Breast cancer (BC) is the most frequent type of solid tumor and the second highest cause of cancer death in women [1]. Treatment of BC is hampered by tumors having a wide molecular heterogeneity, with consequences for relapse risk and response to treatment [2]. Several tumor phenotypes have been identified to date (luminal A, luminal B, HER2+, and triple-negative), depending on the putative molecular targets such as estrogen receptors (ER), progesterone receptors (PR), the human epidermal growth factor 2 receptor (HER2) and Ki-67 level. Triple-negative BC (TNBC) is characterized by the lack of expression of these molecular targets. Patients with this BC subtype and ages younger than 40 present an early risk of relapse and a low survival rate compared to other subtypes [3,4]. Adjuvant radiotherapy (RT) is part of the standard BC treatment due to its effects on loco-regional relapse reduction, as well as the improvement in survival for early-stage to locally advanced BC following conservative surgery or post-mastectomy, with or without regional lymph node involvement [5]. However, the efficacy of RT is not definitively curative, and there are some patients with BC who, over time, develop disease progression (DP) [6,7]. In this context, huge efforts have been focused on investigating the causes of treatment resistance and BC progression with the aim to increase the survival and quality of life of these patients [8]. However, results of these efforts are still inconclusive, and this hampers the design of efficient therapeutic strategies or finding biomarkers that identify individuals at high risk of relapse [9]. We, and other research groups, have reported evidence that antioxidant and inflammation systems are associated with the onset and development of BC, and contribute to resistance-to-treatment, and prognosis [10,11]. We have shown that circulating levels of the enzyme paraoxonase-1 (PON1) are decreased in patients with BC, and other types of cancer, compared to the healthy population [12,13,14,15]. This enzyme degrades lipid peroxides in lipoproteins and cells and plays an important antioxidant role in the organism [16]. Moreover, low serum PON1 concentrations post-RT have been associated with metastatic BC [13]. PON1 participates in the control of inflammation, reducing the capacity of macrophages to oxidize low-density lipoproteins, and downregulating the levels of the pro-inflammatory chemokine (C-C motif) ligand 2 (CCL2) [10]. Cytokines play a key role in carcinogenesis because they are involved in processes such as cell growth, differentiation, proliferation, and migration [17,18]. Interleukin-4 (IL-4), interferon-gamma (IFN-γ), and CCL2 promote tumorigenesis during equilibrium and escape stages. High plasma concentrations of these inflammatory markers have been related to tumor metastasis and poor prognosis [19,20]. However, these associations have been identified using traditional statistical methods, which have several limitations in identifying new variables and generating integrative visualizations [21]. In recent years, technological advances such as the employment of machine learning algorithms have been postulated as accurate methods to find predictive variables in cancer [21,22,23]. For example, previous studies have used random forest for BC early detection according to the clinicopathological features of the patients or Gradient Boosting Machine (GBM) to find and classify predictive variables related to prognosis in patients with different types of cancer [24,25,26]. Although this approach is promising, the evidence related to the identification of circulatory parameters capable of stratifying BC patients with and without DP is scarce.

Hence, in the present study, we investigated the clinical evolution of BC patients and the impact of RT administration on the circulating levels of PON1-related variables, cytokines, and standard biochemical and hematological analytes. Moreover, we identified potential biomarkers of BC prognosis using the GBM algorithm.