Thrombosis is one of the most frequent complications of cancer, with a potential impact on morbidity and mortality, particularly those with acute myeloid leukemia (AML). Therefore, effective thrombosis prevention is a crucial aspect of cancer management. However, preventive measures against thrombosis may carry inherent risks and complications. Consequently, the application of thrombosis prevention should be limited to patients with a reasonable risk of developing thrombosis. This thesis explores the potential of data science (DS) methods for predicting venous thrombosis in patients with acute myeloid leukemia. In order to ascertain which patients are at risk, statistical and machine-learning (ML) algorithms were employed to predict which patients with leukemia will develop thrombosis. Multilayer Perceptron (MLP) was found to be the best fit among the models evaluated, achieving the C statistic of 0.749. We examined which attributes are significant and what role they play in prediction and found six significant parameters: sex of the patient, prior history of thrombotic event, type of therapy, international normalized ratio (INR), Eastern Cooperative Oncology Group (ECOG) performance status, and Hematopoietic Cell Transplantation-specific Comorbidity. These findings suggest that subtle DS techniques can improve the prediction of Thrombosis in AML patients, thereby aiding in individual treatment planning.

Traditional Software-Defined Wide Area Network (SD-WAN) solutions lack adaptive load-balancing mechanisms, leading to inefficient traffic distribution, increased latency, and performance degradation. This paper presents an Application-Aware Dynamic Load Balancing (AADLB) framework designed for open-source SD-WAN environments. The proposed solution enables dynamic traffic routing based on real-time network performance indicators, including CPU utilization, memory usage, connection delay, and packet loss, while considering application-specific requirements. Unlike conventional load-balancing methods, such as Weighted Round Robin (WRR), Weighted Fair Queuing (WFQ), Priority Queuing (PQ), and Deficit Round Robin (DRR), AADLB continuously updates traffic weights based on application requirements and network conditions, ensuring optimal resource allocation and improved Quality of Service (QoS). The AADLB framework leverages a heuristic-based dynamic weight assignment algorithm to redistribute traffic in a multi-cloud environment, mitigating congestion and enhancing system responsiveness. Experimental results demonstrate that compared to these traditional algorithms, the proposed AADLB framework improved CPU utilization by an average of 8.40%, enhanced CPU stability by 76.66%, increased RAM utilization stability by 6.97%, slightly reduced average latency by 2.58%, and significantly enhanced latency consistency by 16.74%. These improvements enhance SD-WAN scalability, optimize bandwidth usage, and reduce operational costs. Our findings highlight the potential of application-aware dynamic load balancing in SD-WAN, offering a cost-effective and scalable alternative to proprietary solutions.

One of the most important adverse consequences of antibiotic use is the development of bacterial resistance. This study investigated the prevalence and antimicrobial resistance patterns of multidrug-resistant (MDR) Escherichia coli in a tertiary care hospital during 2013–2015 and 2024, together with trends in antibiotic consumption and the molecular characteristics of resistance genes in 2024 isolates. Identification and susceptibility testing were performed using the Vitek® 2 system, antibiotic consumption was assessed according to WHO ATC/DDD methodology, and resistance genes were detected by PCR. No significant differences were observed in the total number of isolates, patients, or the proportion of E. coli isolates during 2013–2015. However, the isolation rate of MDR E. coli significantly declined from 36.9% to 30%. Total antibiotic consumption remained stable, with no correlation between consumption and MDR isolation rates, although ampicillin resistance increased significantly. In 2024, MDR E. coli accounted for 24.3% of isolates despite lower antibiotic consumption. Compared with 2015, resistance significantly increased to amoxicillin-clavulanic acid, cefotaxime, cefepime, ceftazidime, ciprofloxacin, and levofloxacin, while ceftriaxone resistance decreased. The blaCTX-M gene was detected in 53% of bloodstream isolates, indicating widespread dissemination of ESBL-producing E. coli and highlighting the need for continuous surveillance and antimicrobial stewardship.

One of the most important adverse consequences of antibiotic use is the development of bacterial resistance. This study investigated the prevalence and antimicrobial resistance patterns of multidrug-resistant (MDR) Escherichia coli in a tertiary care hospital during 2013–2015 and 2024, together with trends in antibiotic consumption and the molecular characteristics of resistance genes in 2024 isolates. Identification and susceptibility testing were performed using the Vitek^® 2 system, antibiotic consumption was assessed according to WHO ATC/DDD methodology, and resistance genes were detected by PCR. No significant differences were observed in the total number of isolates, patients, or the proportion of E. coli isolates during 2013–2015. However, the isolation rate of MDR E. coli significantly declined from 36.9% to 30%. Total antibiotic consumption remained stable, with no correlation between consumption and MDR isolation rates, although ampicillin resistance increased significantly. In 2024, MDR E. coli accounted for 24.3% of isolates despite lower antibiotic consumption. Compared with 2015, resistance significantly increased to amoxicillin–clavulanic acid, cefotaxime, cefepime, ceftazidime, ciprofloxacin, and levofloxacin, while ceftriaxone resistance decreased. The blaCTX-M gene was detected in 53% of bloodstream isolates, indicating widespread dissemination of ESBL-producing E. coli and highlighting the need for continuous surveillance and antimicrobial stewardship.

Background: Spinal deformities in children represent a relevant public health issue, with possible long-term consequences. Timely identification of their determinants is essential for adequate prevention. Methods: This study was a secondary analysis of data from the 2019 Serbian National Health Survey, including 1309 children aged 5–14 years. Logistic regression with LASSO regularization and multiple ML algorithms were tested, with XGBoost selected as the optimal model. Class imbalance was addressed using class weighting and SMOTE. Model interpretability was achieved using SHAP analysis. Results: The prevalence of spinal deformities was 8.6%. Univariable analyses showed that age, poorer self-rated health, chronic illness, recent injuries, and pes planus were significantly associated with spinal deformities. Family-related variables showed no significant associations. Among the evaluated models, XGBoost demonstrated the most stable performance across the applied evaluation metrics and the best balance between predictive performance and interpretability. SHapley Additive exPlanations (SHAP) analysis showed that pes planus was the strongest determinant, followed by age and chronic illness, while socio-demographic and family factors had minimal influence. Conclusion: Explainable machine learning models, particularly XGBoost combined with SHAP, can allow for the identification and interpretation of key determinants of spinal deformities in children. Pes planus was shown to be modifiable and relevant associated determinant, supporting its importance in early screening and preventive strategies.

This paper presents a closed-loop, CPU-aware traffic-control framework for SDN in 5G/6G multi-tenant edge environments based on commodity KVM/OVS infrastructures. It couples fine-grained data-plane telemetry via eBPF with adaptive XDP rate limiting, coordinated by a PID controller in the OVS datapath. Unlike control-plane polling, it provides real-time feedback between CPU utilization and traffic regulation. Experiments in a virtualized multi-tenant OVS testbed (KVM/virtio-net) keep CPU below per-slice CPU targets (e.g., 1.02% for a 3% setpoint), with an under-target bias that avoids overshoot while preserving stable forwarding. We attribute this bias, at light load, to a supply-limited regime, conservative per-slice CPU accounting, and stability-oriented PID tuning, and introduce a low-latency profile that mitigates this bias for latency-sensitive slices. The XDP datapath achieves 1–3 $\mathsf{\mu }$s per-packet processing with 5–10% additional CPU overhead relative to an uninstrumented baseline, while using less CPU than OVS policing at comparable throughput. A 3% per-slice CPU target balances isolation and throughput, while 2% yields stricter isolation at the cost of higher packet loss. Software-based rate limiting can induce cross-slice interference; effective isolation holds below 1 Gbps aggregate load. Above this, shared Linux kernel overhead degrades isolation, causing significant loss; thus, XDP alone cannot ensure line-rate isolation, motivating SmartNICs. The design improves efficiency, predictability, and isolation, laying a foundation for intelligent traffic management in future resource-intensive applications.