Statistical Methods For The Evaluation Of The Effectiveness Of The COVID-19 Vaccine

Statistical analysis of vaccine clinical trials is a cross-cutting topic between clinical medicine and statistics,and the results of vaccine effectiveness are of great interest,so it is important to explore the effectiveness of vaccines using reasonable statistical testing methods.In this study,we will take two clinical trials unfolding with two different endpoints of COVID-19 vaccine effectiveness evaluation as examples,and analyze them with different statistical analysis methods for each endpoint,aiming to compare the changes of COVID-19 vaccine effectiveness evaluation under different treatment methods.Since the handling of missing data and sensitivity analysis remain more conservative in most clinical trials,the importance of sensitivity analysis and missing data handling can be demonstrated by two specific clinical trials.First,a randomized double-blind placebo-controlled clinical trial was designed with confirmed cases of COVID-19 pneumonia as the endpoint,and the data from this trial were statistically analyzed by five different statistical analysis methods(i.e.,Logistic regression model,Poisson distribution with stratification factors,Poisson distribution without stratification factors,Cox proportional risk regression model,and Incidence-based analysis).After comparing the 95% CI and sensitivity analysis,it can be found that: the mean values of the estimated protection rate of this vaccine are 27.90%(E-m FAS)and 28.6%(E-PPS),and the exact Poisson distribution model based on the human annual incidence rate has the smallest lower limit of protection rate and the largest width of 95% CI;the statistical analysis model based on the incidence rate has the largest lower limit of protection rate and the smallest width of 95% CI and the greatest degree of bias.However,the lower limits of the95% CI failed to meet the criterion of superiority.Significantly lower protection rates could be observed in the subgroup analysis for those aged 60 years and older and those with hypertension,which could further explore the vaccine effectiveness in this particular population.Second,a randomized double-blind controlled clinical trial was designed with GMT levels as the surrogate endpoint,and the ratio of GMT in subjects was evaluated using analysis of covariance,analysis of variance,and sensitivity analysis with LOCF filling and MI filling for missing data.The comparison of GMT ratio estimates and 95% CI revealed that: GMT ratios reached the non-inferiority bound in all cases,and it was observed in the sensitivity analysis that the results after MI-filling were closest to those based on the PPS analysis set,while the results after LOCF-filling were more conservative and had wider confidence intervals and the greatest degree of bias.Finally,based on case studies,this paper makes the following four suggestions:(1)to increase the importance of sensitivity analysis;(2)to strengthen the attention to the data preprocessing stage;(3)to refine the division of the population analysis set in statistical analysis;(4)to strengthen the communication and cooperation with experts in the field related to clinical trials.