Output schema

patient_master.csv follows the exact same schema as Registry Forge's master CSV. This page documents each column. If you're already familiar with Registry Forge, you can skim - the only behavioral differences are in the source column (always ccda or empty) and the absence of fhir rows.

Columns

Demographic block (always present)

Column	Required	Notes
last_name	yes	Patient's last name
first_name	yes	Patient's first name
dob	yes	Date of birth, YYYY-MM-DD
patient_id	yes	Canonical patient identifier, format PT-XXXXXXXXXX
mrn	yes	Medical record number when available; blank otherwise
gender	yes	male, female, other, unknown
marital_status	when recorded	Free-text from maritalStatusCode/@displayName

The demographic block is duplicated onto every row of every record, so a filter to one patient's rows is fully self-describing.

Record block

Column	Required	Notes
category	yes	One of: patient, problems, medications, procedures, labs_vitals, allergies, immunizations, careplans, diagnostic_reports, goals, notes, encounters, observations
source	yes	Always ccda in Patient Edition (or empty for patient header rows)
effective_date	when known	YYYY-MM-DD. For the patient row, this is DOB.
end_date	when known	End/resolution date
code	when coded	Primary code value
code_system	when coded	Vocabulary name: SNOMED-CT, RxNorm, LOINC, ICD-10-CM, ICD-9-CM, CPT-4, HCPCS, CVX, etc.
display_name	when known	Human-readable label for the primary code
all_codings_json	when coded	JSON array of every coding on the entry, including translations
value	for labs/vitals	Numeric or text value
unit	for labs/vitals	Units (UCUM strings, usually)
status	when recorded	active, completed, final, etc.
text	varies	Free-text or display name fallback

Provenance

Column	Required	Notes
source_file	yes	Relative path to the source XML file
raw_record_json	yes	Compact JSON of the parsed entry, for reviewer drill-down

Categories

Category	Source section LOINC	Typical content
patient	-	One header row per (patient, document) pair
problems	11450-4	Diagnoses, problem list entries
medications	10160-0	Prescribed and self-reported medications
procedures	47519-4	Surgical and medical procedures
labs_vitals	30954-2, 8716-3	Lab results and vital signs
allergies	48765-2	Drug, food, environmental allergies
immunizations	11369-6	Vaccines administered
careplans	18776-5	Plan of care narratives
diagnostic_reports	18748-4	Imaging and other diagnostic reports
encounters	46239-0, 46240-8	Visits, admissions
goals	61146-7	Patient and care-team goals
notes	29762-2, 10157-6	Social and family history; fallback for narrative-only sections
observations	(other)	Sections we don't recognize map here

Example rows

A condensed look at what real rows in the master CSV contain (synthetic data):

last_name | first_name | dob        | patient_id    | category    | source | effective_date | code      | code_system | display_name        | value | unit
Patient   | Test       | 1965-03-15 | PT-1D2DF76565 | patient     |        | 1965-03-15     |           |             | Test Patient        |       |
Patient   | Test       | 1965-03-15 | PT-1D2DF76565 | problems    | ccda   | 2023-01-01     | 55505003  | SNOMED-CT   | Motor neuron disease|       |
Patient   | Test       | 1965-03-15 | PT-1D2DF76565 | medications | ccda   | 2023-02-01     | 9468      | RxNorm      | Riluzole 50 MG ...  |       |
Patient   | Test       | 1965-03-15 | PT-1D2DF76565 | labs_vitals | ccda   | 2024-06-01     | 2160-0    | LOINC       | Creatinine, serum   | 0.9   | mg/dL

Reading the CSV

Python / pandas

import pandas as pd
df = pd.read_csv('patient_master.csv', encoding='utf-8-sig', dtype=str)

# Counts per category per patient
df.groupby(['patient_id','category']).size().unstack(fill_value=0)

# All medications, sorted by date, for one patient
df.query("patient_id == 'PT-1D2DF76565' and category == 'medications'") \
  .sort_values('effective_date')[['effective_date','code','display_name','text']]

# All LOINC labs across the cohort with numeric values
labs = df[df['category'] == 'labs_vitals'].copy()
labs['value_num'] = pd.to_numeric(labs['value'], errors='coerce')

R

df <- read.csv('patient_master.csv', encoding='UTF-8', stringsAsFactors=FALSE)
table(df$category, df$source)

Excel

The file is UTF-8 with BOM, so it opens cleanly in Excel without an encoding prompt. Add filters via Data → Filter and pivot on patient_id or category.

Compatibility with Registry Forge

Because the schemas match, you can:

• Concatenate a Patient Edition patient_master.csv with a Registry Forge patient_master.csv in pandas (pd.concat).
• Feed our output into the Registry Forge OMOP ETL - it accepts the master CSV as input.
• Use any analytic notebook built for Registry Forge against Patient Edition outputs and vice versa.

The only practical difference is the source column distribution: Registry Forge produces a mix of ccda and fhir rows; Patient Edition produces only ccda. Any downstream code that filters or groups on source will work but find fewer source types.