WHEN TO CLOSE SCREENING / RUN-IN
Eddie Channon, Chirostat Statistical Consulting; Damian McEntegart and Graham Nichols, ClinPhone Group Ltd
RSS/PSI conference July 2005
In trials managed by Interactive Voice Response or other electronic data collection, the status of all patients is always known: ‘ongoing through screening’, ‘screen-failed’ or ‘randomised’, with relevant dates. For ongoing patients, the probability of randomisation can be estimated. The distribution of these probabilities is used to determine when to cut off screening so that N is most likely to be reached but not exceeded by too much. This decision aid could be worthwhile in expensive, long term trials.
The simplest way to calculate the probabilities is to ignore ongoing patients in the numerator and denominator. But more precise solutions require some adjustment for the censored values. A method(1) proposed by one of the authors will be described. This works backwards from the last observed day of randomisation; censored values for ongoing patients are then partitioned into randomised or screen-failed patients based on the experience of future randomised and screen-failed patients.
The screening cut-off day is just an estimate and so a variance is needed. This can be done by simulation based on data from the clinical trial under consideration. A better approach may be to specify plausible distributions for screening times, the recruitment rate over time and the probability of screening success vs screening time. Ideally data from an archive of clinical trials would be used to generate these distributions.
Alternative approaches can be compared using this variance.
1 McEntegart D, When to close screening/run-in. Emails to ISCB membership. 7 April 2004, 08 July 2004; also PSI discussion forum (www.psiweb.org)