All right, so I don't have a cogent theme other than the fact that these are late-breaking nursing studies. So there are three studies that we'll touch on. One in particular I'm going to spend quite a bit of time on, partly because it's a complex study, and I think it has some very interesting findings. So I have no conflicts of interest pertaining to the articles that were discussed in this presentation. I do have some research funding, but it doesn't relate to what we're discussing this morning. And I do want to acknowledge the work of the Year in Review Committee, and especially Dr. John Whitcomb and Dr. Jennifer McAdam, Dr. Whitcomb from Clemson University and Dr. McAdam from Samuel Merritt University, as they assisted with article selection for this session. So the article we're going to talk about was published by Dr. Al-Zaidi from the University of Pittsburgh, published in Nature Medicine, and it's titled Machine Learning for ECG Diagnosis and Risk Stratification of Occlusive Myocardial Infarction. And part of the reason we selected this study is this was a fantastic example of a multidisciplinary study involving emergency medicine, nursing, cardiac medicine, critical care, and this is a artificial intelligence, it's a machine learning study, and I think you've probably heard at this conference there are quite a few speakers talking about the potential for machine learning to improve or augment clinical practice, and I think it's really important to recognize that we have gaps in our knowledge, and so pulling in engineering and mathematicians is very important when it comes to machine learning research. This was also a multinational article involving faculty members from the US and Jordan, Netherlands, and Canada as well. So traditionally, we've always thought of myocardial infarctions as falling into one of two groups, okay, the ST elevation MI. These are patients where there's very clear evidence on their 12 lead EKG that they've got some evidence of ST elevation, and these are patients that have a full occlusion or nearly full occlusion of their coronary vessels that are feeding the myocardium of the heart. And then we've got this other sort of, I call it a catch-all group, non-ST elevation MI. So these are patients that either have a partial occlusion of their coronary vessels, or they're not getting sufficient oxygen, they're often hypoxic ARDS patients, for example, or they may have perfusion issues as well, somebody in shock and they're not getting adequate perfusion of oxygenated blood to their myocardium. But it turns out there's a third group, and this is the group that's been labeled the occlusive MI group. And these are patients in roughly between a quarter to a third of NSTEMI patients would fall into this subcategory. These are patients that do have a full or nearly full occlusion of their coronary vessels, or I should say one of their coronary vessels, but it's not captured on 12 lead EKG. In other words, not that it's not captured, there are some subtle signs, but we don't have that ST elevation that helps us differentiate that they have a fully occlusive process. And these are patients that would benefit from percutaneous interventions in a cardiac catheterization procedure. But if you can't recognize that these patients have this pathophysiologic process going on, then they're not going to be treated in that way. So there are some subtle signatures of OMI that involve the QRST complex, and these are easily missed on a 12 lead EKG. You know, even for experienced clinicians, it can be hard to pick up on some of these very subtle signs. Cardiac biomarkers like high sensitivity troponin can't differentiate OMI until that peak level is reached. And at that point, it's often too late for intervention to salvage the infarcted myocardium. Delays in treatment of this subtype of NSTEMI have led to 14 to 22% excess risk of mortality in this group, so this is pretty significant. So in previous studies, this research team, again, the PI is based at the University of Pittsburgh, helped to create algorithms for artificial intelligence-enabled ECG analysis, and they also demonstrated the feasibility of screening for acute coronary syndrome in the pre-hospital setting. So the purpose of this study was to evaluate the diagnostic accuracy of machine learning for ECG diagnosis and risk stratification of OMI at first medical contact, so that would be out in the field, and in the absence of a clear STEMI pattern on an EKG. And they called their algorithm ECG-SMART, so just remember that as we get into the results, you'll see ECG-SMART listed there. So this was a multi-site perspective cohort study. It involved three cohorts of patients with chest pain. So they first derived this algorithm, and they tested it in a cohort in Pittsburgh, and these were patients that were brought by the Pittsburgh EMS to one of three hospitals in that city. Then there was the first external cohort, so this was now testing this machine learning, this algorithm, with patients that were brought by Orange County EMS to the University of North Carolina Medical Center in Chapel Hill. And then there was a second external cohort, and these were patients brought by the Mecklenburg EMS to Atria Health Hospitals in Charlotte. So patients were excluded if they had duplicate ECGs, and that's, so what you put into a machine learning model impacts the results that you get out, and it can be very confusing for the model if you input multiple different ECGs from the same patient. So for that reason, patients that had duplicate ECGs were excluded from the study. They also excluded patients that had cardiac arrested, had ventricular rhythms, or had confirmed pre-hospital STEMIs. So cases from the Pittsburgh derivation and testing cohort were classified as either positive or negative, and then the model was trained using both. Again, the purpose of the model was to help differentiate between the two. So they did include both positive and negative cases. In order to meet criteria for positive cases, that group had an affected coronary artery with a TIMI flow grade score of zero or one. And if you look over at our table here, TIMI of zero is a complete occlusion of a coronary vessel, and then one would be penetration of obstruction by contrast, but no distal perfusion beyond that obstruction. Or a TIMI flow grade of two with over 70% severe coronary narrowing and a troponin of five to 10 nanograms per ml would also be indicative of OMI. And so in order to establish this TIMI flow grade, these are patients that were catheterized to identify the extent of the occlusion. Negative cases were the absence of OMI, which included all other non-ACS etiologies of chest pain. So you think of maybe muscle injury, reflux, other causes of chest pain, with non-coronary occlusive ACS subtypes. So to identify OMI, the research team selected 73 out of a possible 554 ECG features using some data-driven methods as well as some domain expertise. They did pull in the expertise of cardiologists. There were 10 classifiers that were trained to learn ischemic patterns between the two different groups, the acute coronary syndrome group and then the non-acute coronary syndrome group, and to estimate the probability of OMI. So there was a random forest model that achieved the best bias-variance trade-off for training the model and also internal testing of it. And that team compared the model against the ECG interpretation of practicing clinicians and also against a FDA-approved commercial ECG interpretation system. The name of that system was not published, and I've spoken with the primary investigator and they have a non-disclosure agreement. But he promised me that it is a system that is used in a number of different hospitals in the US. It's cornered a large share of the market. And so this would be when you do a 12-lead EKG and you see the interpretation at the top where it says acute MI. So they were testing it against one of those systems. So using over 7,000 consecutive patients from multiple hospitals, the research team derived an externally validated model, and they used area under the curve. And they found that the area under the curve of the ECG SMART system was 0.91. Now typically for area under the curve, 0.7 to 0.8 would be considered acceptable, 0.8 to 0.9 would be excellent, and then anything over 0.9 is considered outstanding. So this model performed quite well in terms of its discrimination ability. The other groups, the practicing providers and also the commercial interpretation system performed at an acceptable level. So this model proved very effective in ruling out and ruling in OMI cases. The derived risk score boosted the sensitivity of detecting an OMI by 28% and the precision by 32% compared to normal reference standards. So here's a very important point. The purpose of this machine learning isn't necessarily to replace clinical expertise. And they found that combined with the judgment, the clinical judgment of experienced emergency personnel, that the OMI risk score helped correctly reclassify one in three patients with chest pain as having OMI or a non-OMI case. So what are the opportunities? What are the implications of this study? And not just this study, but some of the other studies as well. Some of these will apply as well. But developing, there's the opportunity for nurse researchers to develop or collaborate in the generation of additional machine learning models. And this study was led by a nurse at the University of Pittsburgh. I should say University of Pittsburgh School of Nursing, aimed at identifying cardiac or non-cardiac conditions that may be difficult to detect. So as you've gone through this conference, you've probably seen some of the sessions where they've talked about what are implications of AI and how it can be used to detect stroke, for example, or acute kidney injury. So I think there's a lot of potential implications for this, sepsis was another one I've heard quite a bit of. Also testing future machine learning powered software after it goes to market and testing the effectiveness. And I'm sorry, this will come when we talk about our next study here, testing the effectiveness of the hand grip game with other critical populations. And there are some practicing clinician opportunities as well in terms of discussing the adoption of the software within your facilities. Now, I wouldn't adopt this software based on this singular presentation. I think you have a responsibility as clinicians, as leaders within your organizations to look at the literature and see how does this machine learning compare to other things that are out there. And this has not gone to market yet, but when it does go to market, I think we'll have a responsibility to test it and see how well it works once it has gone to market. And then we'll talk about some of these other things, these hand grip games and the swallowing and oral care program. Okay, so the next article that we'll talk about, and I'm only because of time going to briefly touch on these last two articles, but the next one we'll talk about is an article that was published by Han and his colleagues in intensive and critical care nursing. This was a study in Taiwan, and they looked at the effects of an interactive hand grip game on surgical patients requiring intensive care. And this was a randomized control trial. It was a blinded trial. So their aim with this study was to explore the effects of delivering early mobility programs involving interactive hand grip games to adult patients requiring ICU care in one surgical ICU. And they assess the effects of these games on psychological measures. They looked at depression, anxiety, stress, sleep, and delirium. And they also looked at physical outcomes, specifically at hand grip strength. So they recruited 70 patients with 35 in the intervention group and 35 in the control group. The control group received 20 minutes of passive rehab five days a week, so Monday through Friday, that was delivered by a physical therapist. And then the patients that were allocated to the intervention group received five days a week of passive physical therapy as well, but they also received an individualized 20-minute interactive computer game intervention twice daily for three days. So the goal of this game's kind of an interesting game. And if you go to the article, you can see a picture of this. The goal of the game was to grab fruit that was falling from the top of the computer screen down. So they had a device that they would hold onto and they would squeeze it. And they were asked to try to grab the fruit as it was falling down and to hold it for three seconds. And this was to build up their hand strength. The hope too was that this would help them cognitively. So the patients were encouraged to continue as long as they could without fatigue. And if the patient held the fruit and achieved the thresholds, they received encouraging visual and auditory feedback about the program. The patients in this intervention group had significantly lower scores for depression, anxiety, and stress, but did not significantly improve their sleep quality nor result in less delirium. They also had stronger hand grip over time, which is not surprising. The last study looked at the effects of a swallowing and oral care program on resuming oral feeding and reducing pneumonia in patients following endotracheal extubation. And this too was a randomized controlled trial. This was published by Xiao and colleagues in Critical Care. So adult participants in six medical ICUs in a tertiary care center were randomized to receive either a nurse-administered swallowing and oral care program or usual care. And usual care in this case was our typical oral care that we would provide patients. So this was a seven-day program beginning on the day following extubation. And it was a bundled intervention that involved three different parts. There were oral motor exercises. These were exercises for the lips, the tongue, the jaw, and the cheeks. There were sensory stimuli and lubrication. And so what they actually used was sour-flavored ice pops and pork jerky to generate thermal tactile oral stimulation. That was very interesting. They also used tooth brushing and salivary gland massage as well. And then lastly, the patients received safe swallowing education. They did exclude patients that had neuromuscular disease, who had pre-existing dysphagia before they were intubated, those who had head or neck deformities, those who had pre-existing dysphagia, trach patients, those who were unable to follow verbal instructions, those on contact and droplet precautions such as patients with TB, and also those who were requiring continuous non-invasive ventilation. And so what they ended up finding, they had 145 randomized participants and there were groups of 72 and 73. And they found that by day seven, 51% of the participants who had received the intervention had resumed oral feeding. And that was compared to about 33% of the control participants. So if they completed the program, they were more likely to resume feeding. They also found reduction in pneumonia occurred in 15% of the participants who were in the intervention group versus 35.6% of the control participants. So after controlling for age and intubation longer than six days, they found that those in the intervention group were twice as likely to resume oral feeding within seven days and had 28% lower odds of developing pneumonia. So these are the references. I just, actually, I'm gonna circle back really quick to the implications. So with these studies, considering adoption of hand-grip games to improve patient's mental health and grip strength. You know, maybe it's a system like this, if this system goes to market, but potentially there's other ways that we can very simple ways, very inexpensive ways that we could have patients, you know, engage them cognitively and help improve their grip strength, you know, while they're laying in bed. So this is something that I think could be very easily done. Maybe one of those stress balls, for example. And then consider adoption of the swallowing and oral care program to improve dysphagia and recently extubated patients with a goal of hopefully preventing pneumonia. All right, with that, I will show our references and I will hand it over to Dr. Svetlana. Thank you.

machine learning

ECG diagnosis

occlusive myocardial infarction

nursing studies

patient care innovations