-four-year-old with those terms overlap very little. The ranking could not tell them apart because nobody told it the ages.

What the field knows

Onset is one of the oldest discriminators in clinical genetics. Many conditions are defined partly by it, the infantile, juvenile and adult forms of a single metabolic disorder differ in gene or in variant class, and the differential for a neurological sign at two is a different list from the differential at forty. The Human Phenotype Ontology (Köhler et al., 2021) has a separate branch of onset terms, from antenatal through neonatal, infantile, childhood, juvenile, adult and late onset, and disease annotations in the HPO and in Orphanet carry typical ages of onset. The GA4GH Phenopacket schema (Jacobsen et al., 2022) lets each individual feature carry its own onset, so that a record can say that the ataxia began at eight and the cardiomyopathy was found at eleven.

Ranking methods can use it when they have it. Semantic similarity over HPO terms treats an onset term like any other term, and a specific onset for a specific feature is highly informative, because few diseases predict it. LIRICAL (Robinson et al., 2020) incorporates onset into its likelihood ratios where the data exist. Methods built for specific domains, such as those used in neurological and metabolic referrals, weight age of onset heavily because the clinicians who built them do.

Onset is also the feature most sensitive to how a record is kept. A phenotype captured at referral describes the patient at that moment; the age at which each feature appeared has to be taken from the history, and the history is in the letters. A record that stores features without their onsets has converted a trajectory into a snapshot, and most of the discriminating information in the trajectory is gone.

Course matters as well as onset. Whether a feature was static, progressive or resolved separates a developmental disorder from a degenerative one and a transient finding from a fixed one. The ontology has terms for these too, and the same request forms lose them in the same way.

Where practice breaks

The form has no field for it. A pick list of present features records what the patient has; the onset of each is in the clinician's letter, in free text, and is dropped when the letter is condensed to terms. The laboratory receives ataxia and does not receive eight. The scientist who later wonders about onset has to go back to the letters, if the letters were sent, and most were not.

A single age stands in for a history. Where onset is recorded at all it is one value for the whole case, the age at first presentation, which may be the age at referral rather than the age at which anything began. The man of fifty-four presented at fifty-four; his balance went a decade earlier. The feature-level onsets that would separate conditions are collapsed into a number that separates nothing.

Current age is used instead. The ranking knows the patient is eleven or fifty-four, and a scientist reviewing the list applies that knowledge by hand, discounting childhood-onset conditions for the man and late-onset ones for the girl. The discount is applied to the genes the scientist reaches and not to the ones below the threshold, and it is applied differently by different scientists. Two reviewers of the same list can reach different shortlists from the same current age, and neither records the discount they applied.

Reanalysis loses the trajectory twice. The original record had no onsets; the reanalysis is performed years later against a patient who has acquired new features, with no record of when the new ones appeared either. A progressive condition is being matched against a static list that was wrong about time from the start. Adult patients suffer most from this, since their histories are longest and their letters are scattered across the most services.

How Tulparus handles it

Every finding in the workbench carries its own onset where the clinical text supplies one, read from the letters along with the term and kept with the sentence it came from. The clinician confirming the intake sees the onset beside each finding and can correct it. Where no onset is recorded, the finding carries onset unknown, which is treated as missing rather than as current age.

The phenotype ranker uses onset per feature. A candidate disease whose annotated onset for a feature is far from the patient's recorded onset is moved down, with the mismatch stated; a candidate whose typical onset matches is moved up. The two patients in the opening case would have received different lists from the same three terms, because the terms arrived with their ages.

Missing onsets are gaps with a cheap fix. If the leading candidates are separated mainly by typical onset and the patient's onset for that feature is unknown, the workbench proposes a question to the clinician or the family, which usually takes a phone call, before any further test. The answer is recorded as a finding with its source, and the ranking is recomputed with it.

The case keeps the trajectory. Each finding has a first-recorded date and an onset, and later findings are added with theirs, so a reanalysis sees the order in which features appeared and can match a progressive course against diseases described as progressive in that order.

What you can do tomorrow

The information exists in every clinic letter. The task is to carry it to the laboratory.

  • On the request, put an age beside every feature: 'ataxia (from 8)', 'cardiomyopathy (found at 11, asymptomatic)'. The laboratory can use it even if its form cannot.
  • Distinguish age at onset from age at presentation, and from current age. They are three different numbers and the first is the useful one.
  • When the leading candidates differ mainly in typical onset, ask the family before ordering another test. It is the cheapest discriminator available, and the family usually knows the answer to the month.
  • At reanalysis, add the features that have appeared since, each with its own onset, so that the record describes a course and not a moment. Ask specifically whether anything has resolved as well as what has appeared.

Sources

  1. Köhler S, Gargano M, Matentzoglu N, et al. The Human Phenotype Ontology in 2021. Nucleic Acids Research, 2021.
  2. Jacobsen JOB, Baudis M, Baynam GS, et al. The GA4GH Phenopacket schema defines a computable representation of clinical data. Nature Biotechnology, 2022.
  3. Robinson PN, Ravanmehr V, Jacobsen JOB, et al. Interpretable clinical genomics with a likelihood ratio paradigm. American Journal of Human Genetics, 2020.
  4. Smedley D, Jacobsen JOB, Jäger M, et al. Next-generation diagnostics and disease-gene discovery with the Exomiser. Nature Protocols, 2015.

Same three terms, two patients.See onset change the list.

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