Sepsis: using AI to find improve outcomes

Scientists are trying to develop new treatments by analysing data from sepsis patients
14 September 2021

Interview with 

Anthony Gordon, Imperial College London


A stylised computer network.


Anthony Gordon looks for patterns in how different people’s bodies react to infections so we can pinpoint the best way to treat them. And he’s using artificial intelligence to help him. He's at Imperial College London and spoke with Chris Smith...

Anthony - We're trying to tackle this in a couple of ways. One is looking in the blood at what we call the gene expression, the message that comes from the DNA that Kenny has been talking about, that tells the body how to respond. And we're looking for patterns in that, so that if there are different patterns of how a patient is responding, we may then offer different treatments to those individuals.

Chris - In essence then, when someone's immune system is reacting, if they're going down the pathway that looks like they're developing sepsis, there will be different patterns of genes being turned on and off as their immune system goes into this metabolic tailspin. And you're saying, well, can we spot those?

Anthony - That's right. You've been hearing from the other speakers that it's a very complex system, so we're looking for these patterns. And we're sort of seeing that now, that maybe some patients have a lot of inflammation, and if we can spot that pattern they may benefit from anti-inflammatory treatments. And there may also be patients where their immune system is overwhelmed, and in fact they're a bit immune suppressed, and they need a boost to their immune system to help them fight any further infections. So that's the hope - that we can actually take these very non-specific signs that we usually use to diagnose patients and actually say, no, they're not always the same and can tailor our therapies appropriately.

Chris - What sorts of signals can you bring to bear with your solution and how does AI help?

Anthony - These patterns that you look for, clusters in the data that you as a human can't see, but the computers can look for these patterns. Then if it's something biological like the gene expressions, you can go back and try and understand the biology and maybe develop new therapies. What we're also looking at though is how can AI help us improve the current therapies we're doing? So again, we've used large databases, many tens of thousands of patients and using artificial intelligence where a computer looks at all the decisions made by hundreds of doctors treating all these patients and worked out therapies that we use all the time -things like fluid and what we call vasopressors, drugs that help boost the blood pressure, just looking at which were the best treatments that were offered for all these patients. And try, again, to spot those patterns and then actually offer a sort of aid to doctors to help them just optimise all of these many decisions they make in an intensive care unit.

Chris - Obviously the big hope is that we can follow the old medical mantra of "prevention is better than cure, always". Are there any signals that you can tap into that warn us early a person's going in the wrong direction? And we can therefore effectively change the course for them before they end up in the hands of people in the intensive care unit?

Anthony - Yes. That's the hope, that if you can detect these things earlier, and even if we can't diagnose the infection yet, because it takes a while to grow bugs in a laboratory, the body has recognised that there's an infection, it mounts its immune system response. And if we can tap into that and use the body's own early warning systems, we could intervene earlier and hopefully prevent what may be a simple infection becoming a more severe case of sepsis.

Chris - And is it working?

Anthony - I think we're making good progress in that. Like all these things it takes time to develop, but we've now got the diagnostic devices, the sort of PCR tests that Andy mentioned that we use currently for COVID to detect viruses. Hopefully you can change literally the cartridge that fits in the device instead of looking for virus DNA or RNA, look for the patient's DNA and RNA. And we can do that rapidly, the technology exists to do it in less than an hour, and that's the sort of work we are testing at the moment, making sure it's safe before we bring it into the health system

Chris - Because obviously, just in 30 seconds, agility and speed must be everything here because people can deteriorate so quickly when they go down that slippery slope of sepsis.

Anthony - Yeah. You need a combination of experienced staff, and also if we can use these new technologies, which will be much quicker, I think this can help all clinicians make better decisions and treat patients more effectively and hopefully improve their outcome.


Add a comment