Thank you everybody for coming, we've got a packed house today, so word must have gotten out that this was the session that had the best moderators, so thank you all for coming. No, we've got some excellent presentations here, and we're really excited today to hear some great talks from our lecturers. I want to just remind everybody to please rate the session on the app afterwards, and thank you all for coming. Good morning everyone. Today we have a bunch of great speakers lined up, we have Dr. Chris Yorosco, Thomas Blakeman, Young Ik-An, Ali Ahmed, and Christian Sandrock. We're going to go ahead and start with Dr. Yorosco here. Good morning everyone, thanks for coming out. My name is Chris Yorosco, I'm one of the third year Palm Crit Fellows at Cleveland Clinic. Today I get to talk to you guys about a really exciting simulation study we've been working on, looking at the effects of bronchoscopy on minute ventilation and volumes and pressures during flexible bronchoscopy in the ICU. So we're going to talk briefly about the risks of a performance procedure in the ICU, what the current recommendations for this procedure are, and then going over the study itself. So bronch in the ICU has really increased in frequency. This is for two main reasons. One is the increased need for accurate diagnosis, as immunosuppression is being used more frequently, things like checkpoint inhibitors, getting a clear answer quickly becomes imperative. And the second is the proliferation of single-use bronchoscopy, so we have a procedure that is needed and is much more available, so we have to really start thinking more thoughtfully about the risks of the procedure and what we can do to minimize those. The risks are largely related to the procedure and resistance of airflow. So as you insert the scope into the ET tube, you have an immediate increase in your airway resistance. On the inspiratory side, you're going to have, depending on the vent mode you have, a decrease in the tidal volume delivered. And then on the expiratory side of things, you have a decrease in exhalation leading to volume trap, auto-peep, dynamic hyperinflation, volume trauma, even things up to pneumothorax. So we have a procedure that is being used more frequently. While overall a very safe procedure, it does have some significant and known risks. So we need to think thoughtfully about what we can do to minimize those. Fortunately, there are a few current recommendations out there as to what you can do. But you'll notice largely those are based on the procedure and the instruments being used. So the size of the scope that you're using, the outer diameter and its relation to the inner diameter of the ET tube, making sure you're not occupying too much space, and leading to some more of these major complications. But when it comes to the ventilator itself, there really are very limited to no clear guidelines in any literature between medical ICU, anesthesia, there's really limited data in that realm, which led to our current study. So really looking at, is there an ideal vent mode or setting to help minimize these risks? Our thought is that this is going to be based on the mode itself, your settings of rate, and then back to the procedural instruments themselves, the scope and the size of the ET tube. For the study we're looking at, we're taking away those last two, so the rate and then the scope size. We're going to keep those fixed, at least for this portion of the study, and just focus on that mode of ventilation. And then thinking about the outcomes, what do we need to show a difference and show that one mode is preferable to another? And we looked at this as far as two main outcomes. One is, if you're going to drop the volume delivered, what volume is significant enough to make it that the proceduralist performing is going to have to make some adjustment on the vent? And we thought about this as far as the downstream effects is, if you're decreasing your volume in and decreasing volume out, you're going to start retaining more CO2. So calculating out based on a change in minute ventilation, just an 11% change in your tidal volume could lead to an increase in CO2 by 10 millimeters mercury, which just surveying a number of intensivists, that this would be at least a threshold that they would start making some adjustments. The second significant change that we wanted to evaluate was for, was the increase in auto-PEEP. That becomes a little bit more difficult to set a threshold as, you know, X number of auto-PEEP is now a clinically significant factor, because this is so variable amongst patients. So we were just looking from a baseline value of auto-PEEP, what is the significant jump up, just to show some change. So like I said, this is a simulation-based study using, so the vent we used was a ServOU. The simulator was the adult simulator lung, so the ASL. And then our standard settings of a routine patient, just giving us a nice tidal volume between 420 and 550. And then we were looking at amongst these patients, again, thinking about the intensive care patients, somebody who's in an acute illness, looking at varying levels of compliance. So thinking about your ARDS patients with worsening levels of compliance, and then on your obstructive side of things, increasing level of compliance. And then a change in airway resistance. So for the patients who have more of a compliance issue, having a standard airway, and then increasing that resistance for your COPD-style patients. And then our BRONCs were the AMBU Disposable Scope, which has an outer diameter of 5 and ET2 of 8. So really fitting those prior recommendations of what was a recommended scope to ET2 ratio. And then looking at the vent settings, just kind of at a baseline, volume control versus a pressure control mode, and again, keeping the respiratory rate set at 12. And then like I said, looking at volume change versus auto-peep change, and comparing the two. Here's our setup. So you can see the screen here. You have your ServOU vent. This is connected to a swivel adapter with a BroncoScope being inserted into it, and then that scope being inserted to the distal end of the ET2, which is where you're going to have the largest changes in your pressures. And then another look, just seeing as the ET2 inserting into the ASL, which is essentially the simulator of the patient. So coming to our results, starting just with the volume control mode in the volume, as expected, the 500 delivered every time as we expect. From prior studies this was an issue because a lot of vents would have a peak pressure trigger and you would cut off your volume and you wouldn't get more than 100 cc's in just when the scope was inserted. So some of the vents do have adjustments on that peak pressure limit. So keeping that in mind that we were able to at least maintain the volume delivered that we intended to. When it comes to auto-peep, when you have your worsening levels of compliance, as expected, really no change in auto-peep and nothing really to be significant. With your obstructive lung disease patients at a baseline of a 1.24 pressure and increasing around 50% up to the 1.94, which depending on your patient's situation could be significant. Flipping over to the pressure control modes, as expected, when you insert your scope you have an increase in inhalation flow resistance across the board around a 40% drop in your tidal volume delivered. And then on the auto-peep side of things, again, really nothing to think much about. So just looking at between those two modes, volume versus pressure control, at least in the volume control mode your volume delivered as long as your peak pressures will allow on the vent you're using, which more of the modern vents do allow for, you're going to maintain your volumes and really, you know, depending on the patient population, minimal to potentially significant changes in auto-peep. And then for your pressure control modes, that drastic drop in your tidal volume, which if calculated out the minute ventilation drop, if left in for a number of minutes, can lead to a very significant rise in your PaCO2. So just raising this as a consideration that as bronch becomes more prevalent in the ICU, a lot of places if you don't have somebody who's managing the vent at the same time, just being thoughtful about what settings you're using prior to performing the procedure. Thank you, and I'll leave it for any questions. All right, we've got a couple minutes for questions. Yeah, please make your way up to the microphone there. Sorry, I know it's hard when it's packed. Thanks for your talk. Just my quick question, the ASL 5000 will not mimic the real-time exchange in the dynamics with the LANC patients. I'm sorry? Like, for example, if we put the bronchoscope in the real-time patient, you might have actual real-time airway resistance, for example, bronchospasm. The ASL 5000 will not mimic that change. Correct. So how did you overcome this issue? So those very patient-specific, so with your ADS patients, thinking about that PEEP change, the heterogeneity of the disease, the ASL would not mimic that because it is a single lung simulator setting. So thinking about things like bronchospasm or other complications during the procedure, at least at this point, I don't have a way to simulate that. Okay, so my name is Chris Blakeman. I'm an assistant professor at the University of Cincinnati, and I have nothing to disclose. And the opinions and the findings in this study were all mine. So what I'm going to do, I'm going to show you the evaluation that we did of a rebreathing system that we developed to be used with portable ventilators. So why is it important? So the importance is with, like, military applications or in, like, resource-constrained areas where the oxygen is not either available or is in small quantities, very limited. So also with mechanical ventilation, you need oxygen most of the time, right? So in some areas, or some special military areas across the world, they don't carry oxygen whether the patient needs it or not, right? It's just because of the weight, again, because in the cylinders that you would typically take for, you know, for long-term use, eight cylinders as such, they're logistically, they're challenging. They're heavy, they take up a lot of space, they have a risk of explosion and fire. So that being said, so low-flow alternatives may be the option. So concentrator, chemical oxygen generators, no types of things that put out low-flow oxygen, obviously they're limited as well. And then obviously you want to maximize the low-flow oxygen that you have, right? So hence, that's why we're trying to develop a rebreathing system that will give you a high FiO2 using low levels of oxygen. So the two ventilators we use, they're deployed by the DoD, as you can see them there, the Zol ventilator and the SAVE-2. We tested them both at one liter per minute and three liter per minute oxygen delivery, and then at sea level 8,000 and 16,000 feet of simulated altitude. One note, the SAVE-2 doesn't have the capabilities that the 731 does, just so you know. So there's a little bit difference in the tested settings. The SAVE-2, the ventilation is limited to eight liters a minute. So some of the settings that we had set for our, these are the settings we had for our testing, weren't able to do with that. So a direct comparison between the two ventilators was not, I was not able to do that. But I will show the differences. And then there's the ventilator settings. We had a normal, and then an ARDS setting, and then the CO2 production we simulated with injecting CO2 into the system, and I'll show you where that happened. So here's this canister that is used with the media. So this is the soda lime here. So we have an engineering firm that we use, or partners, that we partner with, and they 3D printed these parts here, and this is just an acrylic tube. Okay? It was refillable, so it was good for evaluation, you didn't have to have ones that you buy and throw away. So the media weight was 775, and the canister weight was 1.55 kilograms when full. So what did we measure? Obviously we want to know what the FiO2 is, right? So we measured peak FiO2 at 30 minutes. The duration of the CO2 absorbent life, so the soda lime, so how long did it last, and we considered that when CO2 rose to above 1%, and it inspired CO2, that was the end of life. You could have probably pushed it to two, but we wanted to be conservative and go to one. We also measured the gas temperature and the humidity, and also the effect of closed suctioning and ventilator disconnect on the FiO2 that was delivered. This is a really busy slide, but this was actually the experiment setup. Real quick, I'm going to try to go through it. So here's the oxygen being put into the reservoir back here, okay? So it's stored here, goes into the inlet of the ventilator, so the ventilator delivers a breath, it goes through the absorber, okay? Obviously this is just a standard six-foot setup, but the way the drawing had to be, it's a little bit out of scale. It goes into this device here, which actually measured the temperature and humidity, and also could measure the CO2 being put into the lung, so that's the CO2 production simulated here, and then this is just a standard test lung, okay? And then back out of the test lung, we're measuring, obviously measuring the O2 and CO2 at both the inspired and expired limb, and the expired gas goes back in to the bag where it mixes with the oxygen that you're putting in, and the cycle just repeats itself, okay? So pretty drawn out, but we tried to simulate actual patients if we could, okay, as well as we could anyway. Also there's a pressure relief valve here that we had to have, any of my anesthesia colleagues know about the pressure relief valve that you have to have when you don't have a net zero negative gas being put in, obviously because we're putting oxygen in here, okay? So that's the setup, and this is what it would look like during transport, so obviously without all the monitoring tools, it would look something like what we have now except with the addition of this. So there's, again, the absorber, back to the patient, and then back to the ventilator, and this is your, whatever low flow, I just chose here to use the cylinder. We used cylinder oxygen in the test, and would feed into here, okay? So it's a little bit more manageable, not as easy, obviously, as just hooking up a portable ventilator without the absorber in line. So the results. So as you can see, on the left is 731, so this is the 3 liter bleed in, and this is the 1 liter bleed in. There's a bit of a decrease at altitude with the 1 liter bleed in, but it really wasn't clinically important. The 731, the peak FO2 for the 3 liter was between 98 and 99 percent, okay? The SAVE 2 was a little bit different, like I said, a lot of variation, 92 percent with the, roughly 92 percent with the 1 liter, and 97 percent with the 3 liter bleed in, okay? And this is at ground level. Humidity. I don't know how much you guys know about the soda lime absorbent, but the reaction when they scrub out CO2, the byproduct is heat and water. So it produces a very high humidity level, as you can see, much higher than what we'd be able to do in just a regular ventilator with a humidifier and heater. So at the lower tidal volume, around 89 percent all the way up to 96, 95, 96 percent humidity. And the absorbent life lasts anywhere between seven and a half to nine hours. And the end results here, so as you can see for suctioning and circuit disconnect, the baselines were pretty comparable, 96, 97, or 99 percent, but a five second suction did not cause much of a reduction, but a 10 second here, you can see how much the FO2 decreased. And also when you go down to the disconnect, the 30 second disconnect really decreased the FO2. And the end results here, so the FO2 was at least 90 percent over all conditions. Obviously it's conserving oxygen, if you use one liter of oxygen for bleed in, right? An e-cylinder will last you, what, one liter a minute, last you 600 minutes? Absorbent life is greater than seven hours, which would be good for transport or prolonged field care. Prolonged suctioning, obviously, and long disconnect times really decreased the FO2. The system may not be compatible with portable ventilators, just because of being able to access the intake, the gas intake for the ventilator, and obviously you have to monitor FO2 and CO2 for safety. Thank you very much. Yeah, that was excellent, and as a former military guy, yeah, that's got so many great applications. Do we have any questions? I have one question. So, you know, there's, again, sort of so many different possibilities of ways that this could be used, you know, in terms of, you know, problems with hypoxia, you know, in flight because of your altitude levels and all these types of things. Have you taken this yet to the folks in charge of the, of like CCAD and those kind of things? Not yet. Now this study was funded by the Air Force, so they will have the results shortly after this meeting. Excellent. So it's, and it's something that's probably going to be looking more for, it's probably going to be cumbersome with that canister the way it is. Maybe in another form it would be better for transport, but prolonged field care would work really well. Excellent. Fantastic. Thank you. Thank you so much. Thank you. Up next we have Dr. Ahn presenting clinical implications of thoracic skeletal muscle volume as predictor, as a predictor of weaning failure in brain injured patients. So good morning and thank you for participating in this session. I'm Dr. Ahn from Ulsan University Hospital in Korea. Today I'd like to discuss quantitative measurements of thoracic skeletal muscle can be the predictor and mechanical ventilation weaning trials in brain injured patient. So brain injured patient often require mechanical ventilator support to prevent complications such as aspiration and to correct hyper or hypochemia and hypoxemia, which can lead to secondary brain damage. Unfortunately, these patients also face higher complication, including ventilators associated pneumonia and increased mortality compared to the general ICU population. Therefore, the safe and timely discontinuation of mechanical ventilation is paramount. However, we encounter a challenge in brain injured patient as traditional cardiopulmonary markers used for weaning preparation may not be directly applicable. This underscores the need for effective predictors specific to this patient population. Our psychopenia is a condition characterized by the generalized loss of skeletal muscle mass and function, while aging is a primary factor in psychopenia development. Other contributors include physical activity, nutrition, and various diseases. Importantly, psychopenia can lead to respiratory muscle dysfunction, impacting the reverberation from mechanical ventilation. Despite its potential significance, psychopenia remains under-emphasized or unrecognized as a predictor of weaning failure in brain injured patient. So we hypothesize that psychopenia is a crucial predictor in determining the success or failure of weaning in this patient population. Our study seeks to investigate its association with ventilator liberation outcomes in brain injured patient. So this is the method. As you can see, we included brain injured patient aged 19 years and older who were treated at a single institution between 2017 and 2019. Patient who can perform daily activities often with assistance from others were considered for inclusion. Patient who passed away or transferred did not undergo the weaning process or has insufficient data were excluded from the study. Finally, a total of 73 patients met the inclusion criteria and were enrolled. Of these, 61 patients were in reverberation success group and 12 were failure group. All patients underwent chest CT before ICU admission. The CT images were optimized for muscle area and the region of interest of the muscle were extracted and digitized. Thoracic skeletal muscle area was measured using images from thoracic vertebra level T1 to 12. The data was reconstituted into a 3D image. Muscle not involved in the breathing were then excluded. All images were analyzed using a program developed based on ImageJ open source. data was calculated using scared body haze resulting in the thoracic skeletal muscle volume index. This study defines sarcopenia as a thoracic skeletal muscle volume index less than the 50th sex-specific percentile. The weaning process was carried out with continuous monitoring of adequate gas exchange, maintaining FiO2 at less than 40% and a PIP below eight centimeter H2O. Hemodynamic stability was ensured up to 24 hours before the weaning prior, attending physicians evaluated weaning parameters which include the rapid shallow breathing index and maximal inspiratory pressure. In our results, comparing the weaning success group and failure group, no significant differences were observed in age, gender, BMI, and core mobilities. However, patient in the reverberation failure group exhibit a significantly lower thoracic skeletal muscle volume index compared to those in the reverberation success group, which means sarcopenia was found to be significantly higher in the reverberation failure group. This table shows that sarcopenia was significantly associated with an increased risk of reverberation failure and also higher SOFA score at the procedure day emerged as an independent predictor for reverberation failure. In our ROC curve analysis, the AUC of thoracic skeletal muscle volume index to predict reverberation failure was 0.840. The optimal cutoff value demonstrates 64% of sensitivity and 89% of specificity. In conclusion, our study suggests that thoracic skeletal muscle volume is an independent predictor of failure to reverberate from mechanical ventilation in patient with brain injury. Early identification of sarcopenia could assist clinician in tailoring ventilator strategies and improve patient care. Thank you for your attention and engagement. Thank you. Can your software measure the diaphragm at the same time that the thoracic muscles? The crew and the diaphragm? Yes. And did you notice any difference? We didn't research about it, so we're researching where we need it. Alright, anybody else? Okay, great. Thank you so much, Dr. An. Our next presenter is Dr. Ali Ahmad. He's presenting extubation failure rates and outcomes in a neurocritical care unit. Hello, everyone. Thank you for participating in this session. Next, we're going to be talking about extubation failure rates and discharge outcomes in a neurocritical care unit. My name is Ali Ahmad, and I'm a neurocritical care and neurophysiology fellow at Baylor College of Medicine. I have no financial disclosures. The study that we'll be discussing is a small study that we did in our ICU, and I'm going to share the results for that study. The extubation failure is pretty tricky to predict in this population, and there's not a lot of data that currently suggests what... Give me one second. All right, so like I said, you know, it's difficult to predict extubation failure. We don't have good predictors currently. And to find that one patient who's going to fail extubation, it adds to the overall cost of healthcare and, you know, we don't know what are the long-term effects of that. All right, so our study was a single center study. We retrospectively looked at the data in our unit between July 2017 and 2018, and all the patients that we looked at were adults who were intubated for a neurological diagnosis primarily and had at least one extubation attempt. And patients who had non-neurological primary diagnosis, missing data, were excluded from our study. So extubation failure in our study was defined as reintubation within 72 hours, and we looked at discharge outcomes by comparing the disposition, defined favorable outcomes as going back to their prior disposition or home or inpatient rehab versus non-favorable outcomes such as skilled nursing or comfort care, palliative hospice, and deceased. Our statistical analysis was done using R, and we looked at ANOVA and Chi-squared tests to see the impact of extubation failures in the variables. We had a total of 175 patients with a slightly female predominance in our sample, and our extubation failure rate was calculated at 9%. The reintubated patients were found to have longer ICU length of stay, which was significant, longer hospital ICU stay, which was significant, and longer tracheostomy rates. The figures on the screen currently depict the ICU and in-hospital stay. You can see the taller rectangle on the right side showing the average days the patient spent in the group that was reintubated, which was significantly higher. On this figure, you can see that we're comparing tracheostomy rates, mortality rates, and discharge outcomes between the group that was re-intubated and not re-intubated. And while there was a significant difference between the first group, the tracheostomy group, our study could not find a difference between the mortality in the two groups and the discharge outcomes, although there was a trend towards unfavorable outcomes in the re-intubated group. All right, so mechanical ventilation in the neuro ICU as opposed to medical ICU is for a lot of times for a low Glasgow Coma scale in the setting of acute cerebral injury, cerebral dysfunction, as opposed to a pure pulmonary cause. And there are multiple studies looking at extubation failure, and the rates are varied in the literature out there, anywhere from 10% from the Project Impact Database by Miltides et al, to as high as 25% in some cohorts. Recently, there was another European study, the ENIO study in 2022, which reported a rate of 19.3%. And they were looking at extubation up to day five for that particular rate. And their rate slightly increased when they increased the duration of how far they went. I think it was 21% when they looked at 28 days. And still there's no consensus on what the ideal limit is or the threshold is where you can cut off that this is after which a patient is said to be re-intubated and has failed extubation. So that is something that needs to be standardized in the current literature and the future literatures. So our rate was on the lower side, at 9%, and this is at 72 hours as a cutoff time. And it could be because longer duration would have caused a higher rate. And that could be a reason why our rate was on the lower side. And other pointers that could explain the lower rate could be the interventions that are done in the neuro ICU, such as IV thrombolytics, endovascular thrombectomy, anti-seizure medications, which can improve medication really quickly and reduce the risk of aspiration pneumonia and possible re-intubation. Again, our study was not powered enough to find a significant discharge outcome difference or a mortality difference, although there was some trend like I had mentioned. And like I said, it was a single center retrospective study, small sample size. We did not have follow-up data available and functional scores were not calculated for this particular abstract, but something that we can look at in the future. At the conclusion, our study had an exubation rate of 9%, which was again lower than what is published currently at the 72 hour mark. Studies have looked at exubation failure anywhere from, some of them even looked at 24 hours, patient re-intubated within 24 hours were deemed exubation failure. Others, mostly between 48 hours and seven days. For our study, we had a longer ICU stay, a longer hospital stay, which is pretty much reflected by the current literature out there as well, and higher tracheostomy rate as well. Although we talked briefly about the ENIO study, which was recently published, there's still a larger need for studies to highlight the at-risk population, help identify and create protocols to prevent this complication. Thank you. All right, we're doing pretty well on time, so we've got time for questions. Yeah, please. Thanks for your talk. Just my question, do you have any subgroup analysis for the reason of the re-intubation failure? Yes, so we looked at the breakdown of our sample size. Unfortunately, with the 175 patients, we didn't have enough participants in each group to where we could have a meaningful subgroup analysis, but the goal would be to get a decent sample size with enough representation of strokes, subarachnoid hemorrhage, subdural hemorrhages, to where we can have a meaningful subgroup analysis, but ours did not show anything that was significantly able to. Right, so just quick, another question. The reason for the re-intubation, is it neurologically related or non-neurological? For example, PE, because most of them, they don't get DVT perflexes. The majority of our patients, everybody who's re-intubated was because of a neurological worsening. Hi, thanks for that. So a couple questions, what are you using for assessing your level of alertness as part of your readiness testing for extubation? That's my main question. The other one, though, might be to track how many of your extubated patients or re-intubated patients get treated for bronchitis, aspiration, or pneumonia, but what are you using for readiness testing for level of alertness? So in addition to your regular SBTs, we do a quick neurological screens with ability to track, follow two-step commands, ability to protrude their trunk, and yeah, those were the main ones for the readiness to extubate. Dr. Ahmed, at your institution, are extubations mostly happening during the day, or is there a 24-hour? We typically just extubate in the, I would say first quarter of the day, so if by noon, if somebody is not extubated, we prefer that we have the full strength of the staff to watch them, re-intubate them if necessary, so most of the extubations were done between 6 a.m. and latest by 1 p.m. or something like that, but that's, you know, I'm just quoting numbers based on my experience. I have a question for you. So, you know, we always say that if you're not re-intubating enough patients, you're not extubating them quickly enough, so after all your study into this, what do you think is the optimal re-intubation rate? What should you be targeting? So, I think, that's a good question. So, you know, obviously you don't want to be re-intubating a lot of patients, so I think the jury's still out there. There needs to be still a better system to get a percentage which is safe, and, you know, we're not running into too many complications down the road, tracheostomies, you know, something that prevents a tracheostomy in a patient that would otherwise have safely been extubated would have been, you know, a good outcome for him. Okay. Excellent, thank you. Thank you. All right. Dr. Sandrock, all right. All right, thank you. Perfect, sorry I wasn't here. I heard you called my name, and apparently I got here so late it was popular. I was sent to the overflow room. So that's a good way to start as a presenter. I thought, I'm out of work, this is good. So we're gonna present a little bit about a lasso derivative predictive model for post-op respiratory failure. This is actually, I'm one of the faculty members at UC Davis. I'm sort of in transition from comfort to pain, so I've got more of an administrative role as time goes on against my will. So you'll see that PSI 11 is playing a bigger role in our work where this really focuses. And Jackie Stocking, who is one of our younger junior faculty who has her K01 award. This is some of the data that got her K01 award, and was also certainly supported by the Chess Foundation as well as one of our chair's clinical research awards. So a little bit about UC Health and the UC3RC. So this started in 2015, and we're essentially a critical care research collaborative group across most of the UCs. So from UC Davis and UCSF all the way down to UCLA and UCSD. So essentially we're looking at patient education, medical student and resident fellow education as well as clinical trials and other research. So we're still continuing to grow. This is the area where Jackie's future K01 is really gonna focus. But our presentation today will look at a single predictive model, really for post-op respiratory failure. So all of us have been sort of called to the bedside for that patient that's come out of the OR and needs to be rapidly re-intubated. Luckily it doesn't happen very often. The definition largely is that they have mechanical ventilation greater than 48 hours post-operatively or they get re-intubated post-operatively at that inflection point of 48 hours. The incidence is very low. So certainly high single digits is about where we're at. Usually it's around one to 2% at most places. And since this is a PSI, it is publicly reported and part of the Visiate scorecard. So most institutions really try and drive that down. It stays relatively low. But the mortality and other outcomes, such as length of stay and cost, tend to be pretty high with this data, which is why it's one of those patient safety indicators. So in our study, this was a single site proof of concept. And what we've done in the past with our work and when it has been multi-centered is it's largely data extraction manually. So you can imagine you're getting post-op respiratory failure cases. You can see we had a little under 24,000 patients in our group. So manually going through this for a low incident project is very difficult, very tedious, very time consuming and hard to yield. So we're actually using a automated data curation machine learning project where we're able to pull the data from these charts and pull patient factors out of Epic. And this was sort of our single site proof of concept, which will now expand out to multiple center sites, which while I'm here, I'll put a plug in that if any of you are interested in joining the project, let me know or I'll give you Jackie's email. And essentially what we looked at was elective surgeries or inclusion criteria, 18 years of age, so non-pediatric elective surgery. Your surgery happened within 24 hours of admission. Exclusion criteria was emergent surgery, prior tracheostomy or respiratory failure going into that case. And essentially our primary and secondary outcomes, primary outcome was the incidence of post-operative respiratory failure. Secondary outcomes were length of stay, time of mechanical ventilation and so forth. And as you can see, we had a really low incidence, a little under 1% overall. So 225 cases were identified and we looked a little bit deeper via machine learning at that for a LASO, not a TED LASO, but a LASO, Variable Selection Regression Analysis. So that uses variable regression as well as regulation. And please don't ask me any more than that. That's about the extent of machine learning I can talk about. We did some cross-validation both manually and with some bootstrapping to make sure that these were accurate and this data was validated. And we certainly looked at some more discrimination, which I'll show you in a second, and a sensitivity analysis. But essentially what we found out, which was some of it not surprising, a little of it relatively surprising, is that the predictors were generally older age population, male, Medicare is the primary payer, not obese, and again, non-obese, which was a little surprising in multiple comorbidities. Longer anesthesia time certainly played a role. Interestingly enough, if they had lower tidal volumes intraoperatively, so particularly lung protective strategies seem to play a risk factor intraoperatively. And then a positive net fluid balance played a role in more morphine equivalents. But interestingly enough, in our descriptive data analysis, a little under half the patients left the OR already intubated. A little more than half were intubated and then subsequently re-intubated with an average time of about 51 hours for those. But obviously we had a 24% in-hospital mortality in that subgroup, which is very high for elective surgery, consistent with published data that we certainly see. So in our LASSO predictive model, you can sort of see the top most frequent hits. Obviously anesthesia duration was one of the biggest net fluid balance in liters, so a positive fluid balance played a role. Anything that was on the cardiovascular system, so CT surgery, again, we're a single center site, so that could be much more surgeon and service dependent, but that was a big risk factor for us as well. Medicare is a primary payer and then the ASA class of greater than three. And then the lesser impacts of these models were certainly what was the FIO2, the mean FIO2 during the procedure. Whether it was a nervous system, musculoskeletal system, use of vasopressors, and then obviously the body mass index down below and the end-tidal CO2 played a role as well. We did a slightly more detailed analysis where we did a temporal split validation to take a look and say, okay, were there any trends or changes over time? That had a nominal predictive value as well. And then we actually had a comorbidity analysis where we looked at 13 variable comorbidities to see if other comorbidities played a role and they really had a nominal analysis as well. So essentially what we can say is that we have an 18 variable predictive model. Some of those such as anesthesia time, ASA class, age, male gender and so forth all played a larger role. And we can use this now in a machine learning process to identify our patients up front, which are gonna be at much higher risk for having postoperative respiratory failure. Again, low in incidence, high in morbidity and severity as well. And this way we can either have some interventions early on, not have our fellows get called at one o'clock in the morning when they're not paying attention or whatever it may be and involve us a bit closer. And right now the next steps in our work where we go is we're looking at obviously a second site validation. So we're gonna across the UCs work with our UC collaborative and do this in a much larger manner to further validate and predict as well. And incorporated more advanced machine learning, which again, as a clinician and an educator, I always get scared of machine learning. So I still have a 45 year old pickup truck because I'm afraid of any electronics. So that scares me, but is the direction we're going and where our work will focus on. So with that, I know we're ahead of schedule. If you guys have any questions, let me know. If you're interested as well, let me know and thanks for your time. And sorry, I missed the beginning introduction as I was banished. Thank you.

predictive model

post-operative respiratory failure

elective surgery

risk factors

machine learning techniques

anesthesia duration

net fluid balance

operating room intubation