Welcome back. This is Kay Dickerson, and I'm speaking in the course Systematic Reviews and Meta Analysis. We're about to begin section F, on Analytic Frameworks. Now this discussion of analytic frameworks, I'm putting in there because I believe it's very important and it's one of the standards that the Institute of Medicine has proposed for systematic reviews, but in this class, because you already have plenty to do, we're making development of an analytic framework optional for your review. I encourage you to try it because it actually is a very helpful framework and my guess is ten years from now, you'll see them in all systematic reviews. But we're limiting how much you do here and so I'm telling you because I think it's important, but it's optional in your own systematic review. So, an analytic framework is just one of the names of what this thing is. It's also called a logic model, and what it does is it links the evidence and how it all fits together in the populations as it relates to outcomes. I'll show you an example, don't worry. But you'll also see terms like logic models, conceptual frameworks, influence diagrams, theoretical frameworks, those are all terms that are used to discuss analytic frameworks. So don't be confused if they're different words. They're really all for the same thing. You'll see analytic frameworks in the evidence-based practice reviews that are done here for example, at Johns Hopkins, and are funded by the AHRQ, the Agency for Healthcare Quality and Research. So why use one of these analytic frameworks anyway, to set up the question that you're asking? What it does when you write down what you're thinking of? How is your intervention going to affect treatment, how is that treatment going to affect the outcome? What other ramifications of the treatment might you expect? It helps you to clarify what exactly you're thinking about. What are the steps in how a person is treated, or diagnosed and then treated? And to help you think about the fact that you may have several questions here, not just one. And therefore that you need to address them separately, because they may be using different sorts of evidence, as I mentioned very early on in this lecture. So especially if you're addressing multiple questions in a review, you want to have analytic framework so that you can parse out the different questions that you're addressing and a complex chain of logic. So what are the components? Well, an analytic framework, as we've been talking about, specifies the population, the interventions or exposures to outcomes, and sometimes you'll see timing, settings and comparators put in. But usually it doesn't focus on that so much. And it clarifies these links between the intermediate outcomes and the health outcomes. Remember when I mentioned hemoglobin A1C, as it would be measured for example in this study of people with diabetes or who are at risk of diabetes. That could be considered by some people an interim outcome or a surrogate outcome, and not the actual outcome that you're interested in, which might be death. It might be amputation, it might be neuropathy or diabetic retinopathy, for example. So it helps you to find when you have a analytic framework, what's an interim outcome along the way, and what is the actual final outcome that you're interested in in your study. These analytic frameworks use arrows and boxes, and squares, and circles that tell you how everything is joined together. And if you're interested in constructing these things, you can look at an article by Cindy Mulrow in Annals of Internal Medicine quite awhile ago in 1997, where she explains how to do this. So here's a typical depiction of a research question. You'd have these boxes with the rounded edges and you might have a square box, each meaning different things. And these numbers represent the key questions that you're asking in your analytic framework or in your series of systematic reviews. And you might have interim outcomes or side outcomes, such as in two. If you're interested in reading more about this, you can read an article by Evelyn Whitlock in American Journal of Preventative Medicine, about ten years ago, 2002. So here is a sample framework and you might start with people who are at risk of a disease, let's say, women over 60 might be at risk of breast cancer, just by being women and over 60. They are screened using mammography. That's a question. Is screening effective in preventing outcomes that we don't want, such as breast cancer mortality? And you can look at both the effectiveness of screening, that is early detection, which is really an interim outcome or even the adverse effects of screening. Were women upset by results that meant they had to come back for another test, for example. You can also see that with early detection and treatment, sometimes with breast cancer people are treated who really wouldn't need to be treated because nothing bad was ever going to happen to them. Unfortunately, we don't know ahead of time who needs to be treated and who doesn't. So that's a tough one to address in a real-life setting. But your early detection then can lead to adverse effects of treatment. So you could treat someone with early breast cancer various ways and that person could have adverse effects simply from, let's say, chemotherapy, nausea and vomiting, hair falling out, various effects. Some forms of treatment, talking about bone marrow transplant for women with breast cancer could kill you, and actually even some of our forms of chemotherapy can kill you in rare situations, and even radiation could kill you over the long term if you follow people long enough. So you get an interim outcome, which might be detection of breast cancer, it might be the type of surgery a person has, it might be progression of disease. Those are all intermediate outcomes where the final outcome that you're interested in is mortality, or perhaps, something else depending on your disease. But by drawing it this way, you can see the different questions, the effectiveness of screening, the possible harm, the effectiveness of early detection, possible harm. You can look at intermediate outcomes, and you would have to define what they are. And you can look at the final outcomes that one might be interested in, reduced morbidity and mortality at a later date. So that would be a sample working format. Again, setting up your question this way, using an analytic framework can be important, just as asking your question in the proper way in those examples we showed to get proper data collection. You don't know what data you want to collect until you have your question properly formatted. The analytic framework helps you to decide what are the questions you're actually asking. What are the interventions, what are the exposures, what are the outcomes, what are the populations. And especially, it will help you to define the final criteria that you're going to use when you're selecting studies for your review. One of the hardest things, and we're struggling with it right now in a systematic review that we're doing, is deciding which data to be abstracted. On one hand, one never wants to abstract too much data because it takes a lot of time, especially checking. You want to do it independently and then check and there might be disagreement. But you also don't want to collect too little data. So making sure that you get just the right amount, like the three bears, too much, too little, just right. But there's no crystal ball that says what it is the right amount so takes a lot of discussion. But getting that question right and having an analytic framework are all steps that help you decide which data will be abstracted as well. So it's a very important part of the whole sequence of events in doing a systematic review. All right, so now we've framed the research question, we're going to have you talk in small groups about this and get your research question straight. As I said at the beginning, this will take probably more than one session, actually, because it's a very important part. And you'll see how many questions there are as you go along. But what we're going to talk about in lecture next is how to develop a search strategy for searching the bibliographic databases, the journals, and clinical trials registries, and all the other sources that are out there that maybe you can't search electronically, that you still have to search by hand. But still, you have to develop a strategy for what you're going to do. And that's the next step, and what we're going to talk about in the next lecture. Thank you very much. [MUSIC]
