NLP: Trilodocs – Artificial intelligence for clinical documentation

Clinical Study Reports (CSRs)

As a specialist in clinical regulatory documentation, Trilogy offers its partners and customers a service of proactive planning, coordination, and ultimately creation of required documents to support review work and thus shorten the required timeline in approval processes.

Clinical study reports, or CSRs, represent an important building block in the drug approval process. After the completion of a clinical trial, this document is prepared based on the previously generated documents, to be finally submitted to the relevant authorities.

The documents required for preparation include the clinical study protocol (CSP), the statistical analysis plan (SAP), and a large set of study-dependent tables (sometimes graphs), brought together in a TFL document.

The protocol is the “cookbook” for the clinical trial. It contains, for example, information about the study design, inclusion, and exclusion criteria for study participants and when which dosage of which drug should be administered. The protocol is always prepared before a study begins.

The statistical analysis plan is also prepared upfront. It provides information on statistical methods to be used in the evaluation of the study.

The TFL document, in turn, is created at the end of the study. It contains tables with evaluations of all measured values recorded during the course of the study. This includes information on reported “adverse events” such as headaches or nausea, but also blood values of the study participants and much more.

Start of the project

A CSR can be divided into two parts: the methodological part and the results-oriented part. For the methodological part, the protocol and the statistical analysis plan are consulted. However, not every paragraph of the report can be traced back to either document. This stems from the close collaboration of the medical authors with study managers, team decisions, queries, and specifications obtained retrospectively.

The outcome-oriented part represents an evaluation of the TFL document. Namely, whether patient X had a mild headache two days after taking the new drug is not relevant. However, it is of very great importance for assessing whether the drug should be approved whether these headaches occurred statistically significantly more often in the experimental group that was administered the drug than in the comparison group that was administered a placebo.

Trilogy Dr. Barry Drees
Dr. Barry Drees, Senior Partner and Founder of Trilogy Writing & Consulting continuously develops the TriloDocs AI system with the expert team of L-One

At the beginning of the project, the methodological part of the CSR was highlighted: What information from the CSP and SAP should be incorporated into the CSR? How can grammatical tenses be adjusted? – These were the questions that L-One’s NLP team got to the bottom of.

Meanwhile,  L-One’s focus is on evaluation or the result-oriented part;  in addition to our Linguistic Data Scientists, we also have specialized developers in the team, to lead the initially strongly NLP-driven project from research to industrial application.

Collaboration with Trilogy Writing & Consulting

Like most NLP projects, the TriloDocs project began with a review of the data: It was necessary to trace the writing process of the medical authors and to ask questions. It became apparent that not all the information we needed was available in the documents provided. Individual decisions by the authors are sometimes based on years of professional experience, and intensive communication is required to meet the core of the customer’s request. Incidentally, this applies to all innovation projects that represent more than a clearly limited development task.

NLP Projekte fuer klinische Dokumentation @ L-One Systems
TriloDocs uses artificial intelligence and domain knowledge to define patterns and adaptively prepare data. This enables the creation of complete, clinically relevant reports. The NLP application has predefined styles that, adapted to customer-specific templates, follow a structure for section headings.

Today, there is a direct exchange with one of the senior partners and weekly remote meetings are held to talk through processes, jointly identify pitfalls, ask questions, and ensure that the common goal is being pursued.

»With the software programmers, we define the rules that should be followed when interpreting the data and then create standard texts that can be used depending on the results [...], which saves the author time and tedious work and shortens the time it takes to get a new drug approved.«

Dr. Barry Drees, Senior Partner of Trilogy Writing & Consulting

Innovative horizon

Although ICH guidelines have been providing directions since 1995, today, 26 years later, standardization and thus digitization in the area of clinical documentation has not progressed as far as one might assume.

As a result, medical authors often have to work their way manually through countless pages of tables, sometimes completing redundant subtasks that a computer could reliably complete in less time. Relieved of this task, the medical writer could devote their time to the exciting aspects of their work with a free mind – reasons for which they originally chose her profession.

TriloDocs will never make the work of medical writers redundant, but it will make it more enjoyable, speed it up, and lead to a higher level of accuracy.

More to read and more about L-One

› More about TriloDocs: The NLP reporting platform for medical editors.

› Big Data and AI solutions at L-One Systems: View our services.

› Why dealing with large amounts of data of written or spoken texts is becoming increasingly important for companies: IT development NLP – Natural Language Processing.

› Contact us: IT project in planning? This is how we can help.