Abstract P1-20-02: A machine learning approach to identify risk factors associated with skeletal-related events following denosumab cessation among patients with bone metastases from breast cancer

Abstract

Abstract Background: The anti-RANKL monoclonal antibody denosumab has been demonstrated to be superior to the bisphosphonate zoledronate in preventing skeletal-related events (SREs) among patients with incident bone metastases (BMs) from solid tumors (STs), including breast cancer. Clinical guidelines recommend the use of a bone-targeting agent for SRE prevention for ≥2 years. However, the denosumab treatment duration is often <1 year in the US. We applied a supervised machine learning approach using real-world data to estimate patient-level SRE risk following cessation of denosumab and determine risk factors associated with increased SRE risk. The method selected prior SREs, shorter denosumab treatment duration, and higher number of clinic visits as the top-ranked risk factors among a diverse group of patients with BMs from STs (Stopeck et al., ASCO 2021). Here, we report the top-ranked risk factors for the subgroup of patients with BMs from breast cancer. Methods: Using the Optum PanTher Electronic Health Record repository, patients diagnosed with incident BMs from a primary ST between January 1, 2007, and September 1, 2019, were evaluated for inclusion in the study. Eligible patients must have received ≥2 consecutive 120 mg denosumab doses on every 4-week (±14 days) schedule and had a minimum follow-up ≥1 year after the last denosumab dose or an SRE occurrence between days 84 and 365 following denosumab cessation. An SRE risk prediction model was developed using extreme gradient boosting and evaluated on an independent test dataset. Multiple variables associated with patient demographics, comorbidities, laboratory values, treatments, and denosumab exposures were examined as potential risk factors for SREs. After denosumab cessation, the impact and relative importance of these factors were extracted from the model using Shapley Additive Explanations (SHAP). In addition, findings from univariate analyses on risk factors with high importance from the breast cancer model were reported. Results: Of 1414 patients who met the inclusion criteria, 563 (40%) had BMs from breast cancer. Following denosumab cessation, 167 (30%) patients in the breast cancer subgroup experienced ≥1 SRE. The breast cancer model performance was meaningful, as evidenced by the area under the receiver operating characteristic (AUROC) score of 73%. SHAP resulted in several significant factors that predicted an increased SRE risk for the subgroup following denosumab cessation, including denosumab treatment duration of ≤8 months, prior SREs, and an average of >2 clinic visits per month (Table). Univariate analyses showed a positive correlation between increased SRE risk and prior SREs, while they revealed an inverse relationship between increased SRE risk and longer durations of denosumab. Conclusion: Shorter denosumab treatment duration, prior SREs, and higher number of clinic visits are top-ranked risk factors associated with SREs after discontinuation of denosumab treatment in patients with BMs from STs, including breast cancer. A machine learning approach to SRE risk factor identification may help clinicians assess the risks of discontinuing denosumab treatment and improve clinical outcomes for patients with BMs from breast cancer. SHAP Risk Factors That Increase SRE Risk 3-12 Months After Denosumab Cessation in Patients With BMsDenosumab therapy decisions: denosumab duration ≤8 months, time to denosumab initiation ≤2 months after BM diagnosisPrior SREs: ≥2 cumulative number of SREs since baseline up to cessationa, SRE occurrence (from denosumab initiation to cessation)Comorbidities: patients with anxietyVisits: >2 average number of visits per month (hospitalization, emergency room, and other visits [excluding nonphysician interaction]), ≥1 hospitalization, ≥1 emergency room visitaBaseline was defined as 180 days before the date of initial BM diagnosis. BM, bone metastasis; SHAP, Shapley Additive Explanations; SRE, skeletal-related event Citation Format: Alison Stopeck, Celestia Higano, David Henry, Basia Bachmann, Marko Rehn, Dionna Jacobson, Benoit Cadieux, Hossam Saad. A machine learning approach to identify risk factors associated with skeletal-related events following denosumab cessation among patients with bone metastases from breast cancer [abstract]. In: Proceedings of the 2021 San Antonio Breast Cancer Symposium; 2021 Dec 7-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2022;82(4 Suppl):Abstract nr P1-20-02.