A Study to Predict Response to Azacitidine or Lenalidomide/Azacitidine Combination Therapy in MDS and AML Patients Using a Computational Biological Modeling (CBM) Approach

Background: Several treatment options are available for patients with myelodysplastic syndromes (MDS) or acute myeloid leukemia (AML); however, the majority of patients will fail treatment or relapse. Predicting response is imperative in determining the right treatment regimen for each patient but currently no validated method exists.

Aim: In this study, we aimed to predict treatment response in patients with MDS or AML using Cellworks' computational biology modeling (CBM) approach.

Methods: In total, 16 patients were included in this analysis. 11 patients were treated on a single-institution phase II study of 5-day intravenous azacitidine (AZA) in patients with previously untreated MDS (Martin et al, AM J Hematol 2009 [NCT00384956]). An additional 5 patients were treated on a single-institution phase I study of daily high-dose lenalidomide (LEN) and 5-day intravenous AZA induction followed by low dose LEN and AZA maintenance in patients with newly diagnosed or relapsed AML (Ramsingh et al, Leukemia 2013 [NCT01016600]). Both treatments were generally well tolerated and showed promising response rates.

For this analysis, whole exome sequencing (SureSelectXT Clinical Research Exome) and CGH microarrays were performed on previously collected samples associated with the respective clinical trials. The genomic data and non-genomic clinical data (age, diagnosis, white blood cell count, hemoglobin, platelet count, and bone marrow blast percentage) were used for CBM and response prediction. Digital drug simulations were conducted by quantitatively measuring drug effect on a cell growth score (proliferation + viability + apoptosis) along with impact on a identified disease specific biomarker unique to each patient. Each patient-specific protein network was screened for the extent by which AZA or LEN+AZA, respectively, reduced disease growth. The CBM tool used simulation prediction and clinical data through weighted decision tree analysis for the final response call. The predicted response was compared with the clinical response to assess for accuracy and predictive value. Patients who had a partial response or better per standard IMW criteria were classified as responders, all others were classified as non-responders.

Results: 16 drug predictions were made by the CBM with 81% accuracy, 86% positive predictive value, 78% negative predictive value, 75% sensitivity and 88% specificity.

In the subset of MDS patients treated with AZA, 4/11(36%) were classified as responders and 7/11 (64%) non-responders per IMW criteria. The median age at study entry was 73 and 6/11 were male. The median number of cycles of AZA administered was 6 (range 1-6). The CBM accurately predicted the clinical outcomes from 9/11 patients including 2/4 responders and 7/7 non-responders.

In the subset of AML patients treated with LEN + AZA, 4/5 (80%) were classified as responders and 1/5 (20%) non-responders per IMW criteria. The median age at study entry was 71 and 1/5 were male. 3 of the patients had newly diagnosed AML and 2 had relapsed disease. The median number of cycles of LEN + AZA administered was 5 (range 1-17). The CBM accurately predicted the clinical outcomes from 4/5 patients including 4/4 responders.

Conclusions: The CBM tested herein, which included genomic abnormalities and clinical data, showed high predictive value of protein network aberrations and clinical outcomes. The study validates the approach to a priori predict response and identify the right therapy option for the patient and could be used to establish eligibility criteria for precision enrollment in drug development trials. In addition, the network method uncovered possible molecular reasons for drug failure and resistance pathways that could be targeted to induce chemo sensitivity. This clinical decision support technology could enable improved effectiveness of therapy and help avoid unneeded toxicity and reduce health care expenses by eliminating unnecessary treatments.