Meta-Analysis - Science method

I just realized the correlations remain very strong (actually, exactly the same) after mean-centering. Then I read this post: https://stats.stackexchange.com/questions/95404/is-centering-a-valid-solution-for-multicollinearity What are the possible solutions to actually address multicolinearity in the context of *multiple-moderator* meta-regression? (Would really appreciate there is R code for me to follow/adapt)

Don't know specific situation,Dare not comment,But you need to analyze from the basic concept of "SMD".

Gokhan Yagiz ,

Thanks for sharing. My question is about the r-value. I saw that you used 0.7. How do I know which value to use in my article?

R statistical environment (R Core Team, 2024) with several flexible packages including my favourite {metafor} package (Viechtbauer, 2010).

1. R Core Team (2024). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/

2. Viechtbauer, W. (2010). Conducting meta-analyses in R with the metafor package. *Journal of Statistical Software, 36*(3), 1–48. https://doi.org/10.18637/jss.v036.i03

"In the U.S., some states guarantee minors full access to gender-affirming medical care while others ban such care outright. Instead, states should follow Europe’s example by adjusting policies based on weighing the emerging clinical evidence, writes independent healthcare analyst Joshua Cohen..."

Many references are given in text by hyperlinks provided.

For resources on statistical approaches for conducting meta-analyses, especially regarding effect size statistics and their combination, you can explore the following:

Hi Mr. Kuswantoro,

I believe my response to Blaine Tonkins might provide you with a comprehensive overview of the topic. Hopefully, it proves useful to you.

You can find it here: [Good Resources for Meta-Analysis Techniques](https://www.researchgate.net/post/Good_Resources_for_Meta-Analysis_Techniques)

Best regards.

Dear professor , I would like to thank you for your answer to my researchgate question .I would like to let you know the title of our paper will be "Meta-analysis of application of particle swarm optimization in geophysical data inverse problem solving". I have collected all papers in this field with this format (about 100 papers ) in 8 categories (a,b,c,d,e,f,g,h):The main applications of PSO to the geophysical inverse problem include the interpretation of: a. vertical electrical sounding (VES) (Fernández-Álvarez et al. 2006; Fernández Martínez et al. 2010a; Pekşen et al. 2014; Cheng et al. 2015; Pace et al. 2019b); b. gravity data (Yuan et al. 2009; Pallero et al. 2015, 2017, 2021; Darisma et al. 2017; Jamasb et al. 2019; Essa and Munschy 2019; Anderson et al. 2020; Essa and Géraud 2020; Essa et al. 2021);c. magnetic data (Liu et al. 2018; Essa and Elhussein 2018, 2020); d. multi-transient electromagnetic data (Olalekan and Di 2017); e. time-domain EM data (Cheng et al. 2015, 2019; Santilano et al. 2018; Pace et al. 2019c; Li et al. 2019; Amato et al. 2021); f. MT data (Shaw and Srivastava 2007; Pace et al. 2017, 2019a, c; Godio and Santilano 2018; Santilano et al. 2018) and radio-MT data (Karcıoğlu and Gürer 2019); g. self-potential data (Santos 2010; Pekşen et al. 2011; Göktürkler and Balkaya 2012; Essa 2019, 2020) and induced polarization (Vinciguerra et al. 2019); h. Rayleigh wave dispersion curve (Song et al. 2012) and full waveform inversion (Aleardi 2019). please guide me how to arrange them for you to analyse result and calculate impact effect of papers and perform our meta analysis method? I am looking forward to hearing from you as soon as possible. Best regards Reza Toushmalani

Good day, Ms. Villacisneros! To answer your question, serological tests for HDV do really help in diagnosing HDV co-infection with Hepatitis B. They help when molecular tests for viral RNA in relation to HDV is not available in some laboratories.

Good day Ms.Sulit!

HBV surface proteins are necessary for the hepatitis D virus (HDV) to produce mature virion particles, which is why HDV is categorized as a satellite virus. As a result, HDV may either co-infect or superinfect with HBV, but it is unable to disseminate or package infectious virions when HBV is not present. Preliminary research using cell cultures and mouse models has shown that encapsulated viruses other than HBV may produce infectious delta virus particles, but as of yet, there is no concrete proof of the therapeutic relevance of these findings. Though they usually recover from both viruses, coinfected patients may experience acute hepatitis. The risk of developing chronic HDV infection and its associated comorbidities increases when an HBV carrier with a chronic infection is superinfected with HDV.

Reference: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9781095/#:~:text=Hepatitis%20D%20virus%20(HDV)%20is,absence%20of%20HBV%20%5B26%5D.

Hello Gustavo Waclawovsky I solved the problem by changing method.tau to "DL" which refers to the DerSimonian-Laird estimator.

The inclusion criteria for selecting studies in the meta-analysis focused on the availability of detailed measurements such as sensitivity, specificity, likelihood ratios, and areas under the summary receiver operating characteristic (SROC) curve for serological tests detecting HDV antibodies. Additionally, studies were required to have performed PCR tests for HDV nucleic acids and HDV antibody tests, with reported numbers of true positives (TP), false positives (FP), true negatives (TN), and false negatives (FN). Studies with a sample size of at least 10 cases and published in English between 1989 and 2023 were included. Conversely, studies with fewer than 10 cases, case reports, review articles, or meta-analyses, as well as repeated studies or those lacking a clear gold standard for HDV diagnosis, were excluded from consideration. These criteria aimed to ensure the inclusion of relevant and methodologically robust studies in the meta-analysis.

Hi,

yes that is possible as long as to inform the model about the nestedness of the data. The key analysis would be a three-level random effects meta-analysis (see my profile for a bunch of applications.

All the best,

Holger

Hello, Ms. Cabrera!

Based on the provided material, HDV RNA method of detection have several restrictions. Such that, PCR testing is only capable of detecting HDV RNA when it is being actively replicated. This poses a problem especially during therapy when replication of viral genome is suppressed. The increased proportion and complementarity of the guanine-cytosine content of HDV RNA has also been determined to have caused challenges in amplification. Furthermore, there had actually been no standard PCR detection technique until recently due to the vast genetic variability of HDV RNA. Hence, results from different laboratories are not comparable.

Yes. But it's important to highlight the differences in population, time period, and potentially methodology in your submission to emphasize the novelty and relevance of your findings.

You can input one-arm comparisons into RevMan. Simply use the pre-intervention mean and SD for the 'control group' and the post-intervention mean and SD for the 'experimental group'.

If you are including one-arm comparison studies and two-arm comparison studies in the same review, you may wish to do seperate meta-analyses for them, or include them all in one meta-analysis, then afterwards do a sub-group analysis spliting the studies.

To conduct a dose-response meta-analysis in STATA 17.0, you can use the "mvmeta" command. Here's a step-by-step guide on how to perform a dose-response meta-analysis in STATA:

Also you can simply perform meta-regression and Bubble plots which are included in the STATA itself. You need to extract the does of the intervention or exposure for each study and then perform a meta regression and bubble plot which will show you the association of dose and assessed outcome perform in your study.

To learn Meta regression and bubble plot you can find useful videos simply giving step by step meta regression in STATA.

Hi Mohamed Jayte do we have more colleagues to work on this systematic review? Should we create a group to start working on this?

Hello Giuseppe Dario Testa. What version of Stata do you have? I ask, because if it is recent enough (v16 or later, IIRC), it will have a suite of new (official) meta-analysis commands, including this one:

PS- I recommend posting questions about how to do X using Stata to Statalist. If you do decide to post there in future, see the very helpful FAQ first.

Findings demonstrate that serological assays are highly accurate in identifying antibodies against HDV, particularly IgM and IgG. Further evidence is need to demonstrate the specific efficacy of serological diagnosis, though. For HDV screening, serological testing is often a good option. For laboratory diagnosis, treatment efficacy analysis, epidemiology, and disease control, the creation of novel serological diagnostic instruments with increased precision and dependability in HDV diagnosis is crucial.

I2 is a measure for quantifying heterogeneity in meta-analyses. It can be interpreted as the ratio of the variance between studies to the total variance in the meta-analysis. I2 is scaled from 0% to 100%, with higher values indicating more heterogeneity. As a rule of thumb: I2 >=50% may represent moderate and I2 >= 75% may represent considerable heterogeneity. Thus, if your calculation is correct, you don't have to worry about inconsistency between the studies in your MA.

I think the type of journal should not a priori be an exclusion factor. You can have poor quality studies published in high impact journals (it always depends on the reviewers' that were selected to assess paper during the peer-reviewed process). Quantification of the studies' quality and risk of bias would be more robust than a priori excluding an eligible study just because it was published in a journal that is thought to be predatory.

My 2 cents.

NB: the definition of what is a predatory journal is still a debated issue.

Dear Martin,

Thank you so much for your recommendations and suggestions! They are much appreciated. This is wonderful.

Take good care,

Nick

In a forest plot, the "line of no effect" typically represents a null effect or a reference point against which the effect sizes of different studies are compared. This line is often drawn at a value of 0 on the x-axis, indicating no effect or no difference between groups or conditions being compared.

In some cases, depending on the context of the data being presented in the forest plot, the line of no effect may be positioned at a value other than 0 or 1. This could occur when:

Overall, while the default position for the line of no effect in a forest plot is typically at 0 on the x-axis, it can be set to other values depending on the specific context and nature of the data being presented.

You can use this:

but polynomial regression is very dangerous to use since it doesn't explain anything and it can never be extrapolated.

Typically you can not pool data from both type of studies as it has different underlying conditions and heterogeneity.

but if you want to consider this option you can approach it considering your research question in different manner like SUBGROUP ANALYSIS, MODERATOR ANALYSIS or SENSISTIVITY ANALYSIS.

By avoiding pooling emergent and elective repair data and considering alternative approaches, you can ensure a more robust and informative meta-analysis.

Always consult with a statistician familiar with meta-analysis techniques for guidance on the most appropriate approach for your specific research question Muhammad Ahmed

Cochrane group recommended to adapt only the observation of the funnel plot

The systematic review on a specific domain can be performed (protocol and search) even if there is no article in the end that are found (meaning that there is no available evidence on the topic under interest).

For meta-analysis, if research studies results can be pooled together for meta-analysis you can theoretically report meta-analysis in n>=2 studies. however, the more studies you get the more analyses you can do.

In many paper in my research area (veterinary science), meta-analysis are conducted when at least 5 papers are available. You can see this paper from Yemisi Takwoingi et al on that topic from a statistical standpoint.

My 2 cents

Takwoingi Y, Guo B, Riley RD, Deeks JJ. Performance of methods for meta-analysis of diagnostic test accuracy with few studies or sparse data. Stat Methods Med Res. 2017 Aug;26(4):1896-1911. doi: 10.1177/0962280215592269. Epub 2015 Jun 26. PMID: 26116616; PMCID: PMC5564999.

While the exact causes of cognitive impairment in MS are unknown, two factors often further impair cognitive performance in patients with the disease: depression and physical disability. Depression often plagues people with MS-related cognitive impairment.

Very nice question. My preferred way to day with it would be trying to categorize the controls under a same umbrella, something in commom, that could make them part of a same category.

You could mention that some or most of the points on the checklist do not apply to a bibliometric study, and comply with those that do. It is worth trying to adapt PRISMA to the scenario of your bibliometric analysis. This would highly enhance reproducibility and replicability and would make it possible for other researchers to build on your research. I hope this helps.

Thanks so much, AbdallFatah

Lucia Sideli : Both need to be transformed

Yousef Shahin : type in the command line: ssc install doiplot

This installs the doiplot module. Then just type doiplot es se, dp. Note es and se are on the transformed scale

There was a recent well-publicised event where an actual legal court case included legal documents prepared by AI that included supposed legal citations to cases that did not ever exist.

So, you have two problems:

(1) Constructing code that does actually work;

(2) Persuading others that you have code that actually works.

Cohen's d in its original form is SMD (although we usually see it in the sample-based form, also given by Cohen). r^2 or R^2 are usually taken as percents/proportions ("percentage/proportion of the variance explained by the grouping variable"). Then, if you convert d to the scale of r by using the traditional formula by Cohen and, if there are more categories than two, you can use the new formulae given in the previous link. When you have transform Cohen's d to the scale of r, you can take the second power and express SMD in the scale of percentages (of some kind).

The interpretation may be make more complex the fact that, when the scales are not equal, two things affect the result: the number of categories in the variable with the narrower scale and the dicrepancy of the cases in different categories. They both affect the magnitude of the deflation or attenuation in the correlation. This causes that, if the proportions of the cases in different categories radically deviate from each other, the scale of r (and R) is no more -1 to +1, but it may range -0.6 to +0.6 (see simulations and publiced datasets in . Then, the challenge is that what means the "percent" of explaining power or r^2 = 0.4? It means "the proportion of what we CAN predict by the given condition of variables". For this you could try the deflation- or attenuation corrected estimates of r (rDC or rAC; see https://link.springer.com/content/pdf/10.1007/s41237-022-00162-2.pdf and ). They strictly reflect the proportion of the observed correlation of the maximum possible correlation. The relation of these latter correlations with Cohen's d is not studied.

Good luck!

JMe

In the scenario you've mentioned, where 10 participants received both a placebo and an intervention, it's important to avoid double counting.

In the context of a split head comparison, each participant serves as their own control, with one side receiving the intervention and the other side receiving the control (placebo). Since the same participant is receiving both the intervention and the control, the total number of participants remains 10.

For the purpose of your analysis, n.i (number of participants receiving the intervention) would be 10 and n.c (number of participants receiving the control) would also be 10. In other words, each participant contributes to both the intervention and control groups, so you should not split the participant numbers.

When conducting your systematic review and meta-analysis, it's important to account for the paired nature of the data due to the split head design. This may involve using statistical techniques that are suitable for paired data, and ensuring that any potential correlations or dependencies between the two sides of the head are appropriately addressed.

I would also look to see if there are any systematic reviews on the PEDro website which is the physiotherapy evidence database. You can read my paper on number needed to treat, where I mention that site.

Thank you James Leigh

I think it is challenging to "pool" results from case-series because of the high risk of bias of this study design. A systematic review would be the first step with an a priori idea of quality indicators of case series.

Then "pooling" should be a priori discussed based on its feasibility and clinical meaning of the different studies you gathered. Pooling poor quality studies together at any price will not give any added value vs a qualitative systematic review.

A Google search on <meta analysis convert adjusted OR to adjusted RR> turns up some hits that may be helpful.

A systematic review is a broader process that involves systematically collecting, appraising, and summarizing all available evidence, while a meta-analysis is a specific statistical technique used within a systematic review to quantitatively synthesize data from multiple studies to produce a more precise estimate of the treatment effect. Often, these two methodologies are used together to provide a comprehensive overview of the existing literature on a particular topic.

Yes, unpublished studies can be included.

I am not sure of the meaning of "overlapping" term you mention. From my perspective, it seems that the same patients could be used twice as separate study findings? This type of problem is depicted in Cochrane training book 6.2.4 section (Chapter 6: Choosing effect measures and computing estimates of effect | Cochrane Training).

This situation could be accounted for in the meta-analysis and would have an effect on the standard error of the estimate (with hierarchical model accounting for repeated measurement it would increase the standard error of your pooled estimate). Practical info should be retrieved in (Chapter 10 “Multilevel” Meta-Analysis | Doing Meta-Analysis in R (bookdown.org)) using metafor package in R environment.

Puthachad Namwaing has given a very comprehensive answer above. An alternative perspective is to consider the following. Imagine we have a HR for two predictors that come from a multivariable model.

i) In the simple model these two predictors are independent of each other they both predict the outcome. The hazard of death in the model is the product of the two factors. One predictor may have a larger effect than the other but they are different sources of variation and both may have an important influence on the outcome.

ii) The size of the HR for each is entirely dependent on the nature / scale of the predictor. Imagine (say) 'weight' as a predictor - the HR associated with a unit change will vary depending on whether weight is measured in g of Kg.

iii) where predictors are categorical a predictor that occurs infrequently (say a rare diagnosis) could have a very large HR associated with it but have very little impact on the overall ability of the model to correctly classify/predict death - for the simple reason that so few people are affected and so omission of this variable does not lead to incorrect prediction for many people.

iv) The accuracy of the estimated effect of each factor will depend on the error associated with it. A 'strong' effect estimated with low precision will be associated with a large prediction error.

I think you'll have to perform 2 different models (or select what is the more realistic/pertinent approach). The beta obtained from your models tells you different things: for continuous measure of exposure this gives you the impact of each unit increase of exposure and the dependent variable (telomere length).

For pesticide as a dichotomous outcome the beta gives you the effect of presence of exposure (independently of its importance) on telomere length.

It makes no sense to put the different studies in the same model because of these differences (like comparing pear and apple).

For anybody interesting by collaboration in this field:

Your approach to adjusting for the two quantitative variables (Age and Weight) in your meta-analysis seems reasonable, and it aligns with a common strategy for addressing heterogeneity by using meta-regression. Let's break down your steps and offer some additional considerations:

In summary, your approach appears technically sound, and adjusting for covariates through meta-regression is a common strategy in meta-analysis. However, always ensure that you have thoroughly checked the assumptions and limitations of your modeling approach and consider seeking expert advice when needed.

Conducting a meta-analysis of mediation using structural equation modeling (SEM) in R can be a complex but rewarding task. The "metasem" package is a valuable resource for such analyses. Here are some additional thoughts and suggestions to help you with the process:

Remember that conducting a meta-analysis, especially involving complex statistical methods like SEM, can be challenging. Take the time to thoroughly understand each step of the process and seek help when needed. Additionally, make use of the resources provided in the "metasem" package documentation and consider reaching out to the package authors for guidance if necessary.

See Borenstein book chapter 7 introduction to metaanalysis

Subgroup analysis investigates whether the results are the same for each group. It is preferable to use continuous data as continuous values. To use categorical data in regression analysis, the fit of the regression is poor unless there are 10 data in the smallest category. If you use multiple variables, check the correlation between them. If the correlation is strong, it is best to select a representative one.

Dear Mariam Elsherbiny

There are many articles and papers that you can search on science direct website specially review papers. By the way Professor, Ali Kaveh has been worked in this field of area so you can read his books and publications.

Studying review papers can help you to obtain to a general vision with enough detail about optimized structural analysis and all kinds of Meta heuristic algorithms are used to this aim.

Good luck.

I'm Interested

Hi Thiago,

If both the 20mg and 40mg effect sizes are measured with respect to the same control group, and thus originate from the same study, I would not perform a subgroup analysis in a standard meta-analysis framework. I'll refer to Chapter 10 of the following book:

"Statistical independence is one of the core assumptions when we pool effect sizes in a meta-analysis. If there is a dependency between effect sizes (i.e. effect sizes are correlated), this can artificially reduce heterogeneity and thus lead to false-positive results."

If the estimates for 20mg and 40mg come from the same study, the estimates are likely correlated, which would lead to a violation of the independence assumption.

If you are still interested in performing a subgroup analysis, you could incorporate it into a multilevel meta-analysis with medication dose (20 mg vs 40mg) as one of the explanatory variables. If you look at equation (10.8) in the book chapter I referred to earlier, you see that a multilevel meta-analysis incorporates a second random-effect: one to account for clustering of effect sizes within the same study. Statistically, this would be the correct way to pool your correlated effect sizes (if you insist to combine them).

Hope this helps!

Kind regards,

Michael

One of the widely used and highly regarded software for conducting meta-analysis is Comprehensive Meta-Analysis (CMA). CMA provides a user-friendly interface, extensive statistical tools, and a comprehensive set of features for conducting meta-analyses across various research domains. Other popular options include RevMan (Review Manager) and Metafor in R, depending on user preferences and familiarity with specific programming languages. Ultimately, the "best" software may depend on the specific needs of the user, the nature of the meta-analysis, and individual preferences regarding interface and functionality.

In meta-analysis for management studies, there's no one-size-fits-all model, but there are steps to follow:

1. Define the Question

2. Gather Studies

3. Assess Studies

4. Combine Results

5. Interpret Findings

Example : For instance, imagine studying different leadership styles in businesses. You'd collect various studies, check their quality, combine their findings, and figure out what they collectively reveal about effective leadership in business.

You need to include all 20. Selecting just 15 significant ones will bias the results. Many years ago the it was common practice to set NS effect sizes as zero in the analysis (which is almost the same as what you propose, but worse) and this is now recognized as a deeply flawed approach.

A major problem in meta-analysis is publication bias, so further excluding NS published effects would bias the results of your analysis (overestimating the effect and possibly underestimating variability). It would result in additional bias over and above the publication bias inherent in the literature.

Comprehensive Meta-Analysis (CMA): This software is widely used for meta-analysis and provides a user-friendly interface. While it may not have built-in AI features, researchers can use external tools or scripts for advanced analyses.

Thank you, Tony. I'll go find these materials.

The Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) is a set of guidelines for reporting systematic reviews and meta-analyses. It was developed by a group of experts in evidence-based medicine and was first published in 2009. The PRISMA guidelines are designed to help authors report their reviews in a clear, transparent, and reproducible way. This is important so that readers can assess the quality of the review and make informed decisions about the evidence.

The PRISMA checklist consists of 27 items that are divided into four sections:

The PRISMA guidelines have been widely adopted by journals and publishers around the world. They are considered to be the gold standard for reporting systematic reviews and meta-analyses.

I recommend checking out reviews by Cochrane Haematology or other Cochrane cancer groups (e.g here https://breastcancer.cochrane.org/news/our-most-cited-reviews). It´s always a good idea to involve practitioners and patients, for example, to identify relevant outcomes.

Just for the sake of completeness: it might be possible if there is a meaningful reference concentration of troponin you could refer to, but I doubt that there is such a value.

Tamatha Zemzars Thank you

Hello Santiago Ferriere Steinert. These two ORs are from different studies, right? How many ORs do you have in total? If I had only the two ORs you describe, I think I would just report them separately. If they were two ORs of a much larger number of ORs, and all but that one were from models that treated the X-variable as continuous, I might compare the OR from the 2x2 table to the pooled estimate of the OR from the other studies. But I think more information is needed. HTH.

Thank you.

What's your definition of "effect size" ?

What you are asking is standard procedure when doing a meta-analysis. When you calculate a Hedges' g (or standardized mean difference), you remove the units (e.g., HA/mm3, g/cc, mmAl, and %) and thus the different studies' values become comparable. Refer to "Introduction to Meta-Analysis" by Borenstein et al., 2021.

For the first question, using multiple databases as you suggest usually help to minimize the risk of missing relevant studies. The risk of missing studies by published year not usually an issue, since published year tend to be well documented and indexed.

However, missing studies due to a narrowed scope while searching for literature is always potential risk. If you have an reasonable knowledge of the field you are planning to review, you might have a range of publications already prior to the reviewing process. If you can find all of these publications during your scoping process then your scope is acceptable. If some of these are missing, you might need to extend the scope further.

Another issue that is difficult to account for is the inclusion of journals/publications that are not indexed in the big databases.

For the second question, one option is to contact authors of these subscription-only publications and request a copy. If some publications are essential yet no access can be granted, the last option is to purchase the publication.

Responding to your first query the number of studies to be included depends on the heterogeneity of the study type you choose, there is no clear cut rule of thumb that can be relied on for example, in a study by Davey et al*. a sample of 22453 meta-analyses, show that the number of studies in a meta-analysis is often relatively small, with a median of 3 studies (Q1-Q3: 2–6), and only 1% of meta-analyses containing 28 studies or more. To compensate the heterogeneity of the studies rigorous review of the methods of the studies should be conducted and to reduce errors statistical methods such as DerSimonian and Laird approach and Hartung, Knapp, Sidik and Jonkman can be used for random effects meta analysis.

*

Yes, it is possible to conduct a meta-SEM using Smart PLS4 software. Smart PLS4 is a software package that is specifically designed for partial least squares (PLS) modeling. PLS is a type of structural equation modeling (SEM) that is well-suited for analyzing data with small sample sizes and/or non-normally distributed data.

To conduct a meta-SEM using Smart PLS4, you will need to collect data from multiple studies that have used the same SEM model. You will then need to prepare the data for PLS analysis. This involves converting the data to a format that is compatible with Smart PLS4 and transforming the data to ensure that it meets the assumptions of PLS.

Once the data is prepared, you can then create a meta-SEM model in Smart PLS4. This involves specifying the latent variables in the model and the relationships between them. You can also specify the measurement models for each latent variable.

Once the model is created, you can then estimate the model parameters using the Smart PLS4 algorithm. Smart PLS4 will generate a variety of outputs, including the estimated path coefficients, standard errors, and p-values. You can then interpret the results of the meta-SEM to identify the relationships between the latent variables in the model.

Here are some additional tips for conducting a meta-SEM using Smart PLS4:

I hope this information is helpful. Good luck with your meta-SEM study!

Dear Dr. Qinyao Wu ,

Based on the situation you described, the preferred approach for me and my team would be to contact the corresponding author. Generally, they are willing to address the questions you have raised.

If we are unable to establish contact, we may consider excluding this paper, as we are well aware that the sample size can affect the weighting of effect sizes in the meta-analysis.

Best regards,

Tzu-Hsiang Peng

I second what Caroliny Hellen Azevedo da Silva said. You need to either find a collaborator or a librarian who can help you to access these databases. There may also be some plans from these company to provide you a temporary access for a specific cost.

Good luck!

As Lowilius Wiyono said, it is difficult to answer specifically to your question.

I think a good point to start with would be an introduction on the meta-analyses methods to know exactly what you have in mind and then you could chose the dedicated software to do it (R, RevMan, ...).

I really like this online book by Harrer which is a thorough overview of meta-analyses and methods that can be used with R sofware (https://bookdown.org/MathiasHarrer/Doing_Meta_Analysis_in_R/intro.html), but I am a R biased user :-)

I would say that by default a random effect meta-analysis would be preferred because it is difficult to say that in your meta-analyses all the observed differences are only due to sampling variability. It is therefore important, I think, to account for the fact that the difference observed is also due to unobserved difference in study design or other characteristics (in fact, with the exception of randomized controlled trial meta-analyses I think random effect would be the default in many situations).

In the random effect, the weight of the studies are not equal. There is always a component which depends on the sample size (within study variance) but you also add a specific term which accounts for between study variance (which is generally called tau-squared).

so the weight depending is on the form:

Weight=1/(Variance_study_i + tau-squared)

You can see that both components are accounted for but now the weight is less influenced by study variance (so larger studies have more impact but much less impact than in fixed effect).

NB: this is a general comment. there may be some specificities for models for a prevalence but this is just to illustrate how it works.

see a specific paper on the 2 types of meta-analyses (Borenstein M, Hedges LV, Higgins JP, Rothstein HR. A basic introduction to fixed-effect and random-effects models for meta-analysis. Res Synth Methods. 2010 Apr;1(2):97-111. doi: 10.1002/jrsm.12. Epub 2010 Nov 21. PMID: 26061376. )

Sébastien Buczinski This is very helpful. Thank you very much

Now coming to your table p-value in the right column of the forest plot is the p-value for the overall test of the treatment effect across all subgroups. It is calculated by combining the results of the individual studies in the meta-analysis. In this case, the p-value is 0.56, which is not statistically significant.

The p-value for the test for effect in the subgroup is the p-value for the test of the null hypothesis that the treatment effect in the subgroup is equal to zero. It is calculated using only the data from the studies in the subgroup. In this case, the p-value for the test for effect in the subgroup is 0.094035, which is statistically significant.

The two p-values are different because of the heterogeneity between the studies in the meta-analysis. The heterogeneity statistic (0.5) is very high, which indicates that there is a lot of variability in the treatment effects across studies. This variability could be due to a number of factors, such as different study designs, different populations of patients, and different treatment regimens.

When there is heterogeneity in the treatment effects across studies, it is more difficult to detect a significant overall treatment effect. This is because the variability in the treatment effects across studies can mask the true effect of the treatment.

In this case, the p-value for the overall test of the treatment effect is not statistically significant, but the p-value for the test for effect in the subgroup is statistically significant. This suggests that the treatment may be effective in the subgroup, but it is not possible to draw a definitive conclusion without further research.

It is important to note that a statistically significant p-value for the test for effect in a subgroup does not necessarily mean that the treatment is clinically effective in that subgroup. It is possible that the difference in the treatment effect is small or that it is not clinically meaningful.

To determine whether the treatment is clinically effective in a subgroup, it is important to consider the magnitude of the difference in the treatment effect and the clinical implications of that difference

You can use STATA metaprop command give power(2) at last, it will come in percent

If the trials you have provide the required data, you CAN include them in a meta-analysis. Whether or not you SHOULD do so is another matter. In a traditional hierarchy of evidence, non-randomised studies are much lower. They are much more prone to bias - so including them risks including highly biased results - rendering the results of your meta-analysis biased. One approach might be to undertake sensitivity analyses to see whether or not the inclusion of such studies alters the point estimate (much) and overall conclusions - but if it does, where does that leave you and if it doesn't what have you gained by including them?

IF this is a topic where non-randomised studies are, pragmatically, the only approach in many circumstances you might proceed with caution - but you must use a suitable approach to assessing the risk of bias for such studies (ROBINS) - and I would suggest doing the sensitivity analysis in any case. Bit remember the old adage - garbage in, garbage out.