User login
SAN DIEGO – A newly developed personalized model that “harnesses the power of artificial intelligence” to predict overall survival and transformation to acute myeloid leukemia (AML) in patients with myelodysplastic syndromes outperforms both the original and revised International Prognostic Scoring Systems (IPSS, IPSS-R), according to Aziz Nazha, MD.
The machine learning model, which was built using clinical and genomic data derived from myelodysplastic syndrome (MDS) patients diagnosed according to 2008 World Health Organization criteria, incorporates information beyond that included in the IPSS and IPSS-R, and provides patient-specific survival probabilities at different time points, Dr. Nazha of Cleveland Clinic reported during a press briefing at the annual meeting of the American Society of Hematology.
The model was developed in a combined training cohort of 1,471 patients from the Cleveland Clinic and Munich Leukemia Laboratory and was validated in a separate cohort of 831 patients from the Moffitt Cancer Center in Tampa, Fla.
The concordance index – a measure for comparing the accuracy of the various models – was 0.80 for overall survival (OS), and 0.78 for AML transformation vs. 0.66 and 0.73, respectively, for IPSS, and 0.67 and 0.73, respectively, for IPSS-R, Dr. Nazha said. The new “geno-clinical” model also outperformed mutations-only analysis, mutations plus cytogenetics analysis, and mutations plus cytogenetics plus age analyses for both OS and AML transformation.
Adding mutational variant allelic frequency did not significantly improve prediction accuracy, he noted.
Dr. Nazha and his colleagues are developing a web application tool that can be used to run the trained model to calculate patient-specific, time-specific OS and AML transformation probabilities. He discussed the new model and its implications for personalized prognosis and treatment in this video interview.
Improved risk assessment helps patients understand their disease and “establish expectations about their journey with their disease,” and it is also extremely important for treating physicians, he said.
“All of our consensus guidelines and treatment recommendations are based on risk,” he explained, noting that the approach varies greatly for higher- and lower-risk patients.
This model represents a potential new focus on “personalized prediction” in addition to the increasing focus on personalized treatment and takes into account the heterogeneous outcomes seen in patients with MDS, he said.
Dr. Nazha reported consultancy for Karyopharma and Tolero, and data-monitoring committee membership for MEI.
SOURCE: Nazha A et al. ASH 2018, Abstract 793.
SAN DIEGO – A newly developed personalized model that “harnesses the power of artificial intelligence” to predict overall survival and transformation to acute myeloid leukemia (AML) in patients with myelodysplastic syndromes outperforms both the original and revised International Prognostic Scoring Systems (IPSS, IPSS-R), according to Aziz Nazha, MD.
The machine learning model, which was built using clinical and genomic data derived from myelodysplastic syndrome (MDS) patients diagnosed according to 2008 World Health Organization criteria, incorporates information beyond that included in the IPSS and IPSS-R, and provides patient-specific survival probabilities at different time points, Dr. Nazha of Cleveland Clinic reported during a press briefing at the annual meeting of the American Society of Hematology.
The model was developed in a combined training cohort of 1,471 patients from the Cleveland Clinic and Munich Leukemia Laboratory and was validated in a separate cohort of 831 patients from the Moffitt Cancer Center in Tampa, Fla.
The concordance index – a measure for comparing the accuracy of the various models – was 0.80 for overall survival (OS), and 0.78 for AML transformation vs. 0.66 and 0.73, respectively, for IPSS, and 0.67 and 0.73, respectively, for IPSS-R, Dr. Nazha said. The new “geno-clinical” model also outperformed mutations-only analysis, mutations plus cytogenetics analysis, and mutations plus cytogenetics plus age analyses for both OS and AML transformation.
Adding mutational variant allelic frequency did not significantly improve prediction accuracy, he noted.
Dr. Nazha and his colleagues are developing a web application tool that can be used to run the trained model to calculate patient-specific, time-specific OS and AML transformation probabilities. He discussed the new model and its implications for personalized prognosis and treatment in this video interview.
Improved risk assessment helps patients understand their disease and “establish expectations about their journey with their disease,” and it is also extremely important for treating physicians, he said.
“All of our consensus guidelines and treatment recommendations are based on risk,” he explained, noting that the approach varies greatly for higher- and lower-risk patients.
This model represents a potential new focus on “personalized prediction” in addition to the increasing focus on personalized treatment and takes into account the heterogeneous outcomes seen in patients with MDS, he said.
Dr. Nazha reported consultancy for Karyopharma and Tolero, and data-monitoring committee membership for MEI.
SOURCE: Nazha A et al. ASH 2018, Abstract 793.
SAN DIEGO – A newly developed personalized model that “harnesses the power of artificial intelligence” to predict overall survival and transformation to acute myeloid leukemia (AML) in patients with myelodysplastic syndromes outperforms both the original and revised International Prognostic Scoring Systems (IPSS, IPSS-R), according to Aziz Nazha, MD.
The machine learning model, which was built using clinical and genomic data derived from myelodysplastic syndrome (MDS) patients diagnosed according to 2008 World Health Organization criteria, incorporates information beyond that included in the IPSS and IPSS-R, and provides patient-specific survival probabilities at different time points, Dr. Nazha of Cleveland Clinic reported during a press briefing at the annual meeting of the American Society of Hematology.
The model was developed in a combined training cohort of 1,471 patients from the Cleveland Clinic and Munich Leukemia Laboratory and was validated in a separate cohort of 831 patients from the Moffitt Cancer Center in Tampa, Fla.
The concordance index – a measure for comparing the accuracy of the various models – was 0.80 for overall survival (OS), and 0.78 for AML transformation vs. 0.66 and 0.73, respectively, for IPSS, and 0.67 and 0.73, respectively, for IPSS-R, Dr. Nazha said. The new “geno-clinical” model also outperformed mutations-only analysis, mutations plus cytogenetics analysis, and mutations plus cytogenetics plus age analyses for both OS and AML transformation.
Adding mutational variant allelic frequency did not significantly improve prediction accuracy, he noted.
Dr. Nazha and his colleagues are developing a web application tool that can be used to run the trained model to calculate patient-specific, time-specific OS and AML transformation probabilities. He discussed the new model and its implications for personalized prognosis and treatment in this video interview.
Improved risk assessment helps patients understand their disease and “establish expectations about their journey with their disease,” and it is also extremely important for treating physicians, he said.
“All of our consensus guidelines and treatment recommendations are based on risk,” he explained, noting that the approach varies greatly for higher- and lower-risk patients.
This model represents a potential new focus on “personalized prediction” in addition to the increasing focus on personalized treatment and takes into account the heterogeneous outcomes seen in patients with MDS, he said.
Dr. Nazha reported consultancy for Karyopharma and Tolero, and data-monitoring committee membership for MEI.
SOURCE: Nazha A et al. ASH 2018, Abstract 793.
REPORTING FROM ASH 2018
