Natural language processing (NLP) is a key component of many machine learning approaches. We’ve looked at (broadly) how it works, now we investigate some of they key use cases in the life sciences industry.
In the first part of this blog post, we looked at how Natural Language Processing works (at a high level, of course). Now, let’s delve into some of the key use cases across the life sciences industry.
At Evaluate, we’ve trained our AI and applied it to a number of quite cool business cases for clients. The broad areas where we’re really seeing the value of NLP deliver are;
Data mining research
- Medical entity recognition and resolution: Given an arbitrary piece of text, NLP can extract different medical entities including symptoms, diseases, or treatments. For a company like Evaluate this is crucial as we need to keep up to date and scan millions of documents to do so. Most might not be useful or contain new information, but if we can recognise the topics, therapies, and conditions then we can ensure the documents are tagged and sent to the right teams for manual review and data entry. By tagging the appropriate sections and auto-filling where we can, we can deliver more up-to-date data for our customers, collected from a wider range of sources that no human team could keep up with – even ours!
- Medical knowledge discovery: NLP can complement structured data by processing different kinds of unstructured medical texts, extracting patterns and the relationships between them. A lot of the interesting knowledge is dilute and difficult to extract when reading one document at a time as a human. Even specialist teams focusing on one area can’t keep up with the rate of new research publications. But NLP can determine the relationships between the entities and categorise and group them, forming links between concepts which can be strengthened each time they are mentioned, and weakened each time they are not. This results in a knowledge graph of concepts and relationships, such as biochemical pathway interactions between drugs, correlations with certain gene variants and so on, with low confidence speculative knowledge as well as high confidence well-supported knowledge.
Clinical Decision Support
- Clinical assertion modelling: Clinical assertion modelling enables healthcare providers to analyse clinical notes and identify whether a patient is experiencing a problem, and whether that problem is present, absent, or conditional. This is particularly useful when analysing large groups of patients together, as it allows you to consider the statistics of groups even if you only have clinical notes to work from.
- Medically relevant response/question suggestion: This involves suggesting responses or questions to physicians who are having a conversation with a patient. The suggestion not only needs to be linguistically and contextually meaningful, but it also needs to have medical validity.
Clinical Trial Matching
Finding the right patients for carrying out clinical trials of new medicines or vaccines can be a challenge for researchers, especially as drugs become more specialised and targeted. NLP-powered systems and computer vision tools can assist healthcare experts with that task. Such systems quickly scan through the past medical records of large numbers patients before confirming whether they are fit to be a part of trials or not.
When we ran a session about NLP at Evaluate recently, one of the first questions asked was “how will this affect our teams?”. While NLP, like any ML-based system, has its own potential biases and ethical concerns, particularly around the source datasets, it is not actually making decisions but instead summarising information, and broadening context for later use by people. As such it is one of the best examples of AI systems working synergistically with humans. Like all applications of AI, the approach must be one that aids human endeavour rather than replacing it. In medical situations, particularly around things like clinical trials, there will always be a need for human validation to ensure critical decisions are correct. But if AI can help speed the path of new treatments from pipeline to patient, it must become a welcome part of any pharma toolkit.
