p-plp-youri 0.0.2

A python alternative of the OHDSI PLP framework

P-PLP

P-PLP is a Python package for patient-level prediction on OMOP CDM data.

It helps you:

• connect to PostgreSQL or DuckDB / Eunomia-style OMOP datasets
• load target and outcome cohorts from Atlas SQL
• generate labels for a time-at-risk window
• build feature datasets
• train and evaluate prediction models

Installation

Install from PyPI:

pip install p_plp

Install locally for development:

pip install -e .[dev]

Quick example

from p_plp.db import get_engine
from p_plp.cohorts import load_atlas_cohort_to_work_table, generate_labels_time_at_risk
from p_plp.feature_engineering import create_covariate_settings, run_feature_query
from p_plp.modeling import train_pipeline, evaluate

engine = get_engine(
    source_name="postgres",
    database_url="postgresql+psycopg2://user:password@localhost:5432/db",
    cdm_schema="cdm",
    work_schema="plp_work",
)

load_atlas_cohort_to_work_table(
    engine,
    sql=target_sql,
    cohort_definition_id=1,
    table_name="target_cohort",
)

load_atlas_cohort_to_work_table(
    engine,
    sql=outcome_sql,
    cohort_definition_id=2,
    table_name="outcome_cohort",
)

labels_df = generate_labels_time_at_risk(engine, risk_start_days=1, risk_end_days=365)

feature_config, base_config = create_covariate_settings(
    engine,
    useDemographicsAge=True,
    useDemographicsGender=True,
    useConditionEraAnyTimePrior=True,
)

dataset_df = run_feature_query(engine, feature_config, base_config)
model, X_test, y_test = train_pipeline(dataset_df, model_name="logreg")
metrics = evaluate(model, X_test, y_test)

More

• Docs: docs/
• Usage guide: docs/usage.md
• API reference: docs/api.md
• Notebooks and examples: notebooks/