Clinical Evidence Prompt Non Small Cell Lung Cancer Dataset

Clinical Evidence Prompt-Completion Pairs for Atezolizumab in NSCLC (IMpower010) AI LLM Dataset

Overview

This dataset contains structured prompt-completion pairs derived from the IMpower010 trial and related CADTH clinical assessments of atezolizumab for non-small cell lung cancer (NSCLC) in the adjuvant setting. It is designed to support AI and LLM (Large Language Model) training, evaluation, and deployment in healthcare, particularly in the oncology decision-support domain.

Each prompt represents a clinically relevant question or query, and each completion provides an evidence-based answer, often directly quoting or summarizing findings from clinical reports, regulatory reviews, or published trial data.

📂 Dataset Format

The dataset follows the standard JSON format for prompt-completion training:


    "prompt": "What was the 3-year DFS rate for the atezolizumab group?",
    "completion": "73.79%"
  },
  {
    "prompt": "Which chemotherapy agents were allowed before randomization in IMpower010?",
    "completion": "Cisplatin plus vinorelbine, docetaxel, gemcitabine, or pemetrexed"
  },
  {
    "prompt": "What percentage of patients receiving atezolizumab had at least one adverse event?",
    "completion": "94.7%"
  },
  {
    "prompt": "What were the most common grade 3 or 4 AEs in the atezolizumab group?",
    "completion": "Increased alanine aminotransferase and abnormal hepatic function"
  },
  {
    "prompt": "What percentage of patients discontinued atezolizumab due to adverse events?",
    "completion": "18.6%"
  },
  {
    "prompt": "Was the IMpower010 trial powered for the Health Canada PD-L1 ≥50% subgroup?",
    "completion": "No"
  },
  {
    "prompt": "What was the main reason Health Canada limited approval to PD-L1 ≥50% patients?",
    "completion": "Uncertainty of clinical benefit in patients with PD-L1 1%-49%"
  },
  {
    "prompt": "What was the most frequently reported serious adverse event with atezolizumab?",
    "completion": "Pyrexia"
  },
  {
    "prompt": "What was the difference in 3-year OS rate between atezolizumab and BSC groups?",
    "completion": "14.27%"
  },
  {
    "prompt": "Which histological subtype was most common in the nonsquamous group?",
    "completion": "Adenocarcinoma"
  },
  {
    "prompt": "What immune-mediated reaction occurred in 14.2% of atezolizumab-treated patients?",
    "completion": "Hypothyroidism"
  },
  {
    "prompt": "What assay was used to assess PD-L1 expression in IMpower010?",
    "completion": "SP263 IHC assay"
  },


  {
    "prompt": "What is the main limitation in interpreting DFS results from the IMpower010 trial for patients with PD-L1 ≥50%?",
    "completion": "The DFS analysis was not included in the statistical hierarchy and lacked multiplicity control, leading to uncertainty in interpreting its statistical significance."
  },
  {
    "prompt": "Why is the efficacy of atezolizumab on overall survival (OS) inconclusive in the IMpower010 trial?",
    "completion": "The OS data were immature at the time of the planned interim analysis, preventing definitive conclusions about the survival benefit."
  },
  {
    "prompt": "Did the IMpower010 trial support clinical benefit of adjuvant atezolizumab despite statistical limitations?",
    "completion": "Yes, clinical experts considered the DFS results to be clinically meaningful and favorable, particularly in patients with PD-L1 ≥50% expression."
  },
  {
    "prompt": "Are the IMpower010 trial results generalizable to real-world clinical settings?",
    "completion": "Yes, clinical experts judged the trial population and outcomes to be generally reflective of real-world NSCLC patients."
  },
  {
    "prompt": "What factors contribute to uncertainty in the efficacy results of the IMpower010 trial?",
    "completion": "Lack of multiplicity adjustment, absence from the statistical testing hierarchy, and immature OS data."
  },

  {
    "prompt": "What is the primary treatment goal for patients with stage IB to IIIA NSCLC?",
    "completion": "The primary goal is to cure and prolong life."
  },
  {
    "prompt": "What is the secondary goal of treating stage IB to IIIA NSCLC?",
    "completion": "To delay disease relapse and prolong disease-free survival."
  },
  {
    "prompt": "What is the gold standard treatment for anatomically resectable NSCLC?",
    "completion": "Surgical resection is the gold standard of care."
  },
  {
    "prompt": "What treatment is used for stage I NSCLC patients who cannot undergo surgery?",
    "completion": "Stereotactic ablative radiation with curative intent."
  },
  {
    "prompt": "When is adjuvant cisplatin-based chemotherapy considered after NSCLC resection?",
    "completion": "For tumors ≥ 4 cm or with lymph node involvement."
  },
  {
    "prompt": "What chemotherapy agents are typically used in adjuvant treatment for resected NSCLC in Canada?",
    "completion": "Cisplatin plus vinorelbine or cisplatin plus pemetrexed."
  },
  {
    "prompt": "What did the LACE meta-analysis show about adjuvant chemotherapy?",
    "completion": "It showed a 5% absolute OS benefit at 5 years (HR = 0.89) and 5.8% DFS benefit at 3 and 5 years."
  },
  {
    "prompt": "What is osimertinib's role in resected NSCLC with EGFR mutations?",
    "completion": "Osimertinib is approved for 3 years after adjuvant chemotherapy, showing DFS benefit in the ADAURA trial."
  },
  {
    "prompt": "What percentage of patients received adjuvant chemotherapy in the European cohort study?",
    "completion": "48% overall; 15.1% in stage IB, 55.1% in stage II, and 71.4% in stage IIIA."
  },
  {
    "prompt": "What are common reasons for not receiving adjuvant chemotherapy after resection?",
    "completion": "Patient refusal, comorbidities, surgical complications, and poor performance status."

Purpose and Use Cases

This dataset serves multiple strategic objectives:

Application Area Description LLM Fine-tuning Adapt general-purpose models for oncology Q&A or guideline summarization. Healthcare QA Systems Power intelligent search or chatbot tools for clinical and regulatory staff. Medical Education and Assessment Train or evaluate LLMs for oncology board prep, CME, or health literacy. Clinical Decision Support Systems Enable EHR-integrated AI tools to assist oncologists with treatment guidance. RAG (Retrieval-Augmented Generation) Use as ground-truth completions in combination with vector search models.

Why This Dataset Matters

Oncology is a high-impact AI use case: Cancer treatment requires nuanced interpretation of clinical trial data, biomarkers, and regulatory guidance. This dataset enables structured learning in this domain.
Evidence-grounded: All completions are rooted in CADTH reviews, Health Canada decisions, or peer-reviewed data, reducing hallucinations in clinical AI applications.
High-value therapeutic area: Atezolizumab is a leading immunotherapy in NSCLC, and understanding its indication, recurrence patterns, and safety is critical to guiding treatment decisions.
Gap in general LLMs: Most open-source or general-purpose LLMs lack precision when asked about oncology-specific clinical nuances—this dataset bridges that gap.

Coverage

The dataset includes prompt-completion entries covering:

Indication population characteristics (stage, PD-L1 status, mutations)
DFS and OS findings in the IMpower010 study
Risk of recurrence and recurrence site breakdown
Comparative insights with osimertinib for EGFR-mutated patients
Chemotherapy regimens used in the study
Harms and adverse events
Generalizability of findings to Canadian practice
Importance and interpretation of surrogate endpoints (e.g., DFS)

Use Case Example Tool Example Outcome Fine-tuning GPT models OpenAI, HuggingFace Train a cancer-specific LLM assistant Retrieval QA System LangChain + Vector DB Build a medical chatbot for clinicians QA Benchmarking Eval frameworks Evaluate accuracy of different LLMs Educational Applications Flashcard apps Use for quiz-based training or medical education

Use Case

Fine-tuning GPT models

Train a cancer-specific LLM assistant

LangChain + Vector DB

Build a medical chatbot for clinicians

QA Benchmarking

Eval frameworks

Evaluate accuracy of different LLMs

Educational Applications

Flashcard apps

Use for quiz-based training or medical educatio