- Posted By: Peter Dutton
I had the opportunity to attend the International Society for Clinical Biostatistics (ISCB) 38th annual conference held in Vigo, Spain this year. This conference attracted attendees from a large number of international countries and covered statistical methodology in many aspects of medicine, not just clinical trials and observation studies.
There was a considerable focus on Joint models, survival analysis, and adaptive trials. I will highlight a few interesting talks.
Correct primary analysis: Steven Senn, as usual, gave a very entertaining opening talk. He stressed that if you stratify your randomisation by a number of covariates then your primary analysis should also include these covariates. He also highlighted issues about publication bias and has come to the conclusion that self-publication is the solution.
Survival analysis: Theodor Adrian Balan gave an interesting talk on non-proportional hazards and unobserved heterogeneity in clustered survival data. He showed that the current models cannot tell the difference between proportional hazard violations and unobserved frailty terms.
Bayesian medians: Peter Fletcher gave a talk comparing the Kaplan-Meier estimates for the median survival time to Bayesian estimates. This problem arose from a single arm study utilising the Bayesian median survival time model as the primary endpoint. In this study, the estimates for the Bayesian median survival time were considerably larger than the Kaplan-Meier estimate. He showed that there is correlation between these estimates but it is not perfect. He also showed that, provided the model assumptions hold for the Bayesian estimates, these have smaller standard errors than the Kaplan-Meier estimates. One interesting point was that the median survival time does not act like a median because the model draws much information from the extreme values.
Adaptive Randomisation: Jordan Elm presented an alternative to block randomisation and minimisation which provided full flexibility in using unequal allocation between treatment arms using a mass urn weighted randomisation scheme. She showed its benefits over simple and block randomisation and indicated it could be used on the back end of any adaptive randomisation method. However, this has not yet been extended to include stratification factors.
The next ISCB will be held in Melbourne, Australia at the end of August next year. This will be a join conference between ISCB and the Statistical Society of Australia. This is likely to be another high calibre conference which will be well worth attending.