Article type
Year
Abstract
Background:In observational studies, the key limitation is the potential for confounding bias because exposures are not random. In the past, researchers used to control the confounding bias by matching, stratification and regression analysis. But when facing unmeasurable confounding bias, it is found that both statistical method bias analysis and additional assumptions are affected by researchers subjectively, as well as too simplified premise assumptions, which makes the unmeasurable confounding bias exert an inestimable impact on the conclusion.
Objectives:Through systematic investigation, to describe the progress of E-value application in observational research, and explore the advantages and limitations of E-value application in observational research.
Methods:
A systematic search was conducted in CNKI and PubMed (from 2018 to 2019), including medical-related observational studies using E-values.
Results:
a total of 180 articles were retrieved and 48 research reports were selected. In these reports, 26 cohort studies, 10 case-control studies, 7 cross-sectional studies, 1 clinical randomized trial and 4 meta-analyses were included. It is worth noting that in all reports, for mixed bias, most of them use 2-3 sensitivity analysis methods with E-value in class, such as stratified analysis, tendency score analysis, instrumental variable analysis, multivariate analysis linear regression or logistic regression or Cox regression. However, E-value is often used as a sensitivity analysis method for unmeasurable confounding bias. Researchers try to further enhance the reliability and robustness of the results with E-value. We found that the most commonly used E-values were about cardiovascular and cerebrovascular diseases, up to 13 studies, followed by tumor diseases and drug use evaluation, both of which had 8 reports. Of course, the application of E-value has become a common sensitivity analysis method in recent years. The investigation shows that the maximum value of E is 38.9, which was mentioned by Carl Michael baraveli et al. In porphyria cutanea tarda increases risk of the patriotic carcinoma and prediction death: a national short study. Among the research exposures involved, disease and drug are the main indicators of exposure, 17 reports mentioned in the former and 12 in the latter. In addition, there are 5 reports using social problems such as psychological problems, policies and regulations as indicators of exposure.
Conclusions:
The application of E-value in sensitivity analysis, especially in observational research, due to the influence of many factors, can not completely and accurately evaluate the correlation of confounding factors with the results，but E-value can make the causal inference of observational research more credible.
Patient or healthcare consumer involvement:
Objectives:Through systematic investigation, to describe the progress of E-value application in observational research, and explore the advantages and limitations of E-value application in observational research.
Methods:
A systematic search was conducted in CNKI and PubMed (from 2018 to 2019), including medical-related observational studies using E-values.
Results:
a total of 180 articles were retrieved and 48 research reports were selected. In these reports, 26 cohort studies, 10 case-control studies, 7 cross-sectional studies, 1 clinical randomized trial and 4 meta-analyses were included. It is worth noting that in all reports, for mixed bias, most of them use 2-3 sensitivity analysis methods with E-value in class, such as stratified analysis, tendency score analysis, instrumental variable analysis, multivariate analysis linear regression or logistic regression or Cox regression. However, E-value is often used as a sensitivity analysis method for unmeasurable confounding bias. Researchers try to further enhance the reliability and robustness of the results with E-value. We found that the most commonly used E-values were about cardiovascular and cerebrovascular diseases, up to 13 studies, followed by tumor diseases and drug use evaluation, both of which had 8 reports. Of course, the application of E-value has become a common sensitivity analysis method in recent years. The investigation shows that the maximum value of E is 38.9, which was mentioned by Carl Michael baraveli et al. In porphyria cutanea tarda increases risk of the patriotic carcinoma and prediction death: a national short study. Among the research exposures involved, disease and drug are the main indicators of exposure, 17 reports mentioned in the former and 12 in the latter. In addition, there are 5 reports using social problems such as psychological problems, policies and regulations as indicators of exposure.
Conclusions:
The application of E-value in sensitivity analysis, especially in observational research, due to the influence of many factors, can not completely and accurately evaluate the correlation of confounding factors with the results，but E-value can make the causal inference of observational research more credible.
Patient or healthcare consumer involvement: