Training for Statisticians
RK Statistics has run several courses for PSI (Statisticians in the Pharmaceutical Industry). These include:
• The Analysis of Time-to-Event Data
• Equivalence and Non-Inferiority
• The Analysis of Binary and Categorical Data
• Statistics for Statistical Programmers
This is a review of the PSI course Analysis of Time-to-Event Data, which appeared in SPIN, Winter 2011. The course took place on 28th-29th September 2011 at Hunton Park.
'Every one of us in the full audience at this 2-day intensive set of lectures by Richard Kay on time-to-event data became increasingly aware of how privileged he/she was to be on this course. Richard Kay has that rare combination of an extremely approachable, non-didactic style together with an impressive mastery of the subject matter, foremost with an eye to practical use, but always grounded in theory. I am not afraid to show my ignorance or failure to understand and so put up my hand and ask questions, and several others in the audience were like-minded: these interruptions were unfailingly carefully and courteously handled by Richard. At the end of the course we were left amazed at how, in the course of two days, pretty well every single facet of the subject of practical relevance had been covered to quite some depth. Darren Jolliffe provided valuable worked SAS examples to illustrate issues raised in Richard's lectures, with helpful print-outs supplied at the end of the course material folder.
The course began from scratch with an introduction to the subject, including hazard functions. survivor functions, parametric and semi-parametric modelling, non-informative and informative censoring. Next Kaplan-Meier plots, survivor functions plotted versus time, were dealt with and alternative approaches to deriving plotted values considered. The K-M curves and their confidence limits were used to compare treatments in a 2-arm clinical trial. This led naturally to the logrank test and the Gehan-Wilcoxon test. This material generated quite a bit of lively lecturer/audience interaction.
Cox's proportional hazrads model was considered in some depth, and its close affinity with the logrank test was mentioned. It was discussed how non-proportional hazards data can virtually always be handled using proportional hazards software but with a 'time' by treament interaction. The advantages of the Accelerated Failure Time model, with alterative model error forms were considered.
Methods for dealing with interval censoring were considered, that is: when the precise time of an event is not recorded, only an interval. These methods are unavoidably cumbersome and the situation happens frequently.
Richard turned his attention to assessing the goodness of fit of a model. Log-log plots, time-axis splits, nested models, quantile-quantile plots were mentioned, together with a practical caution from Richard that proportional hazards model residuals can be just about impossible to interpret.
Modelling competing risks was expertly dealt with, including obtaining meaningful outputs in the presense of strong correlation in immediate model estimates from the data.
Richard gave valuable common-sense advice, based on his extensive industry consultancy experience, on pitfalls in applying multistate models that the unwary can fall into.
Sample size calculation was considered and power was shown dependent of the number of occurred events, so that a trial can be halted as soon as an adequate number of events have taken place.
The lecture finished with a talk on time-to-event cross-over trials.
Sig Johnson (Surrey Clinical Research Centre)
These courses are available to be run on an in-company basis. It is also possible to structure bespoke courses for statisticians covering a range of topics. In addition to the areas listed above, presentations can also be built around specific additional topics, for example; Adaptive Designs, Interim Analyses, Data Monitoring Committees/Data and Safety Monitoring Boards, Observational Studies.
Contact Richard Kay by ‘phone or email if you are interested in these or other statistics courses for your company – contact details at the foot of this page.