Welcome to Natalie Jayram, who will be presenting Adjustment for Risk in the Impact Registry. Dr. Jayram has been involved with IMPACT since she was a fellow in training. Now she serves on the NCDR Oversight Committee and our IMPACT Registry Steering Committee. She's been attending in the Department of Pediatrics and Division of Cardiology at Children's Mercy Hospital in Kansas City, Missouri, and she also is an assistant professor of pediatrics at the University of Missouri, Kansas City. In addition to her clinical work, Dr. Jayram has received training in CV outcomes and has research goals to improve quality of cardiac care for patients that are, for pediatric patients. Welcome, Dr. Jayram. I'd like to thank the conference organizers for the opportunity to speak here today. Over the past several decades, there has been tremendous progress in the care of patients with congenital heart disease with innovations and improvements in care across domains, including innovations in technology that have enhanced diagnostic imaging, advancements in surgical technique, improvement in pre and post-operative care, and importantly, in cardiac catheterization. In its infancy, cardiac catheterization of patients with congenital heart disease was primarily pursued as a means of diagnosing different types of heart defects. However, the past couple of decades have seen tremendous progress in this field of congenital catheterization, and what was once used primarily as a diagnostic tool has evolved into a means of treating certain congenital heart lesions, and in some cases, obviating the need for surgery. With advancement in congenital cardiac catheterization, the need to study and understand the outcomes of patients undergoing catheterization has grown. Efforts to pool data across institutions in the way of multicenter registries date back as far as the 1980s. The valvuloplasty and angioplasty of congenital anomalies, or VACA registry, was a 27-hospital registry where the participating hospitals contributed important information about short-term safety outcomes. The Congenital Cardiovascular Interventional Study Consortium, or CCISC, followed the early VACA efforts and was an investigator-initiated effort to provide outcomes data for percutaneous stenting of coarctation of the aorta. Subsequently, the Mid-Atlantic Group of Interventional Cardiology, or MAGIC, formed and expanded to include 18 centers, including centers in both Europe and the U.S., and sought to understand long-term outcomes data for several catheterization procedures. And finally, the Congenital Catheterization Project on Outcomes, or C3PO for you Star Wars fans, was a multicenter congenital catheterization registry whose work has evolved into some extremely important quality improvement initiatives, including some initiatives that formed the basis of the IMPACT risk model, which is the primary focus of our discussion here today. The creation of the IMPACT registry stemmed from a committed group of pediatric cardiologists who acknowledged that a registry collecting catheterization data from centers across the U.S. performing catheterization on congenital heart disease patients would allow for improved understanding of the outcomes of these procedures through use of pooled data, and inclusive of centers, whether they be small, large, freestanding children's hospitals, or a pediatric program embedded within a larger health care system. As you can see from this graph, IMPACT has grown substantially since its inception in 2011, and now includes centers, about 90 to 100 centers, over the past several years. The IMPACT registry was primarily created as a quality improvement tool, and also serves as a rich repository that has helped our community understand outcomes for congenital heart disease patients through not only the quality improvement programs, but also the research that has come out of this registry. These publications have included two papers that have been dedicated to creation of a risk adjustment model for IMPACT. I wanted to provide a brief background of risk adjustment and why it is important for studying outcomes across hospitals for those who may not be familiar with the concept. As we know, regardless of the medical condition that we are discussing, there's large variation between the types of patients treated and types of procedures performed across hospitals. If we want to study how outcomes differ across institutions, developing a risk adjustment or risk standardization model that accounts for things like age, severity of illness, and medical comorbidities can help us to account for these differences in patient characteristics. Moreover, when we are talking about congenital catheterization, this encompasses a huge number of different types of procedures, each of which is associated with varying risk. Accounting for the risk of the procedure is another important aspect of risk standardization as it pertains to congenital catheterization. To state it more succinctly, risk standardization is a method that allows for more equitable comparison between hospitals by accounting for individual patient or procedural risk factors that put some patients at greater risk of experiencing an adverse event compared to others. Risk adjustment in some ways makes me think about figure skating. In figure skating, scores aren't simply generated on the number of times a skater falls. If this type of scoring system were used, someone like myself, who can barely make it around the rink, could achieve a score similar to an Olympian. In figure skating, the scoring metrics account for the complexity of the routine. If a skater is attempting a high-difficulty performance, that is accounted for in their score, as it is acknowledged that a more difficult routine puts the skater at a higher risk of an error. Similarly, if a hospital is knowingly performing a procedure on a high-risk patient who has a baseline risk of a worse outcome due to age or medical comorbidity, it wouldn't be fair or accurate to compare these outcomes to another hospital who is performing a low-risk procedure on a generally healthy patient. To illustrate the importance of risk adjustment, we can look at a graph like this. Here we see the outcomes from several hospitals contributing data to impact. Centers at the left end of the spectrum would appear to have very good outcomes, and then centers on the right appear to have outcomes with event rates upwards of 20%. What looking at the crude, unadjusted event rate doesn't tell us is whether or not these centers at the left end of the spectrum are really doing much better in terms of their performance, or is it just that these centers are treating a much lower-risk group of patients and performing procedures that are lower risk? Without appropriate methods for addressing the acuity of the patient population at a particular hospital, we may reach inaccurate conclusions regarding a particular hospital's performance. Towards this end, there has been a big push nationally to develop and apply risk standardization methodology, and organizations such as the ACC and the American Heart Association have helped to create and endorse criteria and standards for these types of models. Before we get into the specifics of the impact risk model, I wanted to highlight some of the factors that these guidelines emphasize, as these are things that we had to carefully consider when it came to developing the risk model for the impact registry. First, these guidelines emphasize the need to include clear and explicit definitions of an appropriate sample. This concept applies to both the sample of patients and the sample of providers or hospitals. For the patients, there should be a clear, reproducible, and appropriate method of identifying the people who should be included in the measurement cohort. And for the providers or hospitals, it is important to assure that a patient is appropriately assigned to a particular provider or hospital, which becomes even more important when there is transfer of care between institutions. These guidelines also reinforce the importance of clinical coherence of the model variables. This specifically relates to using clinical judgment and insights from the published literature to help guide the selection of variables. Using clinical judgment and attention to the medical literature should minimize the influence of variables that may reflect idiosyncrasies of individual data sets. These guidelines specifically reference that use of variables that convey non-clinical information like race, ethnicity, or socioeconomic status should generally be avoided as the effect of these variables may be mediated through quality of care, and consequently adjustment for these factors could condone the results. This third bullet refers to not only building a risk model with data that is of high quality, but also data that is current or timely. This next factor is an important consideration in the development of a risk adjustment model is designation of a reference time. What this is referring to is ensuring that there is appropriate time before which covariates are derived and after which the outcome is measured. Events including complications that occur after the reference time or starting time should not be included as covariates in a risk adjustment model. For example, if a patient develops a medical condition in the hospital and this develops after your designated reference time, this should not be included as a covariate. The next point refers to selecting an outcome that is meaningful, whether this be meaningful to patients or providers, and then observing a designated period of time before the outcome assessment. So, if in-hospital outcomes are assessed at time of hospital discharge and a particular hospital discharges patients quickly or transfers them to another facility, this can influence the assessment of the outcome. Assessing for the multilevel organization of the data is also critical. In impact and other NCDR registries, patients are nested within hospitals. In evaluating hospital outcomes, it is the experience of the patients clustered within the hospital that is evaluated. The point is to determine the degree to which the hospital influences the outcomes. Thus, the assumption is that outcomes of patients within the same hospital may be different than outcomes of the same patient in a different hospital. And lastly, these guidelines speak to the importance of being transparent about the methodology that was used for development of these models and how well the model performs. While for some risk models, the model beta coefficients are considered proprietary, transparency regarding the model performance will help to improve public confidence that the model is fair and accurate. As we began our development of the risk model for impact, not only did we use these guidelines as a foundation, we carefully considered work that was relevant to the topic and upon which we could build. Some of the first work done in this space was an effort led by Lisa Bergersen from Boston Children's Hospital and colleagues and was entitled the Congenital Heart Disease Adjustment for Risk or CHARM Methodology. This methodology used the eight-center C3PO dataset that I spoke about earlier. Using this dataset, the investigators identified characteristics that could be associated with an adverse event during or after cardiac catheterization and then built a model which incorporated the variables that they determined to be most predictive. This CHARM model ultimately incorporated patient age, procedure type risk category, and hemodynamic vulnerability as being critical for risk standardization for occurrence of a major adverse event. Procedure type risk category is probably an unfamiliar concept to those who don't work directly in the congenital catheterization space. Basically, this is a concept of grouping different types of congenital catheterization procedures based upon perceived riskiness. Because congenital catheterization encompasses such a large number of different procedures, it isn't possible to adjust for each procedure individually. Procedure type risk groups are a way to group procedures based upon perceived risk and adjust for the risk group rather than adjusting for each procedure individually. And then the last variable included in the CHARM risk model is hemodynamic vulnerability, which was defined as a patient whose hemodynamic parameters during the catheterization met a threshold value for each of the four hemodynamic variables I have listed here. The next step in developing a risk model for the impact registry was to see if some of the work that had been done within CHARM could be directly applied to impact data. In the work I have referenced here, we began creation of an impact risk standardization model for occurrence of an adverse event. In this work, several aspects of the CHARM model were directly applied to impact in order to see if the variables discriminated well with an impact. Specifically, the hemodynamic vulnerability indicators and procedure type risk categories that had been derived in CHARM were applied to impact. While overall this preliminary risk model had reasonable discrimination, we found that the procedure type risk groups and hemodynamic vulnerability indicators derived in CHARM did not perform optimally, likely in part because these variables were developed in a different data set and were developed using a different definition of major adverse events. For these reasons, we felt it was necessary to refine and solidify the risk standardization methodology for impact, including development of new procedure type risk categories and markers of hemodynamic vulnerability using data from impact. To accomplish this task, we used data from Impact Version 1 and looked at all diagnostic and interventional cardiac catheterization procedures from January 2011 until March 2014. Our model endpoint was occurrence of a major adverse event, which included each of the items listed on this slide. And I will point out that while we could not definitively link each of our adverse events to the cardiac catheterization procedure, we did attempt to select adverse events such as device symbolization, for example, that were likely to be a consequence of the procedure. For the development of our procedure type risk categories and markers of hemodynamic vulnerability, we did this using a combination of empiric data as well as expert opinion. And then for model development, we used multivariable hierarchical logistic regression with 70% of the study population designated as the derivation cohort and 30% used for the validation cohort. In order to test the predictive ability of our model, we used a Model C statistic. The C statistic, or concordance statistic, is an indicator of the model's ability to predict the outcome of interest, in this case, major adverse event. Values range from 0.5 to 1. A C statistic of 0.5 would indicate that the model's ability to predict the outcome of interest is no better than chance, whereas a C statistic of 1 would mean that the model perfectly predicts the outcome of interest. From January 2011 through March 2014, we identified 39,725 cardiac catheterization procedures from 74 centers. The event rate in the derivation and validation cohorts was similar at 7.1%. As I said, one of our primary objectives was to recreate the procedure-type risk categories and markers of hemodynamic vulnerability that had initially been created in CHARM with an impact. Using a combination of empiric data and expert opinion, six categories of procedural risk were created. As you can see, the risk of major adverse events increased with increasing procedural complexity. There's a lot on this slide, but I included this just to give you a sense of the types of procedures included in each of the six risk groups. So you can see that, for example, a diagnostic catheterization in an older child is in a lower risk group than a diagnostic catheterization in an infant, and an infant is in a higher risk group than a neonate. For valvuloplasty, for example, aortic or pulmonic valvuloplasty in infants less than 30 days was placed into a higher risk group than in older children. And a device closure of an ASD or PDA is in a lower risk category than a device closure of a VSD. And while these were things that we suspected based upon expert opinion, they were also supported by empiric data. For our other goal of creating new markers of hemodynamic vulnerability, we identified six markers of hemodynamic vulnerability. As you can see here, four of the markers were similar to the original CHARM model. The thresholds used in CHARM were tested, and the same thresholds held up in impact. But in addition to the four markers of hemodynamic vulnerability in CHARM, we identified two additional markers, which were elevation in pulmonary vascular resistance and elevation in pulmonary to systemic flow ratio. Finally, in terms of our final multivariable model, the variables that were identified as being critical for risk standardization and that were included in the final model were procedure type risk group, hemodynamic vulnerability, renal insufficiency, coagulation disorder, and single ventricle physiology. Our final multivariable model had a C-statistic of 0.76, indicating very reasonable model discrimination. This slide shows model calibration in the validation cohort and is designed to determine how well the model performs across a range of outcomes. For well-calibrated models, the slope should approximate 1 with an intercept of 0, both of which were tested. From a practical standpoint, how is this risk model used? This is a screenshot from a sample quarterly report that impact participating sites receive. You can see that the center is presented with their own risk standardized adverse event rate along with the 50th and 90th percentile rates, and the sample center here had a risk adjusted adverse event rate of 9%, giving the center an opportunity to identify areas for improvement in order to decrease this rate. This slide shows the same graph that I showed you earlier showing the unadjusted or observed rate of major adverse events across sites, but now on the right we can see the range in outcomes following risk adjustment, and we can see that this range has narrowed following risk adjustment. Because these event rates on the graph on the right have been risk adjusted, we can be more confident that the centers on the left side of the spectrum are actually the highest performing centers, even when taking into account patient acuity and procedural complexity. And recognition of these high-performing centers is important because this gives us the ability to identify best practices at the high-performing centers and disseminate them across centers caring for children with congenital heart disease undergoing cardiac catheterization. There are certain issues with developing a risk model that are somewhat unique and which presents a unique set of challenges when developing a risk model for impact compared to developing risk models for other NCDR registries. First, as I alluded to earlier, congenital catheterization encompasses a huge number of different types of procedures, each of which has a unique risk. Unlike some of the other registries where we're talking about creating a risk model for one type of procedure, IMPACT captures data on over 200 procedures. I know the font on the table that I included here is small, but this is just a sample page listing a subset of the procedures that can be reported to IMPACT, and this is not even a third of the total procedures for which there's data collected. Because we have a smaller number of patients in the field of congenital heart disease, we want to maximize available data, so that is one of the reasons we chose to create a risk model that encompasses all types of congenital catheterization procedures. The benefit of this is that we have more data upon which to build the model. The tradeoff is that you're potentially sacrificing some of the predictive ability by lumping together procedures that may fundamentally be very different. We did try to account for this, however, by using the procedure type risk groups to lump together procedures that were thought to represent similar risk. There are always discussions about whether it would make sense to create a risk model for some of the most commonly performed congenital catheterization procedures, and the risk model would just be specific to that procedure, but even for procedures that are performed commonly, this leaves us a much smaller cohort for model development. One of the challenges with creation of a risk model for congenital catheterization is also that, fortunately, cardiac catheterization on the whole is relatively safe. That isn't to say that there aren't patients or certain procedures for which there is a significant risk, but by and large, a major adverse event following catheterization still occurs relatively and frequently. Here's a list of the frequency and types of adverse events that occurred in our study cohort. Given the relative safety of these procedures, there's always discussion about how meaningful risk-adjusted adverse events rates are as the sole indicator of quality of care, and there's always discussion about the addition of other metrics that may also be important besides risk-adjusted adverse event rates. One of the other challenges is always in determining what outcomes we want to include in our definition of a major adverse event. I spent some time discussing the guidelines developed by Harlan Premholtz and colleagues for creation of risk models, and one of the key points was to include outcomes that were relevant for both patients and providers. We had some good, healthy discussion as we developed our risk model regarding whether all of the outcomes included in our model truly constituted major adverse events and whether there were other outcomes that could be added or eliminated. One of the most controversial topics was inclusion of death into our risk-adjustment model as a major adverse event. When we initially started this work, there was a strong contingency that felt death should not be included as a major adverse event because, as I discussed, congenital catheterization is generally safe, and many felt that death following cardiac catheterization was just so unlikely to be related to the cardiac catheterization that inclusion of it in the risk model was unfairly penalizing operators and hospitals. This is a study done by Carl Backus and colleagues where their aim was to get at this issue and determine how often deaths in patients with congenital heart disease undergoing catheterization could truly be attributed to the procedure and how often there were other factors at play that were more likely to be associated with patient death. In this study, the authors looked retrospectively at 14,000 cardiac catheterization procedures, and within their cohort, the mortality rate at 30 days was 1.9%. They then went back and examined the medical records of these 279 patients to see if they could get a better understanding with regards to the circumstances of death. And what they found is that only in about 10% of cases was the death within 30 days of catheterization attributed to the catheterization itself. For the remaining 90% of patients, there were other factors at play, including non-cardiac comorbidities, cardiac surgery in the same hospitalization as the catheterization procedure, or high risk features that were either present before the cardiac catheterization or after the cardiac catheterization, keeping in mind that this group here, the high risk features were not attributed to the catheterization procedure itself. Given this data and the personal knowledge and experience from those in the working group, we carefully considered whether it was appropriate to keep death as an outcome in our risk adjustment model. And ultimately, the conclusion that we came to is that the intention of our risk adjustment model is to improve the quality of care of patients with congenital heart disease. And if a particular center has a higher than average risk adjusted adverse event rate, and that event rate is being driven by higher mortality, we felt that information was critically important for the center to have, even if they ultimately evaluated the deaths and determined that the deaths weren't attributable to the catheterization, but rather other circumstances or processes of care. As I noted earlier, the relative safety of congenital catheterization has led some to question whether we should be expanding our outcome reports and looking at other things like procedural efficacy. The challenge with this particular issue is that for some interventional procedures, there may not be commonly accepted definitions of what constitutes procedural success. For some procedures, this may be obvious, but not all catheterization procedures have an agreed upon definition of what constitutes success. And if metrics outlining procedural success aren't feasible, we've discussed are there other outcomes that may provide important insights into the quality of care being provided by an individual center. For example, things like failure to rescue. Failure to rescue for those who may not be familiar with the concept is the idea that a failure to prevent a death resulting from a complication of medical care or from a complication of underlying illness or surgery, in this case catheterization. In this study led by Michael O'Byrne from Children's Hospital of Philadelphia, over 77,000 catheterizations at 91 hospitals with an impact were examined and instances of death within two days following catheterization and following a catheterization related complication were identified. What was found was that a higher catheterization volume was associated with less likelihood of failure to rescue, but higher catheterization volume at the hospital was not associated with odds of occurrence of a major adverse event. The odds of adverse events, however, were associated with patient and procedural level factors. Furthermore, there was no correlation between failure to rescue and pooled adverse events. Moreover, hospital ranks by catheterization volume and failure to rescue were associated with the largest volume hospitals having the lowest risk of failure to rescue. Based upon this, it was felt that failure to rescue may provide a complementary measure of cath lab quality. In terms of where we still need to go with this work, as I said, this model was developed in IMPACT Version 1. There are several important updates in IMPACT Version 2, which include the addition of some additional procedure types and some changes to the reporting structure for major adverse events. It would be important to at least validate performance, if not update and refine the model using Version 2 data. As I mentioned on the prior slides, there are ongoing discussions about additional outcome variables that would be important to consider and which may augment the utility of the current quarterly reports that IMPACT participating sites receive. In conclusion, we have built upon prior work in order to develop risk standardization methodology for the IMPACT Registry that is currently being incorporated into the IMPACT quarterly metric reports. This type of feedback to sites has the potential to improve care for congenital heart disease patients in need of cardiac catheterization if individual centers use this data to understand how their outcomes compare to other centers and work to identify mechanisms to improve their risk-adjusted adverse event rates. And finally, the current risk model will likely need ongoing updates and refinements to really maximize its utility. With that, I'd like to conclude my presentation. I thank you for your attention and I'd be happy to take questions. Okay, Dr. Jaram, thank you so much. That was an excellent presentation. I worked in IMPACT after the model was added to the reports and I didn't learn, I didn't know a lot about some of the development and thinking that went through the model. So that was really helpful, really helpful presentation. Can you comment? There was an interesting discussion about your decision on including death in the model or not including it. That was, I thought, really interesting. But can you comment on why death is equally weighted as an outcome as compared to some of the other outcomes like a valve recurge? Would you ever think about weighting death differently as an outcome? Yeah, Susan. So, you know, again, this was a really hot topic and one that was, generated a lot of vigorous debate amongst all the members of the working group and even after model development. It's something that's asked about very often whenever the topic of the IMPACT risk model comes up. And, you know, just to reinforce, I certainly understand where a lot of the reservations about inclusion of death come from. You know, as I showed you in some of the data in my presentation, death is attributed to the catheterization so infrequently that I certainly understand that hospitals feel like, are we really assessing markers of catheterization related care by including outcomes that may not necessarily be attributable to the catheterization itself. But for our outcome, we chose a composite outcome in which all of the outcome measures are weighted similarly. And I am not sure from a methodologic standpoint with a composite outcome like that, whether or not we would be able to give differential weight to certain outcomes more so than to others. So we chose to, if an individual or if a patient had occurrence of any of the adverse events, we chose to have that all be part of the composite outcome and not separate them out. So we call it any or none in that case. Correct. Okay. Here's the next question. I had a couple jotted down. This model only includes four variables to predict the outcome. Do you think that's sufficient to capture the risk of these complex patients? Yeah, it is. Relatively speaking, there are fewer variables than may be present in some other risk models. I do think with a model C statistic of 0.76, we can say that the model really does a very reasonable job of capturing risk of the individual patients. And I believe should give centers confidence that the risk standardized adverse event rates that they are receiving capture the relative risk of the patients that they are seeing and the procedures that they're performing. That said, one of the inherent challenges of working with observational data and building these types of risk models is that unmeasured confounding is always present, even for the best risk models. So while I think that the present model is a good one, I think that there is always room for improving predictive ability. And when we work on future iterations of the impact risk model, we certainly wouldn't explore whether or not there are important characteristics that we're not including. Yeah, yeah. That's a good answer. And the C index is really, really good. I mean, the two of the variables, I call them like a matrix. So they have so much value in that matrix. It's not just a yes, no, like the patient had diabetes. Correct. Yes. Do you know of any variables or new data elements we need to add to impact in the future that we should consider in model revision? Do you know, just as this congenital heart disease is such a newer avenue in cardiology, is there something that you think we're missing that might stand out to you? It's a good question. And I certainly know that those who were involved in the creation of the impact case report form were very thoughtful and intentional about the variables that were included. And it's always a trade-off trying to collect that information that you think is going to be informative, but also recognizing that this is a huge data collection burden on the individual sites. And so how do you balance collecting those critically important variables with not being overly burdensome to sites and collecting variables that may or may not have the clinical utility? I personally feel that the current variable list as it stands is pretty comprehensive. For patients with congenital heart disease, the anatomy and physiology is so broad. So even two patients who have the same diagnosis, say tetralogy of fallot, can have really different physiologies. And so one of the challenges in our field is trying to place patients in particular boxes and diagnosis alone often won't do it. So I think impact does a really nice job of maybe capturing some of the hemodynamic variables that can sort of capture an individual patient's physiology a little more closely than just the name of their diagnosis alone would. I am certain that as the impact version 3, if and when that rolls out, there will be more thought put into seeing if there are any important clinical variables that we have not included in the registry. But I think the current case report from as it stands is really pretty comprehensive. That's good. Good. Thanks. Okay. Another question is that some centers with low volumes feel the model unfairly penalizes them because even if they have perfect outcomes, like no adverse events, their risk adjusted rate will never be zero. That's a good question, Susan, and one that also does come up a lot when we present our risk model data. Because as you point out, centers that have smaller volumes are, if they, let's say a center does 10 cases per year, that center, even if they have zero adverse event rates, may not necessarily be considered a top performer. And the reason for that is because they have less data to contribute, and so we have to quote unquote borrow information from the larger sites with more experience. And I know that's sort of a hard pill to swallow, maybe for the smaller programs, but I guess I would look at it as that it cuts both ways, right? So that same center that in a particular year, one year may have had zero adverse events and may not necessarily be considered a top performer just because they weren't able to contribute enough, as much data. And so we can't be as confident in their outcomes over a particular year. That same center, the next year, if they had two adverse event rates out of their 10 cases, so an adverse event rate of 20%, I don't think it would be, and that those two adverse event rates were because of bad luck and not bad care, I don't think it would be fair to attribute a 20% adverse event rate to that particular center as well. So I think it cuts both ways. That's right. That's right. And we see those swings that a smaller center may have a zero or a really high adverse event rate. And one thing I've always said is you can't draw a conclusion that you're better than, you know, a big center because your rate is zero. So that's, you know, none of the, the risk standardized rates are never zero. We've, you know, even physicians have trouble looking at that and understanding it because their risk standardized rate is higher than their unadjusted rate. And they get upset about that. Yeah. Okay. Last question. Centers with large volume, you know, they have a higher numerator and a higher denominator. They sometimes feel like they will be unfairly have a higher rate. Do you have any, any comments for them? They feel like sometimes the model over, over predicts and gives them a higher rate. Yeah, that's an interesting question. And I don't know that we have done that particular type of analysis with an impact. So it would be something that would be interesting to look at in the future. But I would say that I, I think for those larger centers, we can be a little more accurate in our predict that in our ability to predict their, their risk adjusted major adverse event rate, because the more data that you contribute, the more kind of precise and accurate we can be in our predictions. And so I think for those larger centers, they should feel like the risk of predicting incorrectly is, is certainly less the more data that you contribute. And so they are, in some ways should have more confidence than, than a small volume center. Yeah. Yeah. And we, we actually studied that in our cath PCI registry because the sites, the high volume sites said they always felt like their rates were higher, higher than their risk standardized rates were higher than they thought they should be. And we, Dr. Peterson at Duke studied just as kind of a subset of large volume centers and PCI. And he found that the model actually almost over predicts and they had a lower, a lower rate than their risk standardized rates were lower than, than, you know, when you, as you compare them to all centers. So it almost over, over predicted for them. Interesting. I'm not, wasn't familiar with that work, but that I will have to look at that and think about for us as we move forward with impact. Yeah. Okay. I think that's it. Dr. Jayram, thank you so much for helping us learn more about the impact model. We're looking forward to improving it in the future and making some adjustments to it, but we appreciate your time. Thank you so much for the opportunity to speak here. I appreciate it. You're welcome. Thank you.

Dr. Natalie Jayram

Adjustment for Risk

IMPACT Registry

catheterization data

congenital heart disease

risk adjustment model

patient characteristics

procedural risk