Two artificial intelligence (AI) models predict higher survival rates after liver transplants via enhancing emergency medical service (EMS) decisions.
An earlier approach to AI called natural language processing (NLP) saw success in improving the survival and better medical outcomes for victims of traffic accidents moved via ambulance – after they underwent liver transplants, according to two recent studies presented during the virtual American College of Surgeons Clinical Congress 2020.
AI models might increase odds of liver transplant survival
Both studies analysed the ways AI can process gigantic amounts of data to help surgeons and other care providers execute crucial decisions at point-of-care facilities, reports MedicalXpress.
One of the studies showed how University of Minnesota researchers applied an approach used before – NLP – to categorise treatment needs and medical interventions for 22,539 motor vehicle crash patients who were moved via EMS personnel to American College of Surgeons verified Level 1 trauma centres in the city of Minnesota.
EMS performance reviews via AI could reduce deaths
The 2016 study – from the National Academies of Sciences, Engineering, and Medicine – showed 20% of deaths from medical injury were potentially preventable. This revealed a lapse in quality, one researchers sought to address.
Performance reviews of EMS teams to identify potentially preventable deaths may improve efforts to minimise unnecessary deaths. "Currently this process for performance review is manual, time-consuming, and expensive," says Christopher James Tignanelli, MD, FACS, and senior author of the first study. "AI allows possible automation of this process."
Natural-language processing AI processes crucial data
NLP is a widely practised AI protocol capable of extracting critical data from written and spoken text. In this study, EMS personnel entered this data as an electronic record – which is part of their job.
Tignanelli is an assistant professor of surgery, division of acute care surgery, at the University of Minnesota Medical School, and also an affiliate faculty at the Institute for Health Informatics at the University of Minnesota.
EMS personnel record reviews could be automated
In the first study, two trauma surgeons manually and independently reviewed a random selection of 1% of patient records – and found where treatment was needed, along with medical interventions.
In analysing the AI system's accuracy, the surgeons' manual determinations were contrasted with those of the NLP determinations. "Overall the algorithm performed with very high accuracy," says Professor Tignanelli, MedicalXpress reports.
Typically, when EMS personnel enter their notes into electronic health records, oversight personnel vet their records to ensure patients received adequate care – usually within a week or so.
"That's quite a labour-intensive process," says presenting author Jacob Swann, MD – a burn and trauma fellow from Regions Hospital in St Paul, Minnesota. "The goal of this project and what it validated was to automate a lot of those notes."
The NLP approach evaluated these notes with an algorithm to differentiate the notes of serious medical interventions from ones that were less serious.
"That can streamline the manual review process," says Swann, reports MedicalXpress. "It's not performed at the accuracy level that would allow you to take the physician out of it and say that AI can determine with complete accuracy if the standard of care was given or not, but it does perform well."
AI enhances performance review, characterising pre-hospital EMS treatment. Source: American College of Surgeons
'Holy Grail' involves automated en-route decision assistance
Swann and colleagues' findings showed that roughly one-quarter (242 of 936) of patients in need of airway intervention actually received one before arriving at the hospital for further treatment – with roughly two-thirds (110 of 170) of those who didn't get necessary intravenous access and required access inside the bone – called intraosseous (IO) access – amid cardiac life support actually got IO access, MedicalXpress reports.
"Being able to identify systemic errors allows you to improve the entire health system," says Swann. "Having the ability to look at large aggregate data and go through 330,000 charts over several minutes with an AI-reading algorithm, to identify specific areas for potential improvement – whether it's getting intravenous access in our patients or having problems with splinting long bone fractures – allows you [to] separate the signal from the noise and then figure out where the problem lies."
To Swann, the 'holy grail' is to develop an AI system capable of listening and observing EMS personnel en route, to monitor and assist them with complex and difficult decision-making processes via real-time assistance.
AI model showed 80% survival one-month after liver transplant
The second AI study involved researchers from Baylor College of Medicine (in Houston), and tested four distinct machine-learning models' capability to predict survival following liver transplants. The two models with a high accuracy for predicting liver transplant survival are called the AdaBoost and Random Forest models.
The Random Forest (RF) model mixes two learning methods, combining the outputs of numerous decision trees and predicting a 'majority wins' outcome, explains Rowland Pettit, lead author for the study and MD-doctoral candidate at Baylor.
The second study carried out the selection of all 109,742 patients who underwent one liver transplant from the United Network of Organ Sharing database – since it was conceived in 1984.
The RF model (a graph) displayed an accuracy – which is the area under the graph's curve – of 80% for predicting survival up to one month, and a 79% chance of survival up to three months, 75% for one year, and 73% for three and five years' survival.
Notably, no other models showed a predicted survival rate greater than 70%.
"The most readily accessible application of these models would be for regulation, providing immediate feedback to clinicians about their outcomes for the past year and how they and their centres performed compared to others," says Pettit. "Being able to accurately predict whether a patient should have survived or not is crucial to then accurately providing feedback."
AI assistance not clinician replacement, but decision tool
This kind of AI model might also integrate with electronic medical record systems – in addition to physician workflows – to generate useful benchmarks, adds Pettit.
"It would be very easy with an integrated model to run predictions for every patient on a liver transplant waiting list in real time and determine the probability of each patient living at one, three or five years," he says.
"This step is not to make the decision for the clinician, but to add a further clinician-assistance decision-making tool to give them quantitative data for use in organ allocation decisions."
As AI is integrated into medical infrastructures and some feel unsure about the idea of handing human wellbeing over to automation, it is important to remember the reason for automation when it comes to AI implementation: it might save lives.
This article was written by Brad Bergan and first appeared in Interesting Engineering.