All right, good morning, everyone. Welcome to our second in-person session for the CAF PCI Registry. Again, my name is Kate Malish, I'm your registry product manager. So today we'll be talking about mitigating acute kidney injury, and you will certainly be learning from the experts. We have Dr. Anezi Uzendu, who is one of the lead investigators for the registry's new AKI risk model, so he'll talk about how that was developed. And then Ellen Keneally, who you remember from yesterday, we liked her so much we had her do two presentations. She is a nurse and RSM for Northwell Health in New York State. So I'm just going to let them jump right in, because we have a lot to talk about. So I'll hand it over to Anezi. Can everyone hear me okay? I'll step away from the mics over here. Thanks, Kate, for the introduction. Just really grateful to be able to talk to you guys today about some of the work we've been doing with the ACC and NCDR's writing group for the new AKI risk model. So my name is Anezi Uzendu, I'm an interventional cardiologist and also a cardiovascular outcomes research fellow at the Mid-America Heart Institute. I really don't have any disclosures today other than that I'm learning, so I wouldn't necessarily consider myself a rock star. I've learned a lot really the last two days here being at the quality summit. This is my first quality summit, and I've kind of come to meet a lot of people and realize that there are people coming from various backgrounds, but the thing I think that's important about quality improvement is that we continue to learn, because as we're learning, we continue to get better. We get better together, which is obviously one of the slogans here for this conference, and that drives patient care and improves patient outcomes. So I'm presenting for the writing group, and I'm very humbled and it's an extreme pleasure of mine to be able to represent them, and these are the writing group members. Also a special thank you to Connie and Kate as we've been kind of working to put this presentation together. So again, we do have members from the audience and people on the live stream with multiple different educational backgrounds, training, experience levels. And so I wanted to make sure that this talk does have some foundational elements, but really is catered to a wide variety of guests, and at the end of the day, I want people to be able to discuss the incidence of AKI after PCI and some of the significant associations, describe the use and importance of risk models, discuss the new risk model itself, the need for it, some of the new definitions we're using, the variables and why we chose those variables, we'll discuss the models themselves, the performance of those models, what the outputs look like on the dashboard, and clinical application. So acute kidney injury after PCI occurs about 7% of procedures, and we know that it's actually associated with more concerning outcomes being bleeding, MI, and death. On the graph here to the right, you can see that in patients that do not have AKI, the rate of death is 0.5%, this is in-hospital, in-hospital events. But if you have acute kidney injury, that increases to 9%, and then if you have kidney injury that actually leads to the need for dialysis, that goes all the way up to 30%. And we're seeing that again with death, with bleeding, and MI. So this is a significant outcome that we want to try and see if we can mitigate. Some of the proposed mechanisms include some of the hemodynamic changes related to intervention. So as you're going up on your balloons, going up on your stunts, you actually include the blood flow going down the coronary artery, that causes some myocardial stunning, there's decrease in actual cardiac output, and then there's decreased perfusion pressure to the kidneys. There's also atherombolic disease, when you're moving your catheters up and down the aorta, they can scrape the plaque off of the aorta, and that can be embolized down into the kidney in the renal vasculature. And then obviously there's this association with contrast, which we used to call contrast-induced nephropathy now, with some of the newer studies that are coming out, we call it more of an association, but we do know, and actually we've used the cath PCI registry to validate some of this, that when we have conserved efforts to decrease contrast exposure, there are decreased AKI rates. So why do we have risk models? You know, as a physician, when I'm doing informed consent with my patients right before a procedure, they may ask me, what is their risk of, you know, really anything that could occur during the procedure itself, am I bleeding to death, and I can use that 7%, which is again the cath PCI mean rate, but for an individual patient, we have the data now to be able to tell them a little bit more accurately what their actual risk is, right? And so being able to do that is one of the benefits of having a risk model. How does our AKI rate compare to others? Using our risk models, we're able to do that and factor in patient variables that increase risk to help us better make comparisons. Is our AKI rate due to patient risk, or is it due to process of care? Are there things that we can do to actually improve our rates? And then are there ways, again, to decrease our AKI rates? So all these things are reasons why we need risk models. And so to kind of summarize, I think of it as we're quantifying risk so that we can educate patients for shared decision making, guide our prior procedural management strategies, and then for benchmarking and quality improvement. So I'm not a statistician. I'm an interventional cardiologist. I've taken a couple stats classes, but the way I think of a model is a model is a mathematical representation of reality, or you can usually think of it as a mathematical representation of any concept, and you can make a model of anything. So a model example today we'll use is the likelihood that someone nods off or kind of dozes off during this presentation today. And so some of our players for this model, we have kidneys on the top left that are asking for help, and that's what we're all here to do. On the top right, we have all-star Steph. On the bottom left, we have Dora the Dreamer. And then on the bottom right, we have a much more handsome version of Dr. Zahn and myself. And so what are the factors that play a role in Dora the Dreamer falling asleep or nodding off during this presentation? So she had an overnight flight. Unfortunately, the hospital didn't allow her to get Thursday off, so she flew in Thursday night, just got in this morning, ate a heavy breakfast. She obviously went into cardiology, so she hates kidneys. And then she has narcolepsy, which is what I thought I suffered from as a second-year medical student. On the other hand, we have all-star Steph. And what's the likelihood of her nodding off during this presentation, right? So she had eight hours of sleep. She had a light breakfast. She had two cups of coffee. And she understands that kidney function is really important to optimize cardiac function. And so she's primed and really ready to kind of stay engaged during the presentation. Now there are factors at play that I'm in control of, right? So if I have a monotone voice, if I'm disorganized, I use terrible examples. All those things would potentially play a role in someone in the audience falling asleep. We wouldn't want to put those things in a model, right? What we want in the model are going to be things that are inherent to the actual patient to try and make sure that we're capturing that risk and using that to guide our processes of care. And so kind of the AKI correlation of this is we would have, for all-star Steph, we'd have someone that's coming in for an elective procedure, has a GFR, which is glomerular filtration rate of 120, which would be normal, and not having diabetes, whereas Dora the Dreamer would be someone coming in with a STEMI, with a GFR of 30, and then diabetes. As you think about it, you know, if I have this presentation, and I have a room full of all-star Stephs, and still the whole room nods off, we'd realize that there's things that I'm doing that are potentially playing a role that are not effective, right? And we need to try and find those things and then try and say, you know, we have a room that's full of people that are primed and ready to be engaged, want to learn, but because of the way you're presenting, it's not effective, right? And people are having bad outcomes. So how do we try and change your process? How do we change your presentation, your examples, whatever the case is, to improve care? And on the other hand, if we have Dora the Dreamer, we have a room full of Dora the Dreamers, and they all stay awake, there's something potentially I'm doing that we should replicate, we should capture, and make sure that we are distributing in other facilities, other hospitals, to make sure that we are optimizing outcomes. So that's really what we're trying to do with these models. So to actually create a model, again, not a statistician, but some basics on kind of regression models themselves, what we do is we have a, this is actually a linear regression equation, AKI itself, for the AKI model, we use a logistic regression, and we use logistic regression when the outcome is binary, so if it's yes, no, but a linear regression equation is a little bit easier to understand. And so Y is our dependent variable or outcome of interest, so this is what we're trying to actually calculate. And so for us, it's AKI, and then X1, X2, X3 are those variables that we think are important and have clinical relevance for this outcome, and so we'd have age, we'd have hypertension, we had GFR, and then the betas in front are the coefficients or the relative importance of each one of those variables in predicting this outcome. So for this model, the reason why we thought we needed it was because the last model was created back in 2014, and so practice patterns are changing, we have new equipment, new stents, really kind of new data guiding how we actually perform PCI, and so we felt like our model should evolve with our evolving practice. The cath PCI versions 5 dataset is a bit more granular, so you guys have been capturing more data elements, and we're going to be able to use that, those extra data elements to better characterize our patients to actually have better fitting models to better characterize patient risk and improve patient outcomes with fair benchmarking. So for the model itself, the definition we used was from our CADIGO and AKIN definitions from our nephrology colleagues, which is an increase in serum creatinine greater than 50% of the original value, or an increase of 0.3. And so there's a lot of debate actually within the group about using this definition, because that's that stage one, very sensitive, 0.3 increase of serum creatinine. We could also potentially have done maybe a 0.5, which has also been described in the literature, or use maybe a stage two or stage three AKI as shown below, but we chose to use the 0.3 and that AKI in stage one, and I'll show you the reasons why. So this is a paper back in 2013-2014 that did show that if you're looking at just that stage one AKI, so that 0.3 absolute increase or 50% relative increase, again there's that increased association with bad outcomes, right, death, bleeding, MI, and that's in hospital outcomes. There was another paper back in 2017 that now linked this to Medicare data to have longer term outcomes outside the hospital, and we could see that if you have that stage one again AKI, there's an increased rate of recurrent AKI, a composite of death, bleeding, and MI, and death in and of itself. And so it's not an insignificant number, and it's something that we want to make sure that we're capturing to improve our processes of care. So again, there was a lot of discussion about this, and I think it's important to think of this as, if we kind of go back to that case example, if instead of nodding off as my example, if I said I want the metric to gauge how well my performance is, or the discussion is today, is if people completely just give up, put their head on the desk, go all the way to sleep, or even let's say they walk out of the room, right? No matter how bad my presentation is, hopefully no one would walk out of the room today or put their head down or something like that. And so I'd have low rates, but I would lose that opportunity to improve my presentation, right? And the same thing here, if we have a less sensitive definition, we lose the opportunity to really try and improve processes of care and help our patients, right? And then again, we do know that this increase is associated with death, with bleeding, with MI, with recurrent AKI. So the study population we used for the new model was from the cath PCI registry. We used all PCIs within the 2020 calendar year. We excluded patients that had a subsequent PCI during admission because some of the changes in serum creatinine maybe relates to other in-hospital factors. We excluded same-day discharges or patients that did not have a post-procedural creatinine. And then we also excluded patients that were already on dialysis. And so that left us with a little under 500,000 patients. And then we split that 500,000 patients into a 70% derivation cohort, which is the cohort in which we used to develop the model, and a 30% validation cohort, which is the cohort that we used to actually test the model and make sure that it's working well. So for the variables themselves, we reviewed the prior model that was published in 2014. Then we also looked at univariate associations, so associations of the new variables that we have in the version 5 dataset with the outcome of AKI. Again, what we wanted to do is we wanted to make sure that we were actually only capturing inherent patient factors, and so we're not avoiding physician practices. And so things like contrast volume, we did not include, and we did not want to include MCS at all. We'll talk a little bit about some of why we ended up including MCS and in what scenarios. And MCS, mechanical circulatory support, sorry. So issues that we kind of had in the working group as we were discussing, we talked about the AKI definition and why we chose that. The GFR equation, so over the last few years, there's been discussion about the inclusion of a race term within estimates for GFR or renal function. And actually, about this time last year, published in the New England Journal of Medicine, our nephrology colleagues did recommend that everyone switch to a GFR equation that did not include a race term. And we were actually able to validate that as well within the cath PCI registry, and we found that GFR equations that did use a race term actually were less accurate in predicting AKI in black patients. And so moving to a GFR equation that did not include a race term was actually better for our models. Mechanical circulatory support. So we did talk about how we don't want to include process of care variables, and that's definitely a process of care variable. However, there was such a strong association with AKI. And we think that the association is probably related to inherent, again, patient factors that are playing a role, right? So you're putting in mechanical circulatory support. These are your impellas, your balloon pumps, and all that. You're putting that in for patients that are really sick, in shock, clinical instability. And so we thought if we put those inherent patient factors within the model, we wouldn't need to include mechanical circulatory support. However, even when we put those in the model, mechanical circulatory support still had a very strong association with AKI. And so the compromise that we made was that if the mechanical circulatory support was in before we started an intervention, before we started a PCI, then it's probably reflecting something, another inherent patient characteristic or variable that we're not capturing in the cath PCI dataset. But it is a factor that shows that that patient is actually sick or not doing well. And our bedside physicians are actually able to see that, but our cath PCI registry is not able to do so. And so we included that if we include MCS prior to the intervention, but no MCS after. The models themselves, we created a model for AKI, for dialysis, and also a bedside integer score to be used for prospective care. So this is the AKI model itself, and you can see here, again, there were about some of these variables. It's actually 13 variables in this kind of reduced model. You can see that, like, let's look at the top two variables. It's age, age plus 10, age plus 10 for if you're above or under 70, and that's actually truly just one variable, which would be age, even though it kind of has two lines there. So 13 actual variables within this model, and it did really well, and we'll kind of talk about the performance. But the most important factors are some of the ones you think of, right? So if it was a salvage, a salvage procedure, if the person was in shock or if they had severe renal dysfunction prior to the intervention, then they're at a higher likelihood of having AKI after the PCI. The dialysis model is very similar. So patients with pre-procedural severe kidney dysfunction, a salvage procedure, or shock were more likely to have AKI that leads to dialysis after the procedure. This was the bedside integer score, and again, you can kind of look. So across the bottom here, you have these scores that were based on those actual odds ratios that are changing out some integers, so it's easier to use at the bedside, and you can see the AKI rates with the increasing scores for both AKI and dialysis. How we look at model performance is, I think, really we look at two factors. Model discrimination, which again, for interventional cardiologists, the way I think of it is, does the model perform well in deciphering if a patient's going to have an AKI event, yes or no, right? And then model calibration says, in patients that the model says are low risk, are we seeing low rates of AKI, whereas on the converse end, if it says that this patient is this patient population is high risk, are we seeing high rates of AKI? So that's model calibration. With that, we look at kind of slopes and intercepts on a graph, whereas with the model discrimination, we look at a C statistic. And so this is the model performance kind of breakdown, and you really want to look at the validation sample, because again, that's the sample that we're testing the model in. You can see that for the full model and the reduced model, we're getting C statistics that are 0.79, really kind of close to 0.8. The original model, or the previous AKI model was a 0.71, so this is much better, and you think of 0.8 as a really, really good model, 0.7 is still a strong model. And so for AKI, the full reduced and bedside models, we're at 0.79, which is really good as far as the performance. The dialysis models are actually even better, and you kind of think about it, the patients that are going to end up on dialysis after an intervention you know beforehand. I mean, again, they have the severe CKD, they have shock, and all those other factors that play a role. This is a graph that actually kind of shows what I was talking about with model calibration. And so on the left-hand side, you see the observed rate. On the left-hand side is kind of the patients that are supposed to have a low rate of AKI. You're seeing a low rate of AKI, and then on the y-axis, you see there that those points are also showing that the expected rate of AKI is low in this patient population, whereas as you go further down on the right on the x-axis, these are the patients that have higher rates of AKI, and you're seeing that our model expected them to have higher rates of AKI. And so you want this line to intercept where it meets the y-axis to be close to 0, and you want the slope of the line to be 1, and that's exactly kind of what we're seeing here. And then for the models, the outputs that you guys will be receiving, just a little bit about kind of the risk-adjusted versus risk-standardized models. So the risk-adjusted model is your kind of standard observed over expected times your unadjusted rate gives you your risk-adjusted model. The risk-standardized model really tries to help factor in things like clustering within sites and then maybe potentially like even small sample sizes, and so it actually gives you a predicted AKI rate over your – a predicted number of AKIs over your expected times the unadjusted rate. So it really kind of helps to, I think, smooth out some of the outliers that you may see from facilities. And so the metrics that you guys will be receiving is on the executive summary it will be primarily the risk-standardized model or the hierarchical model as it is also called, and then the detail line you'll still get the non-hierarchical AKI and dialysis models and the hierarchical or the risk-standardized dialysis model. And so obviously we want to do all these things to help, and Ellen's going to talk about improving care using the feedback you're receiving from the NCDR and using these models. I just want to highlight this paper that was published actually about a week ago in JAMA, and they used – the primary interventions were a patient-centered contrast limit, which used the patient's individual AKI risk, right? And so a lot of people use GFR times some multiplier, 2, 2.5, 3. And as we showed in the model itself, GFR is just one thing within the model that actually predicts the outcome of AKI, right? So there's clinical instability, all these other things. And so this patient-centered contrast threshold actually incorporates all the things that are important for the outcome of AKI or that AKI risk, and gives you a contrast limit or contrast threshold based on that patient's risk. And so the intervention here was some education on the patient's risk, the contrast threshold, and then peri-procedural hydration strategies. And in the intervention period, they were able to reduce their AKI rates from 8.6 to 7.2. And so, really, the hope today is that you guys use these models and go out and save some kidneys. Thank you all. Hi, everyone. I'm Ellen Keneally. I have nothing to disclose, except that I haven't worn high heels in two years. I could fall, so I'm just going to stay right here. I represent Northwell Health. I'm the registry site manager and the nurse team manager covering the cath PCI registry. Northwell Health is New York State's largest healthcare provider and New York State's largest employer. We have over 80,000-plus employees. We have 23 facilities and over 830 outpatient facilities. Our facilities are located in three regions, the western region covering New York City and Westchester, and the eastern and central regions covering all of Long Island. We have 10 facilities out of the 23 with cath labs. The 10 facilities together combined produce almost 10,000 PCIs annually. We are home to the Sandra Atlas Bass Heart Hospital, and it's recognized as one of the top two cardiac surgery programs in the U.S. and Canada, home to the Barbara and Donald Zucker School of Medicine at Hofstra University, soon to be nursing school as well, and we're voted by Glassdoor and Fortune as one of the top 100 companies to work for for diversity. Little look at our volume. We go from 150 PCIs annually to over 2,500. So we can speak to the challenges of a small institution versus a very large institution. This is our acute kidney history look back from 2018 to Q1 of 2022. If you can see, the first four hospitals are either worse or the same. The next three hospitals saw some improvement, and the last hospital, Hospital H, is one of our original facilities. They're not doing much better. And the final two hospitals are our newest facilities, and they're struggling as well. So going back to 2018, it was observed by our physician champion and administrative leaders and cath lab directors across the system that globally, across the system, we weren't doing well in the AKI measure. We were at or below the 50th percentile, and they knew this was an increased risk to our patients, and the physician champion decided to appoint one lead physician. He had some success at another institution to drive a sin reduction program at Northwell. So by 2019, the guidelines were established. The rollout was very disorganized. We started with all facilities, boom, at the same time, and the poor outcomes continued. 2020, not much better, because COVID hit hard, and our facilities were really overwhelmed. Only emergent and urgent PCIs were done, and the outcomes were worse than ever, if it's okay to say that. In 2021, the volumes returned to pre-pandemic levels, and we continued to worry about our AKI rates, which were still not better. So the physician champion again reached out to physicians in one region, and they piloted a new and improved plan, and once there was some success there, then it was rolled out beginning in March of this year. All facilities were on board by May, and we have a dedicated team of RNs abstracting the data and entering it into a dedicated database. The process is monitored now weekly, monthly, and quarterly. And where do we start? Provider engagement. Who will monitor the compliance? What data points are going to be collected? Where will the compliance be tracked? How will the outcomes be reviewed and presented? Just some of the challenges. Engagement buy-in. We provided our physicians, and we continue to provide drill downs. We pinpoint the trends in the AKI patients. We communicate incomplete documentation, which is one of our biggest problems. We initiate compliance checks, and we report process breakdowns at each facility if there are. Improved documentation allows you to focus on the process along the way. Local and system physician champions are also monitoring at all facilities. Who's going to monitor? RN data abstractors track the compliance, the deviations, and the AKIs. We have weekly meetings with the registry site manager and the RNs. That helps to enable the RNs to continue with the process. The RSM shares the data that we receive back to the RNs. The physician champion attends all of our calls to support the RNs, and the corrective actions are applied as needed when we find a process faltering. That's where we focus our attention. What is the initiative going to track? It's going to track pre and post hydration, pre, post, and intra hydration, patient demographics, operator and PCI status, pre and post creatinine, hemoglobin, and GFR, contrast type and volume, and deviations from the protocol, STEMI, heart failure, chronic renal insufficiency, fluid overload, and missing documentation, being the bad one. Where is the data going to be tracked? We track the data in a program called REDCap. The advantages are that it's cloud-based, it's customizable, and it's very simple to mine the data out. REDCap is a secure community. It's creative for knowledge management. REDCap provides security for hackers and breaches and data loss. REDCap is free. All you have to do is request administrative use. How are the outcomes reviewed and presented? The SIN initiative data is presented monthly per hospital. The SIN data is reported on local levels during PCIG conferences. The CAHPS Physician Champion attends all of the QAs, and the NCDR outcomes and SIN data are presented together at the system quarterly meetings. Where we are now. A quick look. Keep in mind that not everybody was on board at the same time. There were facilities on board in January, February, March, and then the rest of the facilities came on board May, end of May, et cetera. So far, we're seeing that almost 61% of the patients in the system globally are on the protocol, and of those patients, 1.5% are having AKI. In the deviation category, it's almost 40%, and 5.9% of the patients that deviated from the protocol are having AKI. The deviation breakdown, as I had mentioned before, it's STEMI, it's fluid overload, it's shock. But the one we were looking at that's the highest percentage is no documentation, that's 36.8%. But again, not everybody was on board at the same time, so we expect that number to really drop down now that we have more of a tighter system in place. A look at your AKI and your protocol deviations as it relates to GFR. 68.7% of the patients. In the deviation category, had a GFR less than 60. 31% had a GFR greater than 60, and that's a category that we think we can really improve upon because those patients were having near normal or maybe not impaired renal function, and so the target group there is to try and get that down as low as possible. Our system-wide average, taking all the hospitals together and looking at it in REDCap and comparing it to Q1 outcome alone, our Q1 average system-wide was 6.7%, and our current raw, unadjusted REDCap average is 3.3%, so we're hopeful that it's working. What's next? We're going to continue to track compliance, communicate, re-educate, meet every week. We're going to follow the adverse events, look to see if we're causing any heart failure or hemoglobin drops. We're going to see if the initiative's working, who's best protected, is there a need to modify or update the guidelines for the initiative, and we're going to continue to validate our AKI outcomes from NCDR against what we're seeing in REDCap, and that's it. Thank you. All right. Thanks so much. We've got some great questions here. The first one with the most votes, and, Nezi, you may be able to answer this, it says, sites with very high rates of same-day discharge after PCI have artificially worse AKI metrics since the healthiest patients are excluded from the denominator. I know you can't speak for ACC, but it says, does ACC have a plan to include same-day discharge rate in the risk model somehow to level this measure better? So you can't speak for ACC, but what are your thoughts on that? Correct. Correct. Yeah, I can't speak for the ACC. I'll start off by saying that. But we actually looked at that. That's a very good, insightful question. I probably should have put that in as one of the points of discussion that we had during the groups themselves. And so just to kind of explain a little bit more, the thought process would be that if in your same-day discharge population, these are going to be probably the most stable patients with the lowest risk of AKI. And so if those people aren't actually going to be included in your rates of AKI, the lowest risk patients are actually being discharged, and you're only left with maybe the medium to high-risk patients. Those are the ones you're actually capturing a post-procedural creatinine on and all that. Those hospitals may artificially look like they're actually performing worse compared to hospitals that actually have lower same-day discharge rates, right? And so they're actually keeping that low-risk cohort to actually kind of improve their numbers. Funny enough, that actually did not actually bear out. So we looked at facility same-day discharge rates and the relationship with their actual AKI rates, and we saw there really wasn't a correlation. And what we're thinking, and actually it might be an area of further research, is that still overall, our adjustments or the patients that we're actually including for same-day discharge probably still have, there's still a proportion that are kind of medium risk and higher risk as well. And so are we actually using the patient's, you know, entire AKI risk when we are considering which patients are actually going for same-day discharge or not? Because the data itself isn't bearing out that those patients, that the facilities that have high same-day discharge rates are actually looking like they're performing worse. But very, very good question, again, something that we thought about and kind of looked at. It was one of our early analyses. Okay, thanks. And sort of a follow-up question. Holly, I'm kind of paraphrasing your question. She brings up the same issue, but poses a question of whether new dialysis is actually a better measure for those high same-day discharge programs. Yeah, new dialysis. I guess, so I think part of it's based on the fact that the first concern, which again is a very valid concern, we weren't, at least that wasn't coming out in the data. New dialysis is actually, the rates of that are extremely low in the dataset, if you think about it. Again, there's that 7% rate of AKI, the rates of dialysis, if I'm remembering correctly, are actually less than 1%. And so, again, if we use a very, it kind of goes back to some of the discussion of if we're using very low, if we lose definitions that aren't sensitive, then we're losing the opportunity to, again, kind of improve our processes. Again, I think it's really just, at the end of the day, we're seeing that facilities that have high same-day discharge rates aren't actually being, seem not to be, are not being penalized by removing them, removing that patient population from your actual metric. And I think further understanding that, again, understanding how facilities are triaging or considering which patients are actually being discharged same-day or not, and is it truly based on only AKI risk or the other factors that are playing a role, may help us understand why that isn't, that concern isn't actually bearing out in the data. Okay, great. What are your thoughts, Anezi, on adding a post-PCI time frame for the AKI definition? The question about, you know, creatinine increases later in a prolonged stay are completely unrelated to the PCI, yet count against this measure. Do you have any thoughts on that? Yeah, so that's, I guess, a mistake on my part to put that in the PowerPoint. So the definitions are that 0.3 increase within 48 hours or a 50% increase over 7 days. So there is, there are some time frames that are actually within the definition that was used. Yeah, we end up, again, kind of just, I think the hospital can be such a crazy setting, right? You make people NPO, they might have got contrast from another procedure, there's so many other things that kind of play a role in PCI and some of the AKI or other events that may occur during the hospitalization. And so to kind of try and control for that and make it as fair as possible, we only wanted to include that first or that initial PCI. And again, there are those stipulations that will have to be within the first, that 0.3 will be within the first 48 hours, and then the 50% increase over the first 7 days. Okay, thanks for that clarification. There's a question about bumps in creatinine. So why are patients with a 0.3 change, for example, from 0.7 to 1.0, not excluded from the risk model? The thought being that their creatinine is still normal. Right. Also a very good question. We kind of actually look at serum creatinine and the change in GFR with changes in serum creatinine. Like, for instance, if serum creatinine changes from 3 to 3.3, the change in GFR is actually minimal, whereas that same 0.3 change is actually quite astounding, or is much more magnified in that 0.7 to 1.0 change. Now again, we see patients that have a 1.0, and it's normal for them. But the change from 0.7 to 0.1 actually is significant. The change in GFR is much wider than in those kind of higher creatinine populations. That's one thing. I think the other kind of consideration is, I guess, I mean, we did look to see what rate of, I guess, how many patients were actually having these small changes where maybe their serum creatinine was like a 0.3 or a 0.4, and then it was going up to a 0.7. What portion of the population are we looking at? And that's actually a very, very small portion of the population with the small baseline serum creatinine. That being said, that GFR change actually is quite high. So that 0.7 to 1.0, that 1.0 isn't normal for that patient, and that GFR is changing drastically for that patient, even though for another patient that 1.0 serum creatinine is normal. And you think about the things that actually dictate a serum creatinine. So one thing is renal function, but we know that age, muscle mass, sex, all those things play a role in serum creatinine. And so 1.1 in an 80-year-old female that's 50 kilograms is a lot different than a 1.0 in someone my size. And so that 0.7 in, again, kind of that older female is normal for her, but a 1.0 isn't. And that 1.0 after her PCI is a decrement in renal function that we want to make sure we're capturing, because, again, it's associated with outcomes. Okay, thank you. I think we have time for one more question. So this is interesting. So your model used 2020 data, and Pat is asking if you think that since it was during COVID, this may have skewed the data at all. Did you take that into consideration? Yeah. I think there was one email. I think that talked about that, and really the thought process then was that renal function itself probably wouldn't change to, like, as long as the PCI was occurring, that the renal function itself probably wasn't changing much as far as the practices, you know, contrast, all that sort of stuff. I think for other metrics, it may matter more, but I think that is an important point. I mean, I think one of the things we could do is, again, that validation cohort was 30 percent of the population in 2020. We could easily make another validation cohort of the 2021 and just kind of make sure that our model performs well there. I suspect it will, just because, again, you wouldn't expect people's renal function to have changed because of the COVID era, and, you know, the effects of contrast and all that would change necessarily due to, you know, kind of the changes in the pandemic and lockdown, but it's something we can take a look at. We can definitely look at validating it in 2021. Okay. Thank you so much. I think I covered most of the questions. We will be around a few minutes afterwards if you have any other questions, but I want to thank Anezzi and Ellen for their time. This was a great, very informative presentation.

acute kidney injury

risk model

percutaneous coronary intervention

patient outcomes

quality improvement

dialysis

provider engagement

COVID-19 pandemic