Example 3.20 Protocol SQGL02: Brimonidine as a neuroprotective agent in acute angle‐closure glaucoma (AACG)
9. STATISTICAL CONSIDERATIONS
Trial size
For sample size determination, it is anticipated that approximately 80% of patients receiving Timolol will experience visual field loss progression at 3 months post laser PI. With the hope that the proportion of Brimonidine treated patients having visual field loss progression will be reduced to 40%, a two‐sided test, with 5% level of significance and power of 90% would require recruitment of 30 patients in each arm (Machin, Campbell, Fayers & Pinol, 19971).
It is anticipated that after randomisation, 10% of the AACG patients may not respond to the initial medical treatment, and will require surgery instead of laser PI, and thus will not continue in the trial. After PI, it is expected that a further 10% of patients may not complete the trial and may withdraw from the study (see Withdrawal from Treatment). Taking into consideration the patients expected to fall out at each stage, we thus require approximately 80 AACG patients (40 patients per treatment group) for the trial.
1Now updated as Machin, Campbell, Tan and Tan (2018).
In this example, the protocol takes note of patient losses due to a relatively large proportion expected to need surgery (rather than laser peripheral iridotomy, PI) and compensates by adjusting upwards the number of patients to recruit. They do not stipulate; however, how these patients will be dealt with in the final analysis and reporting. One option is to regard all these as failures when calculating the proportions in each group with visual field loss at 3 months post PI.
An alternative design possibility would have been to randomise patients after successful PI, so as to avoid recruiting patients who will provide little information on the relative merits of Timolol and Brimonidine. However, the design team may have discussed this (as well as other options) and rejected this for good reasons.
3.10.2 Analysis
As with determining the trial size, the format of the analysis will depend on the type of questions being posed, the trial design, the type of endpoint variables and whether or not characteristics of the patients themselves need to be taken into account. If there is only a single endpoint concerned, then the plan for the subsequent analysis may be described rather succinctly except in circumstances where the analysis may be unusual in format. In some situations, several endpoints will be included so that a careful description of the analytic approach for each needs to be detailed. In these circumstances, multiple statistical tests may be concerned and cognisance of that may need to be taken into account.
Example 3.21 Protocol SQOLP01 (1999): Comparison of steroid with cyclosporine for the topical treatment of oral lichen planus
Analysis plan
Primary endpoints
Analysis of the primary endpoints of response and pain at 4 weeks will be made on an intention‐to‐treat basis.
Clinical response at 4 weeks
Comparison of the observed clinical response rates in the two treatment groups will be made using the χ2‐test for the comparison of two proportions and a 95% confidence interval for the difference reported. In addition, logistic regression analysis will verify if this comparison requires adjustment for imbalance in the baseline clinical assessment values.
Pain score 4 weeks
Comparison of means in the two treatment groups will be made using the t‐test with the appropriate degrees of freedom and a 95% confidence interval for the difference reported. (Should the VAS scores not follow an approximate Normal distribution shape then this analysis may be replaced by the Wilcoxon test). In addition, regression analysis will verify if this comparison requires adjustment for imbalance in the baseline VAS values and clinical assessment values.
Secondary analyses
In those patients for which the marker lesion is measurable by the grid, the mean of the total area remaining affected at 4 weeks for each treatment group will be compared using the t‐test and 95% confidence interval for the difference reported. In addition, regression analysis will verify if this comparison requires adjustment for imbalance in the baseline target lesion area values as well as its location and other clinical assessment values.
A more detailed longitudinal analysis will make use of all the individual measures (maximum 8 per patient) using the area under the curve (AUC) as the summary statistic for each patient. The mean AUC over all patients within the treatment group is calculated and then between treatment comparisons made using the t‐test. A full statistical modelling approach using generalised estimating equations (GEE) is also anticipated. However, details of this latter methodology may have to await examination of the final data available. For example, this methodology would not be appropriate if all patients achieved complete response by week 4 and all subsequent target areas were then zero.
In similar ways, the complete VAS for burning sensation profile will be summarised.
Adverse events
It is not anticipated that a formal statistical analysis comparing the adverse event rates between groups will be conducted. However, full details will be presented and their presence (if any) used to contextualise any observed treatment differences.
Additional analysis
Clinical response at week 8 will also be associated with initial (week 4) response to give an indication of the duration of response and to report any recurrences.
This example details a number of analyses. Some of the repetition could be reduced by replacing ‘adjustment for imbalance in the baseline’ by ‘adjustment for baseline’. Although this is very trivial example, if the number of words can be reduced without loss of clarity, then this reduces the eventual size of the document, facilitates the proof reading, and eases the job of the approving authorities.
3.10.3 Interim analysis
As part of the monitoring of the progress of trials, while recruitment is still ongoing, many protocols include interim looks at the trial data and these may be reviewed by an independent Data and Safety Monitoring Board (DSMB) more details of which are given in Chapter 10. One example of the remit given to a DSMB is in Example 3.22, which is taken from the EXPEL (2016) protocol as described by Kim, Chen, Tay, et al. (2017).
Example 3.22 Protocol EXPEL (2016): Peritoneal lavage after curative gastrectomy
