MM-146 Using Data Science to Characterize Patterns in Cardiovascular Clinical Data: Topological Analysis of Baseline Characteristics in Patients With Relapsed/Refractory Multiple Myeloma Enrolled in Carfilzomib Trials

Context: Oncology clinical trials amass vast amounts of patient data, providing the opportunity to leverage machine learning. Patients with multiple myeloma (MM) often have cardiovascular risk factors and/or established cardiovascular disease. Objective: We used topological analysis to identify characteristics at baseline associated with subsequent cardiovascular adverse events (AEs) in patients with relapsed/refractory MM (RRMM). Design: Retrospective data were pooled from the investigational arm (carfilzomib-treatment) of phase 3 trials (ASPIRE, ARROW, FOCUS, ENDEAVOR) for this post-hoc exploratory analysis. Variables included baseline demographics, vital signs, electrocardiogram results, laboratory values, and comorbidities. Topological data analysis (an analytic approach akin to dimensionality reduction, distinct from traditional multivariate modeling) was used to reveal AE occurrence micropatterns. A decision algorithm accounting for both posterior and prior AE probability was used to find risk factors meeting pre-specified clinical criteria. Results: In the pooled population (n=1,484), grade ≥3 cardiovascular AEs analyzed were cardiac failure (5.1%), ischemic heart disease (IHD, 2.2%), and hypertension (9.2%). Using this modeling approach, patient clusters with higher or lower AE posterior probabilities were derived (cardiac failure: 31.1% [higher probability]/68.9% [lower probability]; IHD: 38.5%/61.5%; hypertension: 32.1%/67.9%), with AE rates enriched in the high-probability group (cardiac failure AE rate was 4x higher in the high-probability group than the low-probability group [10.8% vs. 2.4%]). Grade ≥3 cardiac failure clustered in patients with baseline comorbidity of congestive heart failure, moderate-to-severe chronic kidney disease, prolonged mean QRS ≥120 msec, glucose ≥7.9 mmol/L, and/or Asian race. Grade ≥3 IHD clustered in patients with prior myocardial infarction, glucose ≥7.9 mmol/L, activated partial thromboplastin time ≤20 seconds, and/or creatinine clearance ≤60 mL/min. Grade ≥3 hypertension clustered with systolic blood pressure ≥140 mmHg and/or Asian or African race. Conclusions: This novel topological analysis found that cardiovascular AEs in RRMM patients enrolled in carfilzomib trials clustered into groups with known risk factors. Characteristics associated with cardiovascular AEs were detectable by standard clinical assessment or laboratory tests. Study limitations include the lack of comparator arm, small number of Asian patients (n=102), and that this novel methodology has not been validated in clinical settings. These findings suggest that topological analysis may be a useful tool for leveraging machine learning in clinical trials.