So, in this section, we'll tackle some of the questions you should address when you thinking whether you should put in place the intervention of decision support. I'll be talking mostly about decision support. So, I'll be talking about interventions like these reminders and alerts. But this little panel comes from a much larger set of options that the World Health Organization put out as types of interventions for healthcare providers. We saw lists like these back in the intro module, when we were talking about what modules there are for health environments from Kellogg's to purchase. This listing goes much broader than what we saw before. So, we have health care providers, but it's also health system managers, it's even the patients, and the data services. What I want you to be thinking about in this whole course, is not just the basic function that these modules may be providing, but is there any smarts in them? Are there any decisions that a module wants to support, and as we'll see in later part of this course is, how do you support that? So, should you build or buy it? So, there are two questions here, right? It's whether you need the decision support, the intervention or not? Then should you build it or buy it? I call that a triage decision but really it's two-step type of decisions. Then, you want to be thinking about benefits and then costs. So, the benefits are quite clear, is the patient benefits depends on the number of patients being benefited, it depends on the number of times that the situation comes up for the patients, it depends on the gap between where you are today and where you want to be, it depends on how bad the thing is that you want to fix. In the case of decision support, generally, these are clinical outcomes like death, or morbidity like stroke. So, for instance in the case of drug allergy, how many patients are there who get medication orders at all? What proportion of those patients have drug allergies? What's the difference between the rate of drug allergy bad outcomes today and where you want to be? That would be zero. Then, you don't want any mortality morbidity? None is where you want to be heading. So, from a clinician's point of view, you want to deliver higher quality care at a huge sense of professionalism, you also want efficiency, you want the decision support to help you, not to hinder you, you'd like to improve your personal quality metrics scores. In the case of drug allergies, you want to order quickly, you don't want to hear about the drug allergy like two hours later, you have to go back and fix it and change it. Meanwhile, the patient isn't got the medication, and you certainly do not want a drug allergy bad outcome which is called a "Never" event, both because you want never to happen and also because Medicare will not reimburse you if it happens. The organization, similarly, their focus might be on the quality scores, they also worry about regulatory compliance, they want to lower their liability and again in the case of drug allergies, they again do not want never events. Then, that's strictly financial. They also do care about patients and they don't like hurting patients, separately from the legal and financial implications. But you have to think about costs, and there are basically two costs. You might think about it as the upfront costs and the ongoing costs. So, the upfront costs is the difficulty in doing it at all, the difficulty in just developing it, or the complexity of putting into place, of deploying it and then, the difficulty in maintaining its use. Then in terms of maintenance, we have these issues of false negatives and false positives. So, false negatives, is a case where the machine misses the case that was designed to catch. So, if in fact, you have a drug allergy decision support and a patient with LDA penicillin still gets prescribed penicillin and suffers a bad outcome, that is a missed case. On the other hand, you worry about false positives. If you keep on alerting the doctor after a while they stop listening to the alerts, it's the boy who cried wolf. Every hospital in the United States today is having this experience. The percentage of alerts that are overridden either due to false alarm or to the boy crying wolf is very, very high. Folks from Hermann Memorial, who won the Davies Award several years ago, for having a great system of wrong, they provide this heuristic for how to think about all of this. So, it talks about the patient impact, the organizational impacts, the clinician impact, just as we've been talking about the last few slides, the number of patients positively affected. Again, like we discussed and the gap between where we are and where we are going just like we discussed. Then, the difficulty of addressing the objective, and the cost of addressing objective. Now, for fun I thought, can we put this heuristic score? It's something that's actually computational because after all, we are medical informaticians. There is something called a decision analysis tree. There are decision trees that are used for classification and we will get to that later in the course, but this is a tree that helps you lay out what the decision problem is, and then actually helps calculate what the right thing to do. The way this works is you say, you start off with the left-hand side, you say, okay, what's the context that we're in? Whether or not to deploy? When a certain decision support intervention? Then, you can decide to deploy or not, if you don't deploy, then a whole bunch of people might get affected, and otherwise they will not. If you do deploy, you still have a question about whether the physician will listen to the decision support provided or not. If they do listen, then the rate of bad outcomes should be lessened. So, you can see an age limb at the tree, above the limb you see a label and below the limb is a probability. What you don't see on the right hand side is, how do you assess the value of the outcome? You talk about the cost, you're talking about the number of patients affected with a bad outcomes. There are a number of ways of valuating the outcomes which are not going to go into today. If you do this, I just want to show you that you are actually representing the same things that we saw in the heuristic. So, we've now added the variables that are in this network. With the left-hand side, you see the cost of addressing the objective, between the two limbs you see the gap between as isn't to be, and then you see three variables. The clinician impact at that point as well as the variables use later on the proportion of patients being impacted in the level of impact. So, you can put numbers in, in this case, we're simply looking at the number of bad outcomes and you can see that if you don't deploy, you get 1,000 bad outcomes, if you do deploy, you get 700, doesn't eliminate them altogether, but the gap is 300 and that's a good thing. We like reducing 300 bad outcomes is good. You can also get the same thing for dollars. So, you can see that, if you don't deploy, you're going to end up spending $10 million. If you do deploy you're going to end up spending $7,050,000, which is a combination of the deployment cost plus the cost of each bad outcome. The differences almost three million dollars. So, now having done that assessment, if I ask you, is it worthwhile to put this intervention to place? You could say, hello, its cost benefit is terrific, I'm basically saving three million dollars, its cost saving and I'm getting a positive effect. If you want to do an ROI, you're getting like 59 times the return on investment. It turns out for the actual decision-making ROI is really not important. ROI is great if you're investing in a company, you want to see if I put money in, how much money am I going to get back when the price goes up whatever, but for decision-making, it's really the cost-benefit and maybe the cost effectiveness that matter. So, that's a formal approach, most places will not do formal approaches like I just showed you. Generally, there is a committee that does the heuristic qualitatively. I simply wanted to give you the tools that, if you really wanted to see what the number, how they really play out, there's a tree for you. That might be helpful if you have more than one option to look at. So, if you have five people with different options, each one of them are there only because effective or cost saving. But, you want to compare them against each other, and you may want to do this. When you get the option of doing each one a little bit, which is called portfolio, doesn't really work too well decisions support, either do it or not do it. So, these are tools for you to figure out how to fit whether that you should go ahead with this intervention of this decision support, and this applies whether you're building or buying.
