Predictors of abemaciclib discontinuation in patients with breast cancer: a multicenter retrospective cohort study

Abstract

Objective: This study explored the predictors of abemaciclib discontinuation, a cyclin-dependent kinase 4 and 6 inhibitor, in patients with breast cancer.

Material and methods: Between November 2018 and March 2023, 147 patients with breast cancer treated with abemaciclib at Osaka Medical and Pharmaceutical University Hospital and Kindai University Nara Hospital were included. The exclusion criteria were as follows: lack of blood testing within 2 weeks prior to starting abemaciclib therapy, transfer to another facility after the commencement of abemaciclib therapy, and discontinuation of abemaciclib therapy due to the diagnosis of another cancer. The duration from the initiation of abemaciclib to discontinuation for any reason and to temporary suspension or dose reduction due to adverse events were analyzed as outcome variables using multivariate Cox regression analysis.

Results: Baseline weight < 54 kg, bone metastases, and hemoglobin level ≤ 12.4 g/dL were independent predictors of abemaciclib discontinuation for any reason. The main adverse events leading to abemaciclib discontinuation were liver enzyme elevation and gastrointestinal symptoms. Additionally, focusing on the adverse event of abemaciclib, a baseline weight < 54 kg was an independent predictor of temporary suspension or dose reduction due to adverse events. The most common adverse events leading to temporary suspension or dose reduction were neutropenia and diarrhea.

Conclusion: Patients with lower body weight are more susceptible to the adverse events of abemaciclib, increasing their risk of treatment discontinuation. In such patients, strict monitoring of adverse events and consideration of more frequent medical visits are necessary from the start of abemaciclib therapy.

Keywords: Abemaciclib; Breast cancer; Chemotherapy; Discontinuation; Predictor.

Fig.1

Study diagram. This diagram illustrates the timeline of covariate assessments, exclusion assessments, and follow-up periods for patients treated with abemaciclib. Covariates such as clinical characteristics (HER2, ER, PgR, luminal type, gene mutations, etc.) and laboratory values were assessed within specific windows prior to and at the start of therapy. Patients were followed up from the initiation of abemaciclib therapy until the occurrence of an event, such as therapy discontinuation, or censoring. The exclusion assessment window includes events like transfer to a different hospital or diagnosis of other cancers. BMI, body mass index; ER, estrogen receptor; HER2, human epidermal growth factor receptor 2; LH-RH, luteinizing hormone-releasing hormone; PgR, progesterone receptor. ^aEarliest outcome of interest: discontinuation of abemaciclib therapy, end of the study period

Fig. 2

Kaplan–Meier curve for time to (a) abemaciclib discontinuation due to any reason and (b) abemaciclib temporary suspension or dose reduction due to adverse events

Fig. 3

Sensitivity analysis for predictors of abemaciclib discontinuationdue to any reason. a Density plot of hazard ratios from 10,000 bootstrap samples. b Influence plot for hazard ratios from subsets generated by jackknife resampling. Box plots show the median, third quartile, first quartile, and range of data points outside the quartile range, but excluding outliers. c Cross-validation error as a function of λ. Error bars indicate the standard error of the cross-validation error at each λ value. The hyperparameter α, a measure of the mixture between ridge regression and lasso regression, was set to 0.9 based on the tenfold cross-validation error. The vertical dashed red line corresponds to the value of regularization parameter λ that minimizes cross-validation error. d Regularization path for elastic net-regularized Cox regression

Fig. 4

Sensitivity analysis for predictors of abemaciclib temporary suspension or dose reduction due to adverse events. a Density plot of hazard ratios from 10,000 bootstrap samples. b Influence plot for hazard ratios from subsets generated by jackknife resampling. Box plots show the median, third quartile, first quartile, and range of data points outside the quartile range, but excluding outliers. c Cross-validation error as a function of λ. Error bars indicate the standard error of the cross-validation error at each λ value. The hyperparameter α, a measure of the mixture between ridge regression and lasso regression, was set to 0.7 based on the tenfold cross-validation error. The vertical dashed red line corresponds to the value of regularization parameter λ that minimizes cross-validation error. d Regularization path for elastic net-regularized Cox regression