Synthetic Longitudinal Dataset Generation for Clinical Trial Privacy

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Clinical trial data from Electronic Data Capture (EDC) systems pose privacy concerns due to their sensitive nature, limiting accessibility and hindering research, as conventional de-identification and anonymization techniques are inadequate for ensuring patient privacy, especially with smaller dataset sizes.

Innovation Solution

A method to generate high-fidelity synthetic longitudinal EDC data by embedding feature vectors in a lower dimension space, identifying nearest neighbors, and swapping subsequence patterns, ensuring statistical similarity while maintaining privacy, thus allowing researchers to analyze data without exposing individual patient information.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If conventional de-identification and anonymization techniques are applied to EDC data, then data accessibility is improved, but patient privacy is compromised

Engineering Contradiction:
Improvedata accessibilityVSAvoidpatient privacy risk
Core Design Contradiction:
Ease of operationVSObject-affected harmful factors

Solution Approach 1:

The patent generates synthetic copies of clinical trial data that replicate the statistical properties and patterns of original EDC data without containing actual patient information. This copying approach allows researchers to access and analyze data while patient privacy is preserved, as the synthetic data cannot be traced back to individual patients

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The patent introduces synthetic data as an intermediary between the original sensitive EDC data and research applications. This intermediary layer enables data utilization while maintaining privacy barriers, allowing researchers to work with data that mimics the structure and characteristics of real clinical trial data without direct access to sensitive patient information

Inventive Principle:
Principle #24Intermediary (Mediator)

2Object-affected harmful factors

If deep learning techniques like GANs and VAEs are used to generate synthetic data, then privacy control is improved, but data fidelity deteriorates due to small dataset sizes

Engineering Contradiction:
Improveprivacy controlVSAvoiddata fidelity
Core Design Contradiction:
Object-affected harmful factorsVSManufacturing precision

Solution Approach 1:

The patent changes the approach from deep learning parameter optimization to statistical parameter preservation. It focuses on maintaining key statistical properties (distribution characteristics, event frequencies, temporal patterns) of the original data while generating synthetic samples, which is particularly effective for small clinical trial datasets where deep learning would overfit or fail to learn meaningful patterns

Inventive Principle:
Principle #35Parameter changes

3Productivity

If EDC data is made broadly available for research, then research value is improved, but regulatory and privacy compliance becomes more difficult

Engineering Contradiction:
Improveresearch valueVSAvoidregulatory compliance complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent creates synthetic copies of EDC data that can be freely distributed and used for research purposes without regulatory restrictions. These synthetic datasets maintain the scientific value and statistical properties needed for research while eliminating regulatory and privacy compliance barriers, as they contain no actual patient information

Inventive Principle:
Principle #26Copying

Data Source

PatentEP4447058A1System and method for generating a synthetic longitudinal dataset from an original dataset
Publication Date: 2024.10.16 MEDIDATA SOLUTIONS INC
  • EP4447058A1 patent drawingFigure 1
  • EP4447058A1 patent drawingFigure 2
  • EP4447058A1 patent drawingFigure 3A~3B

AI summary

Generating a synthetic longitudinal dataset includes identifying subsequence patterns in records defining event sequences for patients. Feature vectors are determined, each characterizing a corresponding one of the records, based on the subsequence patterns. The feature vectors are embedded in a lower dimension space. A seed record is iteratively selected from among the records and in each iteration: subsequence patterns are identified in a subset of the records. Instances of subsequence patterns in the seed record are replaced with instances of similar subsequence patterns identified in the subset of the records to form a modified seed record. The iterations are repeated until all of the records have been selected as the seed record. The modified seed records are combined to form the synthetic dataset.