Second, looking for raw materials-Meta-analysis literature retrieval
After determining the production direction, I need to visit the orange planting base to find satisfactory oranges. Because I don't know where oranges meet my needs best, I bought oranges from every town in my hometown This is equivalent to the document retrieval link in the analysis process, searching the database comprehensively and not missing every relevant document.
Third, screening varieties-meta-analysis literature screening.
Find as many varieties of oranges as possible, and then I need to taste and select the varieties of oranges I need one by one according to my plan to determine the source of goods.
First of all, oranges must be full of fruit, excluding a number of oranges with insufficient water. Secondly, oranges must be sweet, excluding a batch of oranges with sweet and sour taste. Finally, the supply cost of oranges should not be higher than per kilogram 10 yuan. Except for the expensive oranges, the rest is what I need.
This process is our literature screening, reading the literature one by one according to the inclusion and exclusion scheme, and determining the included and excluded literature.
Fourthly, conduct procurement evaluation-meta-analysis quality evaluation.
After selecting the raw material supplier, I need to go to the planting base to purchase the oranges I need, and issue a raw material quality grade table, so that my consumers can rest assured to drink my orange juice in the future.
The process of purchasing, selecting and evaluating oranges is our document quality evaluation. The quality of an orange directly affects the quality of a bottle of orange juice, and the quality of the included documents will also directly affect the quality of our final results. Therefore, it is necessary to evaluate the quality of each document, so that readers can understand the quality of the included documents and feel at ease to "eat" this meta-analysis.
Verb (abbreviation of verb) production-peeling and meat removal-meta-analysis data extraction
With raw oranges, the next step is to peel and slice them. We bought a whole box of oranges from the supplier, but only the pulp of oranges went into the juicer, so we had to take it out.
This step is equivalent to information extraction in the research process. The collected documents are all complete papers, but what we really need to analyze is only the research data and research methods in the documents, so we must extract the data through tables.
Six, the production of fruit juice-meta-analysis and statistical processing
With pulp, it's simple. Pour the materials directly into the machine and let the machine juice for us.
Similarly, meta-analysis with data is relatively simple. Just throw the data into the software and let the software analyze it for us. This step is the data synthesis in our meta-analysis.
STATA, r language and RevMan can all do this, and we will introduce them one by one in the next few weeks.
Seven, test-exclude heterogeneity
After juicing is completed, we need to send the orange juice for inspection, so that relevant institutions can detect whether our orange juice contains other impurities and issue a certificate for us, so that consumers can buy our products with confidence.
This step is also needed in meta-analysis, and it is after the conclusion of meta-analysis is drawn through comprehensive analysis. We also need to detect the differences between documents, determine the degree of heterogeneity of documents, and judge whether the conclusions are credible.
Because in the synthesis stage, we mix the data together for analysis, if the two documents we extracted are very different, then we directly mix them for analysis, and the result is very wrong, so there is no research significance.
This step is our heterogeneity test. If the heterogeneity test concludes that there is a large heterogeneity, it is necessary to further analyze the source of heterogeneity, exclude heterogeneity, and re-select different effect models for data synthesis.
Eight, eliminate the risk of fraud
At this time, I got a batch of delicious orange juice, but new problems appeared. How can we ensure the stability of orange juice taste and ensure that there are no other influencing factors in orange juice? For example, how can I make this batch of orange juice reach the right sweetness in case the Zhangsan family cheat me and inject sugar water into oranges?
So, in order to ensure the stability of the product quality after the last batch of production, I bought another batch of oranges, excluded some seemingly unreliable suppliers' raw materials, or joined production lines with different juicing processes to reproduce orange juice.
If the taste changes little every time, it proves that my production process is stable and reliable, and the quality of oranges is uniform, I can realize mass production; On the contrary, if one of them is changed, the taste will change greatly, and the taste of my orange juice will be greatly affected, which shows that one of the influencing factors has a great influence, and it is necessary to further test these influencing factors in order to realize stable batch production.
This is equivalent to sensitivity analysis in meta-analysis. Sensitivity analysis refers to changing the inclusion criteria (especially controversial studies), excluding low-quality studies, and analyzing the same data with different statistical methods/models. , observe the changes of consolidated indicators. If a file is excluded, it is considered a sensitive merge RR, and vice versa.
Nine, eliminate hidden dangers
Finally, when testing the production quality, we only bought a small batch of oranges, and the supplier gave us big and good oranges, and the produced results were stable and reliable, with good taste. What about the other smaller and lower quality oranges that we didn't get? It is not excluded that some suppliers just show the good products to the ordering party for cooperation, and deliberately hide the differences between products. So in the end, we just need to reconfirm whether the quality of suppliers is stable, and deliberately show the good ones and avoid the bad ones. If not, we can achieve mass production.
In meta-analysis, this is the last step in evaluating publication bias. Orange juice tastes good because you choose a good orange that the supplier can give you, but there may not be such a perfect orange in actual production.
Similarly, in medical statistical research, positive results are often easier to publish than negative results, so the documents we include may be because "good quality" documents are provided to us, and we need to consider whether those unpublished "slightly worse" documents will affect our results. It can be tested by evaluating whether the funnel graph is asymmetric, identifying the publication bias, using Begg and Egger methods to test the symmetry of the funnel graph, Trim method and unsafe number method. Follow-up sharing will be held in STATA.
X. End of production research stage
So far, we have basically completed the construction of the factory. Of course, we still need to expand the market, marketing and after-sales to truly realize the operation of the enterprise, which is similar to the subsequent paper writing, format and typesetting of Meta Analysis, but the most important "product" has been produced. As long as the product is hard enough, the follow-up is very simple ~
The next issue of Meta Tread Pit Collection will enter the stage of data synthesis and juicing, and I will share with you how to use RevMan and STATA to juice orange juice. The revolution has not yet succeeded, comrades should study hard with me!