This site is intended for healthcare professionals


Join us for the 8th session of our N Mra Systematic Review and Meta Analysis Series, led by Doctor Connor Gillespie. This session is the second part of a series examining how to conduct a meta-analysis. Doctor Gillespie is a highly accomplished medical professional with experience in neurology and neurosurgery, and a passion for teaching. During this session, he will demonstrate how to employ two popular software programs, Cochrane Revman and R studio, to conduct a meta-analysis, focusing on comparative outcomes. Additionally, he will walk participants through the process of translating their theoretical knowledge into practical application. This valuable, interactive training will equip you with essential tools for your professional development in the field of medical research.
Generated by MedBot


Delve into the NMRA Academy Teaching Series, an enlightening and engaging educational program for those who wish to learn more about how to run systematic review and meta analyses.

This series will be carried out by experts in the fields and by the NMRA committee, and we will be providing you with all the tools needed to be able to carry out your own SRMA.

Join us for this 10-lecture series:

1. Introduction and refining your research question

2. ⁠Writing your protocol and selecting inclusion and exclusion criteria

3. ⁠Creating the search strategy

4. ⁠Screening

5. ⁠Risk of bias assessment

6. ⁠Data extraction and synthesis

7. Meta-analysis part 1

8. Meta-analysis part 2

9. ⁠Interpreting results and writing your paper

10. Getting ready for submission: ⁠referencing and paper formatting

Learning objectives

1. By the end of this session, learners will understand the purpose and principles of conducting a meta-analysis in medical studies. 2. Learners will be able to understand and navigate the Cochrane Revman software for conducting meta-analysis. 3. Learners will develop practical skills in using software tools to input data and generate plots for meta-analysis. 4. Learners will get a basic introduction to R Studio and how to use coding for meta-analysis. 5. Learners will understand how to analyze the results of a meta-analysis, including interpreting forest plots and comparing fixed effect to random effect models.
Generated by MedBot

Related content

Similar communities

View all

Similar events and on demand videos

Computer generated transcript

The following transcript was generated automatically from the content and has not been checked or corrected manually.

Right. Um Hi, everyone and welcome back to the eighth session in N Mra Systematic Review and Meta Analysis Series. Um Joining us again today, we have Doctor Connor Gillespie who's here to do the second part on a series of how to conduct a meta analysis. Um Doctor Connor Gillaspie is a fy two doctor at Cambridge and he graduated medical school from the University of Liverpool with an honors um and also obtained an inter degree in surgery and oncology. He's a former Chair of Neurology and Neurosurgery Interest Group and will also be starting a neurosurgery, a ACF in Cambridge in August. He's published extensively in multiple high and back peer reviewed journals and is passionate about teaching and inspiring the next generation. So I'm just gonna hand over to go now. All right, thank you. Thanks. So I hopefully make this one quick guys should just be a practical session really. So I'm gonna demonstrate just how you can carry out the meta analysis that we did in the theory. I think, to be honest, this will be about 10 to 15 minutes. I'm just gonna kind of show you guys how you translate the knowledge that you guys have into actually doing a meta analysis. It's a bit, um I'll just show you two methods very quickly and, or two software programs and it's gonna focus on the comparative outcomes that we mentioned before. As in, is the outcome a yes or a no such as, you know, has death occurred? Yes, has, or has death occurred? No. And then we'll just look at two different software programs that you could use to do your meta analysis. Probably the two most popular ones. So we'll just get started and take you through the green one. So we'll just be going on my screen for this. So it'll be, there's no prep powerpoint slides or anything. It's just sort of taking you through how you would physically convert your study into a meta analysis. All right. OK. All right. So can you guys uh see my screen? OK. Yeah, that's both cool. So essentially this is the first program. So there will be two programs that I'd recommend you use for your meta analysis calculation. There's gonna one is gonna be Cochrane Revman and the other one is going to be in, in our cock and rev mine is where everyone starts out. So I'd recommend use this tool. Unfortunately, they have recently switched it from a program like a user extension to on this web page which I think is really poorly designed. But we'll have to make do, I'll just show you the example that they have provided, which is sort of looking at inhaled corticosteroids for asthma and they're essentially looking at different doses of corticosteroids and its effect on asthma and using, as we said, ditomus outcome. So like yes or, or no. OK. So more or less assuming you can start off from the get go what you'd have to do, let's say if you've got this as a blank review, all you all you have to do is to, to get essentially this is where they end up. So if we, I'll show you where we need to end up more or less. Oh Good. So no. So essentially here is where they ended up and here is where we want to end up. So with a completed forest plot and meta analysis conducted, you'll start with zero on all these. So it's not very representative but essentially for, for this and to work this out, you don't have to do the full review in Revman. You just have to put in the data. That's important here. I was looking at how you do it. You need a few things. So you need to in Revman, you need to manually upload your studies. So you need to essentially have the study to hand and then use it. So taking you back to kind of the data set that we maybe had, I'll just show you in this program kind of the data that we would have. So it's send you in this top left hand corner, this is in, in R. So we'll, we'll look at R just after this, but more or less we're gonna be using, you know, a database of a number of patients. So in this study, this is like the number of patients who had an intraoperative MRI and this is the number of patients who had a complication after the operation. And then this is people who had the standard treatments and then had some complications. So it's very similar. So you just have the number of participants in total and the number that experience the events and then transect line to this one. You can see that that event is their event is treatment failure, which is defined as a need for systemic corticosteroids. And they have two doses, two groups, one is a group who has an increased dose of inhaled corticosteroids and one group is continuing on the same dose. They've got the study, they've got the number of total participants here and they've got the events in each group. They also include some pa some groups with zero events which they are I guess able to do. And at the end, they have a result here. So essentially to get to here, all you have to do is navigate to your, even if this is blank, you can just name it whatever you need to, you can go to your studies tab and then for, for Revman, you have to manually upload each, each of these studies and this data. So if you go and include it, if you wanted to just include it other study, so let's say uh if you wanted to add one, so it's called New Study two, let's say it was 2023. And then essentially you'd add your references. So this could be the PUB man ID or the PUBMED link. So the journal article, you could just add in pages and stuff. It is not essential for the meta analysis. But if you're doing your whole review in Revman, I recommend it your characteristics. This is just for your method, section of your paper. So not really important, covariate again. Uh I wouldn't worry about these for now your study arm, you can add these a bit later or since you or you more or less aren't in your study on. So you can add it on and then an intervention usually will pop up which you have to, you have to fill this in at the start. So at the start, it'll ask you what your intervention is. So if you have two groups, you have to fill them out at the start and then it will let you pick one. So say if I said, you know these, these are the groups that you're gonna compare. So let's say if I said you know the stable dose and then I added another one cos I wanted to have two groups, one group staying the same that I had data for and one group that doubled that dose of uh inhaled corticosteroids. You have to add his own. And after that, you have to essentially add your, you can add a result. Again, this comes from what you define at the start. So more or less I would add treatment failure. So need for systemic corticosteroids. Again, you'll define the, you can add these as outcomes in your study. So you can just add them using the tool. OK. OK. And now you can, you've got both groups and you can just describe the events in the total. So let's say if I in you study two, I have 10 events and I have 100 people and in a stable group, I have 20 events and I have 200 people and I have maybe I don't know 60 people. And then in theory, in theory, this should crop up here. And then if you go back to your analysis, I'm hoping that when you click on this, it comes up, um it might just need to be refreshed, but essentially that's how you add each study and then it'll pretty much just pop up here and give you a nice result. It's this is quite a sophisticated analysis. So essentially it produces the same forest plot as we described before. Most of the time, you'll be able to kind of edit the analysis and describe what you want, so you can kind of change your filters and your e your effect model. So for me, the most important thing at this point would be to keep most of it the exact same. But then just maybe think about as we described before, changing maybe the fixed effects to a, to a, a random effect. And then I think once we get past that, so back to the anos and yeah, and you can see that it changes, it changes the results just a little bit but more or less. II don't really like the this new Revman Cochrane that they've done. I don't really think it's a good system. It used to be very interactive and very helpful. And I think this one is kind of just very confusing. So it more or less there is a Cochrane handbook that has a guide as to how you can input each data. I'm still figuring out myself, but more or less, it's quite easy. You just upload the studies, fill in the results data and it will automatically generate a plot like this or like this one. So that's more or less, it's more or less Cochrane. It's just a case of looking through your, like looking through the handbooks that they provide and, and go from there if we then look at our studio. So this is our, this is a program that I said I wasn't gonna recommend. But I think if people are really keen to try. It's meta analysis is a good way to get into coding. And this is for people who either don't enjoy Revman or find it difficult, which I did. So our studio is a bit different. Some people might be familiar with it, but it involves coding rather than clicking. So I think Revman, it's based on like a user interface, you click different buttons and different stuff pops up and so you kind of code that way or studio, you have to manually type in your code. So it looks a little bit like the diagram on the top left hand side um makes it makes it quite difficult. It's got a steep learning curve, but I think it's pretty good for doing meta analysis really. It allows you to customize your graph, customize your surgical, like you sort of technique that you use in your analysis. So I kind of show it to people just as a, you know, this is how simple it is really, it doesn't need sophisticated coding and you can actually give it a crack yourself and and most of the time chat GBT can generate the code for you. So essentially in our studio, once you figure out how to use it, you need three things to use it. You need packages. So these are software packages that understand how to generate graphs and what you wanna do. You need to have a data set and you need to have code telling the software packages, what to generate and where. So this is for someone called Gideon, one of my friends that I that I helped on one of his projects for. So you have a few, essentially these are all Google. Like if you just Google meta analysis in our studio packages, it will recommend these and you more or less just run them on your setting. Then what you have to do is you have to import your own data set. So, and then attach it, which is, this is what it is. So if you have a database on your computer, this is kind of what ours looks like. This is just the number of uh you know, just like that Excel spreadsheet that we showed you in the last video or session just has the groups and the number in each. So again, this is intra op MRI number of people. This is the number of complications and this is the people who didn't have intra op MRI and their complication rates and it's on an XL file. So you can pretty much, you click it, you import it, you can import it from Excel or from S PSS or wherever. But I often just use what I'm used to. And if you find the work, it should be. I think it's this one. If you click on it, it should pretty much important, you can go back and you can attach it nice and easy hold you one sec this one. Um I think it's this one actually, then once you've got that dataset attached in the packages, all you need is the code. So this looks quite frightening, but I'll just dissect it down for you on what each one means. Essentially you can get this code from just Google. It's really not hard if you just type a meta analysis of binary outcomes at our studio, even GBT can generate it for you. Essentially what this is is you have. That's great. So in this group, you have this is saying do a meta analysis of binary outcomes. Again, you don't have to remember this if you can all copy this. So do a meta analysis, my events dot E. So this is my number of events in the experimental group. So I can just change L have a look. So this is my experimental group, the group that had intraoperative MRI cos this is my sort of intervention and this is how many developed a complication. So that's gonna be my event group. So I Omri complications, my num my ne is number in experimental group. So like the number in your intervention group. So that's I Omri which is this one here, your stud lab, which is your study label, so which we have here, which is called or just in those spaces. So it's author and year your data. So this is what we've defined before as Gideon me comps files. So that's just saying with the code, use this data set, your sm this is your sort of standard measure. So it's an odds ratio. This is the exact same as the Cochrane tool, which I think is um I'll just show you this is the Cochrane default. And you can see this is an odds ratio that they're using. So you could change this depending on what you want. But in most cases, you'd use a, an odds ratio. You can always use a hazard ratio or like a standardized mean difference. But for now just stick to basics. Your method is mh, so just ignore that. That's a stats one. Mh. Exact again, that's a stats question. Just keep it on the default. And then you can, you can see in R you can kind of modify your which model you're testing. So, you know, we talked about fixed and random effects meta analysis. In the past, we talked about fixed effects and random. So I I've said don't do fixed effects, just do the random effects model and then this bit as well is just all stats that you can keep the same. So obviously, it's a steep learning curve. But let's say, you know, you've managed to figure your way around it and you have a blue line, which means the coat works. Let's say you can just do something called a summary. You can type in summary and then of what you've said, it's called to end up B and it will essentially generate for you some written information about the odds ratios, number of studies, number of observations, number of events, the random effects, odds ratio, the 95% common interval and the P value that you're gonna use. So that's quite nice to have it in writing. But I think obviously we're here for the first plot. So with our, it's not very hard. Again, you just can find code that you can use on online. So most of it's on sort of, well, you can get it from a Google search, but essentially you just can use something called forest, literally just copy paste. I copied this from the internet and you just replace um their sort of labels with your custom ones. So all these are relating to my study labels and this kind of these definitions. OK. So you just grab this and type it in theory to give you, well, it gives you this forest spot here and I'll just show you that for, I'll remove the plot. So you can, so you can see that it works. Yeah. So you can see it generates a plot here and this one gives you the exact same thing as in the previous review. It just gives it in a much nicer and a neater format. So you can see that this one is, is much higher publication quality and it offers a lot more customization as well. But you still get the same results as in you get the numbers, you get the alter ratio there, the confidence interval, the weighting is the same, you get the P value and the heterogeneity value, et cetera, et cetera. So there's two main methods. I think if you're starting first, just do the revman tutorial. And if you're not starting, if you feel like I'm pretty good at this. And II actually like coding, I wanna challenge myself, then you can try it in our. You can also do things like look at the funnel plots in our. So just simply by typing funnel endopin, which it's not specific as in just typing in funnel or going onto Google. If you run that with the same plot, you'll be able to identify a funnel plot as well. So with that, it's really helpful to just visualize all your plots in one. Although it is a higher and much steeper learning curve. I, when Revman, the program was around, it was very good and very easy to navigate. So if you can download the program, as opposed to the web page, I would recommend it. But I do think the the webpage kind of ruined it a bit. So you're gonna have to decide if you wanna do a meta analysis for long term. You can either do it in Revman or you can use another program. But I recommend one of these two. They're quite, once you get the crack on arm I think it's a really easy thing to do and it makes beautiful plots. You know, for example, you could probably, you have infinite customization there. So, you know, I could just quite quickly change this to like, you know, I could quite quickly, I think that should work. You can change the color of your plots. You know, you can change, you can change pretty much any aspect of a clot and you can also just quickly switch into random effects models as well and true and fixed effects depending on what you prefer. I think I should change it to yellow. Yeah. So really those are the key things that you need to generate. Remember that while default is gonna be run of effects model, which we'll put in rather than a true effect, he is gonna be 46%. So that justifies to moderate. So all sort of, well, it's low to moderate. So it justifies the decision to do run of effects. And we can see we've just got a number of events and the total participants in each group. And the model just does the rest for us because a really meta analysis can be done in approximately two or three minutes if you have the right dataset, the right code or just the right user interface. But it takes a lot of time and it takes a lot of skill to understand what you're doing and understand when is a good time. I think a lot of people just rush into meta analysis and meta analyze anything as soon as they get the code. But you do need to be aware of it and you need to have a think about, is it the right time? And am I doing it correctly? Because some, you know, statisticians will really pull you up on this and you just have to make sure what you've done is pretty sound methodologically and that you could justify it if you had to speak to your senior about it or something to that effect. No one. All right. So I think, and I think that's kind of the, that's kind of the main section. What I would try and do in the interim is just give yourselves like sort of 8 to 10 minutes for general questions. Uh Because I think that's gonna be more helpful as far as your specific questions for the projects. I just, obviously, I can't use your screen and I can't take 24 different people through how to use Revman or how to use our studio. But hopefully it kind of shows that, you know, with a bit of bit of learning, a bit of a learning curve, you can do quite good meta analysis and you can do good graphs. You just have to type in the data and have the right program. Uh But yeah, that, that's, that's pretty much the the walking session. I'm aware there's a football match going on. So might be keen to watch that. But yeah, if ever, if anyone has any questions, I'll hang around for another couple of minutes and yeah, if it happens to be relating to the reviews, that's great. If it's just general, I'm also happy to answer that as well. And that's, yeah, that's, that's, that's all I got for you today. And yeah, I don't see any questions popping up just now, but I see you put your email in the chat earlier so if they have any questions they can reach, reach out to you through your email. Yeah, is that ok? Yeah, that's fine. Yeah, that's fine. Yeah, so feel free to drop me an email guys and uh thank you for your time. Thank you.