Sepsis alert saves lives in hospital

Could electronic medical record systems help to spot sepsis risk sooner?
26 November 2019

Interview with 

Kate Honeyford, Imperial College London


Close up of a doctor's coat, with a stethoscope and a pocket full of pens


Increasingly, hospitals around the world are replacing paper-based notes in favour of electronic medical record systems. As well as the advantages of being able to better read what doctors have written, these computer-based systems can also spot subtle trends in the data being stored indicating that a patient could be about to deteriorate. And it can do this sooner than a human, and sound the alarm. Kate Honeyford has been looking at the results from a healthcare trust in London, and she spoke to Chris Smith...

Kate -  When the hospital trust adopted an electronic health record system, they had the opportunity to introduce a sepsis alert, and we were analysing whether the introduction of that alert improves the outcomes for patients. Sepsis is an overwhelming and life threatening response to infection. So some people might have heard of blood poisoning or septicaemia, and that's when the bacteria are in the blood, and that can develop into sepsis, which is this overwhelming response.

Chris - And how might the electronic patient record spot when a person's at risk of that happening?

Kate - So it can pick up on the normal signs of infection, like a high or low temperature, changes in blood pressure and heart rate. But it can also look at blood test results in the system. And then, using a formula, it can determine whether or not a patient is at risk of developing sepsis.

Chris - Okay. So talk me through the actual methodology. What did you do, what did you measure, and what did you find?

Kate - So the trust switched the alert from a silent running period, which was just kind of ticking over in the data, to a live period where clinicians could see it. And that meant we could analyse the data as something called a 'natural experiment', which makes statisticians really happy. So that was the first part of our methodology.

Chris - So in other words, you had data collected before the system went live and you could see what was happening to people. Then you've got the system going live, it's alerting clinicians and you can ask how did they change their behaviour, and presumably you can compare mortality, or how ill people are when the clinicians are not being alerted before it goes live, with afterwards to see if you're making a difference.

Kate - Yes. So what we found was a 35% increase in the chance of receiving antibiotics within an hour, when the alert was live and a 24% lower odds of death for the same group of patients.

Chris - How do you ascribe that benefit to the alert though? How do you know that wasn't just happening by chance?

Kate - Well, we're fairly confident it wasn't happening by chance because we used statistical methodology, which takes into account differences in patient characteristics in the two groups of patients.

Chris - The thing with these alerts though is that doesn't it still rely on some proactivity on the part of the doctor to go and check on that patient's records to see the alert? It's not like a bleep going off; "Oh dear, bed thirteen's in trouble. You better go and review them."

Kate - Yeah, that is true. However, it also alerts to nurses, so nurses, if they're plugged into the system, if they're in their computer, the alert will fire and they then can contact the doctors to come and have a look at the patient.

Chris - Given that this really costs nothing to implement, because we're already collecting all of the data that's being used by the system, it's merely just crunching it together and generating an outcome that can be made known to medical staff. It begs the question, what's not to like?

Kate - I think what's not to like is that if you have too many alerts in a system, people will become desensitised to the alerts. So if you have lots of alerts, people won't know which one to respond to and how to deal with it. So all organisations need to be quite careful in deciding, well, which alerts are going to have the most impact on patient outcomes.

Chris - Do you think that the sensitivity and the specificity of these sorts of measures could be improved as we learn more about other things we can monitor, so if we bring more things to the party, we look at more parameters about the patient, that actually the sensitivity will go up and that problem goes away with over calling the alerts.

Kate - Yeah, definitely, we need to make the patient history much more part of the thresholds within the alert, without question, that is the next stage of what we're looking at.

Chris - And are we in position to do that with these new electronic patient records systems that are increasingly being implemented and with good reason, across the UK and elsewhere?

Kate - Yes. The technology is there. It's just a question of refining the alerts, and refining the clinical input into them to make sure that that is what happens. Because what we don't want is patients to be missed because it's not working quite as we expect it to.

Chris - And, at the end of the day, this all comes down to money. These computer platforms are not cheap. A big hospital will spend money running into hundreds of millions on these systems. Does this translate into a saving, ultimately?

Kate - So length of stay has been shown to be reduced by the introduction of alerts like these, but also with an era of antibiotic stewardship, where we're trying to only use antibiotics when we absolutely need to, coupling the alert with antibiotic guidelines, with treatment plans, can actually improve the use of medication. So we may not see a financial cost saving straight away, but we may see improved uses of antibiotics and other medications.


Add a comment