Image Credit: Adobe Stock Images/GraphicRoyalty.com
Estrogen receptor–positive (ER+) breast cancer—which is when cancer cells with receptors attach to estrogen—make up more than 70% of all breast cancer diagnoses. Receptors must be found in more than 10% of cancers in order to be considered ER.+
To combat this, the use of endocrine therapy, along with the creation oral endocrine drugs, is able to provide patients with alternatives to the standard intravenous chemotherapy, but a challenge in medication adherence persists.
Although this challenge of adherence and persistence is well-recognized in the space, it is difficult for it to be evaluated with a clinical trial; however, a retrospective cohort study published in JAMA Network Open1 sought to determine the characteristics of French female patients with localized breast cancer that are associated with greatest possibility of nonpersistence to prescribed endocrine therapy.
The study featured data from the French National Health Data System (Système National des Données de Santé [SNDS]) database between January 2013 and December 2018; this aforementioned database includes reimbursement data for the French Health System, accounting for 66 million people, or 98% of the total French population, while serving as “a completely unbiased source of data,” according to study investigators.
It’s also important to note that the models created for this study were merely trained using available data in the reimbursement history. At the point, using reimbursement history, a care pathway was created that allowed the investigators to calculate both adherence and persistence.
Among the 229,695 female patients (median [IQR] age, 63 [52-72] years) with localized breast cancer that were included in this study, a deep learning model derived from gated-recurrent unit architecture was utilized to uncover incidences of nonpersistence or nonadherence. This particular model determined an area under the receiving operating curve of 0.71 for persistence and 0.73 for adherence.
Using the Shapley Additive Explanations values, a better understanding into the real each of different features played over the model’s decision added more clarity. Patients older than 70 years of age, with past nonadherence, taking more than one treatment in the previous three months, and with low income, presented a greater risk nonpersistence episodes. Given this determination, age and past nonadherence—along with consistency of past adherence—were the main components of the nonadherence model.
Various limitations arose along the way, including the fact that the data was limited to French health insurance reimbursement data. Also, the aforementioned deep learning model that was used to measure adherence focused on results that were measured via pharmacy claims data, which could cause bias in adherence classification for those patients who decide not to take their medications.
As a result, the investigators concluded that, “In this study, a model was able to estimate the risk of nonpersistence and nonadherence among French female patients with localized breast cancer. Additionally, we identified factors, such as age, past behavior, and number of treatments, that may be associated with risk of nonpersistence and nonadherence. These results highlighted the potential to dynamically anticipate nonadherence and nonpersistence in women undergoing endocrine therapy for early breast cancer.
“The emergence of telemonitoring and digital patient tracking presents an exciting opportunity to integrate models into clinical practice. This integration could refine follow-up and support strategies for patients. A prospective evaluation or clinical trial is essential to more precisely assess the outcomes of such technological innovations. This cohort analysis demonstrated the feasibility of estimating the risk of nonadherence and nonpersistence using pharmacy purchase behaviors.”
Reference
1. Rinder P, Marcille T, Sinel-Boucher P, et al. Dynamic Projection of Medication Nonpersistence and Nonadherence Among Patients With Early Breast Cancer. JAMA Netw Open. 2024;7(5):e2411909. doi:10.1001/jamanetworkopen.2024.11909
