HRMeta分析-Cancer Care Ontario .pptVIP

Meta-analysis of Hazard Ratios Key References Parmar M.K.B., et al. Extracting summary statistics to perform meta-analysis of the published literature for survival endpoints. Statist Med. 1998. 7; 2815-34. Michiels S., et al. Meta-analysis when only the median survival times are known: a comparison with individual patient data results. Three ways to compare time-to-event outcomes Point in time – Odds Ratio Median time – Ratio of medians Hazard Ratio Point in time Odds ratio or other measure at a single point in time Does not take censoring into account Discards data Choice of time point can change results May be misleading if survival curves cross or are erratic Median Time Ratio of median times-to-event No concern about time point selection, but still somewhat arbitrary Ignores censoring Discards data May be misleading Methods to calculate variance unclear Hazard Ratio The ratio of the survival functions of both treatment arms Accounts for censoring Includes all data More tolerant of strange curve behaviour Describes all of patient experience Michiels et al study Used individual patient data from 13 meta-analyses to directly compare the three methods Found 4 of 13 results discordant using OR vs. HR Found 5 of 13 results discordant using MR vs. HR “…neither the median ratio nor the OR can be recommended as a surrogate method for analyzing time to event outcomes.” In other words… Conducting a meta-analysis using point-in-time or median times may be worse than doing nothing at all If in error, likely to result in finding no significant effect when one exists So why not HR’s? Because the *$^%*$ researchers don’t report it! Give p-values but no HR’s Give HR’s but no confidence intervals Give none of the above, only survival curves More of a problem the older the study Point-in-time measures are accessible in almost every study, so this method has been commonly used due to convenience Parmar Toolbox Parmar et al provides a tool box of methods to get the info you n