Skip to content

2026 Data Refresh

Companion to the ALS TDI OMOP Data Set release notes. This page covers what a recipient needs to work with the 2026 release and how it differs from the 2025 release (version 0.1.0). Read it before comparing the two.

1. Scope

The release covers 708 ARC participants with linked biological samples, drawn from a registry of 2,014. The mapped instruments and concept sets are unchanged from version 0.1.0: 43 measurement concepts (35 laboratory analytes and 8 self-reported ALS gene concepts) and the same condition concept set. The personal medical history grid is now taken from all four survey rounds rather than round 1 only (see Section 3).

procedure_occurrence, device_exposure, specimen, note and the remaining CDM domains are not populated. Tables conform to the CDM v5.4 column set and are accepted directly by Achilles, the Data Quality Dashboard and ATLAS.

Data vintage: registry surveys 2026-07-05 · central-laboratory results 2026-09-11 · EHR documents 2026-08-05 · release assembled 2026-09-13.

2. Vocabulary

Built against the Athena v5.0 release of 27 February 2026 (version 0.1.0 used 30-AUG-2024). Load that edition or later; earlier bundles lack OMOP Genomic and PPI, so the ALS gene concepts will not resolve. Vocabularies in use: SNOMED, LOINC (in measurement_source_value on EHR rows), RxNorm, NUCC, UCUM, OMOP Genomic, PPI.

The six ARC custom concepts (2000000057–2000000062, 2000000396) are local by design and must be loaded alongside the standard tables.

ALS gene concepts: SOD1 35948140 · C9orf72 35954626 · TARDBP 35964178 · FUS 19643404 · NEK1 35957944 · PFN1 35949229 · SPG11 35950925 · VCP 35958302.

3. Structure and semantics

Three measurement sources, machine-distinguishable. measurement_type_concept_id partitions the table: 32856 central laboratory (56,445 rows, 420 participants), 32817 EHR-derived (16,934 rows, 47 participants; new in this release), 32862 self-reported gene results (80 rows, 75 participants). EHR rows carry the registry analyte concepts with the original LOINC code in measurement_source_value, are already deduplicated, and are the only rows with measurement_datetime populated. EHR-derived visits carry care_site_id 21; all others carry 20.

Every event links to a visit. visit_occurrence (22,969 rows) spans five classes — survey, ALSFRS-R, participant data, blood draw, EHR lab — and 100% of measurement and observation rows resolve to a visit of the matching class. Every event date falls within its participant's observation period.

Observation periods span all recorded event dates, deliberately including retrospective self-reported dates such as symptom onset and medication history — the prodrome interval matters in ALS. Death dates are the one exception: they are year-imputed (nine are placeholders) and are excluded from the period calculation, so a period may end before a participant's recorded death date. No other period ends after the extraction date. Participants with no dated event (10 of 708) carry no period. If your person-time method requires data-capture contact only, derive spans by filtering events on the type concepts above.

Survey items are current-value; ALSFRS-R is longitudinal. Apart from ALSFRS-R and conditions, each survey item carries one current instance per participant — the most recent informative answer. Diagnosis, El Escorial certainty and onset site resolved this way for the first time in this release, so version 0.1.0 reports different values for them: replace it rather than pooling. Multi-select items (family history, occupational industry, onset site) carry one row per selected value. Interval-anchored follow-up questions ("hospitalized in the past 3 months?") are outside the mapped footprint of both releases. ALSFRS-R is the longitudinal record: one row per scale item per assessment, 17,313 assessments over 670 participants; same-day duplicate assessments are retained.

Conditions take affirmatives from all rounds: one row per participant per condition, dated to the round that first reported it (485 rows, 304 participants). A later blank is silence, not retraction — the module contains no explicit negations. Note the lifestyle smoking item maps round 1's lifetime-history question only; the follow-up rounds ask a different question (current smoking) that is not mapped.

Medications are deduplicated (3,142 rows, 529 participants); version 0.1.0 emitted one row per source re-export, so its counts are inflated roughly fourfold and are not comparable. Imputed death dates use 31 December of the death year, capped at the extraction date.

4. Known limitations

  • Nine death rows carry the placeholder date 1900-01-01 — a death with an unknown date, not an event in 1900.
  • unit_concept_id is 0 on dimensionless ratios by design (A/G, BUN/creatinine); elsewhere it is populated on 94.3% of central-laboratory and 90.1% of EHR rows. Rows whose source unit could not be established are excluded rather than delivered unitless.
  • Bulbar and trunk onset sites map to value_as_concept_id = 0 with the verbatim site in value_source_value.
  • Free-text "other" write-in fields are not mapped (family history and personal medical history forms).
  • A small number of participants carry no diagnosis row (only recorded answer was uninformative) or no onset site (onset is taken from the New Enrollee Survey only).
  • EHR encounter structure is not imported; EHR-derived visits are date-based extract visits.

5. Working with these data

Situation What to know What to do
Joining registry blood draws to EHR labs Both carry the same 35 registry analyte concepts; LOINC is in measurement_source_value Join on measurement_concept_id
Distinguishing data sources Type concepts partition sources: 32856 lab, 32817 EHR, 32862 self-report Filter on the type concept
Person-time or incidence Periods include retrospective self-reported dates by design Derive contact-based spans via type concepts if needed
Longitudinal history of a survey item Only ALSFRS-R is longitudinal Other items reflect the most recent informative answer
ethnicity_concept_id = 0 Means unknown, by design Treat 0 as unknown
Units vs. version 0.1.0 Unit labels and concepts were corrected in this release (hemoglobin g/dL, mmol/L electrolytes, cells/uL differentials, ALT labelling) See the DIVERGENCES documentation before cross-release comparison
Date formats Registry dates are US format, EHR dates are ISO 8601 Parse explicitly
Any comparison with version 0.1.0 Diagnosis, certainty, onset site, measurements, medications and visit structure all differ Replace it with this release
Loading vocabularies Gene and custom concepts need the 27-FEB-2026 release plus the supplied local concepts Load both

Working from the raw survey exports instead: yes/no columns sometimes carry an HTML checkmark instead of "Yes", and free text mixes three apostrophe encodings — normalize both.

6. Provenance

The registry-side transform and the EHR side (built with the open-source Registry Forge ALS; preprint: Boyce D, et al., medRxiv 2026.06.02.26354637) are documented in the METHODS and QC files that accompany the delivery. Documents containing participant names are excluded by policy and are never a pipeline input. Detailed documentation and person-level inclusion accounting are available to data recipients on request.

7. Caution

These data are participant self-report plus central-laboratory results plus patient-mediated EHR extracts — not a clinician-adjudicated chart review.

Questions and data problem reports: dboyce@als.net.