mlazniewski/trialbench-combined

    # TrialBench: Clinical Trial Outcome Prediction Dataset

TrialBench is a curated dataset collection designed to support machine learning research on clinical trial outcome prediction. It includes multiple tasks relevant to the analysis of trial success, safety, and patient behavior, extracted and preprocessed from publicly available clinical trial data.

Dataset Structure

This repository contains multiple configurations, each corresponding to a specific prediction task and clinical trial phase. The tasks include:

• trialbench-mortality: Predict mortality rate and binary mortality outcome across Phases 1–4.
• trialbench-adverse-events: Predict serious adverse event rates and binary event flags across Phases 1–4.
• trialbench-patient-dropout: Predict patient dropout rates and binary dropout labels across Phases 1–4.

Each configuration includes a single DataFrame converted into Hugging Face DatasetDict with a train split, containing the following columns:

Column Name	Description
NCT_ID	ClinicalTrials.gov registry identifier
*_rate	Outcome rate (e.g. mortality_rate, droupout_rate, etc.)
label	Binary classification label derived from rate thresholds
phase	Phase of the clinical trial (Phase1, Phase2, ...)

Usage

from datasets import load_dataset

# Load a specific configuration
ds = load_dataset("mlazniewski/trialbench-combined", name="trialbench-mortality")
df = ds["train"].to_pandas()