 We can have the next slide. I was going to say that I had a line all prepared about being near the end and how I'd be reiterating what everyone said, and then Mark took that line. So now we're into a Mandelbrot fractal, apparently. Something like that. But what I'm going to try to do is briefly go over some, comment a bit on some of what Mark just talked about, but really, next slide, please, focus on some of his future directions. Next. We've been talking about clinical decision support throughout this workshop, but I don't think we've really talked about some of the different types, and these are just a couple of examples of types, and there's actually probably more examples of different types. But I think these, maybe these three types of clinical decision support are what's important, some of the important ones when we're talking about genomic medicine. So we have reminders or sometimes order sets where you might be reminded to, for instance, obtain a particular test. We've talked a lot about the interpretations that could be explaining the meaning of results, but it may also include more details about the science behind them. And we also talked about things that might be alerts and recommendations at particular points in the clinical process. And I think some of the issues around those different types of clinical decision support will be different, which I want to go into. And we talked about other people in the past mentioned, talked about the importance of implementation. I'm probably going to focus on that and the issues of workflow. Why don't we, next please. So Mark talked about some of the challenges of CDS, the Genomic Medicine. One of the challenges, as we talked about before, Zach talked about it and is getting the data from the EHR. Zach mentioned the problems that some of the systems don't have the data. Other people have said that the data are not collected or they're not in the right form. They're not standards for what is being collected or how it's represented. And they may not be in structured form to be able to be used appropriately. I think it's important that we think about, and we've all mentioned this as a problem. And one of the suggestions was to work with the EHR vendors. But I think there's another group of stakeholders that we should be thinking of working with. And that's the clinical leadership who actually has some control over what is recorded and what isn't recorded in the EHR. Because if it's not recorded, it's going to be very difficult to get it out. So I think it's important as we move our PGX projects forward that we work very closely with clinical leadership. In terms of distribution, I'm not going to say much other than the clinical decision support consortium is using a service-oriented architecture approach, which does show promise. And I think Ken Kaomotos on our panel, and I'm sure he'll have something to say about that because he's also done seminal work in that area. Mark mentioned the question of what is the accuracy across some of the use cases that you'll be looking for in trying to study that. I think it's really important to study it because in most other areas where people are looking for processes that transcend different content, they tend not to find common processes or it's very difficult to find it. Often the content determines the differences in how those processes work. So when you develop the use cases, I think it's really important to understand what their content and context is and use that to study what is and what isn't generalizable across them. And again, we all want to see impact on outcomes. But again, the literature on non-genomic clinical decision support has shown much more of an effect on process and a more limited effect on outcomes. Partly, I think that's because the CDS is aimed at the clinician and between what the clinician does and what happens to the patient, there's a lot of room for variation. So I think in thinking about outcomes, we should think about outcomes but also target CDS to processes that are most likely to relate to outcomes and be very conscious of doing that so that we can actually look at both process and outcomes. Next slide, please. There are two additional challenges which have also been brought up today. One is the need for reinterpretation, revision, and maintenance. And that's really in terms of reinterpreting the genomic analysis, might be the patient data as well. And the CDS also will need revision as this is done. So that is a continual need to look at all of those things again. It's particularly important, and I thank Mark for actually mentioning this when we were discussing the slides in advance, that the versioning of the CDS is going to be very important because the decisions are being made on particular versions. So we need to attend to all of those areas. And then I want to spend a little time on some of the implementation challenges. Next slide, please. What I'm going to do is take the different types of CDS that I mentioned and talk about what some of the issues are, in particular focusing on workflow issues. So if we look at reminder and order sets, generally those are going to be automatically displayed to the clinician. And the clinician presumably understands the test and the results or whatever the reminder is and can choose whether or not to do it. So the key issue here is the workflow. Where in the workflow do these reminders occur? How are they done? Are they different actually for genomic or non-genomic medicine? I don't know if we know that, but studying not just how to get the reminders into the system, but how and when to display it is probably one of the most important things there. But that assumes that the clinician understands what the reminder is for. Next slide. In the interpretation issue, the interpretation type of clinical decision support, and that might be automatic or on demand. We talked about the info buttons. You could display things automatically in context or they could be available for the clinician to select and study in context. And what's important with that is basically to provide an explanation for the clinicians. But we're assuming now that they're unfamiliar with the data. Otherwise, they might not be seeking that explanation. And the key issue there is not so much workflow as much as ease of access in use. And the challenges of the kind of explanation that we have to provide make that a really key issue because, again, if it's going to be extremely time consuming, it's not going to be used well and it may not be scalable. I know in the, I meant to comment that the special supplement in genetics and medicine in the October issue addressed many of these issues. And one of the, and they also brought up many of these same points, but the point is if it's going to be scalable, it's got that you cannot have lengthy, personalized interpretations of all of the results as is currently mostly the case. So what we need to think about is how can we provide those interpretations in a way that makes it easy for the clinicians to find and to use. Next please. Again, we talked about the alerts and the recommendations and the workflow issue there is assuming that the clinician is knowledgeable about what the recommendation means. And again, it becomes ease of use, but also, again, and this is best practices for clinical decision support, does it include specific recommendations? Does it not just say that the patient is likely to have a reaction to a certain medication, but does it tell them the alternative? So all of these implementation issues need to be attended to as we actually get to implementation. Next slide please. If we talk about the portals and the use of portals in THR, which I think is actually a very good idea because certainly with the meaningful use requirements, they're going to become more common, or at least portals probably will, independent to THR through portals. THRs that are untethered may or may not increase. But again, the information complexity is really a challenge both for patients and for the physicians. So that we need to attend to how to display, how to represent the information on portals. We've talked about the difficulties in explaining things to patients where consent forms may or may not have made the impact they're supposed to, or patients don't understand consent forms. If we're talking about use automating some of these processes and displaying them for patients, I think a key research area to address is how we actually represent the data, especially for patients. And again, because some patients will not use the portals no matter what, we cannot rely on them as the only source of information. So as we go ahead and develop the ways of presenting results to patients in the portals, we also have to have alternative forms. Next slide. So I think those were my main points. My conclusions are that both CDS and patient portals for non-genomic medicine because of the meaningful use regulations are going to become a routine part of care. And that we can, that's a good approach to take to also use with genomic medicine. We can learn from some of the non-genomic CDS implementation approaches. Clinician education and the complexity of the information is a challenge for automation. And that's for clinicians and especially for patients. And what we have to do is plan for a regular review, reinterpretation and revisions at multiple levels. And that needs to be built into our implementation plans. And that's about it.