In a recent study, researchers used a machine learning approach to determine that genomic biomarkers are capable of predicting a lack of response to hypomethylating agents (HMAs) in some patients with myelodysplastic syndromes (MDS). Results were reported in JCO Precision Oncology.
Prior to the initiation of treatment, 433 patients in this study were screened using a 29-gene panel for identification of myeloid-associated mutations. The researchers developed a market basket analysis (MBA) algorithm based on mutation data and responses to HMA therapy among these patients, who were assigned to the training cohort. The model was then tested with data from the randomized phase 2/3 SWOG S1117 clinical trial (ClinicalTrials.gov Identifier: NCT01522976).
In the training cohort, azacitidine was used alone or in combination with other agents in 53% of patients, and decitabine was used alone or in combination with other agents in 47%. There was a median of 3 mutations (range, 0-9) per patient, with 41% of patients showing at least 3 mutations per sample.
Continue Reading
The training cohort had a 43% overall response rate to HMAs. The median overall survival (OS) rate was 19.5 months for the total training cohort; it was 29.5 months for patients who responded to HMA therapy and 18.9 months for patients who did not experience a response. Median study follow-up was 30 months.
With the MBA algorithm used in this study, the researchers uncovered 8 combinations of the ASXL1, NF1, EZH2, TET2, RUNX1, SRSF2, and BCOR genes that could function as biomarkers for predicting HMA resistance.
When used to estimate responses for the 30% of patients in the training cohort who possessed the biomarkers identified in this study, the model accuracy rate was 87%. The accuracy rate was 93% when the model was tested using data from patients with these biomarkers in the SWOG S1117 trial population.
“Our study highlights the importance of machine learning algorithms…in translating genomic data into useful clinical tools that can be used by physicians in the clinic,” the investigators wrote.
Reference
- Nazha A, Sekeres MA, Bejar R, et al. Genomic biomarkers to predict resistance to hypomethylating agents in patients with myelodysplastic syndromes using artificial intelligence [published online September 20, 2019]. JCO Precis Oncol. doi:10.1200/PO.19.00119
