Welcome back. We have talked about what is a meta-analysis. And now we have to ask ourselves the question, Why Do a Meta-analysis? Well, there are a couple reasons you want to do a Meta-analysis. Meta-analysis can help you to increase the power and precision if the studies your putting to a meta-analysis are homogenous. We will be able to detect effect as statistically significant with narrower confidence intervals. If the studies are homogeneous. And the meta-analytical result, which is the diamond on the bottom of your forest plot, will quantify the effect size as well as their uncertainty. By doing a meta-analysis, we will be able to reduce problems of interpretation due to sampling variation. And I will show you examples of what do I mean by that. And we will be able to look at all studies together. And asses the homogeneity or heterogeneity of the results. And quantify between study variation. Meta-analysis can also help you to answer questions not posed by individual studies. For example, factors that differ across studies. If you don't look at them together to begin with, how will you know that the studies are different? You will be able to look at the comparative effectiveness of multiple interventions for the same condition. And settle controversies arising from conflicting studies, as well as generating new hypothesis. Here is a famous example that looks at intravenous streptokinase for acute myocardia infarction. And the outcome here is the three month mortality. You will see there are many studies. I believe there are 38 altogether that have evaluated this question. But by looking at individual studies, and particularly those studies in blue, those are smaller studies. Most of them have a 95% confidence interval crossing the risk ratio of one, which is the null value. So many of the early studies are pretty small. They were not powered to look at mortality in three month. So, we're still uncertain whether intravenous streptokinase is effective by looking at those smaller studies. And there are five studies in spring. Those are the larger studies. And by combining all these studies together, which is the diamond on the bottom, you will see that it shows streptokinase is very effective in lowering the mortality in three month. So meta-analysis helps you increase the power and precision and quantify the effect size in uncertainty. And this question cannot be answered by individual smaller studies. This is another example where we can use meta analysis, or more specifically meta regression to examine factors that may differ across studies. And here on the plot, each bubble is one study. And again, the size of the bubble is proportional to the weight that the particular study is taking. And the example we're using here is the effectiveness of toothpaste. Whether the effectiveness of toothpaste depends on the baseline population levels of cavities. So the y-axis of the plot is the preventable fraction of new cavities, so the larger the fraction The more effective the toothpaste is. And on the x-axis you see the baseline cavities. So the question is lets say the more baseline cavities you have is using toothpaste more effective. And here we can see the regression line through all these studies which is reflected that there's a red line on this plot. And what I can tell from this plot is, it seems to me the more baseline cavities you have the more fraction of new cavities that the toothpaste can prevent. So you can use meta regression technique to examine factors that differ across studies. And answer questions that cannot be answered by individual study. Because for each individual study, you will only have one baseline level of cavities. In another example i'm going to show you is to use meta-analysis or network meta-analysis more specifically, to examine the comparative effectiveness of medical interventions. I use the example of medical interventions for our primary open angle glaucoma. As a matter of fact there are four different classes of drugs you can use and classes of drugs are color coded. Within each class there are more than one Drug. Okay, so, here we have 11 active drug, plus the placebo and no treatment. And the size of the circle, each circle represent one drug and the size of it is proportional to how many patients have been randomized to that particular drug. When there's a line connecting two dots or two circles, meaning that those two treatment have been compared against each other in a randomized controlled trial setting. So not all drugs have been compared against each other. For example, if you focus on the Upper left corner, you will see that Letanoprost has never been compared to Carteolol. Here is the question that clinicians need answered. When a patient comes to them, for a drug to lower their eye pressure. They have to pick from these four different classes, and from these 11 drugs, which one to prescribe. Okay? And can we do analysis to answer that question? The conventional pair-wise meta-analysis will only compare two drugs at a time. So, that analysis won't be able to tell you of those 11 drugs, which is the most effective. And then, we have to use network meta-analysis to answer that question. What is the comparative effectiveness of multiple interventions for the same condition? Can we rank those interventions? And what is the probability that a drug will be ranked as the best, second best, or the third best, or the worst? So, network meta-analysis is the technique to address that question. And we will talk about formally in a separate lecture. Systematic review and meta analysis can also be used to generate new hypothesis. And this is a long quote from a review published in the Cochran Library. Its not new to us that by the end of scientific paper that the authors concluded more research is needed. So for example here they say, well-designed, long term randomized trials are urgently needed. But don't stop there, you have to talk about in the new study, what should the authors include. So what should be the design for the new study? So the authors continue saying, ideally, studies aiming at comparing different treatment procedures should attempt to standardize all parameters potentially effecting the outcome. In particular, factors such as initial lesion size, the patients characteristics, tooth type and location, the operator's skill, clinical procedures, magnification devices, instrumentation and materials. So on and so forth. So by doing a systematic review, you will see the deficiencies in the existing literature or the existing evidence, and you can come up with the recommendation for the new studies. And the best design for the new study. When not to do a meta-analysis? Well, we have heard of garbage in, garbage out. A meta-analysis is only as good as the studies in it. Let's say all the studies included in a meta-analysis are biased or have deficiency in the design Hoarding. You're going to get a very narrow confidence interval around the combination of the biased studies, which is worse han the biased studies on their own. So, if you're throwing a bunch of bad studies into a meta-analysis you're going to get a very precise, but biased result. Beware of reporting biases. As we have heard from other lectures that not all studies are reported. And the studies with the significant results are more likely to be reported than studies with negative results. And people have talked about mixing apples and oranges. Well, it's not very useful for learning about apples, although useful for learning about the fruit. Again, this goes back to the aim of your systematic review. How broad is your research question? Are you trying to say this class of the drug works, or are you trying to say this particular drug, at this particular dose, for this particular patient group, is effective? So, there are different questions, and depending on the scope of your question, you're going to include different studies. They might be similar in one way or another but in most part, they will be slightly different. So studies must address the same question, though the question can and usually must be broader than the question posed in the individual study. What does a meta-analysis entail? So, to do a meta-analysis you have to answer all these questions. Which comparisons should be made? Again, this refers back to whether we're comparing two interventions or we're comparing more than two interventions for the same condition. Which study results should be used in each comparison? Are we using mortality of three months, mortality of one year, or quality of life at 24 months? You have to choose what are your outcomes. For each outcome, what is the best summary of effect for each comparison? So, are we using risk ratio, auth ratio, or the difference in means? We have to decide, are the results of studies similar when each comparison? And that can be done by plotting our studies together on a forest plot. As well as by examining the characteristics of the participants, of the interventions, the outcomes, as well as the risk bias of the studies included in your meta-analysis. And how reliable are those summaries? Again, that refers back to the risk bias of individual study. In this section we have talked about why to do a meta analysis and different aims or objectives of the study questions really determines what type of meta analysis your going to do. In meta analysis is not the solution for every single question in a systematic review does not have to have a meta analysis.