Dissertation
Pacing modification by incorporation of lag in early phase designs
University of Iowa
Doctor of Philosophy (PhD), University of Iowa
Autumn 2019
DOI: 10.17077/etd.005225
Abstract
Efficient and accurate first-in-human clinical studies are essential for establishing appropriate doses for novel drugs. By selecting doses that are tolerable and have good potential to accomplish treatment goals in preliminary smaller scale studies, larger studies have a better chance at confirming whether the drug should be available to the general population. To be successful, the dose of the investigational drug must be safe, that is, it must be nontoxic. Establishing safety of the compound in humans is the goal of Phase I trials. Since most compounds are assumed to have increasing efficacy with higher doses, the goal of finding the maximum tolerable dose (MTD) is often the goal of this phase. Furthermore, a single trial can be done to establish both safety and efficacy, called a Phase I/II trial. In these trials, the dose-finding designs consider both safety and efficacy outcomes to recommend a dose. This dissertation is motivated by an early phase clinical trial in acute ischemic stroke, which experienced unexpected rapid recruitment and an underutilization of available study participants. The current methods for early phase clinical trials are reviewed and their limitations regarding accommodation of recruitment rates is explored. Then several designs are developed and evaluated to address this concern for phase I and phase I/II.
In these early studies, groups of participants are given doses in a structured manner and observed for any adverse reactions. Based on the data collected, the dose-response relationship can be characterized and used to decide which dose to give to the next group of participants in a structured manner. One method used in phase I is the continual reassessment method (CRM: O'Quigley, Pepe & Fisher, 1990), which was originally used for the motivating clinical trial used for this dissertation. This design bases dose escalation on safety outcomes on information gathered from participants enrolled in the study who have completed follow up by using a Bayesian statistical model to update the probability of toxicity at each dose. To accomplish this, the CRM requires that the participants in a dose cohort complete full follow up before the next cohort can start enrollment. For the motivating NN104 (RHAPSODY) study, this necessity of delaying enrollment of participants originally led to a higher than expected number of participants being assigned by default to the placebo group (Lyden et al., 2016; Lyden et al., 2018). To accommodate rapid recruitment rates, several modified designs have been proposed.
One alternative is to use the time-to-event CRM (TITE-CRM: Cheung & Chappell, 2000). In this design, patients can enter the study at any time and the dose that they are assigned is determined using information from all patients currently in the study, weighted by their percentage of completed follow-up. However, this design can possibly have an increased risk of treating patients at unsafe doses in the situation of rapid recruitment and late onset dose limiting toxicities (Bekele et al., 2007).The Lag-CRM design follows the CRM design but also allows for an additional cohort which does not require the immediate previous cohort to complete follow up before beginning enrollment. A simulation study comparing the CRM, Lag-CRM, and the TITE-CRM showed that the Lag-CRM was able to safely and correctly select the MTD at comparable rate to established designs. The study also showed that the Lag-CRM is not as influenced by rapid recruitment rates and late onset DLT as the TITE-CRM. This supports that modifying when participants can be enrolled and when collected information can be used to guide dose escalation for subsequent cohorts may be useful in better utilizing available study participants.
Several modifications to the established designs were developed to modify the pace of participant inclusion to the study and the statistical modeling, with the goal of better accommodating rapid recruitment and late DLT onset. The first of these designs allows for continuous enrollment of study participants, but only uses data in the statistical model from participants who have completed a certain amount of follow up, which is specified as part of the study design. The next design allows for enrollment of a new dose cohort only after all currently enrolled participants have completed a certain amount of follow up and bases dose escalation decisions on all available data. The final design is a modification of Lag-CRM design, where instead of completely excluding the previous cohort this information is instead included if the participants in the previous cohort have completed a specified amount of follow up. These designs potentially allow for the improved utilization of available participants but are based on using statistical models that allow for partial information. A simulation study was conducted to compare the performance of these designs. The performance varied based on the setting for the specified amount of follow up, with settings requiring lower amounts of follow-up being more negatively influenced by rapid recruitment and late onset DLTs. However, the lower setting allowed for a shorter study with fewer fall through participants. The results suggest that compared to the Lag-CRM, these designs had a much higher potential to be adversely influenced by recruitment rates and DLT onset time and do not necessarily present a definitively better performance.
Finally, the CRM and Lag-CRM were compared in the phase I/II setting, using both toxicity and efficacy information to make dose escalation decisions. The bivariate probit model was used for this comparison. A simulation study was conducted to facilitate the comparison of designs. The results suggest that the bivariate Lag-CRM has a slightly lower ability to select the best nontoxic dose or even select a final dose that is nontoxic, but it did reduce the number of fall through participants and the study duration. The results also suggest that the Lag-CRM may be sensitive to lower sample size.
Details
- Title: Subtitle
- Pacing modification by incorporation of lag in early phase designs
- Creators
- Anne E Welhaven
- Contributors
- Christopher Coffey (Advisor)Eric Foster (Advisor)Emine Bayman (Committee Member)Grant Brown (Committee Member)Jacob Oleson (Committee Member)Mitchell Thomann (Committee Member)
- Resource Type
- Dissertation
- Degree Awarded
- Doctor of Philosophy (PhD), University of Iowa
- Degree in
- Biostatistics
- Date degree season
- Autumn 2019
- DOI
- 10.17077/etd.005225
- Publisher
- University of Iowa
- Number of pages
- xiv, 182 pages
- Copyright
- Copyright 2019 Anne E Welhaven
- Language
- English
- Description illustrations
- color illustrations
- Description bibliographic
- Includes bibliographical references (pages 106-113)
- Public Abstract (ETD)
Efficient and accurate first-in-human clinical studies are essential for establishing appropriate doses for novel drugs. By selecting doses that are tolerable and have good potential to accomplish treatment goals in preliminary smaller scale studies, larger studies have a better chance at confirming whether the drug should be available to the general population.
In these early studies, groups of participants are given doses in a structured manner and observed for any adverse reactions. Based on the data collected, the dose-response relationship can be characterized and used to decide which dose to give to the next group of participants in a structured manner. The continual reassessment method (CRM) is an example of this kind of study and was used for the motivating clinical study in this dissertation. One drawback to this method is that all participants need to complete full follow up before the next group can enter the study, which prevented new arrivals from being enrolled efficiently. One additional design, the time-to-event CRM (TITE), allows for the inclusion of information for participants who have completed part of the follow-up. However, this design has poor performance when participants arrive rapidly, and adverse reactions occur later in the follow-up time. To try to accommodate the rapid recruitment rates, several modifications to the method were proposed and evaluated. First, the Lag-CRM allowed for a group of participants to be enrolled and based dosing decisions on all patients except the previous group, so that previous group did not need to complete follow up. This modification is shown to have comparable ability to select the best dose, while reducing the study duration.
Other designs modified when participants could be enrolled and when the collected participant data could be used in the dose-response characterization. These designs relied on the inclusion of partial information and showed poorer performance when less information was available, although they were effective at reducing study duration and number of excluded participants. For each of these designs, the balance is between speed and accuracy. Modifications that are effective at creating a shorter study have higher potential to miss the best dose. Lastly, Lag-CRM is evaluated in the setting where both safety and efficacy information is collected from participants. The results of this evaluation show that in this setting, the Lag-CRM shows a reduced ability to select the best dose compared to established designs and may be sensitive to how many participants are included in the study.
- Academic Unit
- Biostatistics
- Record Identifier
- 9983779397802771
Metrics
33 File views/ downloads
65 Record Views