OneEHR
EHR AI platform MedX-PKU/OneEHR

Data Model

OneEHR reads plain CSV files. The core input is a longitudinal event table; static covariates and labels are separate tables keyed by patient.

Overview

Table Required Purpose
dynamic.csv Yes Longitudinal event table in long format
static.csv No Patient-level covariates such as demographics or baseline features
label.csv No Label events for one or more prediction tasks 
[dataset]
dynamic = "data/dynamic.csv"
static = "data/static.csv"
label = "data/label.csv"

dynamic.csv

dynamic.csv is one row per observed event.

Column Type Description
patient_id string Patient identifier
event_time datetime Timestamp parseable by pandas.to_datetime
code string Measurement, diagnosis, procedure, medication, or feature name
value numeric or string Observed value

Example:

patient_id,event_time,code,value
P001,2023-01-01 08:00,heart_rate,72
P001,2023-01-01 08:00,blood_pressure_sys,120
P001,2023-01-01 08:00,diagnosis,A01
P001,2023-01-02 10:00,heart_rate,80
P002,2023-01-01 09:30,heart_rate,68
P002,2023-01-01 09:30,lab_glucose,5.4

Preprocessing bins events into fixed time windows such as 1h, 6h, or 1d. Numeric event values are aggregated with preprocess.numeric_strategy. Categorical values are encoded with preprocess.categorical_strategy.

static.csv

static.csv has one row per patient.

Column Type Description
patient_id string Patient identifier matching dynamic.csv
Other columns numeric or string Static covariates

Example:

patient_id,age,sex,insurance
P001,65,M,Medicare
P002,42,F,Private
P003,78,M,Medicaid

Numeric static columns become num__* features. Categorical static columns become cat__*__* one-hot features. Models with static branches receive these features as a separate tensor; tabular models receive them in the flattened feature matrix.

label.csv

label.csv is a long-format label table. A single file can contain multiple label types through label_code.

Column Type Description
patient_id string Patient identifier
label_time datetime When the label is observed
label_code string Label type, for example outcome or los
label_value numeric Label value

Example:

patient_id,label_time,label_code,label_value
P001,2023-01-05,outcome,1
P001,2023-01-05,los,4.5
P002,2023-01-03,outcome,0
P002,2023-01-03,los,2.0

Patient-level tasks use one label per patient. Time-level tasks align labels to binned time windows.

Flow Through A Run

dynamic.csv --+
              +--> preprocess --> binned features + labels --> train --> test --> analyze
static.csv  --+
label.csv   --+

Downstream commands read the saved run artifacts instead of reading the raw CSV files again. This keeps training, testing, analysis, and plotting tied to the same preprocessed data and split.