 Hi, my name is Bernardo Sousa-Pinto. I'm an assistant professor at the Faculty of Medicine of the University of Porto in Portugal. I will present a tutorial on estim raw, a package which estimates raw data for fold table cells frequencies from risk ratios, risk differences, and odds ratios. Primary studies often present for binary outcomes pre-calculated effect size measures only. However, there may be cases where for meta-analytical purposes, we would rather have access to underlying raw data. It may happen that provided effect size measure by the primary study may not be the one that we are planning to pull or do meta-analysis, or that effect size measure that a specific primary study provides may differ from the effect size measures provided by the remaining primary studies. In those cases, being able to estimate raw data, these cells for fold tables would be of great value. Let me show you an example. This study is assessing the association between allergic disorders and anxiety or depression in children. In particular, we are interested for meta-analytical purposes on the association between allergic rhinitis and anxiety. However, we only have the pre-calculated odds ratio. We can see it in table number three. It's here with the corresponding confidence interval. And we aim to do a meta-analysis of risk ratios. And off-note, it is important to let you know that this is a cohort study, so we can actually compute the risk ratios, but we do not have them here as pre-calculated effect size measures, nor we have the raw data to compute them, at least directly provided in the studies. We do not have that all information we need directly in the paper. We know the odds ratio, which is 1.7. We know the confidence interval, which goes from 1 to 2.8. We know the number of children developing the outcome, the event, in this case, it is anxiety. So it's 83, and we can have them here. We also know the number of exposed children. That is the number of children who have allergic rhinitis. And in this case, it is 119. So we may write it down. And we know the total number of assessed children. This is present here in the text, and this is of 546, so we can also have that information. Of course, we can estimate the number of children without anxiety. We can estimate the children without rhinitis, the number of unexplosed children, but we do not have much more direct information in the paper. So we still do not know what are the numbers of children who are exposed and develop the outcome or who are exposed but not develop the outcome, who were not exposed but develop the outcome, and who neither were exposed and not develop the outcome. But with these ingredients, we can actually estimate the values on these cells. These are the ingredients we need to estimate raw data using the Estim-Raw package, which is based on formula provided in a paper by the Pieter Antoni et al. So let's do it. So we will load the Estim-Raw package and this package has a function which is called Estim-Raw with an underscore. And here we have all the information we need. We know that the ES argument, which corresponds to the effect size, corresponds to 1.7, which is the value of the odds ratio. For the LB argument, we have the value of 1.0. This corresponds to the lower bound of the confidence interval for the odds ratio. We also know what is the upper bound of the confidence interval for the odds ratio. And that is the UB argument and it is of 2.8. Then we know the number of children who had allergic rhinitis were exposed and this corresponds to the M1 argument and which is of 119. For the M2 argument, corresponding to the number of participants without exposure, in this case, the number of children without allergic rhinitis, the number is of 427. We have that information here. Then we also know the number of participants who developed the outcome. That is the number of children who had anxiety. In this case, there were 83 children. And finally, we need to specify the measure argument. And in this case, we have the effect size measure being an odds ratio. So measure equals to OR. We could also have risk ratios, R, R or risk differences, R, D. But in the case of risk differences, we should always insert the effect size measure and the bounds of the confidence interval as proportions and not as percentages. So, running this line of code, we notice that we only have one solution and these are the raw data that lead to the effect size measure which was calculated to that odds ratio of 1.7. So we have now all the data we need. But let us imagine that we did not have the total number of children with anxiety. That is, the total number of participants who have developed the outcome measure. That would have happened. In this case, we would run exactly that same line of code but we would not have the E1 argument because we did not have information on that. And in this case, we can see that we have more uncertainty. We have two solutions which would be correct and we would have to rely on our clinical knowledge to know which solution was more plausible. And then within its solution, we have also some uncertainty because we have point estimates but we also have minimum and maximum values that each cell could take. And this is based on the fact that there may be roundings in the decimal places. So for this case, we would need to consider this uncertainty possibly by doing a sensitivity analysis. If we present results with a higher number of decimal places, if that's possible, in this case was not because everything was presented with a decimal place, that helps increasing the certainty or the precision of obtained results. This package can also be used through a shiny web page at estimeraw.me.up.pt. Here, we can easily introduce the available data and obtain the estimates corresponding to raw data cells. For instance, let us imagine we have a risk ratio of 0.6 with a lower bound of the confidence interval being of 0.4 and the upper bound being of 0.9. We are dealing here with one decimal place. We have as the number of exposed that of 352 and as the number of non-exposed we have 376. Then we can see that we have for each cell the point estimates and the minimum and maximum values. We have a lot of uncertainty here. Now let us increase the number of decimal places. Imagine we have the risk ratio being presented as 0.61 with a lower bound being of 0.40 and the upper bound of 0.93. Here, we are dealing with two decimal places and all the remaining values stay the same. Here, we can see that the uncertainty decreases. We can have a point estimate of 30, but minimum value of 29 and the maximum value of 31. The difference is of one unit upper or lower. If we have information on the number of events, then we get only one solution. Once again, this is dependent on the information we have on the primary studies. So thank you very much for watching this tutorial. If you have any question or if you have any suggestion on how can we improve the estimator package, please let me know. Please send me an email. Thank you very much.