Dossier bot for automated value dossier generation

About the client

Our client is MArS—a leading consulting agency for healthcare companies in Central and Western Europe. One of its jobs is supporting pharmaceuticals in their reimbursement submissions. To this end, the agency distills elaborate reports, often over 400 pages long each, into summarized drug dossiers for healthcare systems. HW.Tech supported the agency in the development of a time- and cost-saving dossier creation solution based on an AI-powered platform for text summarization—DO-BO®.

Problem statement

A pharmaceutical company usually creates up to 10 dossiers annually. Due to the sheer volume of raw data and the number of tables medical writers must analyze, a single dossier requires many man-months and hundreds of thousands of dollars of investment.

• Writers must sift through vast amounts of data to find relevant information often scattered across clinical study reports and additional data packages, such as those from systematic literature searches.
• They need expertise in the health systems of different jurisdictions, as each system requires a country-specific submission, including documentation in the local language.
• Completing a drug dossier can take up to a year in some cases.

Challenge

A drug reimbursement dossier comprises four to five modules tailored to specific requirements. The writer’s key challenge is finding, screening, and summarizing complex medical data dispersed across lengthy reports. The limitations of existing AI tools hinder the ability to automate or simplify this complex process. They have trouble capturing the clinical report content, including embedded images and tables, often leading to factual errors and information omissions. Furthermore, the use of most platforms is restricted by confidentiality requirements, as sensitive information cannot be uploaded.

• Data complexity. Report summarization requires manual extraction and review due to vast amounts of unstructured digital texts.
• Complex terminology. Medical jargon must be simplified and correctly translated so regulatory agencies can understand the report.
• Format compatibility. Managers must verify all information imported from PDFs, images, and TLFs (tables, listings, and figures).
• Requirement compliance. Reimbursement dossiers have to align with varying international jurisdictions and specific template structures.
• Factual accuracy. AI summaries might have hidden inaccuracies, making the drug reimbursement dossier ineligible for reimbursement approval.
• Contextual misinterpretation. Given the intricacy of clinical reports, untrained AI might misread the trial goals or methodology.
• Data omission. Most AI tools struggle to prioritize information and often ignore important facts and details in text summaries.

Solution

We started the project as a Proof of Concept (PoC) for an intuitive platform that integrates AI technologies to speed up the creation of drug reimbursement dossiers. The model simplifies the search, extraction, summarization, and formatting of medical data. Additionally, it incorporates a machine-learning algorithm that learns from the processed documents and improves output quality over time.

• Dossier template. The platform allows identifying sections of the dossier that summarize, reword, or duplicate information.
• Self-learning AI model. Our LLM retrains every 24 hours based on user searches, edits, and manual input.
• Intelligent search. Allows users to locate information in text, graphs, images, and tables based on various search criteria.
• Table formatting. Organizes data into structured tables by aligning column names in source documents based on search queries.
• Image extraction. OCR (optical character recognition) organizes information extracted from documents into tables and graphs.
• Contextual understanding. Semantic analysis and natural language processing help simplify complex terms into plain language.
• Accurate referencing. The extracted information is cross-referenced with the source document, which helps verify its accuracy.
• Collaboration tools. Multiple users can review the summarization results and easily adjust the search criteria for fine-tuning.

Result - DO-BO®

The prototype streamlines time-consuming aspects of dossier creation, such as searching, extracting, rewording, and formatting. DO-BO®, the AI-powered platform, helps generate accurate documents aligning with international standards and payer requirements much faster. The current focus is to expand the platform’s capabilities with each new version, with the ultimate goal of fully automating the process of creating drug reimbursement dossiers.

• The solution streamlines dossier creation and accelerates analysis by up to 60%.
• The AI automation tool greatly cuts the dossier creation costs.
• Semantic analysis tools transform jargon-heavy text into plain, straightforward summaries translated into multiple languages.
• The platform improves document quality by eliminating human error, including typos, incorrect terminology, and misformatting.
• Robust authorization mechanisms, role-based access, and encrypted communications protect confidential information.
• The platform’s intuitive interface and customizable structure make dossier creation easy for users with diverse technical skills.