| Background:Individual patient data (IPD) meta-analyses (MAs), which involve the central collection,checking, and re-analysis of updated IPD, have been described as the gold standard ofsystematic review. However, this approach is not always practical, often due to economic,resource, or time constraints. Some meta-analyses use summary data, which are suppliedby the trialists, but more often than not, meta-analyses are performed by extracting the datafrom the published literature. This approach is prone to many biases, such as reporting bias,publication bias, and patient exclusion bias. Time to event outcomes are most appropriatelyanalyzed by calculating HRs where individual durations of survival are used to calculatethe overall instantaneous risk of event on experimental compared with control intervention.Such analyses are most easily done using individual patient data and indeed many IPDmeta-analyses are done primarily because time to event analyses are essential to theproject.For each trial, the HR and its variance are derived from the log rank statistic, calculatedwith time to event for individual patients. The pooled logHR is a weighted average of thelogHRs with the weights inversely proportional to the variances of the studies.Asymptotically, the logHR follows a standard normal distribution under the null hypothesisof no treatment effect. The overall pooled HR represents the overall risk of dying on theexperimental treatment compared with the control treatment. However, the combined HRonly convey that the relative risk of the event between experimental group and controlgroup. The actual benefit of the treatment was not expressed vividly. The clinicians oftenmore concern about the direct effect of a treatment rather than the relative hazard risk. Inthese circumstances, median survival time can convey more direct information of thebenefit of a treatment.Aim:This thesis aims to explore an alternative method using median survival times whenconducting meta-analysis using survival data, which may be more readily available inpublications and a more vivid index in the survival analysis. We planned to conduct astatistical model for synthesis of median survival time in meta-analysis by generic inversevariance method weighted by the variance of lnHRs or the event and sample size of experiment or control group. After testifying the methods of statistical model for synthesisof median survival time in meta-analysis, the actual meta-analysis will be done byproviding the combined median survival time。Methods:Given individual patient data meta-analyses is considered as the gold standard ofsystematic review, and there is few individual patient data available in the publishedsurvival analysis due to economic, resource, or time constraints. We employed simulateddata for statistical modeling for the synthesis of median survival time in meta-analysis. Thegold standard of median survival time is also computed from the simulated data. Then wecompared the combined median survival time by statistical model and the gold standardusing correlation coefficient method, Bland-Altman method and Allowable Total Error andLimits for Erroneous Results (ATE/LER) zones method.After verifying the statistical methods, the combined median survival time will becomputed for the meta-analysis of small molecule small molecule epidermal growth factorreceptor tyrosine kinase inhibitors (EGFR TKIs) in continuous or switch maintenancetherapy for non-small-cell lung cancer (NSCLC).Results:(1) The study generated of survival times from the exponential distribution for constanthazards. The simulated variables are group, number, survival time and survival outcome.The raw data were manipulated in two ways,1) The hazard ratios and their95%confidence intervals, the median survival time in each experiment and control group, thesample size and event number were derived from the raw data. This information was usedto simulate the extracted data form published literature when carrying out meta-analysis ofsurvival analysis. The summarized data were also used to statistically modeling for thesynthesis of the median survival time.2) The raw data were also combined to construct theIPD dataset. The survival information was derived from the IPD dataset. The mediansurvival time in the two groups is considered to be the gold standard to compare with themedian survival time computed from the models.Five simulated datasets were generated which are M20S500, M10S500, M10S200,M5S400and M5S200(M: meta-analysis; the number behind M: the number of studyincluded in the meta-analysis; S: study; the number behind S: the number of individualpatients included in the study). For each IPD dataset, there are10000,5000,2000or1000patients in each meta-analysis, which might included20,10or5studies. Five hundred times simulation was conducted to obtain the standard median survival time and computedmedian survival time from the model.(2) Statistical modeling: the model was constructed based on the weighted average method.To perform a meta-analysis and statistical modeling for the synthesis of median survivaltime we compute an effect size and variance for each study, and then compute a weightedmean of these effect sizes. The variances for each study in this study are as follow:The95%confidence interval of HR and the event number and sample size of each groupwere used to calculate the variance.(3) The measurement agreement between the median survival time derived from themodels and the gold standard:1)HR confidence interval weighted methods: for the data ofM20S500the correlation coefficient (r) between the results of the two methods is0.99andp<0.001; The95%confidence interval of deviation between calculated and gold standardmedian survival time is ranged from-1.2744to1.6390in Bland-Altman test; When theclinical cutoff was set to2, which represent a small difference of median survival timebetween calculated one and gold standard, the percentage of all the plots falling within theATE zones and LER Zones is99.3%.As for the simulated data of M10S500, M10S200,M5S400and M5S200, the similar results were obtained. All the r between the results of thetwo methods is above0.99and p<0.001; The95%confidence interval of deviationbetween calculated and gold standard median survival time are al ranged from-3to3inBland-Altman test; The percentage of all the plots falling within the ATE zones and LERZones is above86.4%.2) Event and sample size weighted method: for the data ofM20S500the correlation coefficient (r) between the results of the two methods is0.99andp<0.001; The95%confidence interval of deviation between calculated and gold standardmedian survival time is ranged from-2.1571to1.5818in Bland-Altman test; When theclinical cutoff was set to2, which represent a small difference of median survival timebetween calculated one and gold standard, the percentage of all the plots falling within theATE zones and LER Zones is98.7%.As for the simulated data of M10S500, M10S200,M5S400and M5S200, the similar results were obtained. All the r between the results of thetwo methods is above0.99and p<0.001; The95%confidence interval of deviation between calculated and gold standard median survival time are al ranged from-5to5inBland-Altman test; The percentage of all the plots falling within the ATE zones and LERZones is above78.6%.(4) As for the actual meta-analysis of survival analysis: Trials, which compared mostfrequently used small molecule EGFR TKIs with respective control involving maintenancetreatment of patients with NSCLC, were retrieved. Data from the trials for overall survival(OS), progression-free survival (PFS), adverse events (AE) and QoL were gathered andsummarized. Nine trials including6,655patients were identified. Continuous maintenancewith small molecule EGFR TKIs failed to improve OS (Hazard ratio (HR)1.05,95%confidence interval (CI)0.98-1.14; p=0.186) or PFS (HR0.88,95%CI0.74-1.05;p=0.165). The median survival time is10.24month for experiment group and10.55monthfor control group. The median progression time is5.36month for experiment group and5.18month for control group. In contrast, switch maintenance with small molecule EGFRTKIs significantly improved OS (HR0.87,95%CI0.80-0.95; p=0.001) and PFS (HR0.75,95%CI0.69-0.81; p<0.001). The median survival time is12.18month for experimentgroup and11.43month for control group. The median progression time is3.22month forexperiment group and2.42month for control group. Subgroup analysis found thatstatistically significant improvement of switch maintenance therapy on OS and PFS infemale, Asians, never smokers, patients with adenocarcinoma, with EGFR mutations andin patients with stable disease at baseline pre-randomization. The effect on OS wassignificantly greater in trials with switch than continuous maintenance with small moleculeEGFR TKIs (0.87vs1.05; interaction p=0.0013). Post-maintenance treatment balancedbetween experimental and control arms in this meta-analysis. Maintenance therapy wasassociated with a little higher rate of AE but better trend of QoL.Conclusions:The simulated datasets was generated for survival analysis; the variables in the raw datawere group variable, survival time, survival outcome. Five simulated datasets weregenerated which are M20S500, M10S500, M10S200, M5S400AND M5S200. For eachIPD dataset, there are10000,5000,2000or1000patients in each meta-analysis, whichmight included20,10or5studies. Five hundred times simulation was conducted to obtainthe standard median survival time and computed median survival time from the model. Thestatistical model for the synthesis of median survival time weighted by the confidenceinterval of HRs are more properly than weighted by the events and sample size of each groups. As for the actual meta-analysis about the effect and risk of small molecule EGFRTKIs for patients with NSCLC, the small molecule EGFR TKIs improved OS and PFS inswitch maintenance therapy. It was suggested that the appropriate drugs administrationapproaches and specified sensitive patients should be considered when choosing smallmolecule EGFR TKIs for maintenance therapy.
|
PDF Full Text Request