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

Post market surveillance (PMS), a critical part of drug and medical device safety, involves many repetitive and tedious tasks, such as reviewing thousands of research papers for adverse effects.

Now, artificial intelligence (AI) is automating parts of post market surveillance, which is great news, enabling PMS professionals to spend their valuable time and expertise on more fruitful and interesting tasks.

The post market surveillance process can broadly be broken into four parts: 

  1. Data collection, such as collecting case and clinical trial reports, scientific literature and even social media reports
  2. Data entry, including reviewing collected data and extracting all relevant data, such as adverse events
  3. Data analysis, such as signal detection
  4. Action, inclduing causality assessment and market communications

In this post we’ll focus on steps 1 and 2. They are particularly susceptible to automation by AI, as the tasks involved are more manual, repetitive and error-prone. AI can best help the PMS processes when you already have access to a lot of data, which you need to slice and dice to draw conclusions.

8 Iris.ai modules to automate post market surveillance

We’ve split the PMS automation process into 8 modules:

A. Literature matching from description

B. Dynamic document evaluation

C. Advanced literature list refinement

D. Highlighting relevant information

E. Document cluster summarization

F. Extraction of key data points

G. Surveillance, monitoring and alert

H. Integrations with your tools or UX

A. Literature discovery

Original literature searches in pharmacovigilance are based on keywords, and require extensive boolean logic and in-depth expertise in the subject matter — a process which is really time-consuming.

This is particularly hard if you’re researching a new type of product or competitors’ products, and you don’t have in-depth expertise of the product and its context.

Using AI, you can automatically find the documents you need by inserting any written text on the topic, such as product and disease descriptions, patents, case reports, clinical trials, etc. There’s no need to find the optimal search query.

The machine will return a list of relevant documents sorted by relevance score. (If you’re curious about how the technology works, check out our research paper.)

B. Dynamic document evaluation

If you don’t know specifically what type of data you’re looking for, the dynamic document evaluation module is perfect. 

This module makes your list of documents dynamic; as you go through the documents one by one, selecting each as relevant or not, the module continuously re-orders unevaluated documents, making the list a prioritised overview of what you’re looking for.

C. Advanced literature list refinement

However, if you know specifically what type of data you’re looking for, the advanced literature list refinement module is most helpful.

Narrowing down major literature collections requires people with in-depth understanding of the subject and context, however AI filters search results based on Context matching and Entity tracking. (See the venn diagram below for an illustration.)

Context matching is basically a written description, e.g. a disease description, product specification and practical context, that helps the machine find the most relevant content.

Entity tracking is a specific object that you can precisely name, informing the machine about what to focus the results on. Entities include genetic mutation, disease variant, chemical formulas and product names.

The advanced literature refinement combines the Context matching and Entity tracking (in whatever combination you prefer, for example two Entities in one Context or as below, two Contexts and one Entity), and returns a precise list of documents that you can read.

👉 Read part II, on how to extra data from text and tables and literature surveillance.

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

👉 Do you have any questions?