Meta-analysis of survival data is becoming more and more popular. The data could be extracted from the original literature, such as hazard ratio (HR) and its 95% confidence interval, the difference of actual frequency and theoretical frequency (O - E) and its standard deviation. The data can be used to calculate the combined HR using Review Manager (RevMan), Stata and R softwares. RevMan software is easy to learn, but there are some limitations. Stata and R software are powerful and flexible, and they are able to draw a variety of graphics, however, they need to be programmed to achieve.
In systematic reviews and meta-analyses, time-to-event outcomes were mostly analysed using hazard ratios (HR). It was neglected transformation of the data so that some wrong outcomes were gained. This study introduces how to use Stata and R software to calculate the HR correctly if the report presents HR and confidence intervals were gained.
Hazard ratio (HR) is usually regarded as the effect size in survival studies. Meanwhile, it is supposed to be perfect for pooling results in the meta-analysis of survival data. However, it does not function usually due to absence of original data for pooling HR. As a compromise method, entering data from reading Kaplan-Meier curves and follow-up times into the calculation spreadsheet can also be used to obtain related survival data. But related study on the subject is scarce, and opinions are inconsistent. Accordingly, we conduct this study to further illustrate the procedure in details.
Network meta-analyses (NMA) of survival data often rely on the proportional hazards (PH) assumption, however, this assumption fails when survival curves intersect. With the emergence of innovative therapies such as immunotherapy, the importance of NMA based on non-proportional hazards (non-PH) in the current evidence-based medicine evaluation of oncology drugs has become increasingly prominent. Fractional polynomial (FP) models do not rely on the assumption of PH, which can flexibly capture the characteristics of survival curves, and the corresponding fitting effects are better than those of the PH models. This study introduced a complete workflow in R for NMA using FP models with non-PH.
Survival data include the occurrence and duration of an event. As most survival data are distributed irregularly, the Kaplan-Meier method is often used in survival analysis; however, studies usually only report the Kaplan-Meier curve and median survival time and do not provide the original survival data, which creates issues for subsequent secondary research. This study introduced a systematic method whereby image processing software and R software were used to process and extract survival data from published Kaplan-Meier curves. It also introduced the specific steps required to obtain survival data using an example to show the accuracy and feasibility of the extraction method and provided references for the effective secondary use of survival data.
With the increase in the number of single-arm clinical trials and lack of head-to-head clinical studies, the application of unadjusted indirect comparisons and network meta-analysis methods has been limited. Matching-adjusted indirect comparison (MAIC) is an alternative method to fully utilize individual patient data from one study and balance potential bias caused by baseline characteristics differences in different trials through propensity score matching with aggregated data reported in other studies, and complete the comparison of the efficacy between target interventions. This study introduced the concept and principles of MAIC. In addition, we demonstrated how to use the anchored MAIC method based on R language for survival data, which has been widely used in anti-cancer drug evaluation. This study aimed to provide an alternative method to inform evidence-based decisions.