Shemra Rizzo, of the Statistics Department at the University of California, Riverside, will present:
“Meta-Analysis of Odds Ratios with Missing Counts Estimated using Kaplan Meier Curves”
A social tea will be held at 3:00 p.m. in A434 Mayo. All are welcome.
A meta-analysis of aggregate odds-ratios assumes binomially distributed numbers of events in a treatment and control group and requires the number of events (i.e. deaths) and non-events (i.e. survivors) to be extracted from published papers. These data are often not available in the publications due to loss to follow-up. When the Kaplan-Meier (KM) survival plot is available, it is common practice to extract the survival probability from the plot and multiply it by the baseline sample size to infer the number of deaths and survivors. The naive approach introduces these estimates as real extracted data in the meta-analysis leading to over-certain and potentially inaccurate results. Accounting for the uncertainty introduced from these estimations is difficult as KM curves are typically published without variance information. We propose a method to estimate the missing KM variance using follow-up summary statistics and to introduce it into the meta-analysis using effective sample sizes of the studies. A simulation study shows that our model outperforms the naive approach in terms of the coverage of the 95% confidence interval. We use real and simulated data to illustrate our method.