Dose-response relationship model has been widely used in epidemiology studies, as well as in evidence-based medicine area. In dose-response meta-analysis, the results are highly depended on the raw data. However, many primary studies did not provide sufficient data and led the difficulties in data analysis. The efficiency and response rate of collecting the raw data from original authors were always low, thus, evaluating and transforming the missing data is very important. In this paper, we summarized several types of missing data, and introduced how to estimate the missing data and transform the effect measure using the existed information.
When investing the relationship between independent and dependent variables in dose-response meta-analysis, the common method is to fit a regression function. A well-established model should take both linear and non-linear relationship into consideration. Traditional linear dose-response meta-analysis model showed poor applicability since it was based on simple linear function. We introduced a piecewise linear function into dose-response meta-analysis model which overcame this problem. In this paper, we will give a detailed discussion on traditional linear and piecewise linear regression model in dose-response meta-analysis.
ObjectiveTo systematically review the dose-response relationship between body mass index (BMI) and the risk of stroke. MethodsPubMed, EMbase, Web of Science, The Cochrane Library, CBM, VIP, WanFang Data and CNKI databases were electronically searched to collect studies on BMI and the risk of stroke from inception to December 2021. Two reviewers independently screened literature, extracted data and assessed the risk of bias of included studies; then, meta-analysis was performed by using Stata 16.0 software, and the dose-response relationship between BMI and risk of stroke was analyzed by using restricted cubic spline function and generalized least squares estimation (GLST). ResultsA total of 19 studies involving 3 689 589 patients were included. The results of meta-analysis showed that compared with normal BMI, overweight (RR=1.28, 95%CI 1.19 to 1.39, P<0.01) and obesity (RR=1.41, 95%CI 1.15 to 1.72, P<0.01) had a higher risk of stroke. Dose-response meta-analysis suggested that there was no significant non-linear relationship between BMI and stroke risk (nonlinear test P=0.318), and linear trend showed that the risk of stroke increased by 4% for each unit increase in BMI (RR=1.04, 95%CI 1.03 to 1.05, P<0.01). ConclusionCurrent evidence suggests that increased BMI is associated with an increased risk of stroke. Due to limited quality and quantity of the included studies, more high-quality studies are needed to verify the above conclusion.
ObjectiveTo systematically evaluate the dose-response relationship between coffee consumption and liver cancer risk. MethodsThe PubMed, Web of Science, Cochrane Library, EMbase, CNKI, VIP, WanFang Data, and CBM databases were searched from inception to December 2022. Two reviewers independently screened literature, extracted data and assessed the risk of bias of the included studies. Meta-analysis was then performed by using Stata 17.0 software. ResultsFifteen studies (11 cohort studies and 4 case-control studies) involving 557 259 participants were included. The results of meta-analysis showed that coffee consumption was significantly negatively associated with the risk of liver cancer (RR=0.39, 95%CI 0.27 to 0.57, P<0.01). The dose-response meta-analysis showed a non-linear dose-response relationship between coffee consumption and the risk of liver cancer (P<0.01). Compared with people who did not drink coffee, people who drank 1 cup of coffee a day had a 25% lower risk of liver cancer (RR=0.75, 95%CI 0.67 to 0.83), and people who drank 2 cups of coffee a day had a 38% lower risk of liver cancer (RR=0.62, 95%CI 0.56 to 0.70). The risk of liver cancer decreased by 45% (RR=0.55, 95%CI 0.48 to 0.62) for 3 cups of coffee and by 51% (RR=0.49, 95%CI 0.43 to 0.56) for 4 cups of coffee. ConclusionCurrent evidence suggests that there is a nonlinear dose-response relationship between coffee consumption and the risk of liver cancer. These results indicate that habitual coffee consumption is a protective factor for liver cancer. Due to the limited quality and quantity of the included studies, more high quality studies are needed to verify the above conclusion.
Dose-response meta-analysis, as a subset of meta-analysis, plays an important role in dealing with the relationship between exposure level and risk of diseases. Traditional models limited in linear regression between the independent variables and the dependent variable. With the development of methodology and functional model, Nonlinear regression method was applied to dose-response meta-analysis, such as restricted cubic spline regression, quadratic B-spline regression. However, in these methods, the term and order of the independent variables have been assigned that may not suit for any trend distribution and it may lead to over fitting. Flexible fraction polynomial regression is a good method to solve this problem, which modelling a flexible fraction polynomial and choosing the best fitting model by using the likelihood-ratio test for a more accurate evaluation. In this article, we will discuss how to conduct a dose-response meta-analysis by flexible fraction polynomial.
As a valid method in systematic review, dose-response meta-analysis is widely used in investigating the relationship between independent variable and dependent variable, and which usually based on observational studies. With large sample size, observational studies can provide a reasonable amount of statistical power for meta-analysis. However, due to the design defects of observational studies, they tend to introduce many kinds of biases, which may influence the final results that make them deviation from the truth. Given the dead zone of methodology, there is no any bias adjusting method in dose-response meta-analysis. In this article, we will introduce some bias adjusting methods from other observational-study-based meta-analysis and make them suit for dose-response meta-analysis, and then compare the advantages and disadvantages of these methods.
ObjectiveTo systematically review the dose-response relationship between body mass index (BMI) and all-cause mortality in the elderly with frailty.MethodsPubMed, EMbase, Web of Science, CNKI, VIP, WanFang Data, and CBM databases were electronically searched to collect cohort studies on the association of BMI and mortality in frail adults from inception to November 2019. Two reviewers independently screened literature, extracted data and assessed risk bias of included studies; Stata 15.0 software was then used to analyze the dose-response analysis of BMI and mortality by restricted cubic spline function and generalized least squares method.ResultsA total of 4 cohort studies involving 12 861 frail adults were included. Meta-analysis results showed that compared with normal BMI, the frail elderly who were overweight (HR=0.80, 95%CI 0.74 to 0.88, P<0.001) and obese (HR=0.89, 95%CI 0.79 to 1.00, P=0.047) had lower all-cause mortality. The results of dose-response meta-analysis showed that there was a non-linear relationship between BMI and all-cause mortality in the elderly with frailty (P value for nonlinearity was 0.035), for which the elderly with frailty had a BMI nadir of 27.5-31.9 kg/m2. For linear trends, and when BMI was less than 27.5 kg/m2, the risk of all-cause death was reduced by 4% for every 1 kg/m2 increase in BMI (RR=0.96, 95%CI 0.90 to 1.03, P=0.320), when BMI was greater than 27.5 kg/m2, the risk of all-cause death increased by 4% for every 1 kg/m2 increase in BMI (RR=1.04, 95%CI 1.03 to 1.05, P<0.001).ConclusionsThere is a paradox of obesity and a significant nonlinear relationship between BMI and all-cause mortality in the frailty elderly, with the lowest all-cause mortality in the frailty elderly at BMI 27.5-31.9 kg/m2. Due to limited quality and quantity of the included studies, more high quality studies are needed to verify the above conclusions.
In evidence-based practice and decision, dose-response meta-analysis has been concerned by many scholars. It can provide unique dose-response relationship between exposure and disease, with a high grade of evidence among observational-study based meta-analysis. Thus, it is important to clearly understand this type of meta-analysis on software implementations. Currently, there are different software for dose-response meta-analysis with various characteristics. In this paper, we will focus on how to conduct dose-response meta-analysis by Stata, R and SAS software, which including a brief introduction, the process of calculation, the graph drawing, the generalization, and some examples of the processes.
Dose-response meta-analysis is being increasingly applied in evidence production and clinical decision. The research method, synthesizing certain dose-specific effects across studies with the same target question by a certain types of weighting schedule to get a mean dose-response effect, is to reflect the dose-response relationship between certain exposure and outcome. Currently, the most popular method for dose-response meta-analysis is based on the classical "two-stage approach", with the advantage that it allows fixed- or random-effect model, according to the amount of heterogeneity in the model. There are two types of random-effect model available for dose-response meta-analysis, that is, the generally model and the coefficient-correlation-adjusted model. In this article, we briefly introduce two models and illustrate how they are applied in Stata software, which is expected to provide theoretical foundation for evidence-based practice.
According to the heterogeneity between dose-response data across different studies and the potential nonlinear trend within the dose-response relationship, there are several models for trend estimation from summarized dose-response data, with applications to meta-analysis. However, up to now, there is no guideline of conducting a metaanalysis of dose-response data. After summarizing the previous papers, this paper focuses on how to select the right model for conducting a meta-analysis of dose-response data based on the heterogeneity across different studies, the goodness of fit, and the P value of overall association between exposure and event. Then a preliminary statistical process of conducting a meta-analysis of dose-response data is proposed.
