8 modules for automating post market surveillance, using AI — PART II

This post is the sequel and final post on how artificial intelligence is automating post market surveillance. Read the first post.

D. Highlighting relevant information

As I’m sure you’re aware of, it’s really hard for humans to process large amounts of text. This module highlights relevant text in documents, based on your input in modules A-C. See example below.

Highlight relevant information in documents, speeding up post market surveillance

E. Document cluster summarization

There are two types of document summarizations in artificial intelligence: extractive and abstractive summarization. In the former instance, the machine copies and combines complete sentences from the original text (the easy solution), and in the latter it rewrites the original text into a short, accurate and fluent summary (the complex but better solution).

A document summarization machine enables researchers to quickly get the overview of what’s in multiple documents, and will also help in the further narrative writing.

F. Extraction of key data points

One of the more time-consuming and repetitive tasks in post market surveillance is extracting and contextualizing key data points from texts, tables and figures, for example adverse effects, from thousands of documents. 

The good news is that the repetitive nature of this process makes it easier for a machine to do — it can literally extract data from thousands of documents in a couple of minutes. 

In short, the module does the following:

  1. Extracts entities and data from text (anything with a name)
  2. Extract data from tables
  3. Links and matches all data to entities and the required output layout

Step number 1 and 2 are fairly straight forward, but the real magic and value lies in step 3. The machine links the data in text and tables to entities, and places the extracted data in the correct rows and columns of your results sheet.

As far as we know, we at Iris.ai are the first in the world who have achieved this.

Oh, and the Extract module also comes with confidence assessments so you know where to do manual spot checks of the results.

G. Surveillance, monitoring and alert in post market surveillance

There’s a constant stream of data on adverse effects, which it’s important to stay up to date on. Therefore, we recommend you to set up automated updates, which is easy to set up once you’ve run through the process described in this post.

H. Integrations with your tools or own UX

We all work in different ways, which is why you can integrate these modules with APIs for your software, receive the results simply in Excel/CSV, or visualize them in Iris.ai’s own interface. 

How to get started?

Your process is the starting point. We recommend you to ask yourself what parts of your post market surveillance processes are the most time consuming, repetitive and error-prone.  

If you decide to work with us at Iris.ai, our machine would quickly train on your domain. First, we’d ask for a description of your area of operation. Second, together we’d create a simple output data layout for the data extraction, based on how you want the output to look like (headers, etc.). Finally, we’d sit down with you after the first results to validate the output quality.

Consider your provider

There are a plethora of providers “out there”, and we urge you to carefully choose a provider that supports your specific post market surveillance challenges. Here are a couple of questions to keep in mind:

  • What are the training requirements to get started?
  • Do you prefer a provider that has specialized in scientific texts specifically or an out of the box provider with less know-how into PMS processes?
  • Does the tools come with explainability and confidence measuring, so you can trust the results?
  • What is the return of investment?

👉 Sign up to our newsletter for more content about digitizing R&D.

👉 Do you have any questions?