Population-health pipeline

Five supplier feeds, a validated warehouse, nine quality measures, longitudinal patient documents and a de-identified copy - run on a synthetic network of 5,000 patients over five scheduled days, and graded against an answer key. This page is built from that run. Repository · run evidence · design decisions

Rows delivered

51,485

18 files from 6 supplier sources

Planted defects caught

269 / 269

by the rule each was planted for

Measure cells matching

117 / 117

dbt vs an independent implementation

dbt nodes passing

54 / 54

models, seed and tests, last build

Leak-gate checks

12 / 12

before the demo copy is released

Five scheduled days

Each run sees only the files that had arrived by its date.

day	what happens	files loaded	rows held	gate	freshness	measures rebuilt
2026-02-16	First weekly extract: a year of history from all five suppliers.	6	90	all published	all on time	yes
2026-02-23	Second extract: late registrations released, appointment statuses updated, late 2025 lab results and claims restate the closed year.	6	15 (released 35)	all published	all on time	yes
2026-03-02	Third extract: the laboratory's new layout (IFCC units) is recognised; the clinic appointments file arrives truncated and is stopped; the pharmacy file is due.	4 + 1 stopped	5	held: appointments	appointments stuck; pharmacy due	yes
2026-03-03	The clinic file is re-delivered and published; the pharmacy file is now late.	1	0	all published	pharmacy late	yes
2026-03-03	The same day again: nothing new, nothing changes.	0	0	nothing new	pharmacy late	no (nothing changed)

What the rules caught

Every planted defect, by the rule it was planted for, held, repaired or flagged by that rule - and, graded separately, nothing else touched.

Planted defects caught, by rule

Table

rule	what it checks	planted	caught
RX-012	Late claims for a closed measurement year are announced	40	40
LAB-003	Medical record numbers keep their leading zeros	30	30
ROS-011	Phone numbers that cannot be standardised are blanked	30	30
LAB-013	Late rows for a closed measurement year are announced	26	26
APT-009	Patient is on the roster	15	15
ROS-007	Clinic codes are upper case	15	15
SCR-010	Patient is on the roster	15	15
RX-010	Patient is on the roster	12	12
ROS-008	Attributed clinic exists	9	9
ROS-010	One row per record number	8	8
SCR-012	Platform totals agree with the item responses	8	8
APT-006	Status code is mapped	6	6
SCR-007	Item responses are whole numbers within the instrument's range	6	6
APT-011	Visit type is mapped	5	5
LAB-008	Result is a number or a censored number	5	5
LAB-011	Patient is on the roster	5	5
RX-006	Days supply is a whole number from 1 to 365	5	5
duplicates	Exact duplicates inside a file are merged, counted once	5	5
LAB-010	Result is clinically plausible	4	4
SCR-009	Administration mode is recognised	4	4
APT-007	Clinic exists	3	3
RX-008	Drug is in a tracked class	3	3
APT-008	Completed or missed appointments are not in the future	2	2
LAB-009	Unit is valid for the analyte	2	2
ROS-005	Birth date is a real date	2	2
SCR-006	Screening date is not after the extract	2	2
SCR-011	One version of each screening per file	2	2

Measures, 2025

Network rates from the dbt marts. Each is also computed by an independent Python implementation from the generator's own data; the two agree at every clinic, and adherence agrees patient by patient.

Network rate by measure

Table

measure	numerator	denominator	rate
Adherence: diabetes medications	204	323	63.2%
Adherence: RAS antagonists	434	690	62.9%
Adherence: statins	349	553	63.1%
Follow-up after a positive depression screen	285	407	70.0%
Depression screening	2,612	3,520	74.2%
Diabetes: HbA1c below 8.0%	141	307	45.9%
Diabetes: HbA1c above 9.0% or not tested	114	307	37.1%
Diabetes: HbA1c tested	273	307	88.9%
Social needs screening	1,316	3,289	40.0%

Restated by late data (day 2)

Late lab results and late-adjudicated claims for 2025 arrived after the year was reported.

measure	day 1	day 2
Diabetes: HbA1c tested	272 / 307 (88.6%)	273 / 307 (88.9%)
Adherence: diabetes medications	203 / 322 (63.0%)	204 / 323 (63.2%)
Adherence: statins	346 / 553 (62.6%)	349 / 553 (63.1%)

Depression screening, by language and payer

Relative rate compares a group with everyone else in its stratifier. Small cells are hidden, with a complementary cell where needed.

group	eligible	rate	relative rate
language group: English	2,391	78.4%	1.199
language group: Other	–	suppressed (complementary)
language group: Spanish	706	65.9%	0.863
language group: Unknown	–	suppressed (small cell)
payer type: commercial	1,166	73.8%	0.993
payer type: government	1,692	73.6%	0.985
payer type: marketplace	345	75.4%	1.017
payer type: self_pay	317	77.3%	1.046

Freshness

Per supplier source: did the delivery arrive on time, and did what arrived reach published? Day 3 and day 4.

2026-03-02

feed	source	delivery	pipeline	newest published	detail
appointments	clinic_scheduling	on time	stuck	2026-02-20	appointments_main_20260302.csv blocked by APT-001
appointments	mobile_unit	on time	stuck	2026-02-20	appointments_mobile_20260302.csv waiting at merged
labs	reference_lab	on time	ok	2026-02-27
pharmacy	pharmacy_claims	due	ok	2026-02-22	expected today (2026-03-02), not yet arrived
roster	ehr_roster	on time	ok	2026-03-02
screenings	screening_platform	on time	ok	2026-02-27

2026-03-03

feed	source	delivery	pipeline	newest published	detail
appointments	clinic_scheduling	on time	ok	2026-02-27
appointments	mobile_unit	on time	ok	2026-02-27
labs	reference_lab	on time	ok	2026-02-27
pharmacy	pharmacy_claims	late	ok	2026-02-22	expected by 2026-03-02, 1 day(s) late
roster	ehr_roster	on time	ok	2026-03-02
screenings	screening_platform	on time	ok	2026-02-27

De-identification

The same patient in the restricted documents and in the released demo copy (both synthetic). Every date moved by one shift; clinical values unchanged.

Restricted

{
 "_id": "01301982",
 "demographics": {
  "firstName": "Katherine",
  "lastName": "Allen",
  "birthDate": "1959-09-25",
  "phone": "5555665004",
  "email": "katherine.allen23@example.org"
 },
 "address": {
  "city": "Easton",
  "line1": "70 Sunset Ln",
  "zip": "70303"
 },
 "careTeam": [
  {
   "name": "Aisha Ramirez",
   "role": "primary_care"
  },
  {
   "name": "Ana Flores",
   "role": "care_manager"
  }
 ],
 "latestPHQ9": {
  "band": "minimal",
  "complete": true,
  "date": "2025-01-21",
  "positive": false,
  "score": 1
 }
}

Released demo copy

{
 "_id": "D9395744",
 "demographics": {
  "firstName": "Kai",
  "lastName": "Stonebridge",
  "birthDate": "1959-09-19",
  "phone": "5559506210",
  "email": "kai.stonebridge71@example.org"
 },
 "address": {
  "city": "Easton",
  "line1": "1791 Foundry Walk",
  "zip": "70316"
 },
 "careTeam": [
  {
   "name": "Marlow Birchmore",
   "role": "primary_care"
  },
  {
   "name": "Arden Jessamine",
   "role": "care_manager"
  }
 ],
 "latestPHQ9": {
  "band": "minimal",
  "complete": true,
  "date": "2025-01-15",
  "positive": false,
  "score": 1
 }
}

Leak gate

check	result	detail
no real record number anywhere	pass	searched 301,620 strings
no real full name anywhere	pass	searched 301,620 strings
no real e-mail anywhere	pass	searched 301,620 strings
no real phone anywhere	pass	searched 301,620 strings
no real street address anywhere	pass	searched 301,620 strings
every demo patient differs from the real one	pass	4,997 patients compared
dates moved by one consistent, non-zero shift	pass	birth dates checked against shifts
policy transforms every registered identifier path	pass	17 paths covered
patients point at clinics, worklists at patients	pass	all references resolve
extract carries only allowed columns	pass	ok
no age above 89 and no birth year for 90+	pass	ok
restricted ZIP3 prefixes replaced by 000	pass	ok

Synthetic data only. Built by python -m pophealth demo --docs docs. LOINC content (codes only) is copyright Regenstrief Institute, Inc., used under the LOINC licence.