Artificial Intelligence Technology: How to Most Effectively and Efficiently Mitigate Risk When Meeting Data Transparency Requirements

May 20, 2016The Clinical Informatics News Best Practices Awards program recognizes outstanding examples of applied strategic innovation—partnerships, deployments, and collaborations that manifestly improve the clinical trial process. This year, a team of expert judges evaluated submissions to the program and announced ten finalists. Three grand prize winners were named at the 2016 Summit for Clinical Ops Executives event in Miami, Florida. We have invited the 2016 finalists to present the projects that so impressed our judges in hopes that these best practices can make wide-reaching impacts on the industry.   

--Allison Proffitt, Editor

By Kelley Kendle

New European and U.S. clinical trial data disclosure initiatives – such as European Medicines Agency (EMA) Policy 70, which went into effect early in 2015 – have created additional compliance requirements for biopharmaceutical companies. As part of Policy 70 and the related Policy 43, sponsors are required to make publicly available their clinical study information after redacting all patient-protected data (PPD) and company-confidential information (CCI) from their clinical study reports (CSRs) and summary-level submission documents. Sponsors risk potential litigation should any confidential information be exposed. As each CSR can be 5,000-30,000 pages long with 600 words per page, they can total 75-150 million words apiece. Redacting documents of that size accurately and efficiently is a monumental task laden with risk.

Ensuring Precise Redaction of CSRs

A top-10 pharmaceutical company committed to publishing CSRs for clinical outcome trials for all its marketed medicines, dating back to the formation of the company at its last merger. As a result, it needed to prepare 2,000+ reports for disclosure.

Before those reports could be made publicly available, all PPD and CCI had to be redacted. The sponsor developed guidances governing the redaction of those data within its reports. However, the volume of work was more than it could handle internally.

Initially, the sponsor engaged a contractor who offered a manual solution to this large-scale redaction project. But in its first year, the sponsor’s contractor had redacted only five reports. The sponsor also found it difficult to assess the accuracy of the contractor’s PPD redaction, as there was no mechanism that explained the rationale for text removal. As a result of those slow turnaround times and poor quality, the sponsor’s transparency team concluded that a manual process was not a viable solution. The sponsor also had zero tolerance for accidental exposure of PPD as it could lead to litigation.

Consequently, the sponsor decided to investigate Synchrogenix’s artificial intelligence- (AI-) enabled redaction solution and consultancy services. Synchrogenix’s unique AI engine is built on natural language processing and recognition. As such, the engine can identify individual words, parts of speech, word groupings, and phrasing combinations automatically and, therefore, determine context. It is the industry’s only AI-assisted technology platform developed for regulatory writing. This engine can be configured to identify, PPD and CCI automatically, the same way a human would be trained to identify them. The process of identifying and redacting sensitive information therefore becomes automated and significantly more accurate.

Working in tandem with this technology is a team of domain-specific experts. Synchrogenix’s redaction team performs a thorough quality control (QC) review of every document, ensuring the highest level of accuracy. Findings and the resolution of findings are documented on a quality review checklist and then integrated into the system and used to refine the tool’s accuracy in real time.

Within three months, Synchrogenix completed the sponsor’s on-boarding – configured the AI system to follow the sponsor’s specific set of rules, trained a dedicated review team to understand and review against those rules, and redacted several test reports prior to going live. Once live, Synchrogenix released 20-30 units (approximately 30 CSRs) per week to meet the 2,000+ report volume requirement within two years.

Three-quarters of the way into the project, the sponsor determined that its CSRs contained an often copyrighted document: the clinical outcome assessment (COA). The sponsor did not hold the COA copyright and, therefore, had to ensure retroactively that this information was removed from its CSRs. Synchrogenix was able to configure its AI system to identify and redact COAs, using a database of copyrighted COAs that it developed. Within five months, Synchrogenix retroactively reviewed all of the CSRs that had already been released in the prior nine months and prospectively implemented the new COA redaction process in all CSRs going forward.

Partnering with Synchrogenix, the sponsor progressed from a run rate of five redacted reports per year to almost 1,500 reports per year, at a quality of 99.99 percent non-PPD disclosure accuracy.

Managing Freedom of Information Requests Efficiently

In addition to providing qualified scientific researchers with access to anonymized patient data and full CSRs from its clinical trials, another top-10 pharmaceutical company, had to manage multiple EMA Policy 43 Freedom of Information (FOI) requests with very short notice.

As a result, the sponsor required multiple document types to be made available to the public, each requiring redaction following separate business rules. This requirement made a fully manual effort nearly impossible.

By employing Synchrogenix’s unique AI engine, and its team of domain-specific experts, the sponsor obtained an accurate product across seven document types.

Redaction of several of the sponsor’s document types was driven by its need to comply with EMA Policy 43. Through this FOI policy, researchers can request any documents used by the EMA in its drug approval decision-making process. Once a request is deemed appropriate, the EMA redacts the requested document. The sponsor found some inconsistencies in the way that the regulators redacted its data and was able to use Synchrogenix-generated documents to cross check and correct PPD and CCI redactions.

Synchrogenix managed each Policy 43 redaction request within 48 to 96 hours, depending on the document size, and provided more thorough feedback to the EMA regarding additional suggested redactions for its documents.

The increased recognition of the importance of providing clinical trial results to both the public and scientific community has spurred the growth of transparency and disclosure mandates. Synchrogenix’s AI solution enables companies to satisfy these mandates while minimizing their risk of disclosing confidential information.

Kelley Kendle is president of Synchrogenix, Certara’s regulatory and medical consultancy.