Limits...
A Bayesian adaptive design for biomarker trials with linked treatments.

Wason JM, Abraham JE, Baird RD, Gournaris I, Vallier AL, Brenton JD, Earl HM, Mander AP - Br. J. Cancer (2015)

Bottom Line: Response to treatments is highly heterogeneous in cancer.At interim analyses, data from treated patients are used to update the allocation probabilities.Our proposed design has high power to detect treatment effects if the pairings of treatment with biomarker are correct, but also performs well when alternative pairings are true.

View Article: PubMed Central - PubMed

Affiliation: MRC Biostatistics Unit Hub for Trials Methodology Research, Cambridge, UK.

ABSTRACT

Background: Response to treatments is highly heterogeneous in cancer. Increased availability of biomarkers and targeted treatments has led to the need for trial designs that efficiently test new treatments in biomarker-stratified patient subgroups.

Methods: We propose a novel Bayesian adaptive randomisation (BAR) design for use in multi-arm phase II trials where biomarkers exist that are potentially predictive of a linked treatment's effect. The design is motivated in part by two phase II trials that are currently in development. The design starts by randomising patients to the control treatment or to experimental treatments that the biomarker profile suggests should be active. At interim analyses, data from treated patients are used to update the allocation probabilities. If the linked treatments are effective, the allocation remains high; if ineffective, the allocation changes over the course of the trial to unlinked treatments that are more effective.

Results: Our proposed design has high power to detect treatment effects if the pairings of treatment with biomarker are correct, but also performs well when alternative pairings are true. The design is consistently more powerful than parallel-groups stratified trials.

Conclusions: This BAR design is a powerful approach to use when there are pairings of biomarkers with treatments available for testing simultaneously.

Show MeSH

Related in: MedlinePlus

Mean allocation probability for: (A) linked-BAR design and (B) non-linked BAR design as trial progresses for a B1-positive patient when T2 provides benefit in B1-positive patients and T1 is detrimental in B1-positive patients. Lines represent the average over 2500 replicates.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4559835&req=5

fig2: Mean allocation probability for: (A) linked-BAR design and (B) non-linked BAR design as trial progresses for a B1-positive patient when T2 provides benefit in B1-positive patients and T1 is detrimental in B1-positive patients. Lines represent the average over 2500 replicates.

Mentions: Figure 2 shows the second case. In this case the linked-BAR approach starts allocating half of B1-positive patients to T1, but this drastically reduces after the first interim analysis and continues to decline as more data is gathered. The average allocation to T2, the superior treatment, increases as time goes on. The non-linked BAR approach performs better in this case because the allocation to T1 starts lower and drops more quickly due to a non-informative prior distribution.


A Bayesian adaptive design for biomarker trials with linked treatments.

Wason JM, Abraham JE, Baird RD, Gournaris I, Vallier AL, Brenton JD, Earl HM, Mander AP - Br. J. Cancer (2015)

Mean allocation probability for: (A) linked-BAR design and (B) non-linked BAR design as trial progresses for a B1-positive patient when T2 provides benefit in B1-positive patients and T1 is detrimental in B1-positive patients. Lines represent the average over 2500 replicates.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4559835&req=5

fig2: Mean allocation probability for: (A) linked-BAR design and (B) non-linked BAR design as trial progresses for a B1-positive patient when T2 provides benefit in B1-positive patients and T1 is detrimental in B1-positive patients. Lines represent the average over 2500 replicates.
Mentions: Figure 2 shows the second case. In this case the linked-BAR approach starts allocating half of B1-positive patients to T1, but this drastically reduces after the first interim analysis and continues to decline as more data is gathered. The average allocation to T2, the superior treatment, increases as time goes on. The non-linked BAR approach performs better in this case because the allocation to T1 starts lower and drops more quickly due to a non-informative prior distribution.

Bottom Line: Response to treatments is highly heterogeneous in cancer.At interim analyses, data from treated patients are used to update the allocation probabilities.Our proposed design has high power to detect treatment effects if the pairings of treatment with biomarker are correct, but also performs well when alternative pairings are true.

View Article: PubMed Central - PubMed

Affiliation: MRC Biostatistics Unit Hub for Trials Methodology Research, Cambridge, UK.

ABSTRACT

Background: Response to treatments is highly heterogeneous in cancer. Increased availability of biomarkers and targeted treatments has led to the need for trial designs that efficiently test new treatments in biomarker-stratified patient subgroups.

Methods: We propose a novel Bayesian adaptive randomisation (BAR) design for use in multi-arm phase II trials where biomarkers exist that are potentially predictive of a linked treatment's effect. The design is motivated in part by two phase II trials that are currently in development. The design starts by randomising patients to the control treatment or to experimental treatments that the biomarker profile suggests should be active. At interim analyses, data from treated patients are used to update the allocation probabilities. If the linked treatments are effective, the allocation remains high; if ineffective, the allocation changes over the course of the trial to unlinked treatments that are more effective.

Results: Our proposed design has high power to detect treatment effects if the pairings of treatment with biomarker are correct, but also performs well when alternative pairings are true. The design is consistently more powerful than parallel-groups stratified trials.

Conclusions: This BAR design is a powerful approach to use when there are pairings of biomarkers with treatments available for testing simultaneously.

Show MeSH
Related in: MedlinePlus