Limits...
A comprehensive comparison of the continual reassessment method to the standard 3 + 3 dose escalation scheme in Phase I dose-finding studies.

Iasonos A, Wilton AS, Riedel ER, Seshan VE, Spriggs DR - Clin Trials (2008)

Bottom Line: We focused on methods with practical design features that are of interest to clinicians.However, there are several alternative CRM-based designs that are not investigated in this manuscript, and hence our results are not generalizable to other designs.We show that CRM-based methods are an improvement over the SM in terms of accuracy and optimal dose allocation in almost all cases, except when the true dose is among the lower levels.

View Article: PubMed Central - PubMed

Affiliation: Department of Epidemiology and Biostatistics, Memorial Sloan Kettering Cancer Center, New York, NY 10021, USA. iasonosa@mskcc.org

ABSTRACT

Background: An extensive literature has covered the statistical properties of the Continual Reassessment Method (CRM) and the modifications of this method. While there are some applications of CRM designs in recent Phase I trials, the standard method (SM) of escalating doses after three patients with an option for an additional three patients SM remains very popular, mainly due to its simplicity. From a practical perspective, clinicians are interested in designs that can estimate the MTD using fewer patients for a fixed number of doses, or can test more dose levels for a given sample size.

Purpose: This article compares CRM-based methods with the SM in terms of the number of patients needed to reach the MTD, total sample size required, and trial duration.

Methods: The comparisons are performed under two alternative schemes: a fixed or a varying sample approach with the implementation of a stopping rule. The stopping rule halts the trial if the confidence interval around the MTD is within a pre-specified bound. Our simulations evaluated several CRM-based methods under different scenarios by varying the number of dose levels from five to eight and the location of the true MTD.

Results: CRM and SM are comparable in terms of how fast they reach the MTD and the total sample size required when testing a limited number of dose levels (

Limitations: We focused on methods with practical design features that are of interest to clinicians. However, there are several alternative CRM-based designs that are not investigated in this manuscript, and hence our results are not generalizable to other designs.

Conclusions: We show that CRM-based methods are an improvement over the SM in terms of accuracy and optimal dose allocation in almost all cases, except when the true dose is among the lower levels.

Show MeSH

Related in: MedlinePlus

Median trial duration across simulations. Each panel represents a scenario under four different accrual rates (3, 2, 1.5, or 1 patient/month). The sample sizes are shown at the bottom: CRM, E-CRM and CRM-I are using a fixed sample of 20 patients. G-CRM and SM use the stopping rules as listed in section ‘Methods’ under subsection ‘Design and endpoints’
© Copyright Policy - open-access
Related In: Results  -  Collection

License 1 - License 2 - License 3
getmorefigures.php?uid=PMC2637378&req=5

fig2-1740774508096474: Median trial duration across simulations. Each panel represents a scenario under four different accrual rates (3, 2, 1.5, or 1 patient/month). The sample sizes are shown at the bottom: CRM, E-CRM and CRM-I are using a fixed sample of 20 patients. G-CRM and SM use the stopping rules as listed in section ‘Methods’ under subsection ‘Design and endpoints’

Mentions: We report the median trial duration over 1000 simulations using the fixed sample approach, except for G-CRM and SM where the sample size varies. Since R-CRM and CRM assign one patient at a time using a fixed sample size, the trial duration of R-CRM as calculated in simulations was very close to CRM. Thus we report CRM only. We added a modified design, denoted as CRM-I, that allows for one patient with incomplete DLT information (delayed response) - a modification that allows CRM to proceed to the next patient's assignment without having observed the last patient's toxicity data. At each time point the assignment for patient j depends on data from j - 2 patients, where j ≥ 3. If patient j arrives after patient j − 1 was fully observed, the response from patient j - 1 is used. CRM-I starts at the lowest dose level and prevents escalation of more than one dose level. The results for the five designs under each of the eight scenarios and using varying accrual rates are shown in Figure 2. It is evident that when accrual rate is two or three patients/month CRM and E-CRM take on average five months longer to complete than the other three methods. As accruals become less frequent, then CRM-based methods and SM become closer in trial duration. In S4–6, as a result of a larger sample size needed to evaluate eight doses, SM and G-CRM take longer than the other CRM methods when accrual is less than two patients/month. The concept that CRM always takes longer to complete than SM because it accrues one patient at time is not correct. The longer the wait between patient accruals, the closer the two methods are in trial duration. Depending on the number of dose levels tested, CRM may have shorter trial duration than SM. CRM-I greatly decreases trial duration over CRM especially for trials when a higher accrual rate is anticipated. CRM-I performs similarly to the original CRM in accuracy and to E-CRM in terms of the other endpoints presented in this article (data shown in supplemental table). These results show that we can use CRM-I in practice with reduced trial duration without sacrificing other clinical endpoints such as accuracy and safety.


A comprehensive comparison of the continual reassessment method to the standard 3 + 3 dose escalation scheme in Phase I dose-finding studies.

Iasonos A, Wilton AS, Riedel ER, Seshan VE, Spriggs DR - Clin Trials (2008)

Median trial duration across simulations. Each panel represents a scenario under four different accrual rates (3, 2, 1.5, or 1 patient/month). The sample sizes are shown at the bottom: CRM, E-CRM and CRM-I are using a fixed sample of 20 patients. G-CRM and SM use the stopping rules as listed in section ‘Methods’ under subsection ‘Design and endpoints’
© Copyright Policy - open-access
Related In: Results  -  Collection

License 1 - License 2 - License 3
Show All Figures
getmorefigures.php?uid=PMC2637378&req=5

fig2-1740774508096474: Median trial duration across simulations. Each panel represents a scenario under four different accrual rates (3, 2, 1.5, or 1 patient/month). The sample sizes are shown at the bottom: CRM, E-CRM and CRM-I are using a fixed sample of 20 patients. G-CRM and SM use the stopping rules as listed in section ‘Methods’ under subsection ‘Design and endpoints’
Mentions: We report the median trial duration over 1000 simulations using the fixed sample approach, except for G-CRM and SM where the sample size varies. Since R-CRM and CRM assign one patient at a time using a fixed sample size, the trial duration of R-CRM as calculated in simulations was very close to CRM. Thus we report CRM only. We added a modified design, denoted as CRM-I, that allows for one patient with incomplete DLT information (delayed response) - a modification that allows CRM to proceed to the next patient's assignment without having observed the last patient's toxicity data. At each time point the assignment for patient j depends on data from j - 2 patients, where j ≥ 3. If patient j arrives after patient j − 1 was fully observed, the response from patient j - 1 is used. CRM-I starts at the lowest dose level and prevents escalation of more than one dose level. The results for the five designs under each of the eight scenarios and using varying accrual rates are shown in Figure 2. It is evident that when accrual rate is two or three patients/month CRM and E-CRM take on average five months longer to complete than the other three methods. As accruals become less frequent, then CRM-based methods and SM become closer in trial duration. In S4–6, as a result of a larger sample size needed to evaluate eight doses, SM and G-CRM take longer than the other CRM methods when accrual is less than two patients/month. The concept that CRM always takes longer to complete than SM because it accrues one patient at time is not correct. The longer the wait between patient accruals, the closer the two methods are in trial duration. Depending on the number of dose levels tested, CRM may have shorter trial duration than SM. CRM-I greatly decreases trial duration over CRM especially for trials when a higher accrual rate is anticipated. CRM-I performs similarly to the original CRM in accuracy and to E-CRM in terms of the other endpoints presented in this article (data shown in supplemental table). These results show that we can use CRM-I in practice with reduced trial duration without sacrificing other clinical endpoints such as accuracy and safety.

Bottom Line: We focused on methods with practical design features that are of interest to clinicians.However, there are several alternative CRM-based designs that are not investigated in this manuscript, and hence our results are not generalizable to other designs.We show that CRM-based methods are an improvement over the SM in terms of accuracy and optimal dose allocation in almost all cases, except when the true dose is among the lower levels.

View Article: PubMed Central - PubMed

Affiliation: Department of Epidemiology and Biostatistics, Memorial Sloan Kettering Cancer Center, New York, NY 10021, USA. iasonosa@mskcc.org

ABSTRACT

Background: An extensive literature has covered the statistical properties of the Continual Reassessment Method (CRM) and the modifications of this method. While there are some applications of CRM designs in recent Phase I trials, the standard method (SM) of escalating doses after three patients with an option for an additional three patients SM remains very popular, mainly due to its simplicity. From a practical perspective, clinicians are interested in designs that can estimate the MTD using fewer patients for a fixed number of doses, or can test more dose levels for a given sample size.

Purpose: This article compares CRM-based methods with the SM in terms of the number of patients needed to reach the MTD, total sample size required, and trial duration.

Methods: The comparisons are performed under two alternative schemes: a fixed or a varying sample approach with the implementation of a stopping rule. The stopping rule halts the trial if the confidence interval around the MTD is within a pre-specified bound. Our simulations evaluated several CRM-based methods under different scenarios by varying the number of dose levels from five to eight and the location of the true MTD.

Results: CRM and SM are comparable in terms of how fast they reach the MTD and the total sample size required when testing a limited number of dose levels (

Limitations: We focused on methods with practical design features that are of interest to clinicians. However, there are several alternative CRM-based designs that are not investigated in this manuscript, and hence our results are not generalizable to other designs.

Conclusions: We show that CRM-based methods are an improvement over the SM in terms of accuracy and optimal dose allocation in almost all cases, except when the true dose is among the lower levels.

Show MeSH
Related in: MedlinePlus