CARDIOVASCULAR RISK STRATIFICATION USING UNSUPERVISED MACHINE LEARNING IN BREAST CANCER PATIENTS, INITIATING ANTHRACYCLINE-BASED TREATMENTS
Author(s)
Udim Damachi, MS1, Manu Murali Mysore, MD2, Mathangi Gopalakrishnan, MS, PhD1, Susan Dosreis, PhD1, Eberechukwu Onukwugha, MSc, PhD1, Wendy Camelo Castillo, MD, MSc, PhD1.
1University of Maryland, Baltimore School of Pharmacy, Baltimore, MD, USA, 2University of Maryland School of Medicine, Baltimore, MD, USA.
1University of Maryland, Baltimore School of Pharmacy, Baltimore, MD, USA, 2University of Maryland School of Medicine, Baltimore, MD, USA.
OBJECTIVES: Approximately 39% of women with breast cancer have concurrent cardiovascular risk factors (CVRF) before anthracycline-based treatment (ABT). Concurrent CVRF may have a synergistic effect on anthracycline-induced cardiotoxicity (AIC) after ABT initiation. We explored whether unsupervised machine learning could identify latent subgroups defined by distinct CVRF profiles and assessed their association with AIC-related outcomes.
METHODS: We derived a retrospective cohort from 2006-2024 using a 25% random sample of the IQVIA PharMetrics® Plus for Academics US health plan claims data. Women 18-84 years with a claim for breast cancer diagnosis and breast cancer surgery six months before ABT initiation (index date) comprised the cohort. K-prototype clustering identified subgroups with varying levels of CVRF. We included age, cancer treatments, CVRF, cardiovascular conditions, and treatment as variables in our clustering algorithm. The optimal number of clusters was determined using the silhouette method, and Cox proportional hazards models were used to estimate the risk of AIC-related (heart failure or cardiac arrhythmia) hospitalizations and emergency department (ED) visits across the identified clusters.
RESULTS: A total of 5,896 women with a mean age of 52.4 years (SD 9.7) comprised the eligible cohort. We identified four clusters. Cluster 1 consisted predominantly of women aged 50-59 years with a low prevalence of cardiovascular conditions; Cluster 2 included women with a high prevalence of CVRF; Cluster 3 included mostly younger women (mean:43.4 years-old) with low prevalence of cardiovascular conditions; and Cluster 4 comprised women ≥60 years-old with hypertension and antihypertensive therapy. Clusters 1, 2, and 4 demonstrated a higher hazards of AIC-related hospitalizations and ED visits relative to Cluster 3.
CONCLUSIONS: Using real-world data and K-prototype clustering, we identified distinct cardiovascular risk subgroups among women initiating ABT. Distinct cluster separation was largely driven by the combination of age and CVRF burden.
METHODS: We derived a retrospective cohort from 2006-2024 using a 25% random sample of the IQVIA PharMetrics® Plus for Academics US health plan claims data. Women 18-84 years with a claim for breast cancer diagnosis and breast cancer surgery six months before ABT initiation (index date) comprised the cohort. K-prototype clustering identified subgroups with varying levels of CVRF. We included age, cancer treatments, CVRF, cardiovascular conditions, and treatment as variables in our clustering algorithm. The optimal number of clusters was determined using the silhouette method, and Cox proportional hazards models were used to estimate the risk of AIC-related (heart failure or cardiac arrhythmia) hospitalizations and emergency department (ED) visits across the identified clusters.
RESULTS: A total of 5,896 women with a mean age of 52.4 years (SD 9.7) comprised the eligible cohort. We identified four clusters. Cluster 1 consisted predominantly of women aged 50-59 years with a low prevalence of cardiovascular conditions; Cluster 2 included women with a high prevalence of CVRF; Cluster 3 included mostly younger women (mean:43.4 years-old) with low prevalence of cardiovascular conditions; and Cluster 4 comprised women ≥60 years-old with hypertension and antihypertensive therapy. Clusters 1, 2, and 4 demonstrated a higher hazards of AIC-related hospitalizations and ED visits relative to Cluster 3.
CONCLUSIONS: Using real-world data and K-prototype clustering, we identified distinct cardiovascular risk subgroups among women initiating ABT. Distinct cluster separation was largely driven by the combination of age and CVRF burden.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
RWD91
Topic
Real World Data & Information Systems
Disease
SDC: Cardiovascular Disorders (including MI, Stroke, Circulatory), SDC: Oncology