OBJECTIVES

Adherence to treatment and persistence of patients participating in clinical studies are known to have a significant impact on clinical studies. Both of these factors affect observed treatment response, likelihood of success, study cost and duration. We used simulation to model the complex interactions between these factors and quantify their impact on study outcomes.

METHODS

A series of simulations were performed based on results taken from a randomized, placebo-controlled blood-pressure trial (The Lancet Vol 358 2001). Using these data, virtual patient populations were simulated for a range of scenarios with differing levels of adherence and persistence. We generated further scenarios to measure any bias introduced due to correlation between non-persistence and lack of efficacy on the likelihood of success and on the observed treatment response.

RESULTS

Non-persistence had the greatest impact on study power which reduced by up to 35% due to patients dropping out. Likewise, low levels of adherence also resulted in reduced statistical power resulting in sample sizes increasing by 33%. When a correlation exists between non-persistence and lack of efficacy the statistical power reduces even further, the estimate of the treatment effect is reduced by 31%.

CONCLUSIONS

We used simulation to evaluate the impact of adherence and persistence. While the inter-relationships between these factors are complex, the simulations showed these factors have a considerable impact on study power and treatment effect. High levels of non-persistence and non-adherence result in studies needing to be larger in order to maintain a high probability of success. In addition, when non-persistence is linked to lack of efficacy, the observed treatment response is biased as non-responders are likely to drop out. This leads to an over-estimation of treatment response as well as an inflated placebo response.