Research On Statistical Methods To Explore The Level Of Vaccine Protective Antibodies Based On Immunogenicity Data

Backround An immunological surrogate endpoints is a vaccine-induced immune response(either humoral or cellular immunity)that predicts protection against clinical endpoints(infection or disease),and can be used to evaluate vaccine efficacy in clinical vaccine trials.Compared to field efficacy trials observing clinical endpoints,immunological vaccine trials could reduce the sample size or shorten the duration of a trial,which promote the license and development of new candidate vaccines.For these reasons,establishing immunological surrogate endpoints is one of the 14 Grand Challenges of the Global Health of National Institutes of Health(NIH)and the Bill and Melinda Gates Foundation.Objective Validate and compare the statistical methods used in the existing literature to determine the correlation between immunogenicity and protection effectiveness,and explore the application conditions of the method through actual data and simulation data.Further,explore new methods that can be used to determine the correlation between immunogenicity and protective effects,and apply actual and simulated data to evaluate new methods.Method The traditional and new statistical models that could be used for evaluation of surrogate endpoints has been evaluated by fitting actual data and the simulated data under different scenes.The predictive value of the antibody has been used to explore the application conditions of different models.Result ROC model,Threshold model and Logit model were applied to fit the clinical data of the immunogenicity subgroup come from EV71 phase ? clinical trial.The results showed that Logit model and threshold model had a high level of antibody which could protect the subjects from disease with a target protection rate of 50%(121.05U/ml for Logit model and 93.75U/ml for threshold model),while the antibody level determined by the ROC model was much lower(46.88U/ml).Simulation result showed that ROC model and threshold model were less affected by sample size compared to the Logit model.And the Logit model was more accurate in the studies with larger sample size and a higher exposure rate.Bayes model was applied to fit the clinical data of immunogenicity subgroup of EV71 phase ? clinical trial.The results showed that subjects with antibody level above 27.95U/ml could be protected from disease with a target protection rate of 50%.This level was much lower than the results obtained by traditional models(121.05U/ml for Logit model,93.75U/ml for threshold model and 46.88U/ml for ROC model).This conclusion was consistent with the theoretical significance of the Bayes Logit method.The simulation result showed that the Bayesian Logit model had certain dependence on the prior information of parameter gamma,and under the premise of an accurate prior of the parameter Gama,the Bayesian Logit model can accurately estimate the threshold of protection.Although the results were affected by the sample size and the exposure rate parameters to a certain extent,but the bias was quite moderate.If the sample size is large enough and the exposure rate is high enough,the Bayesian Logit model would be very efficient.Accelerated failure time model and COX model were applied to fit the clinical data of immunogenicity subgroup of EV71 phase ? clinical trial.Results showed that when the antibody level was above 76.09U/ml and 97.03U/ml,subjects could be protected from disease with a target protection rate of 50%,respectively.Results of simulation study showed that the baseline distribution of survival time followed logarithmic Logistic distribution.The point estimate of the predicted antibody levels calculated by the COX model and the Accelerated failure time model using logarithmic Logistic distribution as the base survival time distribution was quite accurate.Although both two models were affected by the exposure rate and sample size,the bias was still acceptable.Conclusion Traditional models were consistent.The ROC model and the threshold model were less affected by sample size compared to the Logit model.And the Logit model was more accurate in the studies with larger sample size and a higher exposure rate.Bayesian Logit model had certain dependence on the prior information of parameter gamma,and under the premise of an accurate prior of the parameter Gama,the Bayesian Logit model can accurately estimate the threshold of protection.Although the results were affected by the sample size and the exposure rate parameters to a certain extent,the bias was quite moderate.If the sample size is large enough and the exposure rate is high enough,the Bayesian Logit model would be very efficient.The point estimate of the predicted antibody levels calculated by the COX model and the Accelerated failure time model using logarithmic Logistic distribution as the base survival time distribution was quite accurate.Although both two models were affected by the exposure rate and sample size,the bias was still acceptable.