A Study Of Combining IPD And AgD In Network Meta Regression Model

Background: To provide valuable evidence for decision making,when direct comparisons are not available in existing studies,or when the risk of bias in direct comparisons is high,network meta-analysis and indirect comparisons can use common interventions as controls,make comparisons between multiple interventions to inform decisions,and potentially improve the statistical power and accuracy of analysis.Indirect comparisons can link interventions in different studies through common controls,and due to the large bias of standard indirect comparisons,the main methods used at this stage are adjusted indirect comparisons,which can be adjusted for study populations with only aggregate data(Ag D)by using study populations containing with individual patient data(IPD).The main methods are matched-adjusted indirect comparisons and simulated treatment methods.These two methods are only applicable to comparisons between two studies and the results are only applicable to the population studied with aggregated data.Therefore the proposed mesh meta-analysis approach is able to combine direct and indirect evidence to extend the research and treatment network and to explore more complex evidence.Classical network meta-analysis allows the analysis of aggregate data,but there are modifying effect factors that affect the analysis of the intervention’s effect strength.Network meta-regression adjusts for important effect modifiers,but the lack of individual patient data introduces aggregation bias.The "gold standard" individual patient data network meta analysis(IPD-NMA)of individual case data can analyze all IPD in the body of evidence,but in practice it is difficult to obtain all accurate raw study data in the body of evidence,and missing and inaccurate data can introduce bias.Therefore,there is a need to develop a reticulated meta-regression analysis model that combines IPD with Ag D to improve the accuracy and efficiency of network meta-analysis.Objectives:(1)To develop a statistical analysis model that combines IPD with Ag D in a network meta-analysis based on existing studies;(2)Using simulation studies and empirical research methods to investigate the advantages of the new model compared with other population adjustment methods in indirect comparisons and to explore the effects of the new model in the application of survival outcome data.Method: Through theoretical studies,a network meta-regression model that simultaneously handles IPD and Ag D was developed based on existing studies.And through three studys,the model performance was investigated and compared with other population adjustment methods.1.Simulated study: Simulated indirect comparison datasets were constructed containing 2 studies of 3 interventions,one of which provides individual patient data.Four study scenarios were set for different sample sizes,strength of different effect modification effects,whether the shared effect modifier hypothesis holds,and strength of different covariate associations.Effect measures were selected for bias,standard error,and coverage probability.We applied the new model of matching adjusted indirect comparisons,simulated treatment comparisons and standard indirect comparisons were analyzed to examine the performance of different methods under different scenarios.2.Empirical study: Data were obtained from published studies on psoriasis treatment,containing 2 series of 4 studies with 6 interventions,one series providing IPD and the other only including Ag D.Some data were first taken for comparison of population adjustment methods in different indirect comparisons,and then a new model was applied to the entire study network to investigate the statistical power of the model.3.Applied study: A simulated dataset of indirect comparisons of survival outcome data was constructed,including 2 studies of 3 interventions.It was assumed that one of the studies contain only Ag Dand apply the new model for analysis,and that the complete data would be applied to an individual patient data network meta-analysis.Standard indirect comparisons would be conducted for reference.The log risk ratio was chosen as an indicator of the outcome estimate.The results and performance of the new model analysis using only partial IPD were compared with the IPD-NMA and with the true values.Results:(1)Simulation study: In different scenarios,it was found that both the new model and STC are regression-based,the results obtained are more similar,but the simulated treatment approach is more difficult to apply to larger study networks.MAIC used a re-weighting approach that did not allow for inference,so the results performed worse and in some cases even weaker than the standard indirect comparison.(2)Empirical study: The estimation power of indirect comparisons can be improved by population adjustment methods,the model combining two levels of data can use more data information,its confidence intervals were smaller,the results were more reliable,and it can be compared across populations and is not limited to aggregated level populations.The new model can compare different interventions in each study population and the outcome estimates are similar to the observed values,with a better ability to reflect the true situation and good extrapolation of results.(3)Implication study: The performance of the two models was comparable,in the study population with aggregate data,the lack of individual level data did not reduce their precision too much.For values that could be observed in the study population,the log-risk ratios estimated by the two methods had almost no error,and for values that could not be directly observed,the estimates of the two methods differed slightly,but the difference was not significant and both were similar to the true values.The estimates of the model parameters by the two methods differed little and were both close to the true values,and were considered to more accurately recover the true situation of the parameters.Conclusion: The network meta-regression model based on the combination of IPD and Ag D had high statistical performance in ideal or simulated constructed scenarios.It takes into account the effect of the distribution of effect modifiers on the effect of interventions,accounts for inter-and intra-study heterogeneity,provides a model with greater statistical accuracy and stability,and could be applied to larger study networks.It is expected that further research will lead to practical applications in health technology assessment in the future to provide more accurate results for health decision making.