Article type
Abstract
Background
High-quality systematic review and meta-analysis is a proven method that sits at the top of the evidence pyramid, which could improve statistical efficiency. In recent years, multiple studies have been published on constructing prediction models based on systematic review and meta-analysis. However, there are no studies summarizing the methods used.
Objective
To summarize methods and key steps for constructing prediction model based on systematic review and meta-analysis.
Methods
We searched PubMed, Embase, Web of Science, Wanfang Data and China National Knowledge Infrastructure until April 2023 to identify prediction models that developed based on systematic review and meta-analysis. We have no language restrictions.
Results
We included 94 eligible studies. Of these studies, there were three methods based on meta-analysis to obtain predictors included in the prediction model. The first method (n=59) is obtaining statically significant risk factors directly based on meta-analysis. The second method (n=4) is combining results from the first method with expert opinions or clinical experience to determine predictors to be included in the model. The third method (n=31) is to make a questionnaire based on the results from the first or second method for clinical investigation, so as to obtain participants’ data, and conduct a multi-factor analysis to identify predictors. The key steps of developing prediction models include: formulating a clinical question, choosing appropriate type of prediction model, collecting and processing data, selecting variable (based on meta-analysis or multi-factor analysis), developing prediction model, and presenting the full prediction model.
Conclusions
The key steps of developing prediction model based on meta-analysis were generally the same as those of traditional prediction model development methods, and the main difference were variable selection methods. The former usually obtained statistically significant risk factors based on meta-analysis.
High-quality systematic review and meta-analysis is a proven method that sits at the top of the evidence pyramid, which could improve statistical efficiency. In recent years, multiple studies have been published on constructing prediction models based on systematic review and meta-analysis. However, there are no studies summarizing the methods used.
Objective
To summarize methods and key steps for constructing prediction model based on systematic review and meta-analysis.
Methods
We searched PubMed, Embase, Web of Science, Wanfang Data and China National Knowledge Infrastructure until April 2023 to identify prediction models that developed based on systematic review and meta-analysis. We have no language restrictions.
Results
We included 94 eligible studies. Of these studies, there were three methods based on meta-analysis to obtain predictors included in the prediction model. The first method (n=59) is obtaining statically significant risk factors directly based on meta-analysis. The second method (n=4) is combining results from the first method with expert opinions or clinical experience to determine predictors to be included in the model. The third method (n=31) is to make a questionnaire based on the results from the first or second method for clinical investigation, so as to obtain participants’ data, and conduct a multi-factor analysis to identify predictors. The key steps of developing prediction models include: formulating a clinical question, choosing appropriate type of prediction model, collecting and processing data, selecting variable (based on meta-analysis or multi-factor analysis), developing prediction model, and presenting the full prediction model.
Conclusions
The key steps of developing prediction model based on meta-analysis were generally the same as those of traditional prediction model development methods, and the main difference were variable selection methods. The former usually obtained statistically significant risk factors based on meta-analysis.