JMIR Publications

ABSTRACT

Background:

Oral medications are commonly used in the treatment of breast cancer (BC) patients, despite a high association with non-adherence rates. Since treatment compliance is fundamental to achieving optimal benefit from treatments, finding ways to effectively improve adherence to medication in BC patients is of utmost importance. Artificial Intelligence (AI) is being widely applied to healthcare, and medication adherence is no exception.

Objective:

This review aims to offer an overview of the state-of-the-art contribution of AI to medication non-adherence among BC patients, suggesting future research directions, and highlighting existing gaps.

Methods:

Three databases (PubMed, Embase, and Scopus) were systematically searched to identify studies applying AI to predict or support medication adherence among BC patients. The systematic review was conducted following PRISMA guidelines and was registered in PROSPERO database. The risk of bias and the applicability of the included studies were evaluated using the Prediction Model Risk of Bias Assessment Tool and the Downs and Black's methodological quality scale.

Results:

Seven studies were included in the review. Most of them (k = 6) focused on developing machine learning models to predict medication non-adherence among patients with BC. These studies employed a range of techniques, including logistic regression, artificial neural networks, and random forest. Model performance varied widely, with AUC values ranging from 0.61 to 1.00. Predictors of non-adherence were clustered into four main groups: clinical, disease, and treatment-related factors (e.g., side-effects, comorbidities); behavioural factors (e.g., prior non-adherence); psychosocial factors (e.g., quality of life, self-efficacy); and sociodemographic factors (e.g., age, income). The only intervention study identified evaluated an AI-based chatbot and reported promising results, showing a 20% increase in adherence among participants who engaged with its reminder feature. Overall, all included studies were at high risk of bias, mainly due to the absence of model calibration or insufficient reporting, and their findings should therefore be interpreted with caution. A further limitation was the complete lack of attention to implementation: predictive accuracy alone is insufficient, and future work must also address actionability, safety, and cost-effectiveness to enable real clinical use. Progress in this area will require coordinated efforts among researchers, developers, clinicians, and policymakers to support the responsible development and implementation of these tools into routine care.

Conclusions:

This review shows that AI may emerge as a promising resource for identifying and addressing medication non-adherence in BC. However, its contribution remains limited by gaps in implementation and by the scarce attention given to patient engagement. Moving forward, AI-based tools will require careful, ethically informed development and robust planning across all stages, from design to clinical use, if they are to become reliable and meaningful component of routine care.