Background: Survival meta-analysis combines the results of a series of clinical trials that have compared two treatments in terms of survival. Its statistical calculations can be based either on the aggregate information published in the trials or on individual patient data. In this second case, the survival information on the patients is retrieved by contacting the authors of the various studies. This paper describes an "intermediate" approach for survival meta-analysis in which the individual survival times are estimated in approximated form without the need to contact the authors of the trials.
Method description: In the analysis of a survival curve, the intermediate method examines the survival graph and reconstructs the individual survival times from the height of the steps of the curve and from other information reported in the study. This procedure is repeated for all curves introduced in the meta-analysis. Finally, the comparison between the two treatments under examination is carried out using non-meta-analytic traditional survival statistics (e.g. log-rank test, Cox analysis).
Examples: Two examples are described in which the individual survival times are reconstructed from the graphs of the survival curves. For comparing the relative performance between the intermediate method and the true individual patient data approach, another two data sets are considered wherein the true survival times of the patients are compared with the approximated survival times estimated by the intermediate method. The concordance between true and approximated values was excellent in both cases (Kendall's W =0.996 in the data set example based on 17 patients: Kendall's W = 0.996 in the second data set based on 87 patients).
Conclusions: The main advantage of the intermediate method is that its application is straightforward and does not require a complex and costly co-operation among researchers from different centres. Hence, this intermediate methodology can contribute to a more widespread use of meta-analysis in comparing treatments that affect survival. While further study is need to better understand to what extent approximated survival times might differ from the corresponding true values, our preliminary data on this point are however very encouraging.