AI Agents for TLF Generation

Tables, Listings, and Figures (TLFs) are ways to analyze and summarize datasets of a clinical study into an easily readable format; they are created with the help of statistical programmmers and biostatisticians from the designated departments. After validating with the study statistician, medical writers use TLFs to create documents like clinical study reports (CSRs). Some TLFs are straightforward and easy to interpret, while others may require writing an extensive discussion to explain them. In addition to their value in CSRs, TLFs are used to answer regulatory questions and support publications based on the clinical data contained in them.

Creating TLFs is a tiresome process that involves frequent quality checks to verify the validity of the included information. Not only that, medical writers could request changes in the output, which requires additional time to rewrite the code that produced the TLF documents. This means new rounds of validation and quality checks to confirm the correctness of the new outcome.

In addition, medical writers spend significant amounts of time flipping back and forth between TLFs to interpret their meanings, confirm their correctness, and ensure consistency. All these processes lead to delays in delivering the final CSR to health authorities. The outcome is a considerable lag phase in providing life-saving medications and treatments to those needing them the most.

Statistical Analysis System (SAS) is a piece of software that is widely used to compile files and documents submitted to regulatory authorities. Not to be confused with SaaS (or software-as-a-service), people involved in clinical trial reporting will often refer to SAS as the “gold standard” in manipulating clinical study datasets. It has gotten to the point where there is a general misconception that regulators strictly require the use of SAS for clinical trial reporting, but this is not the case. More specifically, SAS is widely used for statistically processing clinical datasets to create Tables, Listings, and Figures (TLFs).

Regardless if you’re using SAS or R, datasets need to be extracted and structured in various formats. In addition, the outcome needs to be formatted with respect to font, dimension, and other aspects to fit a template.

With SAS, a combination of statements, like PROC MEANS/PROC SUMMARY/PROC UNIVARIATE, are used to summarize datasets into descriptive statistics, such as mean, median, sum, and other metrics like standard deviation. In addition, prompts such as PROC GENMOD/PROC MIXED are used to define the order in which data populate and display statistical measures for longitudinal data (information related to the same subject across various time points). Finally, data are fed into the PROC REPORT prompt to produce the reports. To extract the outcome in an RTF format, extra styling commands need to be added.

Using R, the same job can be done via multiple pathways. For example, a code that calculates each summary can be written and then arranged to get descriptive statistics. Or, source data could be rearranged using numeric and character variables for a time point; then, results are wrapped with dedicated functions.

In addition to the TLFs solution, Narrativa offers other tools that support clinical research and regulatory documentation and submission professionals. The redaction and anonymization solution is an example of these tools. This solution scans thousands of documents within few minutes to anonymize and redact critical information that could compromise the privacy of trial participants.

There are several bottleneck processes at different stages that cause delays in creating TLFs. For example, creating TLFs requires validating the output by two different statistical programmers. As reassuring as it sounds to have two people validate one output, it is actually a cause for concern. This is because, statistically speaking, they may end up making the same mistake and further delay output creation.

Other bottleneck processes arise from frequent quality checks needed to verify the data included in TLFs. This could mean requests from medical writers to change certain information or how they are presented. The outcome is further delays in CSR submission, as validation and quality checks must be done again. Eventually, this translates into significant delays in creating CSRs and presenting them to health authorities.

Once you access the platform, you will be given multiple options to create TLFs according to your preferences. All you need is to select the inputs you would like to have in the outcome. Also, you can choose the output in several available formats, like rich text format (RTF) or a Microsoft Office Word file.