Abstract:We consider event-driven clinical trials, where the analysis is performed once a pre-determined number of clinical events has been reached. For example, these events could be progression in oncology or a stroke in cardiovascular trials. At the interim stage, one of the main tasks is predicting the number of events over time and the time to reach specific milestones, where we need to account for events that may occur not only in patients already recruited and are followed-up but also in patients yet to be recruited. Therefore, in such trials we need to model patient recruitment and event counts together. In the paper we develop a new analytic approach which accounts for the opportunity of patients to be cured, as well as for them to dropout and be lost to follow-up. Recruitment is modelled using a Poisson-gamma model developed in previous publications. When considering the occurrence of events, we assume that the time to the main event and the time to dropout are independent random variables, and we have developed a few advanced models with cure using exponential, Weibull and log-normal distributions. This technique is supported by well developed, tested and documented software. The results are illustrated using simulation and a real dataset with reference to the developed software.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is how to predict the number of clinical events that occur at a specific time point or during a specific period in event - driven clinical trials, and estimate the time required to reach a specific milestone. Such trials are usually analyzed after a predetermined number of clinical events are reached, such as disease progression in oncology or stroke events in cardiovascular trials.
At the interim stage, one of the main tasks is to predict the number of events that change over time and the time required to reach a specific milestone. This requires considering not only the events that may occur in patients who have been recruited and are being followed up, but also the events that may occur in patients who have not yet been recruited. Therefore, in such trials, it is necessary to model the patient recruitment process and event counting simultaneously.
The paper proposes a new analytical method that takes into account the possibility of patients being cured, as well as the situation of patients dropping out or being lost to follow - up. Specifically:
- **Recruitment model**: Use the Poisson - Gamma model to model the patient recruitment process, which can describe the variation in recruitment rates in different centers.
- **Event - occurrence model**: Assume that the time of the main event occurrence and the time of dropping out are independent random variables, and develop several advanced models, using the exponential distribution, Weibull distribution, and log - normal distribution to describe these times.
- **Interim prediction**: At the interim stage, estimate the parameters of different models by the maximum - likelihood estimation method, and derive the predictive distribution of the number of future events. These distributions are given in closed form, so Monte Carlo simulation is not required.
The paper also introduces the developed techniques and R tools, which can be used to predict the number of events that change over time and the time to stop the trial, and provide the mean and prediction intervals. The research results are illustrated by Monte Carlo simulation and real - data sets.
In summary, this paper aims to provide a comprehensive method for more accurately predicting event occurrences in event - driven clinical trials, while taking into account patient recruitment, dropping out, and the possibility of being cured.