Glossary

Data sources

ACS

American Community Survey — the US Census Bureau's annual population and demographics survey

ADA

American Diabetes Association — publisher of the standard US diabetes treatment guidelines (Standards of Care)

BLS

Bureau of Labor Statistics — the US government agency that publishes employment and wage data

CDC

Centers for Disease Control and Prevention — the US federal public-health agency

CPS ASEC

Current Population Survey, Annual Social and Economic Supplement — a US Census/BLS survey of income and education

KFF

KFF (formerly the Kaiser Family Foundation) — a US health-policy research organization

MEPS

Medical Expenditure Panel Survey — a US survey of healthcare use and costs

MITRE

a US nonprofit research organization (the maker of Synthea)

NDSS

National Diabetes Surveillance System — the CDC's diabetes-statistics program

NHANES

National Health and Nutrition Examination Survey — a large US government health survey run by the CDC

NHIS

National Health Interview Survey — a CDC household health survey

NIDDK

National Institute of Diabetes and Digestive and Kidney Diseases — a US National Institutes of Health institute

UKPDS

UK Prospective Diabetes Study — a landmark long-term diabetes study

US Census 2020

the 2020 US Decennial Census — the national population count

USPS L005

US Postal Service product L005 — the official ZIP-code reference data

USPS SCF ranges

US Postal Service Sectional Center Facility ranges — the geographic blocks of ZIP codes assigned to each state

USRDS

United States Renal Data System — the national kidney-disease statistics registry

VEHSS

Vision and Eye Health Surveillance System — a CDC eye-health data source

Clinical & health

A1c

A1c (HbA1c) — a blood test showing average blood sugar over ~3 months, used to diagnose and track diabetes

ACC/AHA

American College of Cardiology / American Heart Association — their guideline defines high blood pressure as 130/80 or higher

antihypertensive

a blood-pressure-lowering medication

biological vs clinical stage

two ways of rating disease severity: biological = how far the disease has actually progressed in the body; clinical = how far it has been detected/diagnosed

BMI

Body Mass Index — a weight-for-height ratio used to screen for under/overweight (weight kg ÷ (height m)²)

CKD

Chronic Kidney Disease — long-term loss of kidney function (stage 1 = mild, stage 5 = kidney failure)

claims

insurance billing records for medical services

comorbidity / co-occurrence

an additional disease present alongside the main one, and how often two conditions appear together

CVD / cardiovascular event

cardiovascular disease — heart and blood-vessel disease; a cardiovascular event is a heart attack or stroke

detection lag

the years a patient had diabetes before it was diagnosed

eGFR

estimated glomerular filtration rate — a number measuring how well the kidneys filter blood (lower = worse)

EHR

Electronic Health Record — the digital medical chart a clinic or hospital keeps on a patient

EIN

Employer Identification Number — a business's federal (IRS) tax ID; the one in the data is fake

ESRD

end-stage renal disease — kidney failure needing dialysis or a transplant

HbA1c / glycohemoglobin

hemoglobin A1c — the formal lab name for the average-blood-sugar (A1c) test

hyperlipidemia

high blood cholesterol/fats

hypertension

high blood pressure

insulin

a diabetes medication; in the data it is only present for diagnosed diabetics

KDIGO

Kidney Disease: Improving Global Outcomes — the international body that sets kidney-disease staging guidelines

marketplace insurance

health insurance bought through the government ACA exchange (HealthCare.gov), vs employer/Medicare/Medicaid

Medicare

the US government health-insurance program, mainly for people aged 65+

microalbuminuria / macroalbuminuria / albuminuria

protein (albumin) leaking into the urine, a sign of kidney damage (micro = small amounts/early, macro = large amounts/advanced)

microvascular complications

small-blood-vessel damage from diabetes, affecting eyes, kidneys and nerves

nephropathy

diabetic kidney damage (to the kidneys' filtering units)

neuropathy

nerve damage (a common diabetes complication causing numbness/pain, often in the feet)

prediabetes

blood sugar higher than normal but not yet in the diabetic range

prevalence

the share of people in a group who have a given condition

retinopathy

diabetic eye damage to the retina that can cause vision loss

SSN / ssn_last_four

Social Security Number; ssn_last_four is the last four digits of a fake one (not a real SSN)

statin

a cholesterol-lowering medication

systolic / diastolic

the top (systolic, heart beating) and bottom (diastolic, heart resting) numbers of a blood-pressure reading

T2DM

Type 2 Diabetes Mellitus — the common, mostly adult-onset form of diabetes

urbanicity

how urban vs rural an area is (urban / suburban / rural)

vitals

basic body measurements such as blood pressure, heart rate, height and weight

Statistics & validation

birthday paradox

the statistical fact that random duplicates appear sooner than intuition expects when values come from a limited set

co-vary / jointly-distributed / joint realism

attributes are generated together so they relate realistically (e.g. age, weight and blood sugar move together as in real people), not as independent random columns

cohort

a defined group of people sharing a characteristic (here, e.g. people at one diabetes severity level)

conditioned on

each value is chosen based on the other values already set for the same person, so they fit together realistically

correlation / correlated / r

how strongly two values move together (r = correlation coefficient: 0 = none, 1 = perfectly together)

Cramér's V

a 0-to-1 score of how strongly two categorical attributes are related (0 = unrelated, 1 = fully tied together)

cross-field invariants

rules that must always hold between fields (e.g. the ZIP code belongs to the state; BMI matches height and weight)

distribution

the spread/shape of values for a field — how often each value occurs across the population

dose-response

more of the cause produces more of the effect (e.g. worse kidney stage leads to higher mortality)

entropy

a measure of how much variety/randomness a field carries

incidence rate

how often a disease newly occurs in a population (vs prevalence, which is how many currently have it)

k-anonymity / quasi-identifiers

a privacy measure: every record looks identical to at least k−1 others on the fields (quasi-identifiers, e.g. ZIP + birth date + sex) that could single someone out

Kolmogorov–Smirnov (KS)

a 0-to-1 statistic measuring how far apart two value distributions are (lower = closer; 0 = identical)

longitudinal vs cross-sectional

longitudinal = tracking the same person over time (a whole history); cross-sectional = a single point-in-time snapshot of many people

Mantel–Haenszel (age-adjusted)

a method that combines results across age groups so age differences don't distort the link

marginal / marginals / marginal distribution

the overall percentage of one attribute across the whole population (e.g. the share of people with diabetes)

MEC-weighted microdata

individual-level survey records weighted (MEC = NHANES Mobile Examination Center sample) to represent the whole US population

microdata

the raw record-by-record survey responses (individual people's data), not just summary totals

monotone / monotonic

values move in only one direction — each later/worse stage is higher than the one before, never dipping back

odds ratio (OR)

how many times more likely a condition is in one group vs another (1 = no difference)

PCA / effective dimensionality

Principal Component Analysis — a measure of how many genuinely independent directions of variation the data has

pp (percentage points)

percentage points — the arithmetic gap between two percentages (e.g. 40% vs 44% is 4 pp)

reference distributions

the published real-world statistics (how values spread in the actual US population) used as the blueprint

sampling variance

normal random fluctuation from drawing a limited sample, not a real flaw

two-sample classifier (C2ST) / AUC

a machine-learning test of how distinguishable fake is from real (AUC: 0.5 = indistinguishable, 1.0 = perfectly separable)

Woolf–Haldane confidence interval

a standard statistical method for the uncertainty range around an odds ratio

μ (mu)

the Greek letter mu — statistics shorthand for the average (mean) value

χ² (chi-squared)

a statistical test of whether observed counts match an even/expected spread

Formats & technical

API

Application Programming Interface — a way for your software to request data from this service automatically

argon2id / hash / salted

argon2id is a strong one-way password-scrambling (hashing) algorithm; a hash can't be reversed to the original; salting mixes in a random value so identical passwords don't store the same hash

async / asynchronous

you submit a request now and download the result later, instead of getting it back instantly

C-CDA

Consolidated Clinical Document Architecture — a standard XML format for US clinical summary documents

CI / CI fixtures

CI = Continuous Integration, the automated build-and-test pipeline run on every code change; fixtures = the fixed sample data those tests run against

CSV

comma-separated values — a plain spreadsheet/table file that opens in Excel or Google Sheets

diffusion models

AI models that generate data by gradually removing random noise (the technique behind many AI image generators)

drift

data quietly changing from one run to the next, so tests become inconsistent

edge network

servers spread worldwide so the site loads quickly from a location near each user

endpoint

a specific URL your software calls to use the service

feature matrix / ML features

the table of input columns/variables fed into a machine-learning model

FHIR / FHIR R4

Fast Healthcare Interoperability Resources (Release 4) — the standard format health-IT systems use to exchange medical records

FHIR resources (Patient/Condition/MedicationRequest/Coverage/Encounter)

the named FHIR record types — e.g. MedicationRequest = a prescription, Coverage = insurance, Encounter = a clinical visit

GANs

generative adversarial networks — an AI model type where two networks compete to produce realistic synthetic data

interoperability

different health systems being able to share and understand each other's data

JSONL

JSON Lines — a text file with one JSON record per line

MCP

Model Context Protocol — a standard that lets AI assistants connect to this service as a tool

MD5

a short digital fingerprint of a file; matching fingerprints prove two runs produced identical data

ML / machine learning

training computer models to learn patterns from data

NDJSON

newline-delimited JSON — one JSON record per line, the standard format for bulk FHIR export

OpenAPI spec

a standard machine-readable description of the API's endpoints and parameters (for developers)

OTP

One-Time Password — a single-use sign-in code emailed to you each login (two-factor authentication)

QA

Quality Assurance — the software-testing function

schema

the set of data fields/columns in each record and their format/structure

SKU

Stock Keeping Unit — a product's inventory ID code

SQL

Structured Query Language — the standard language for relational databases (here, exporting as database insert statements)

staging / non-production / production data

production = the live system real users touch; staging and non-production are pre-production test/dev copies; production data = the real, live data from the running system

test fixtures

the fixed sample data a software test runs against

TLS / encryption in transit

Transport Layer Security — encryption that protects data while it travels between your browser and the server (the browser lock icon)

ULID

Universally Unique Lexicographically Sortable Identifier — a unique, time-ordered record ID

US-Core

the US-specific set of FHIR rules defining required data elements for US health records

UTC

Coordinated Universal Time — the global reference clock the daily quota resets on (midnight UTC)

UUID

Universally Unique Identifier — a random ID with virtually no chance of repeating

variational autoencoders

an AI model that compresses data into a compact form and then generates new lookalike samples

Product

byte-identical

exactly the same output every time, down to the last character

calibrated / calibration

tuned so the synthetic numbers match real-world statistics from trusted US data sources

data dictionary / data domains

a reference list of every field, what it means and its format; domains are the categories the fields group into (identity, geography, health, etc.)

deterministic / deterministic by seed

re-running with the same seed always produces the exact same data (repeatable, not random between runs)

GOLD

the top internal validation grade — passes all the data-quality layers

manifest / dataset manifest

a small summary file describing how a dataset was generated, including its seed

seed

a starting number you choose; the same seed always regenerates the exact same dataset

synthetic data

computer-generated fake records that look statistically realistic but describe no real person

Legal & compliance

data minimisation

collecting only the data actually needed and no more

data portability

the right to receive your data in a reusable format so you can move it elsewhere

data subject

the real, identifiable person a piece of personal data is about (a GDPR term)

de-identification / anonymized data

real data with names and other identifiers stripped or masked so individuals can't be recognized — contrasted with fully made-up synthetic data

DPDP

Digital Personal Data Protection Act — India's data-privacy law

GDPR

General Data Protection Regulation — the EU's data-privacy law

HIPAA

Health Insurance Portability and Accountability Act — the US law protecting patient health-data privacy

legitimate interests / performance of a contract / legal obligation

the lawful reasons (legal bases) privacy law allows for using your data — e.g. to deliver the service, for sensible business needs, or because the law requires it

masking / tokenized

hiding or replacing sensitive values in real records with stand-in placeholders (e.g. blacking out a real SSN)

PHI / protected health information

real, identifiable medical data about a real person that privacy law protects

PII / personally identifiable information

data that can identify a real person (name, address, SSN, etc.)

processor

a company that handles data on our behalf and under our instructions (a GDPR/DPDP role)

Pvt Ltd

Private Limited — an Indian company type (similar to a US LLC/Inc.)

re-identification risk / residual linkage risk

the leftover chance someone could match a supposedly anonymized record back to a real person, often by cross-matching another dataset

Safe Harbor / Expert Determination

HIPAA's two official de-identification methods: Safe Harbor removes 18 specified identifiers; Expert Determination has a statistician certify the re-identification risk is very small

Standard Contractual Clauses

EU-approved legal contract terms that protect personal data sent to other countries