Abstract:
|
In medical sciences, a disease is associated with multiple risk/protective factors. While many studies report multiple factors, nearly all meta-analyses separately synthesized the association between each factor and the disease condition of interest. Since each study may only report a subset of all factors, results from separate meta-analyses may not be comparable as each may use different subpopulation. This limits our ability to select most important factors for the design of a multifactor intervention program. From the perspective of missing data analysis, we propose a novel network meta-analysis method to jointly model multiple factors (NMA-MF) simultaneously. By borrowing information across multiple factors, NMA-MF greatly improves statistical efficiency and reduce potential biases. As within-study correlations between multiple factors are commonly unknown from published articles, we introduce a hybrid random-effects model to perform NMA-MF, which effectively accounts for both within- and between-study correlations. The performance of the proposed method is studied using simulations, and illustrated using a real dataset of 8 risk factors on pterygium synthesizing 29 studies.
|