Could two or three minutes of documentation have shortened a 16-year search?
I used Google and published information to recognize a pattern in my son Charlie's phenotype, years before brain tissue testing identified a mosaic SLC35A2 variant. This teaching case asks why those useful observations were not prominent and reusable in his accessible health record.
By Danielle Boyce, MPH, DPA, data scientist, informatician, and Charlie's mother. The case history is my own account as his parent. Evidence reviewed October 21, 2026. Mosaicism narrative, sources, evidence-comparison interface, granularity check, and chain-of-safeguards diagram updated October 22, 2026.
This is a small ask, and a parent has already shown it can be done. The findings that narrowed Charlie's search were visible to anyone who examined him, and the search itself was done by hand, by a parent, with public resources; other parents in the rare epilepsy community have done the same. Recording a few specific findings once, during the genetic evaluation of a child with a severe and poorly understood condition, adds minutes, not hours, to a workup that already exists. In this case the result was a shortlist concentrated in one disorder family whose specialist clinics see suspected cases as well as confirmed ones, which is enough to justify a referral on strong suspicion alone. A health system that runs decision support and AI on its records is well placed to do what one parent did with a search engine, and for anyone building OMOP phenotypes, the same specific terms are what make a rare condition this simple to narrow.
A 16-year diagnostic odyssey
Charlie developed infantile spasms and later lived with severe developmental and intellectual disability. I considered findings beyond seizures, including low muscle tone, inverted nipples, and facial features, and first used the phenotype to build a much smaller shortlist that included glycosylation-related genes alongside others. I then used published evidence of somatic mosaicism to prioritize that shortlist further, which helped bring SLC35A2 into focus, several years before his diagnosis.
Earlier genetic testing did not provide the answer. Testing of brain tissue obtained during epilepsy surgery later identified mosaicism in SLC35A2. The exact variant classification and clinical interpretation belong to his laboratory report and specialist team, and this page does not restate them.
How the mosaicism hypothesis entered the search
Literature and discussions with other parents drew my attention to variants found in epilepsy surgery tissue that peripheral testing could miss, and I used this as a hypothesis years before Charlie's tissue result. It is why I personally went looking for a way to have his brain tissue tested. Community discussions helped identify a question to investigate, and published evidence supported the gene prioritization.
Useful phenotyping had already happened, so why was the information so difficult to find and reuse?
The argument this example supports
A small set of well-described findings can justify looking beyond a general epilepsy label and seeking neurogenetics or metabolic genetics review.
Phenotype information collected in the examination, family history, genetics order, or laboratory report should remain visible and reusable. An order form sent to a laboratory is also a potential source of clinical data.
Claims this example cannot support
It cannot show that an earlier referral would have produced a diagnosis, prevented disability, or led to effective treatment. The current annotation snapshot is an illustration, not a reconstruction of the evidence available at each point in Charlie's life.
The goal is a reliable route from a family's observation to specialist review and reusable data.
A few findings can change the referral question
Build a shortlist from the entry evidence, low muscle tone, and inverted nipples, then compare additional findings with the source evidence. Missing annotations should prompt review, not make the shortlist disappear. The third control shows what the same search returns when the record describes each finding by its parent category instead of the specific term. Counts describe this annotation snapshot, not diagnoses or probabilities.
For an EHR cohort: infantile / epileptic spasms OR West syndrome OR documented hypsarrhythmia can trigger review. The annotation demo implements epileptic spasm (including infantile spasms) OR hypsarrhythmia. West syndrome is an EHR diagnosis route, not a separate HPO feature counted here.
Additional findings: compare evidence
These selections add evidence to compare. In the default view, they do not remove genes from the basic shortlist.
Unchecked means “not selected,” never “absent.” Choose only documented findings. Wide nose, broad nasal tip, and wide nasal bridge are different anatomical features, not interchangeable labels.
Specific terms are matched, including their more specific descendants in HPO.
Matching rules and source selection
The strict audit reproduces the original filter. It can reach zero because the annotation records are incomplete or differ in coverage. It cannot rule out a diagnosis.
Records remain separate even when they describe the same disorder, and the sources differ in annotation coverage, so these are not counts of unique disorders.
Green identifies genes linked to a matching record whose name explicitly identifies a glycosylation disorder. Alphabetical order is not a clinical ranking.
Case example: why SLC35A2 can appear to disappear
For the selected findings, compare the two SLC35A2-CDG source records below. A missing annotation does not mean Charlie lacks the feature or that the disease cannot include it.
| Selected finding | OMIM:300896 | ORPHA:356961 |
|---|
This case reference shows both sources independently of the source filter. The records are not merged into the shortlist counts. “No positive annotation here” describes this snapshot, not a confirmed negative finding. This comparison does not establish a diagnosis or attribute every feature to a brain mosaic variant.
Compare shortlist records and their evidence
Frequency fields describe source reports, not this child's probability. A disease record can link several genes without establishing every feature for every gene, variant, or patient.
4. Consider published mosaicism evidence
In my original search, phenotype matching produced a mixed shortlist, not exclusively CDG genes. I then prioritized the candidates for which I found relevant reports of disease-causing somatic mosaicism, and this additional reasoning helped focus my search on SLC35A2.
The genes in this demonstration are not a reconstruction of my original shortlist. Mosaicism is not applied as a numerical filter here, because doing that would require a dated, systematic review of every candidate, and finding no report for a gene is different from evidence that the gene cannot cause disease through mosaicism. See supporting literature and dates.
Why this example stops at referral
In this snapshot, the broad flag plus low muscle tone and inverted nipples matches source disease records linked to ALG2, ATN1, DPAGT1, EIF4A2, MGAT2, MSL2, and SLC35A2. ALG2, DPAGT1, MGAT2, and SLC35A2 each have an explicitly named CDG record; ATN1, EIF4A2, and MSL2 do not. The same findings can occur outside CDG, and a person with CDG need not have all of them.
In the optional strict audit, adding thick eyebrows makes the source-record intersection empty. The default evidence-comparison view keeps the basic shortlist visible and shows the missing annotations. An empty strict intersection is not evidence that Charlie's observed phenotype is impossible. For SLC35A2-CDG, the inverted-nipple annotation occurs in the Orphanet record, while several facial annotations occur in the OMIM record. Wide nose and broad nasal tip have no positive annotation in either source record in this snapshot, while wide nasal bridge is annotated in OMIM, and the interface keeps these distinctions visible. A production system may harmonize verified equivalent disease records, retain provenance, and use clinically validated phenotype matching.
Earlier review is a meaningful outcome
The proposed action is a review of the phenotype and prior testing, with referral to neurogenetics and, when indicated, a metabolic geneticist or CDG specialist.
Recommendations already support genetic evaluation
NSGC, published online 2022: recommends genetic testing for unexplained epilepsy at any age. Exome/genome sequencing or a multigene panel are first-tier options, with a conditional preference for exome/genome sequencing. Qualified interpretation and genetic counseling are part of the recommendation. Source
ILAE Genetics Commission, 2022: recommends exome or genome sequencing, including copy-number analysis, as first-line testing in most cases under genetic evaluation, and periodic reevaluation when no molecular diagnosis is obtained. Source
ACMG, 2021: recommends considering exome/genome sequencing as a first- or second-tier test for pediatric congenital anomalies, developmental delay, or intellectual disability. Source
These recommendations have different scopes. They do not mandate one universal HPO checklist or a particular Epic, FHIR, or OMOP implementation. Retaining structured phenotype data across these systems is the proposal in this teaching case.
Interpret the sample and the method
A panel tests selected genes. An exome primarily assesses coding regions. A genome examines more sequence, but no assay detects or interprets every variant. Depth, coverage, bioinformatics, and the sampled tissue affect detection.
Brain-limited mosaicism may be undetectable in blood or a cheek sample. A cheek swab is a specimen, not a separate type of sequencing. It can reveal mosaicism when the variant is present in the sampled cells and the assay is sufficiently sensitive. An unrevealing cheek-swab result does not by itself establish mosaicism, and it cannot exclude a variant confined to brain tissue. Reanalysis can help when knowledge changes, but it cannot recover a variant absent from the sampled tissue. Testing archived tissue from a clinically indicated operation can be considered by specialists. This example does not propose surgery solely to obtain DNA.
Keep related presentations distinct
SLC35A2-CDG and brain somatic SLC35A2-associated epilepsy are related but should not be treated as interchangeable phenotypes. MOGHE is a histopathologic diagnosis, and a brain variant alone does not establish MOGHE or show that every systemic feature has the same cause.
A normal transferrin glycosylation result does not exclude SLC35A2-CDG, and blood screening may not reflect a brain-restricted process. Specialists choose the appropriate biochemical and molecular workup. Sources
Why specialist access can change the conversation
D-galactose supplementation has shown promising findings in small studies: ten people with SLC35A2-CDG in an 18-week study, and twelve postsurgical patients with MOGHE in a six-month pilot, six with a detected SLC35A2 brain variant. The latter reported gastrointestinal adverse effects in two participants. These studies do not establish universal efficacy, long-term safety, or a benefit Charlie would necessarily have received earlier. Studies
“Medical food” should not be used here as shorthand for a proven, risk-free treatment for every CDG. Treatment selection and monitoring require the relevant specialist. The defensible lost opportunity is earlier expert evaluation and discussion of evolving evidence.
What has to line up between a parent's observation and the clinic
Charlie's referral depended on a chain of ordinary steps, and any one of them can quietly fail. The diagram borrows the Swiss cheese model from safety science: each slice is a safeguard, each hole is a way that safeguard can fail, and a missed opportunity passes all the way through only when the holes line up. Select a slice to read how it fails, what would repair it, and where it touches Charlie's story. Source
The slices are illustrative safeguards drawn from my account of Charlie's care. They are not an audit of any care site, and not a claim that every layer failed.
Why the exact words change the answer
When a clinician writes "inverted nipples" in a note, a person reading the chart understands it. For a computer to reuse the finding, whether for a decision-support rule, a research extract, or a tool like the one on this page, the finding has to be stored as a term from a shared dictionary. The dictionary used here is the Human Phenotype Ontology (HPO), and it is arranged as a tree. "Inverted nipples" sits under "Abnormal nipple morphology", which sits under "Abnormality of the breast". Every step up the tree keeps the statement true and throws away detail.
"Dysmorphic facial features", the wording many notes use, is a synonym of "Abnormal facial shape". In HPO 2026-09-01 that term is a sibling of the eyebrow and nose branches, not their parent, so a record that says only "dysmorphic features" does not merely lose detail: it does not match "Thick eyebrow" or "Wide nasal bridge" at all. The table shows what each level of wording does to the shortlist in this snapshot. The rows are computed from the same files and method as the example, and the control in Try the example applies the same roll-up to any selection.
| How the record describes the findings | Terms matched | Linked genes | Where SLC35A2 sits |
|---|
Counts are properties of this annotation snapshot (HPO 2026-09-01, disease annotations 2026-09-02, files retrieved October 21, 2026), not estimates of clinical sensitivity. In an OMOP ETL the same loss happens when a source term is mapped to a broader standard concept, so map to the most specific standard concept available, keep the source term and vocabulary version, document every term that could not be mapped, and test the mapping against known cases before relying on it.
What OMOP, FHIR, and the record can hold, and a pilot that could test it
What OMOP can support
OMOP is not limited to billing diagnoses. CDM v5.4 includes CONDITION_OCCURRENCE, OBSERVATION, MEASUREMENT, NOTE, and NOTE_NLP. Route each record using the standard concept's domain and the relevant conventions. Do not assume every HPO finding belongs in the same table.
Retain the original source term, terminology version, and mapping provenance. Preserve unmapped terms using permitted source fields and a documented companion representation where needed. Availability in a vocabulary does not prove that a particular term is standard, mapped, or loaded at a site.
FHIR can represent phenotypic observations. USCDI includes clinical notes. Whether a particular EHR installation exports the needed structured findings must be measured. Neither a portal download nor a document's presence proves complete coverage. Standards
A small set of reusable fields
For each finding, retain the precise term and identifier, present / absent / uncertain / not assessed, onset or observation date, severity when defined, who reported or confirmed it, and the source document or encounter.
For genetic testing, retain the sample type, method, relevant detection limitations, result classification, and date. Keep suspected conditions separate from established diagnoses.
The interface can be a brief checklist prefilled from existing work. A family can contribute observations, and a clinician can verify them. The design goal is less duplicate entry; actual workload and accuracy must be tested.
An OHDSI pilot that could answer this
- Select a small, reviewed set of rare epilepsy cases, including diagnosed and unresolved cases.
- Trace the same findings through notes, genetics requisitions, results, accessible exchanges, and OMOP.
- Compare a broad spasms-related flag with confirmed cases. Retain the entry route, onset, timing, and negation. Do not require a visit on which every feature was recorded.
- Measure capture completeness, preservation of meaning, review burden, false flags, missed cases, referral completion, and time to review.
- Publish the concept sets, vocabulary versions, mappings, and validation results. Extend the approach to other rare diseases.
Verified phenotype identifiers: plain-language findings and their HPO terms
Preferred labels below were checked against HPO 2026-09-01. Plain-language display text is kept separately from the official term. Generic global developmental delay and intellectual disability are distinct, and a more specific severity term should be retained when supported.
| Plain-language finding | Preferred HPO label | Identifier |
|---|
Profound intellectual disability is HP:0002187. “Dysmorphic facial features” is a synonym of Abnormal facial shape, HP:0001999. Broad terms preserve less detail than the individual findings.
ICD-10-CM context: the billing codes near these findings and why they are not phenotype equivalents
| Code | FY2027 description | Use in this case |
|---|---|---|
| G40.82 | Epileptic spasms | A nonbillable category. The complete codes are G40.821 through G40.824. |
| G40.821 / G40.822 | Not intractable, with / without status epilepticus | Select only the code supported by the record. |
| G40.823 / G40.824 | Intractable, with / without status epilepticus | A historical category alone may miss differently coded cases. |
| F73 | Profound intellectual disabilities | Distinct from unspecified severity or developmental delay. |
| P94.2 | Congenital hypotonia | Context-specific. Do not assign to every hypotonia finding. |
| N64.53 | Retraction of nipple | Do not automatically equate with the congenital phenotype. |
| Q83.8 | Other congenital malformations of breast | Broader category, not a one-to-one HPO substitute. |
Descriptions verified from the CMS FY2027 code file, effective October 1, 2026. This is terminology documentation, not billing advice or a complete executable OMOP concept set. Validate standard concepts, descendants, domain assignments, and source mappings in the site's vocabulary release.
HPO, Mondo, and the OMOP CDM have different roles
HPO defines phenotype terms. HPO disease annotation files connect disease records to reported findings and their evidence. Mondo harmonizes disease entities and identifiers. OMOP structures observational data. A disease identifier does not itself encode a patient's phenotype.
For SLC35A2-CDG, useful disease identifiers include MONDO:0010478, OMIM:300896, and ORPHA:356961. These are not interchangeable with a finding of brain mosaicism or a histologic diagnosis of MOGHE.
The SLC35A2 source records and their annotation gaps
| Finding | OMIM:300896 | ORPHA:356961 |
|---|
“Not annotated here” is not a biological negative. The nipple annotation is “Occasional” (5 to 29%) in the Orphanet source. It is not present in every person with SLC35A2-CDG. Annotations do not establish co-occurrence of all selected features in one individual.
Matching method, source versions, and downloads
The disease corpus includes source records with at least one eligible gene link and an annotation under seizure or hypsarrhythmia. HPO descendant expansion follows is_a edges. We exclude NOT annotations and excluded-frequency annotations. OMIM links marked MENDELIAN and supplied Orphanet links are retained. The latter are labeled UNKNOWN for association type in the gene-link file, which should not be read as a clinical validity grade.
Exact AND filtering is intentionally transparent for teaching. It is not a validated clinical algorithm. It ignores age, inheritance, feature frequency, negative findings, variant mechanism, and many potential etiologies. It is unsuitable for ruling out diagnoses. A clinical tool would require broader evidence, graded matching, and independent validation.
HPO ontology: 2026-09-01. Disease annotations: 2026-09-02. Files retrieved October 21, 2026. The gene-link file has no embedded release date; its exact hash is retained. Counts apply to these files and this method only.
Selected sources: guidelines, HPO files, mosaicism literature, standards
- Human Phenotype Ontology: annotations and downloads. Exact official file URLs and SHA-256 hashes appear above. Primary source for term labels, ancestry, disease findings, frequencies, and references. Original source annotations remain attributable to HPO and its contributing OMIM / Orphanet records.
- ILAE EpilepsyDiagnosis.org: infantile epileptic spasms syndrome. Spasms, syndrome criteria, and EEG context. Hypsarrhythmia is an EEG pattern, not a synonym for every epileptic spasm.
- Smith et al. Genetic testing and counseling for the unexplained epilepsies. NSGC guideline, online 2022 / journal 2023, PMID 36281494.
- Krey et al. Current practice in diagnostic genetic testing of the epilepsies. ILAE Genetics Commission, 2022, PMID 35830287.
- Manickam et al. Exome and genome sequencing for pediatric patients with congenital anomalies or intellectual disability. ACMG guideline, 2021, PMID 34211152.
- Ng et al. Mosaicism of the UDP-galactose transporter SLC35A2 causes a congenital disorder of glycosylation. 2013, PMID 23561849, described two affected males with somatic mosaicism. Winawer et al. Somatic SLC35A2 variants in the brain are associated with intractable neocortical epilepsy. 2018, PMID 29679388, reported variants in resected brain that were undetected in matched blood. These establish that relevant evidence was published years before Charlie's diagnosis. They are supporting references, not a verified list of the exact papers I read. Sdano et al. Clinical utility of chromosomal microarray analysis of DNA from buccal cells. 2014, PMID 25120037, illustrates that buccal specimens can detect some mosaic abnormalities missed in blood. Its cases concern chromosomal abnormalities, not SLC35A2.
- Bonduelle et al. Frequent SLC35A2 brain mosaicism in MOGHE. Acta Neuropathologica Communications, 2021. Also see FCDGC's SLC35A2-CDG description for systemic phenotype and limitations of transferrin screening.
- Witters et al. Clinical and biochemical improvement with galactose supplementation in SLC35A2-CDG. 2020, ten participants. Aledo-Serrano et al. D-galactose supplementation for MOGHE. 2023 pilot, twelve participants. These are small uncontrolled studies.
- OHDSI OMOP CDM v5.4; HL7 FHIR R4 Observation; USCDI Clinical Notes; GA4GH Phenopackets. These describe representational capacity, not implementation at Charlie's care sites.
- CMS ICD-10-CM files. FY2027 code descriptions verified. NCBI MedGen: SLC35A2-CDG supplies disease cross-references.
- Reason. Human error: models and management. BMJ, 2000. The data-journey application is an illustrative adaptation.