CBDB Amyloid & Comorbidity Schema (amyloid_comorbidity)
Last updated: 2026-10-30
Overview
amyloid_comorbidity collects Alzheimer's disease biomarkers, anti-amyloid treatment and comorbidities for
every CBDB person (158,474 people), ready to join to the research schema on person_id.
| Topic | What is in it |
|---|---|
| Amyloid PET | every amyloid PET exam with its read (positive / negative / ...), Centiloid and tracer; every PET order in any status |
| CSF and plasma | AD biomarker lab orders with the lab's verdict and values (Abeta42, t-tau, p-tau181, ATI, ratios, p-tau217, AD-Detect score) |
| Clinical notes | results stated in notes (including tests done outside), APOE genotype, tau PET, NfL / GFAP |
| Person summary | amyloid tested / positive ever, first positive date, latest status, per source |
| Anti-amyloid drugs | lecanemab, donanemab, aducanumab exposure with first-dose date and infusion counts |
| Cognitive scores | dated MMSE, CDR global, CDR sum of boxes, the six CDR boxes and MoCA from Dementia Clinic notes, ADRC visits and Epic flowsheets |
| Leqembi cohort | lecanemab cases and lecanemab-eligible untreated controls, with T0 |
| Comorbidities | Table 1 conditions (ICD-10) and appropriate-use exclusions, dated |
Access. The trail database user can SELECT all tables listed on this page.
Real dates. Unlike the
researchschema, where dates are converted to ages, this schema carries real calendar dates (exam, collection, note and diagnosis dates), andamy_pet_readscarries the imaging accession number. Treat the data as identifiable under your IRB protocol and data use agreement. To de-identify, compute ages or days relative to a study T0 before exporting results.
At a glance (2026-10-30)
| Persons | |
|---|---|
| Ever tested for amyloid (PET, CSF, plasma, or a result in notes) | 4,118 |
| Ever amyloid positive, any source | 1,942 |
| Ever positive in structured data (PET report or CSF / plasma lab) | 1,596 |
| Positive only in notes (usually a test done elsewhere) | 346 |
| Positive by source: PET 958, CSF 640, plasma 66, notes 1,391 (one person can count in several) | |
| APOE genotype stated in Dementia Clinic notes | 589 |
| Leqembi cases / eligible untreated controls | 354 / 977 |
Key concepts
Verdict vocabulary (used across tables)
| Verdict | Meaning |
|---|---|
positive / negative |
stated read or lab interpretation |
borderline |
legacy ADMARK CSF "Borderline"; never folded into positive or negative |
indeterminate |
equivocal / intermediate / conflicting |
reduced |
post-treatment PET with reduced but still present amyloid (use Centiloid) |
scanned_only |
legacy CSF whose result exists only as a scanned report (not readable) |
no_result |
order with no result |
"Ever" vs. "at T0". amy_person summarizes a person over all time. A treated patient whose amyloid cleared
has any_pos_ever = 1 but status_latest = 'negative'. For status at a study date, use amy_events (recipe below)
or the *_by_t0 columns of lec_cohort.
Structured vs. note-stated. Structured results come from PET reports and lab orders in Epic. Note-stated
results are tests quoted in clinical notes, often done elsewhere. They were extracted two ways: rules over all
post-2020 CBDB notes (method = 'regex'), and a large language model (Claude, through the institution's approved
AWS route) over Dementia Clinic notes (method = 'llm'). Use the structured_* columns for a strict definition.
Raw values do not pool across assays. The old and new ADMARK CSF tests report Abeta42 on different scales. Quest and Labcorp plasma Abeta42/40 use different cut-offs (normal >= 0.170 vs > 0.102). Combine verdicts, not values.
Tables
| Table | One row per | Use it for |
|---|---|---|
amy_person |
person | amyloid status, first positive date, per-source summary |
amy_events |
test result (any source) | status at or before any date |
amy_pet_reads |
amyloid PET exam | reads, Centiloid, tracer |
amy_pet_orders |
PET order, any status | what was ordered, performed, reported, and whether it became an amyloid exam |
amy_fluid_results |
CSF / plasma order | lab verdicts and values |
amy_note_results |
distinct note-stated amyloid result | amyloid PET / CSF / plasma results from notes |
amy_note_llm_results |
distinct LLM-read result | everything the LLM read: amyloid, tau PET, APOE, NfL, GFAP |
amy_apoe_person |
person with APOE in notes | APOE genotype, e4 copies |
amy_mab_exposure |
person × drug | lecanemab / donanemab / aducanumab exposure |
cog_scores |
person × test × date × score | MMSE / CDR / CDR-SB / CDR boxes / MoCA over time |
lec_cohort |
Leqembi case or control | the case / control study cohort |
int_lec_eligibility |
person with MCI / AD dx or drug record | eligibility components, funnels |
cmb_dx_events |
person × condition × dated code | comorbidity in any window |
cmb_person_condition |
person × condition | first / last date, counts by source |
cmb_person_flags |
person | wide ever_<cond> / first_<cond>_date flags |
reference: amyloid_procs, amyloid_fluid_components, comorbidity_icd10_map |
code | test whitelist, lab components, condition code lists |
amy_person
One row per CBDB person.
| Column | Meaning |
|---|---|
amyloid_tested_ever |
1 if any amyloid PET, CSF / plasma order or note-stated result |
any_pos_ever, first_pos_date |
ever positive from any source, and the first positive date |
structured_pos_ever, first_structured_pos_date |
the same, PET reports and lab results only |
note_only_pos |
positive in notes with no positive structured result |
status_latest, status_source |
latest status: latest PET read if positive / negative, else latest CSF verdict, else latest plasma (AD-Detect), else latest note verdict; indeterminate if tested but none apply; not_tested |
pet_pos_ever, first_pet_pos_date, latest_pet_read, latest_pet_date, n_pet_exams |
PET summary |
first_centiloid, latest_centiloid (+ dates) |
Centiloid from PET reports |
csf_pos_ever, latest_csf_verdict, has_csf_values, csf_scanned_only_ever, n_csf_orders |
CSF summary |
plasma_pos_ever, latest_plasma_verdict, has_plasma_values, n_plasma_orders |
plasma summary |
note_pos_ever, note_neg_ever, latest_note_verdict, n_note_results |
note summary |
amy_events
All amyloid evidence in one long table: one row per PET exam, lab order, or distinct note-stated result.
| Column | Meaning |
|---|---|
source |
pet_report, lab, note |
modality |
PET, CSF, PLASMA |
assay |
tracer (PET), assay name (lab), or extraction method / rule (note) |
event_date, date_source |
exam / collection date; for notes the stated test date, else the first note stating it (an upper bound) |
verdict |
see the vocabulary above |
centiloid, centiloid_prior_inline |
Centiloid; a prior value quoted inline in a follow-up report |
ab42, ttau, ptau181, ati, plasma_ptau217, plasma_ab4240, addetect_score |
values where present |
is_structured |
1 for PET reports and labs, 0 for notes |
amy_pet_reads
One row per amyloid PET exam (identified by tracer name in the report, not by order name; most are ordered as a
generic PET-CT BRAIN).
| Column | Meaning |
|---|---|
exam_date, accession_num, description |
exam date (end of exam), accession, order description |
tracer |
florbetaben, florbetapir, flutemetamol, or unnamed (Centiloid only) |
read_class |
positive, negative, reduced, indeterminate, research_no_read, unclassified, not_amyloid |
centiloid, cl_prior_inline |
Centiloid of this exam; a prior value quoted in the same report |
research_read, outside_scan |
research protocol scan; outside study posted to PACS |
Reads are rule-based on the report impression and checked against each report's own Centiloid (positive reads
average about 82, negative about −3). Exams run to the latest Epic pull (2026-10-26). An exam is here only when it was
performed (exam-end time in Epic) and its report names an amyloid tracer or a Centiloid; for every PET order, including
those that never became an amyloid exam, see amy_pet_orders.
amy_pet_orders
One row per PET order in any status (68,471 orders, 10,900 people): the denominator behind amy_pet_reads. An order
can be canceled, not yet performed, an outside scan archived without a Columbia report (ARCHIVE ONLY NM PET), a
research scan without a clinical read (PET-CT RESEARCH), or an extra order line of a scan whose report sits on
another order.
| Column | Meaning |
|---|---|
description, order_status, order_class |
order name, Epic status (Completed / Sent / Canceled), class |
ordering_date, exam_end_date, event_date |
order date; exam-end date; exam date, else procedure start, else order date |
pet_area |
brain, body (oncology, cardiac, other body PET) or unspecified (limited area, research, archive, outside), from the order name |
tracer, tracer_source |
amyloid, tau, fdg or unknown, from the report when it names one (report), else the order name (order_name) |
canceled, performed, has_report |
canceled; exam-end time set and not canceled; a radiology report exists |
in_amy_pet_reads, read_class, centiloid |
the order is an amyloid exam in amy_pet_reads, with its read |
For example, among the 212 Leqembi cases at least 15 months past T0, 76 have a non-canceled brain PET order after that
point, 62 have a report, and 60 have an amyloid PET in amy_pet_reads.
amy_fluid_results
One row per non-canceled CSF or plasma amyloid order.
| Column | Meaning |
|---|---|
modality, assay, proc_name, collect_date |
CSF / PLASMA, assay (csf_admark_legacy, csf_admark, plasma_addetect, ...), collection date |
verdict, verdict_raw |
normalized and raw lab interpretation |
ab42, ttau, ptau181, ati, ptau181_ab42, ttau_ab42 |
CSF values (pg/mL; ATI and ratios unitless) |
addetect_score, plasma_ptau217, plasma_ab4240, plasma_ptau181 |
plasma values |
Readable legacy CSF results start in 2021. Earlier legacy CSF exists only as scanned reports (scanned_only).
amy_note_results
One row per distinct amyloid PET / CSF / plasma result stated in notes (repeats from copy-forward collapsed).
method = regex (all post-2020 notes) or llm (Dementia Clinic notes). The two methods are not de-duplicated
against each other, so a result found by both appears twice. This does not affect "ever positive" flags.
amy_note_llm_results
One row per distinct result the LLM read in Dementia Clinic notes (2020–2025), for amyloid PET, tau PET, CSF, plasma, APOE, NfL and GFAP. Only statements of actual results are here, not plans, pending tests or eligibility criteria.
| Column | Meaning |
|---|---|
modality |
PET, TAU_PET, CSF, PLASMA, APOE, NFL, GFAP |
verdict, verdict_source |
stated verdict, or (legacy ADMARK values with no stated interpretation) the vendor cut-off rule |
apoe_genotype |
e.g. e3/e4, e4_carrier |
test_name, performed_elsewhere, confidence |
test as written, outside test, model confidence |
| value columns | centiloid, suvr, csf_*, plasma_*, nfl, gfap as written |
test_date, date_precision, event_date, date_source |
stated test date (day / month / year), else first note date |
n_notes, first_note_date, last_note_date |
how often and when it was restated |
amy_apoe_person
APOE genotype per person from Dementia Clinic notes: apoe_genotype (a full genotype, e4_carrier /
e4_noncarrier when only carrier status is stated, or conflict when notes disagree) and e4_copies (0 / 1 / 2).
Structured APOE lab orders are not included yet.
amy_mab_exposure
One row per person × drug (lecanemab, donanemab, aducanumab), from Epic medication orders and infusion
(MAR) records.
| Column | Meaning |
|---|---|
t0, t0_source, t0_reliable |
first dose: first infusion (MAR), else first order (ORDER, e.g. the subcutaneous autoinjector), else an outside-medication entry (OUTSIDE_MED, t0_reliable = 0: started elsewhere before that date) |
infusion_days, first_infusion_date, last_infusion_date |
infusion days with a given / started administration |
order_site |
Columbia only, Columbia + another site, Weill Cornell only, NYP Medical Group only, Outside med only, ... |
ever_sc, ever_investigational |
subcutaneous formulation; trial formulation |
cog_scores
Dated cognitive scores, one row per person × source × test × date × score. Each row is one administration of a test on the day it was done; visits with no score have no row (nothing is carried forward).
| Column | Meaning |
|---|---|
source |
note (Dementia Clinic notes, read by a language model), adrc (ADRC UDS visits, to 2024-11), flowsheet (Epic flowsheets, Leqembi cohort only) |
test |
mmse (Folstein), cdr (CDR global), cdr_sb (CDR sum of boxes, 0–18), the six CDR boxes cdr_memory, cdr_orientation, cdr_judgment, cdr_community, cdr_home_hobbies, cdr_personal_care (0 / 0.5 / 1 / 2 / 3), moca |
score_date, date_approx |
test date; date_approx = true when the note gives only a month or year (the 1st is used) |
score, max_score |
the score; MMSE out of another total (partial or video administration, e.g. 22 / 28) keeps that total (15–30; other denominators are extraction errors, such as a date read as the denominator, and are dropped) |
in_range |
meets the Leqembi early-AD criterion on its own (MMSE >= 20 / 30, CDR 0.5–1, MoCA >= 17); null for partial MMSE, cdr_sb and the boxes |
n_notes |
notes giving this score on this date |
| Source | MMSE | CDR | CDR-SB | MoCA |
|---|---|---|---|---|
| Notes (rows / people) | 13,413 / 7,405 | 8,819 / 3,856 | 4,448 / 2,344 | not extracted |
| ADRC | 3,238 / 539 | 4,803 / 1,108 | 4,803 / 1,108 | 2,161 / 768 |
| Epic flowsheets | 191 / 125 | 1 / 1 | 27 / 17 |
Each CDR box has about 4,440 note rows (2,345 people; personal care 4,658) and one row per ADRC visit (4,803 / 1,108).
Rows of one CDR rating share person_id, source and score_date; pivot on test to get the boxes side by side.
Dates. A note score is dated by the test date written in the note; only when none is written is the note date used (about 45% of note MMSE and 60% of note CDR in the Leqembi cohort). Notes often restate an earlier visit's result, so the note date alone would misdate it. Copies of one result across notes collapse to one row.
Copied-forward MMSE. Later notes repeat the first visit's total in their summary ("MMSE was 10/30 with 0/3 recall") without a date. A note MMSE with fewer than two sub-item domains, no written date, and the same score the person already had on an earlier date is treated as that earlier result and dropped. Without this rule MMSE looks unchanged over time. CDR is rated at each visit and is kept as written.
Not the raw table. research.neuro_note_mmse_cdr is the per-note extraction behind the note rows: it still has
every restated and copied score, and research.adrc_values has ages instead of dates. Use cog_scores for
trajectories.
lec_cohort
The Leqembi (lecanemab) case / control cohort, one row per person.
Definitions. Cases are everyone ever exposed to lecanemab; T0 is the first dose. Controls are lecanemab-eligible patients who never received lecanemab, donanemab or aducanumab. To be eligible, all of these must hold:
- Early-stage AD: an MCI diagnosis (G31.84), or an AD diagnosis (G30) plus early symptoms (MMSE >= 20 out of 30,
CDR global 0.5 to 1, or MoCA >= 17, from
cog_scoreson or after 2023-01-01). AD patients with no score since 2023 are judged on their last score before 2023 (met_pre2023kept,not_met_pre2023not eligible); with no score ever they are kept withcognitive_criterion = 'unknown'. - Amyloid positive from any source (
first_amy_pos_date). Usefirst_amy_pos_structured_datefor structured-only. - Eligibility maintained after 2023: an MCI or AD diagnosis dated on or after 2023-01-01.
The control T0 is the later of 2023-01-01 and the date all criteria first held. Controls whose scores within 3 months of T0 are all out of range had progressed past early AD and are not included.
| Column | Meaning |
|---|---|
arm, t0, t0_source, t0_reliable |
case / control and index date |
cohort_tier, cohort_tier_name |
sampling priority: 1 Dementia Clinic, 2 ADRD, 3 Pre-AD, 4 other |
early_ad_path, cognitive_criterion |
MCI, AD, AD_cognition_pre2023, AD_cognition_unknown; met / not_met / met_pre2023 / not_met_pre2023 / unknown |
amy_pos_by_t0, amy_pos_structured_by_t0, mci_ad_dx_by_t0 |
status at T0 |
max_mmse_2023, min_cdr_global_2023, max_moca_2023 |
best MMSE / CDR global / MoCA since 2023 |
n_cog_dates_after_t0 |
dates with any cognitive score after T0 |
order_site, infusion_days, ever_sc, other_amab_ever |
treatment details (cases) |
aup_stroke_12m, aup_ich_ever, aup_excluded |
appropriate-use exclusions at T0, as flags only: stroke / TIA in the prior 12 months (strict codes, encounter diagnoses), intracranial hemorrhage ever; either |
trial_participant |
any investigational anti-amyloid order (clinical trial dosing) |
last_record_date, has_followup, followup_days |
last visit, diagnosis, Dementia Clinic note or cognitive score; any record after T0; days from T0 |
| Dementia Clinic | ADRD | Pre-AD | Other | Total | |
|---|---|---|---|---|---|
| Cases | 337 | 6 | 11 | 354 | |
| Controls | 825 | 46 | 98 | 8 | 977 |
Visits, diagnoses, notes, imaging and MMSE / CDR for this cohort run to October 2026 (2026-10 refresh).
int_lec_eligibility
The eligibility components for every person with an MCI / AD diagnosis or an anti-amyloid drug record (diagnosis dates, MMSE / CDR since 2023, amyloid positivity, eligibility path and date). Use it to build your own variant of the definition, or a funnel.
cmb_dx_events, cmb_person_condition, cmb_person_flags
Comorbidities from CBDB ICD-10 diagnoses (billing, encounter, problem list). cond_group is table1 (30 Table 1
conditions such as htn, diabetes, depression, sleep_apnea), aup (appropriate-use exclusions such as
aup_stroke_strict, aup_ich) or lec_elig (mci, ad). Code lists are in comorbidity_icd10_map.
cmb_dx_events is dated (problem-list entries have an end date when resolved), so any window can be applied.
Example queries
Amyloid status of your cohort, ever:
SELECT c.person_id, a.any_pos_ever, a.first_pos_date, a.structured_pos_ever, a.status_latest
FROM my_cohort c
LEFT JOIN amyloid_comorbidity.amy_person a USING (person_id);
Amyloid status at a study T0 (latest result on or before T0, per source):
WITH t0 AS (SELECT person_id, t0 FROM my_cohort),
pre AS (
SELECT e.*, row_number() OVER (PARTITION BY e.person_id, e.source, e.modality ORDER BY e.event_date DESC) AS rn
FROM amyloid_comorbidity.amy_events e JOIN t0 USING (person_id)
WHERE e.event_date <= t0.t0 AND e.verdict IN ('positive', 'negative', 'borderline')
)
SELECT person_id,
max(CASE WHEN source = 'pet_report' AND rn = 1 THEN verdict END) AS pet_pre,
max(CASE WHEN modality = 'CSF' AND source = 'lab' AND rn = 1 THEN verdict END) AS csf_pre,
max(CASE WHEN verdict = 'positive' THEN 1 ELSE 0 END) AS any_pos_pre
FROM pre GROUP BY person_id;
The Leqembi cohort, Dementia Clinic first, with age at T0 from the research schema:
SELECT l.person_id, l.arm, l.t0, l.cohort_tier_name, l.early_ad_path,
extract(year FROM l.t0) - p.year_of_birth AS approx_age_at_t0
FROM amyloid_comorbidity.lec_cohort l
JOIN research.person p USING (person_id)
WHERE l.cohort_tier = 1;
MMSE / CDR-SB trajectory around T0, in months (baseline = last score on or before T0):
SELECT c.person_id, c.arm, s.test, s.source, s.score_date,
round((s.score_date - c.t0) / 30.44, 1) AS months_from_t0, s.score, s.date_approx
FROM amyloid_comorbidity.lec_cohort c
JOIN amyloid_comorbidity.cog_scores s USING (person_id)
WHERE s.test IN ('mmse', 'cdr_sb') AND (s.test <> 'mmse' OR s.max_score = 30)
ORDER BY c.person_id, s.test, s.score_date;
Comorbidities in the 2 years before T0 (problem-list entries count while open):
SELECT c.person_id, d.cond_key, 1 AS present_2y
FROM amyloid_comorbidity.lec_cohort c
JOIN amyloid_comorbidity.cmb_dx_events d USING (person_id)
WHERE d.cond_group = 'table1'
AND d.start_date <= c.t0
AND coalesce(d.end_date, CASE WHEN d.source_type = 'problem_list' THEN date '9999-12-31' ELSE d.start_date END)
>= c.t0 - 730
GROUP BY 1, 2;
Coverage and caveats
- Sources. PET, CSF, plasma and anti-amyloid drugs come from Epic (Clarity):
ORDER_NARRATIVE/ORDER_IMPRESSION,ORDER_RESULTS/ORDER_RES_COMMENT,ORDER_MED/MAR_ADMIN_INFO. Diagnoses come from CBDBcondition_occurrence. Notes come from CBDBcbdb_notes(post-2020) andneuro_clinic_notes. - Time coverage. For the Leqembi cohort, diagnoses, visits, notes, imaging and MMSE / CDR run to October 2026 (2026-10 refresh). For other CBDB persons notes and diagnoses stop in late 2025 until the full CBDB refresh. Epic labs, PET and drug records run to the latest extract (October 2026). Recent eligibility can be under-counted.
- Persons. Epic-derived tables cover CBDB persons with an Epic patient ID (about 2,600 legacy-system persons have none).
- MMSE / CDR / MoCA come from Dementia Clinic notes, ADRC visits (to 2024-11) and Epic flowsheets (Leqembi
cohort). MoCA is not extracted from notes, so it exists only from ADRC and flowsheets. MMSE is not given at every
visit; copied-forward MMSE is removed (see
cog_scores). - Note-stated test dates are those written in the note; otherwise the first note stating the result is used, which is an upper bound of the test date.
- Comorbidity counts include billing diagnoses.
I67.82("cerebral ischemia", common on imaging bills) makes small-vessel disease and stroke / TIA higher than encounter-only counts. Restrict onsource_typeif needed. - CBDB "control" cohort. 19 people in the CBDB control cohort (no ADRD diagnosis) have a positive amyloid result: possible preclinical AD or a cohort-flag error. Review them before using that cohort as a negative reference.