Treatment simulation could help personalize myeloma therapy

With the help of gene expression signatures, a simulated treatment learning model identified which patients with multiple myeloma would benefit most from treatment with bortezomib or lenalidomide, researchers reported in Nature Communications.

The study included 910 participants across three phase 3 trials. In all, 20% would have a 100% greater-than-average progression-free survival (PFS) benefit from bortezomib, while 31% would have a 200% greater-than-average PFS benefit from lenalidomide, wrote Joske Ubels of University Center Utrecht, the Netherlands, and her colleagues.

The genetic heterogeneity of cancer and risk of treatment necessitate tools that “predict – at the moment of diagnosis – which patients will benefit most from a certain treatment,” the researchers wrote. While gene expression signatures can predict a favorable or adverse prognosis, they do not account for the effect of treatment on survival.

“The key idea of simulated treatment learning is that a patient’s treatment benefit can be estimated by comparing [his or her] survival to a set of genetically similar patients [who] received the comparator treatment,” they noted.

To do so, the researchers applied an algorithm called GESTURE to combined data from the TT2 (Total Therapy 2 for Multiple Myeloma), TT3, and HOVON-65/GMMG-HD4 trials. These trials compared bortezomib or lenalidomide with conventional therapies for multiple myeloma. The model identified 180 patients (20%) for whom bortezomib would produce a 100% greater PFS benefit than in the study population as a whole. Conversely, lenalidomide would produce a 200% greater PFS benefit in 31% of patients.

The simulated treatment learning model “can derive clinically actionable gene expression signatures that enable a more personalized approach to treatment,” the researchers concluded. The method requires a large dataset but could be useful for trials that have missed their primary endpoint, such as the CheckMate-026 trial of nivolumab. The next step is to see if the model makes useful treatment predictions for other cancers. The code needed to train and validate the model is available at github.com/jubels/GESTURE.

The Van Herk Fellowship provided support. The lenalidomide dataset was created as part of the Multiple Myeloma Research Foundation Personalized Medicine Initiative. Dr. Ubels and one coinvestigator are employees of SkylineDx; another coinvestigator served on its advisory board. The others reported having no relevant conflicts of interest.

SOURCE: Ubels J et al. Nat Commun. 2018 Jul 27. doi: 10.1038/s41467-018-05348-5.

With the help of gene expression signatures, a simulated treatment learning model identified which patients with multiple myeloma would benefit most from treatment with bortezomib or lenalidomide, researchers reported in Nature Communications.

The study included 910 participants across three phase 3 trials. In all, 20% would have a 100% greater-than-average progression-free survival (PFS) benefit from bortezomib, while 31% would have a 200% greater-than-average PFS benefit from lenalidomide, wrote Joske Ubels of University Center Utrecht, the Netherlands, and her colleagues.

The genetic heterogeneity of cancer and risk of treatment necessitate tools that “predict – at the moment of diagnosis – which patients will benefit most from a certain treatment,” the researchers wrote. While gene expression signatures can predict a favorable or adverse prognosis, they do not account for the effect of treatment on survival.

“The key idea of simulated treatment learning is that a patient’s treatment benefit can be estimated by comparing [his or her] survival to a set of genetically similar patients [who] received the comparator treatment,” they noted.

To do so, the researchers applied an algorithm called GESTURE to combined data from the TT2 (Total Therapy 2 for Multiple Myeloma), TT3, and HOVON-65/GMMG-HD4 trials. These trials compared bortezomib or lenalidomide with conventional therapies for multiple myeloma. The model identified 180 patients (20%) for whom bortezomib would produce a 100% greater PFS benefit than in the study population as a whole. Conversely, lenalidomide would produce a 200% greater PFS benefit in 31% of patients.

The simulated treatment learning model “can derive clinically actionable gene expression signatures that enable a more personalized approach to treatment,” the researchers concluded. The method requires a large dataset but could be useful for trials that have missed their primary endpoint, such as the CheckMate-026 trial of nivolumab. The next step is to see if the model makes useful treatment predictions for other cancers. The code needed to train and validate the model is available at github.com/jubels/GESTURE.

The Van Herk Fellowship provided support. The lenalidomide dataset was created as part of the Multiple Myeloma Research Foundation Personalized Medicine Initiative. Dr. Ubels and one coinvestigator are employees of SkylineDx; another coinvestigator served on its advisory board. The others reported having no relevant conflicts of interest.

SOURCE: Ubels J et al. Nat Commun. 2018 Jul 27. doi: 10.1038/s41467-018-05348-5.

With the help of gene expression signatures, a simulated treatment learning model identified which patients with multiple myeloma would benefit most from treatment with bortezomib or lenalidomide, researchers reported in Nature Communications.

The study included 910 participants across three phase 3 trials. In all, 20% would have a 100% greater-than-average progression-free survival (PFS) benefit from bortezomib, while 31% would have a 200% greater-than-average PFS benefit from lenalidomide, wrote Joske Ubels of University Center Utrecht, the Netherlands, and her colleagues.

The genetic heterogeneity of cancer and risk of treatment necessitate tools that “predict – at the moment of diagnosis – which patients will benefit most from a certain treatment,” the researchers wrote. While gene expression signatures can predict a favorable or adverse prognosis, they do not account for the effect of treatment on survival.

“The key idea of simulated treatment learning is that a patient’s treatment benefit can be estimated by comparing [his or her] survival to a set of genetically similar patients [who] received the comparator treatment,” they noted.

To do so, the researchers applied an algorithm called GESTURE to combined data from the TT2 (Total Therapy 2 for Multiple Myeloma), TT3, and HOVON-65/GMMG-HD4 trials. These trials compared bortezomib or lenalidomide with conventional therapies for multiple myeloma. The model identified 180 patients (20%) for whom bortezomib would produce a 100% greater PFS benefit than in the study population as a whole. Conversely, lenalidomide would produce a 200% greater PFS benefit in 31% of patients.

The simulated treatment learning model “can derive clinically actionable gene expression signatures that enable a more personalized approach to treatment,” the researchers concluded. The method requires a large dataset but could be useful for trials that have missed their primary endpoint, such as the CheckMate-026 trial of nivolumab. The next step is to see if the model makes useful treatment predictions for other cancers. The code needed to train and validate the model is available at github.com/jubels/GESTURE.

The Van Herk Fellowship provided support. The lenalidomide dataset was created as part of the Multiple Myeloma Research Foundation Personalized Medicine Initiative. Dr. Ubels and one coinvestigator are employees of SkylineDx; another coinvestigator served on its advisory board. The others reported having no relevant conflicts of interest.

SOURCE: Ubels J et al. Nat Commun. 2018 Jul 27. doi: 10.1038/s41467-018-05348-5.