ObjectiveTo introduce a meta-analysis of linear or nonlinear multilevel models using SAS MIXED and SAS NLMIXED.MethodsA systematic review performed to evaluate the risk of local recurrence in patients with cervical cancer treated with radical chemoradiotherapy and adjuvant surgery published by Shim et al. was selected as an illustration. An SAS software was used to implement meta-analysis based on linear or nonlinear multi-level models, and programming codes were provided.ResultsIn the absence of covariates, the OR combined effect values of PROC MIXED based on the bivariate random effects model and PROC NLMIXED of the nonlinear mixed effects model were 0.63 (95%CI 0.46 to 0.87, P=0.005 7) and 0.60 (95%CI 0.39 to 0.81, P=0.000 3), respectively. In the case of covariates, the bivariate random effects model and the nonlinear mixed effects model provided an effect value of OR=0.65 (95%CI 0.47 to 0.91, P=0.011) and 0.59 (95%CI 0.38 to 0.80, P=0.000 3). Covariate OR effect values were 2.70 (95%CI 0.16 to 45.23, P>0.05) and 1.86 (95%CI −0.07 to 3.79, P=0.06).ConclusionsThe meta-analysis results of the SAS NLMIXED nonlinear mixed-effects model are similar to those of the SAS MIXED linear mixed-effects model. PROC NLMIXED has powerful programming capability and nonlinear mixed-effects model has flexible modeling capabilities for sparse data. Therefore, PROC NLMIXED will play an increasingly important role in meta-analysis.
The meta-analysis of rare binary data is a difficulty in the field of medical research, and its methodology remains immature. The traditional meta-analysis technique is based on the normal-normal model of fixed effects analysis or random-effects analysis, however there are methodological problems in this method. Stijnen proposed an exact within-study likelihood models (EWLM) meta-analysis technique based on the generalized linear mixed model (GLMM), including the binomial-normal model (BN) and Hypergeometric-normal model (HNM), which can be used to achieve random effects meta-analysis of rare binary data. This paper introduces the model in detail and its implementation in SAS software with examples to provide relevant SAS code.