In **Meta Analysis 3**, we aim to provide participants with a solid foundation in advanced meta-analysis and systematic reviews. The training encompasses several key objectives. Firstly, we'll explore advanced statistical techniques for meta-analysis, enhancing your data synthesis skills. Then, we'll delve into the assessment and mitigation of publication bias, a critical aspect of research integrity. You'll also gain insight into effectively interpreting and presenting your meta-analysis results. Additionally, we'll introduce you to software tools that streamline the process. Finally, you'll learn the essential steps in conducting and reporting comprehensive systematic reviews. These objectives collectively prepare you for a deeper understanding of meta-analysis and systematic reviews, enabling you to conduct research with confidence and precision.

# Module III: Meta Analysis 3

*visibility*319 views

*card_membership*Claim your certificate

# Summary

This on-demand teaching session for medical professionals will cover the important steps for meta analysis. It will go over how to establish a team and define eligibility criteria, search strategy and protocols, and the tools of quality assessment to assess risk of bias. It will then move on to discuss what data to extract, the importance of sample size,heterogeneity tests, and the difference between fixed model and random model when conducting a meta analysis depending on the presence of heterogeneity. Finally, it will cover the use of meta regression for sub group analysis to examine sources of heterogeneity.

*smart_toy*Generated by MedBot

# Description

# Learning objectives

Learning Objectives:

- Describe what is involved in conducting a systematic review
- Explain the different measures of association used in meta-analysis
- Identify sources of heterogeneity between the papers during meta-analysis
- Calculate the effect size of a given study
- Analyze the results of a meta-analysis to identify gender differences and explain the implications of these differences

*smart_toy*Generated by MedBot

*card_membership*

# Related content

# Similar communities

*arrow_forward*

# Similar events and on demand videos

#### Computer generated transcript

Warning!

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

Good morning doc will begin at 1135. Just not much, a little difference. Ah re yes, nosing most here. Uh Why you came here? Yes. Come in. Uh Today will, will be our last session uh concerning analysis, analysis. Uh What are the important steps, steps to do them? And uh what are the main test uh that we run um the meta analysis and later more deep in the meta analysis, how we go for meta analysis to do uh first to assess heterogeneity between the included articles and to study subgroup analysis and me a uh trying to find what are the cause or what are the sort of heterogeneity and need to assess the obligation bias between the included S uh OK. Uh OK. What, what, what, what, what was done in the mole one? Uh We, we go first how we define our, how to establish our team, how to define our eligibility criteria and our search strategy. And what are the information if you are uh extract them from included articles and then to write our protocols and to register it in for uh after we write the protocol and we get the acceptance from there. We can work by running a strategy and uh to, to collect all the references that we found them uh uh between databases for me and baby scholar me uh scopes of science. At least we need for database. We, we, we collect them all in one file. Then we go to eliminate uh Ahmad you are seeing in my life. Uh No, no, I didn't. Uh We're seeing the uh and the review about because OK. OK. OK. The present. Yes, present. OK. Oh No, you uh mm Then we, we go from a it may duplicate based on a back cortex. And the and the second uh we saw how we first do our uh screening based on pattern abstract and full text and then how to go to do our flow chart to draw our flow chart and prisma flow chart. And then to apply all data collection and to use the uh tools of quality assessment to study the risk of bias that are present in each article uh that will be included in our meta analysis. What you going to tackle today? They in our last session. Uh all that extraction quality assessment everything. So and then give you statistical information from each R yeah. It now we have a table of data extraction. We have a table of data extraction that contain everything all the information, all the sociodemographics information, all the tools that was used in each article and all the statistical information and uh risk uh ps ratio, risk ratio, uh confident beta confidence interval, everything mean difference, everything that was mentioned in the in each article. No, no. What is our step I to conduct a descriptive uh sent the sentences step number uh 19. Now we are going to stop on a systematic review. How we are going to continue to a meta analysis. Then we are going to stop on a systematic review. All the people had a qualitative paper and now the biggest part is descriptive, descriptive analysis, kill our analysis, descriptive and to explain about each article and to compare each article in a qualitative way and a bill in a descriptive way without numbers. However, to continue to a analysis, our very minimal but our part in the in the results just all in all we have had the included articles DD gets and highest population, highest percentage of female and article. And in a very minimal way, normally descriptive is very summarized. That's our paragraph and the results go ahead. However, is the our our best could give you the results here. I bought the samples so that we are going to take it today. We can go to meta analysis and we will stop on step number steps between systematic review and meta, however, a meta analysis analysis and do a and how but the government a quantitative analysis and to continue to do a meta analysis and systemically, you know, what we are going to see, we are going to see how much step is. Step number 20 how to, how to, how, how to judge that we going to decide if we are going to continue a meta analysis, we continue to meta analysis how to explore the gen first, how to second how to explore it. And I will see how to measure hydro heterogeneity. Second to assess what are the source of heterogeneity to check the publication bias, our reporting bias, check the quality of the evidence and to update report and the publication for whole steps in the if we decide a meta analysis is a lot, it's not clear you are normal, no sub publication, no need normal of evidence wall, a risk of bias wall heterogenic. Uh OK. OK. Now the step number five good to me analyze or no prior to the step as I mentioned before, analysis have the same steps. Step number 19, that's analysis. We need normal uh protocol. We need a strategy. We need normal screening title and abstract full depth at the team might decide whether the data gathered for for each outcome is suitable for pooling using quantitative methods or no by combining data from different studies. The sample size. Why? OK. Why it's important to me analysis? We are they me analysis here under the highest evidence because NANA we obtained the highest sample size in size. Each, each paper, one sample size. OK. The sample size increase, generating a more statistical power and improving estimate of the size of the effect. And it has the potential of resolving uncertainty. Primary studies, studies, lung cancer and smoking, lung cancer. Two studies, no smoking didn't cause lung cancer. And II resolved I resolve the debate. General data bill me analysis is really the decision to pull depends on heterogenic. And if I will go to meta analysis, I know depends our paper or our included articles are heterogeneous. So heterogeneous based, if there are heterogeneous based are multiple levels. First, I will look at the study characteristics of each one. The truth is a and intervention. Uh We had the drug physiotherapy which is one of paper or analysis and I can do that intervention. It's very different magnet and observation. It's very different magnet uh sociodemographic characteristics. Also, it's important age, gender, geographic location and it's very important, geographical location and geographical in the social demographic criteria. That's very important. Tell it, tell it, uh tell it uh tell it goes off in case control paper, cross sectional paper work. OK. And I usually I do it, I combine them. OK. I will do is I can, I can but it's better. That's usually you can do but it's better. Uh Four about the results can be different level of adjustment and level of adjustment. And what I adjustment, age, daily adjustment, age, gender, uh ethnicity, adjustment, age, gender B level of adjustment, very heterogeneous either also it's a source of heterogeneity, uh sex yearly. Who will also it's important who would measure of association measure of association? Uh to be honest, since you the slide im I will go very quickly measure of association and how I will measure the association between the exposure and the outcome. How the intervention and the outcome I can measure it, it depends a lot type of the outcome or type of the export But am ratio like an a category variable. But risk ratio is that a cohort study? But the mean is that continuous variable. But uh uh the combat sub regression can also say continuous but the regression but the correlation or correlation. But I have many measure of association. But how does she also not to combine them? Because a, a new tool calculator is a effect size, calculator, effect size, effect size calculator is a, is a mean difference. For example, I was a mean difference. Uh My kidney uh linear regression, efficient linear regression uh measure of association to measure usually and to be honest, you know, uh it's more about it's, it's subjective, it's more subjective. OK. OK. I meta analysis. A heterogeneous and the heterogeneity between the papers. The two main two main, two main tests must use the C square test test with AQ uh I square. I had some think it's OK. Go ahead. Oh OK. OK. I uh IQ test examined the null hypothesis that all study are evaluating the same effect but may not always accurately detected or genetic. However, I square represent the percentage of variation between the sample estimate that is due to genetic other than sampling error. Yeah, Chi Chi square test usually study no hypothesis. He H zero between the studies. The truth is a hydro effect Magne it's different out kill me no difference. The sample is to me not, the sampling error can also be a value depends based values and the heterogeneity effect size between papers is I square between zero and 40%. Gen between 3060. Still heterogeneity, moderate heterogeneity, let's say substantial heterogeneity. Usually the cut off for seven for 70% but I have an important heterogeneity. Hello and, and, and in my articles, what I should do uh I should. Oh is she the statistic is small fixed model? A small random model? Why? Why? What is the difference between a fixed model and model? But I had the heterogeneity. Why it's important to assess it? Because our type of statistical analysis model, the model 12, it's important to assess what are the source of heterogenic is a can set up a gender measure of association heterogeneity. We assess that you use chi square and I square I squares are represent a difference. The estimated effect size. He usually for 70%. Very considerable heterogeneity model. Finally, very software application will be able to perform a meta analysis and start and website website uh application. Uh a very simple but advanced meta analysis with me reg uh and I usually or meta easy with Excel but just me a meta analysis but unusual what I, what I use with with comprehensive, OK. I'm gonna get home comprehensive meta analysis. OK. OK. It's most CIA and they usually a law to do or meta analysis who will do application is more comprehensive meta analysis, meta analysis, meta regression have a group analysis, uh uh multivaried meta regression with us uh that are very advanced the measure of association we can to transform them like effect. Uh OK. That's what I said uh analysis into consideration the heterogeneity by now results without taking into consideration the heterogeneity. But uh I need to explore heterogeneity papers, meta analysis. They do have a sheet and really it's a hard work. Why it's hard work, it's hard work and it's hard work. A team, it's hard work. A it's a hard work uh team also extra, extra but extraction all the information that you want to include them. The subgroup analysis with metal aggression, sub group analysis. So metal A as simply subgroup analysis here categoric variable by no metal A continuous vy age age is a continuous variable age. Bama me a gender mama subgroup analysis. Y gender Bama two group analysis to a few by it by now meta regression. Uh Population study study group analysis of category uh G heterogeneity type of intervention and subgroup analysis based on gender prove a few different different significant value difference. One of the source of my heterogenic gender, OK. Three. It's a bit more analysis uh advance also with discussion by then you will discuss why the gender can, yeah and why the gender can affect the results and the female different and the male, our male and the female and I am talking and no uh smoking cause lung cancer and general results and no uh significant smoking kill lung cancer risk ratio who are using as a risk ratio, risk ratio in lung cancer. And the heterogenetic papers subgroup analysis, male, female and by significant and significant. However, and it's one of the uh uh source of heterogeneity. The statistics discussion I need to discuss it. Gender affect the results, affected path physiology and female and the risk a male and hormonal muscle and environment and no risk in cancer. I have triggered lung cancer but, but to discuss the G AM the difference amyl heal effect that will be value significant and source of he and to explore source to try to explain the whole hard work. And also our hard part I would explain for each variable leg can be as significant, age, gender, significant and a significant but the age gender as the association between smoking does not cancer. Ok. Uh 2224 statistical example. OK. Uh concerning step number 2, 22 it's checking the reporting bias out publication bias. How the publication bias through channel 10 through the funnel plot fun a geomet uh yes, geometrical visual uh uh PLT before uh is actually publication by using the test that a value command based on value than the uh OK. Zero point hi can under 0.5 had it can under 0.7. OK. And it's yeah. Mm the studies and the studies association and the studies and the positive association for them in both it is a plus micro and publication. However, it a proof and handle on the plot. The class on asymmetry, the majority of papers under a positive odds ratio, a positive log odds ratio. My majority of my papers were under a positive association and a high publications as Kill M and the high public book. But there and public publication wise man usually in general example, smoking and lung cancer and smoking cause lung cancer. When the 10% of the paper go, no smoking didn't cause lung cancer. No publication was papers. Significant result. They didn't publish it a positive result. They didn't publish a publication. It's a positive association. However, a paper negative association, a paper without association but are not published. How does she from? We Kill me symmetrical. Kill low publication wise, Kill me asymmetrical, kill high publication step number 24 and last step and no one ready to submit the study for publication of the interval. Since beginning of that database is greater than 6 to 12 months. OK. And a search be uh be December. We have a little analysis ana be July or August for six months. We keep papers that were, that were published. So I need a another search soon as behind the six months sure was published behind the six months. A included list a kid as I mentioned a while and uh guidelines, prisma guidelines, music guidelines, prima guidelines, uh any questions at all? OK. How often last questions? No. Uh I need to let my feet sh it something or something. All the steps to go more into statistics into a sub group analysis or uh meal aggression. If anyone on the question of should be the second part step. 23 step 23. It's about publication wise, I guess. 20 to see. Oh and my wife left 21. My stuff from the it's I guess uh checking the evidence, I guess that. Thank you. As I know it's uh check, check the quality of evidence that it sounds not important. Sounds important any other question? Ok. Now um the second part uh and a break and, and trying statistics still 10% of the statistics on analysis to go more about the about the about everything, but I don't have the really a bit low percent men. Uh many, many statistics on the metasis. OK? Because it's very deep www selfie and you can specialist. Uh I'm sorry, I'm trying to go fast. My email to contact me. It's hard. OK. Uh The first, the 1st, 1st plot will be among a summary of the results of each article. Yeah, I did we study uh our atopic about uh is a uh how, how to try to find an example? Uh is OK. She, I mean my search strategy I run out, I mean my screening and six studies. OK. Six studies before should be relation between fever six RC Y clinical trials. What? And the six studies. The each study, the data extraction Hakan and the intervention group, I come under a control group high and the number of events. My total Y study number one can end the 20 per participant intervention or 20 participant control adominal intervention. How much is the unkown fever? However, we do not control some of I don't feel OK. So another first before reading each study results, these are the effect size, effect size with trans, the data from each article, one variable. He affects us, us. Standardized mean difference, mean difference, standardized mean difference, mean difference, man mean control, not similar intervention. OK. Now let me know. OK, you mean difference? OK. Hold with me and, and the intervention now is immune and be controlled. I want to assess uh keep the acid Vitamin D on the pressure. OK. And the tube rooms and the control winter, the control group, the Vitamin D intervention, the Vitamin D group. OK. And no under control. The gets to 97. Standard deviation, 2.31 the cases I under the intervention total and score five PHQ nine ma standard deviation 1.3 uh uh last year sessions, deviation man, the general population, a the true value of general population. A it deviation will I can predict in general population confidence interval? Ok. The confidence. Keep with me. Sorry back, I back. Let uh minus standard deviation plus standard deviation confidence comes in as 1.3. It's uh 3.7 hi limit minimum with maximum 7/6 0.3 like over 6.3. Yeah, confidence the PP nine population. OK. A sample in one of 10 persons when all 10 persons had that to me. And OK, five confidence interval. Uh sorry, standardized deviation. DP 1.3 for an elephant and the whole as person under a mean of five and a mean like the H PHP 9/9 under a mean of five. OK. Uh deviation one point uh confidence interval 3.7 last 6.3 between 3.6 and 6.3. I am 95%. Uh Let me confidence and let us know 95%. I am confident that the need of the total population is uh between 3.6 and 6.3 to he. So she like to come back here. She like to come back here. I'm back here and us, us. But the, and I understand that but the effect of Vitamin D on depression and group control intervention intervention 51.3. OK. Uh but the is a and a significant value of the meta analysis and study number one, we'll start number two, ma study number 286 and protocol values there. OK. How to, how I want to combine study number one or number two usually difference. A difference. Some daughter stan die. Oh, I mean difference. She find it by that. I don't know. I mean difference. Hai ho mean that intervention not as me control as simply as that. Ok. However, MD difference. Don worry about it. Honey, lenal intervention, not a similar control. You in the horn. It's divided. Standard deviation is the standard deviation like that. Yeah, I had into consideration, sorry, had into consideration left deviation. However, you mean the first can mean on the standard deviation. Usually it's better to understand height one. So S MD. Yeah, I bought 100 7, sorry on five minus seven. Divided are 2.3 plus 1.3. Ok. Usually how do you have to control this intervention? Additional control it out and study number one. Number two, IM DS 30. Number two, I will combine S MD study number one. Study. Number two, a general, I will pull the S MD like two stops uh intervention control. OK. How are you gonna see them control back here? Ok. It's more tender. The study, sorry. OK. The study of the pla cannot be used with inflammatory markers on the patient with depression and patient without depression. For our study, you can study number one. So, I mean, 16.3 standard deviation is 6 10% and 10%. And depression and the nondepressed deviation to 18.4 16.3. Now it's 18.4 to 2.05 deviation has minus 2.0. Usually it is significant. Ok. Total tac 3.9 and to like patient, non depressed and then depressed, 3.95 difference ma it's a significant. OK. That's a significant uh sorry. Uh 3.95 M in the patient antidepression. If you want to markers, I am nondepressed concentration. It's 3.9 micro, micro, micro, micro, micro, micro, no significant confidence mean difference and minus 2.5 to confidence. OK. And minus 7.63 0.42 0.5 to 3.95 presentation to come back here. Each study. Uh I hate talking details about each study where who will mean different side effect size or pull the result? Uh Just, just English. Uh I OK. With P value between the studies usually. Uh ba hi doc for Yeah. What the heck. Why? Little J it's important, let J it's important because it affect, it affect the way that the mean uh size. Uh A standardized error of standardized uh deviation of standardized ad E and the utility utility. It's more about the clinical choice, the whole value and I will hold the value. Heal part. Follow each to study. You mean depression, not deviation, total depression, the one no depressed or control. That should mean I understand deviation system. Our program weight, the weight study and study contribute to the total value and the total value here is 3.9 a study contribute, study contribute 5.1% and contribution, however, was 10%. It depends a factor. The major factor is S 100. OK? It's contribute 10%. OK? And factor will a whole size for human weight and contribute general uh results. OK. The way heterogeneity affect weight. Yeah, its effect at the paper or the study contribute to the general population D one T square, he like I square or T square zero. Heterogeneity. Can each article and the results I will study number one and the results number two in square, they contribute to general results. OK? V 12 and study number 11112 H 11 plus T square. The factors that assess the heterogenic, for example, also heterogeneity, its affect our standard error. OK? And how II would skip. It's more, it's more advanced. I will go I it's formula I, it's 100%. A QM uh degree of freedom. Q also large 15 mo and be 25 and slow. It. Caution I score is not a measure of absolute heterogeneity I square. Tell us what proportion of the observed dis dis dispersion reflect difference between two score rather than random sampling error? Um Tronic sorry. OK. What happened when heterogeneity is significant? And I 44 comes in, I have heterogeneity. Uh a difference between trials exist, it might, it might be invalid to pull the results. And can you know my is involved because on, on high heterogeneity, it's described our variation and to investigate our source pathogen with me, why, how we investigate our pa usually subgroup analysis and meta regression subgroup analysis is for categorical variable and metal regression for continuous value. This is an example of a meta analysis. I mean a subgroup analysis and for the only for the total OK. And subtotal male to female has been horn test for some group difference I PP value. It's significant and you have a significant difference between male and female. Male M 0.2 with female minus 0.400 0.20 0.4 final H with results. As I'm gonna have discussion, I want to explain why if female 0.4 can if female difference than me and how to try explanation that would explanation why can. And if you made difference. For example, I an example from one of my studies, Depression and tell me L but meta analysis, I go for subgroup analysis. She subgroup analysis. Just a measurement technique. Q PCR and southern blood. No significant will my discussion and Q PCR can tell results. So I tried to explain that but also I am trying to explain factors really affect my results. And pr leg significant difference in analysis of like 10.2 Q PCR. So I want to try to explain why Q PCR affect significantly all my results of the discussion of comorbidities, comorbidities, comorbidities, comorbidities, significant M 0.2. And I want to try to explain why patient comorbidities can hand them a significant result. Patient significant result. Also, I want to try to explain whether she be my discussion again is four C variable, however, continuous variable, I do metal regression and the metal regression meals. Yeah. And now when II do a random effect model, the predictive effect mean effects. And so I will put the effect and I want to get it effect size like I try to do from each two effect size like paper to translate it how effect size. But the how, how will handle to effect size like a common effect size like to try to combine them B aggression NCC. But the is variable we try to shoot to see if he variable ri results. Yeah. For example, one of my studies, depression and still under a significant association between them. But he age with AAA continuous variable age age results results but efficient, efficient hold it is 0.01 minus, it's a positive, efficient and a standardized mean difference. Defect exposure is a negative confluent my act effect. So that is a positive and effect size for my my discussion or age contribute to heterogeneity. But also I want to explain this age. Can the association, this can high strength of association between the exposure and out. This is however, is a uh uh fixed effect model random effect model. I will add the variable of errors, the variable also regression. Uh also also it's show us metal regression. Yeah. And um one of the exercise can mean age, the mean age of the participant, will he age 24 with 34 percent of female kill percentage of female am affect size. And I want to explain and OK, percentage of female contribute to heterogeneity. But also I want to explain less percentage of female association between smoking and lung cancer, of association between depression and tell me like, OK. Hi. Uh I guess right now about fixed effect model or random effect model. I know me analysis of effect model. She had a fixed effect model. I assume that all the studies in the meta analysis are included have a 12 effect size and the observed vig medication among this study is caused by sampling error or chance. OK. Yeah. Fixed model. And I'm assuming and all the meta analysis have a true effect difference by and difference by is because of the sampling error or just for chance. However, the a high heterogeneity model and the high heterogeneity very early. And a difference is not because only sampling error or chance and the a difference, I assume the model random effect that different study exhibit substantial diversity. And the two effect size may vary from study to study fixed effect model sampling error or chance I fixed model random model. Then I just because of sampling error or chance, we have a significant difference, we have a diverse, we have a difference of what had my positive result negative. So, so I want to include the difference by our statistics show fixed model. I didn't include the difference between study and my statistics. However, by random, I think the effect value a fixed model and random model and it will fix more mean by however random model mean plus diversity by they will, they, they will take into consideration the heterogeneity during calculation of the pool effect size. Uh How about the antibiotic? Um And I put your details coming home. OK. OK. Hold on, I'm back here but also will fix effect the weight assigned to depend on the study positions. It weight is equal to the inverse of its variants and the way it area one VV. And he variants the larger the sample size in H I, the smaller the variance of the effect side and the larger the corresponding as analysis I will will effect model kill me all all bigger study contribute more to the estimation of the summary effect on our assigned larger way where the smaller study convey less information and our small way. However, with random effect, the summary effect is estimated at a weighted average and the weight assigned to each observed effect size is equal. The inverse of the variance plus an additional variance component that reflect heterogenic and, and weight. The fixed model into consideration heterogenic weight. OK. I need the, yeah, they fixed motor in a way. Her one divided a variants le B and BB random effect and one divided by the various study plus heterogenic into consideration my weight of the study study study sample size that other uh other source of hygenic for that reflect the heterogeneity. The summary, OK. The summary estimate under the obtains more information from the larger. Oh my God. OK. The summary estimate under the obtains more information from the larger and more precise study, but the distribution of the weight, it's not as much contrasted as under the effect. And home can distribution of the weight based on aamp size, home based as simple size and heterogenic fixed model. The model plus 1.96 deviation or lower limit. 1.6961 0.696. Uh OK. He l mean like funny le le error sampling error. OK. Yes. G 1 g 23123 circle. Who will mean he and he pulled me so he pulled me 1123 plus one equation. OK. Sorry doc. Uh Hi. OK. However, the government fixed the Rando effect model G equal the standardized error plus the however, OK. Difference about E three E one E two epsilon, one epsilon two epsilon three. OK. The distribution of the true effect size one E two E three. And thank you any question. OK, write down my email. Anyone have any question you can contact me or come in. Also anyone have any question? Can you I know? Thank you a clear uh four or five sessions at home. Can you clear uh comprehensive? Oh, thank you. Um uh research it easy. OK. The invitation in. Ok. What, so any introduction to research actor, a specific systematic review? Yes. Thank you very much. Thank you very much and have a good day. Have a good Sunday. Bye bye and bye bye.