Article type
Year
Abstract
Introduction: Quantitative statistical combination of survival data from different studies is an increasingly accepted approach to synthesise research findings in oncology. The hazard ratio (HR) is a generally suggested statistic to summarise time to event outcomes and its use should be recommended in meta-analyses using literature data (MAL) as well.
Objective: To provide a description of statistical methods used in MALs based on comparative studies in oncology.
Methods: MALs, published from 1985 were identified by Medline. MALs were considered eligible for inclusion if they were based on prospective, comparative studies and provided complete quantitative results. The following variables were examined: type of disease, outcome, summary statistics, variance computation, approach for allowance of censoring.
Results: Fifteen MALs combining results from trials on Head and Neck, Glioma, Sarcoma, Lymphoma, Myeloma, NSCLC, SCLC, Gastric and Colorectal cancer were included, with a total of seventeen analyses reported. Regarding the statistic used for outcome evaluation 4 of the 17 considered the total number of deaths, 5 survivors at fixed time-points, 3 calculated the survival at different time-points, 2 the median survival time and only 3 used the hazard rate. The preferred summary statistics were odds ratio (9), hazard ratio (3), relative risk (2), risk difference (1) and difference in log median survival (2). In 4 cases only allowance for censoring was made. An exploratory analysis of gastric cancer RCTs showed that different approaches yielded controversial results.
Discussion: This study underlines the great heterogeneity of approaches for summarising results of studies using survival data. Summary statistics using information easily available from published reports, not specifically developed for survival data, are being used most of the time. This, in our opinion, may compromise the reliability of results from MALs. Research on the best methods for data extraction and pooling of survival data for MALs is warranted.
Objective: To provide a description of statistical methods used in MALs based on comparative studies in oncology.
Methods: MALs, published from 1985 were identified by Medline. MALs were considered eligible for inclusion if they were based on prospective, comparative studies and provided complete quantitative results. The following variables were examined: type of disease, outcome, summary statistics, variance computation, approach for allowance of censoring.
Results: Fifteen MALs combining results from trials on Head and Neck, Glioma, Sarcoma, Lymphoma, Myeloma, NSCLC, SCLC, Gastric and Colorectal cancer were included, with a total of seventeen analyses reported. Regarding the statistic used for outcome evaluation 4 of the 17 considered the total number of deaths, 5 survivors at fixed time-points, 3 calculated the survival at different time-points, 2 the median survival time and only 3 used the hazard rate. The preferred summary statistics were odds ratio (9), hazard ratio (3), relative risk (2), risk difference (1) and difference in log median survival (2). In 4 cases only allowance for censoring was made. An exploratory analysis of gastric cancer RCTs showed that different approaches yielded controversial results.
Discussion: This study underlines the great heterogeneity of approaches for summarising results of studies using survival data. Summary statistics using information easily available from published reports, not specifically developed for survival data, are being used most of the time. This, in our opinion, may compromise the reliability of results from MALs. Research on the best methods for data extraction and pooling of survival data for MALs is warranted.