File size: 4,165 Bytes
4766213
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d5f425b
4766213
 
 
 
 
 
 
 
 
 
 
 
fdd6554
4766213
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
---
license: apache-2.0
language:
- en
tags:
- medical
- biomedicine
- clinical-trials
- large-language-models
- supervised-finetuning
- evidence-based-medicine
- trial-design
- systematic-review

config_names:
- study_search
- study_screening
- evidence_summarization
- trial_design
- sample_size_estimation
- trial_completion_assessment

configs:
- config_name: study_search
  data_files:
  - split: train
    path: sft_study_search_data_cleaned.parquet

- config_name: study_screening
  data_files:
  - split: train
    path: sft_study_screening_data.parquet

- config_name: evidence_summarization
  data_files:
  - split: train
    path: sft_evidence_summarization_data.parquet

- config_name: trial_design
  data_files:
  - split: train
    path: sft_design_data.parquet

- config_name: sample_size_estimation
  data_files:
  - split: train
    path: sft_sample_size_data.parquet

- config_name: trial_completion_assessment
  data_files:
  - split: train
    path: sft_trial_completion_assessment_data.parquet
---
# TrialPanorama: Supervised Fine-Tuning Data for Clinical Research LLMs

## Dataset Summary

**TrialPanorama SFT** is a large-scale, task-oriented **supervised fine-tuning (SFT) dataset** designed to train large language models for **end-to-end clinical research and trial development workflows**.

The dataset is derived from **TrialPanorama**, a structured clinical research resource aggregating **1.6M+ clinical trial records** across global registries and linking them with biomedical ontologies and supporting literature. It focuses on transforming raw clinical trial data and curated evidence into **instruction–response pairs** that reflect realistic, expert-level research tasks.

The dataset supports training LLMs to operate as **clinical research assistants** capable of systematic literature review, trial design reasoning, and evidence-based decision making.

---

## Supported Training Tasks

Each task is released as a separate dataset configuration.

| Task (config) | Description |
|---|---|
| `study_search` | Given a clinical research question, retrieve and justify relevant studies from large trial and literature corpora. |
| `study_screening` | Perform inclusion/exclusion decisions for candidate studies based on eligibility criteria and study metadata. |
| `evidence_summarization` | Synthesize structured and unstructured trial evidence into concise, faithful summaries. |
| `trial_design` | Generate or refine clinical trial designs, including arms, interventions, and eligibility criteria. |
| `sample_size_estimation` | Estimate appropriate sample sizes under specified statistical and design assumptions. |
| `trial_completion_assessment` | Assess trial completion likelihood and rationalize risks using trial design and historical evidence. |

---

## Data Characteristics

- **Instruction-following format** suitable for SFT
- Grounded in **real clinical trial records**
- Emphasizes **clinical reasoning**, not surface text generation
- Covers both **systematic review** and **trial design & optimization** tasks
- Designed to support **generalist and agentic LLM training**

All files are provided in **Apache Parquet** format.

---

## Typical Fields

Each record may include:
- Task-specific **instruction or prompt**
- Structured **context** (trial metadata, eligibility criteria, outcomes, phase)
- **Model response targets** written or validated by domain experts
- Task and difficulty metadata

Exact schemas vary by task.

---

## Intended Use

This dataset is intended for:
- **Supervised fine-tuning of LLMs** for clinical research tasks
- Training **research-oriented AI agents** for trial planning and evidence synthesis
- Building domain-adapted models for **systematic review automation**
- Academic benchmarking of clinical reasoning capabilities

### Not Intended For

- Model evaluation (see TrialPanorama benchmarks for evaluation)
- Clinical decision making
- Direct medical or regulatory use

---

## How to Load

Load a specific training task via its configuration name:

```python
from datasets import load_dataset

ds = load_dataset("TrialPanorama/dataset", "study_screening")
ds["train"]