Adopting Machine Learning to Predict Breast Cancer Patients Adherence with Lifestyle Recommendations and Quality of Life Outcomes

Anna Crispo¹Maria Elisabetta Pagnano²Agnese Bonfigli²Leandro Pecchia²Assunta Luongo¹Giuseppe Porciello¹Melania Prete^3*Sergio Coluccia⁴Luca Bacco²Sara Vitale¹Elvira Palumbo¹Paolo Giaccone²Rosa Pica¹Maria Grimaldi¹Marco Cascella⁵Ernesta Cavalcanti¹Anita Minopoli¹Michelino De Laurentiis¹Massimo Libra⁶Jerry Polesel⁷Samuele Massarut⁷Egidio Celentano¹Livia Augustin¹

• ¹Istituto Nazionale Tumori IRCCS Fondazione Pascale, Naples, Italy
• ²Universita Campus Bio-Medico di Roma, Rome, Italy
• ³G. Pascale National Cancer Institute Foundation (IRCCS), Naples, Italy
• ⁴Universita degli Studi di Milano, Milan, Italy
• ⁵Universita degli Studi di Salerno, Fisciano, Italy
• ⁶Universita degli Studi di Catania, Catania, Italy
• ⁷Centro di Riferimento Oncologico IRCCS, Aviano, Italy

Healthy lifestyle behaviors and improved quality of life have been associated with better prognoses in breast cancer survivors. However, sustaining behavioral changes remains challenging; therefore, identifying effective components of lifestyle education programs is essential to enhance adherence, improve quality of life, and facilitate their integration into clinical practice. This study aimed to predict patient adherence to a lifestyle intervention of diet, physical activity, and vitamin D supplementation and to forecast the most frequent Health-Related Quality of Life over the subsequent three measurements. A total of 316 breast cancer survivors were included in the analysis. Adherence was modeled as a multi-label time series classification task, with compliance recorded on a three-point scale for each treatment component at quarterly intervals over one year. Health-Related Quality of Life was predicted by evaluating first-year adherence data to estimate the mean score over the subsequent three measurements. The dataset was split into 70% for training and 30% for evaluation. Random forest classifiers were employed for adherence prediction, achieving accuracy of up to 81%. An XGBoost regressor was used for Health-Related quality of life prediction, and it was compared to a baseline linear regression model. XGBoost demonstrated superior predictive performance, achieving an R-squared value of 0.62. These findings highlight the promise of machine learning techniques in supporting personalized medicine. Advanced predictive models may aid in identifying patients at risk of non-adherence, enabling early interventions, and improving long-term outcomes through tailored lifestyle strategies for breast cancer survivors.