Overview

Example Project

How does non-adherence affect the results of a study investigating the efficacy of a treatment designed to prevent strokes?

A drug regime for hypertension is being investigated and a long-term study planned to demonstrate that the drug combination effectively reduces the number of subjects having a stroke. The study is planned to follow subjects for 4 years and hence it is expected that a fairly large number of subjects will not adhere to the treatment regime and may also drop out over time. This simulation is designed to investigate the effect that different levels of non-adherence and different levels of drop out have on the observed treatment effect on the risk of stroke. The result is shown through the probability of study success to meet criteria as defined below.

Adherence is difficult to measure, as it depends on the subject self-reporting to deviating from the regime. Some studies use pill counting to measure adherence. Adherence can include any deviations from the set regime, some examples include:

1. Subjects taking double doses one day because they forgot to take them the day before, subjects forgetting and so having gaps e.g., taking one dose a day instead of two.

2. Subjects losing motivation, and so after a set amount of time stopping taking treatment altogether but remaining in the study.

For simplicity this simulation sets up different levels of adherence by assuming that a reduction in adherence equates to a reduction in the effectiveness of the treatment.

This tutorial shows how to build up the KerusCloud simulation in stages for a single Variable Scenario but multiple Advanced Option Scenarios. Variable Scenarios are defined to examine the effect that different distribution settings will have on the study outcome and Advanced Option Scenarios are defined to examine “real-world” data effects such as missing or truncated data. The simulation also demonstrates how KerusCloud can explore what type of trial design to use, by examining how results would change if an interim analysis was introduced 80% of the way through recruitment, with an opportunity to stop at the interim analysis for futility if particular criteria are met at the interim.

For more detailed instructions on how to use KerusCloud refer to the KerusCloud User Guide.

Background Information

The variable parameter estimates used in this tutorial were taken from an article published in the Lancet journal: Vol 358 Sept 29 2001 ‘Randomised trial of a perindopril-based blood-pressure-lowering regimen among 6105 individuals with previous stroke or transient ischaemic attack’. The study described was set up as a collaboration between many groups with patients recruited from around the world. The collaboration was named PROGRESS and funded by grants from Servier, the Health Research Council of New Zealand, and the National Health and Medical Research Council of Australia. The results of the study were used to give estimates for the KerusCloud simulation and to assess the effect that non-adherence would have on study outcomes.

Event Rate

The estimates for the event rate were taken from Figure 10: percentage of stroke events for the combination therapy. This showed that over the course of the study 14.4% of subjects had a stroke in the placebo group and 8.5% in the combination therapy group. An assumption was made that subjects not adhering to their treatment regime would have a reduced level of protection from the active treatment. Thus, if a subject was only taking 70% of their pills the risk of stroke would increase from 0.085 (8.5%) to 0.103 (10.3%). If a subject was only adhering half the time and hence only getting half the expected benefit, then the risk of stroke would increase from 0.085 to 0.1145 (11.45%). The risk of stroke in the placebo group remains at 0.144 (14.4%) regardless of whether the subject was adhering or not due to the belief that missing doses of placebo won’t have any true effect on the risk of stroke.

Blood pressure

The estimates for average blood pressure were taken from the text. Claims that under combination treatment the systolic blood pressure reduced by 12.3 from a placebo value of 147, the estimate for SD was taken from the baseline characteristics table as 19. An assumption was made that subjects not adhering to their treatment regime would have a reduced level of protection from the active treatment. Thus, if a subject was only taking 70% of their pills the average blood pressure would increase from 134.7 to 138.39. If a subject was only adhering half the time and hence only getting half the expected benefit, then the average systolic blood pressure would increase from 134.7 to 140.85. The average response of subjects in the placebo group remains at 147 regardless of whether the subject was adhering or not due to the belief that missing doses of placebo won’t have any true effect on blood pressure.