The Medisch Spectrum Twente (MST) is the hospital of the city of Enschede. It is a top-clinical center offering secondary and limited tertiary care..
Healthcare systems are increasingly reliant on accurate cost forecasting tools to support strategic management and cost control. Healthcare costs are growing and are vulnerable to external systemic shocks and structural breaks, such as those induced by the COVID-19 pandemic. This systematic literature review examines the development in time-series forecasting (TSF) applied to healthcare cost, with a lens on service contracts. Following the PRISMA 2020 framework, a systematic search was conducted across Scopus, IEEE, and ACM databases, yielding 422 records. After screening and eligibility assessment, 50 studies published between 2020 and 2025 were included in the final synthesis. The results show an evolution from classical univariate models (i.e., ARIMA) toward multivariate, hybrid, and ensemble-based machine learning (ML) approaches such as random forest and XGBoost. Within this shift toward more complex models, neural network approaches often outperform traditional methods for long-term horizons but require extensive preprocessing, greater computational power, and larger data volumes while diminishing the explainability of the model. Only a few studies incorporate exogenous shocks, revealing a persistent gap in adaptive and explainable models for healthcare cost forecasting. Evaluation practices remain inconsistent, often not mentioning or lacking time-aware validation. Building on taxonomies proposed in prior TSF literature and surveys, we introduce an eight-step synthesis framework integrating data preparation, model selection, validation, and drift monitoring into a single pipeline. This synthesis highlights the need for more research on multi-modal data integration, domain-informed interpretability, and adaptive input and retraining strategies.
Investigate associations between brain pathology (pTDP-43 inclusions and microglial activation) and cognitive and behavioural impairment in patients with amyotrophic lateral sclerosis (ALS). Based on comprehensive neuropsychological examination and behavioural assessment, 21 ALS patients of whom post mortem brain tissue was obtained, were classified as having 1) no cognitive and/or behavioural impairment (pure motor ALS), 2) mild cognitive and/or behavioural impairment (ALSci/bi), and 3) ALS with behavioural variant frontotemporal dementia (ALS-bvFTD). Immunohistochemical staining of pTDP-43 and HLA-DR-defined microglial activation was semi-quantitatively assessed in grey and/or white matter of the prefrontal cortex, thalamus, hippocampus, and motor cortex. Fourteen patients had pure motor ALS, four patients had ALSci/bi, and three patients had ALS-bvFTD. pTDP-43 pathology in the grey matter of the prefrontal cortex and gyrus dentatus differed between groups, especially between pure motor ALS and ALS-bvFTD. For each extra-motor brain region, pTDP-43 severity was highest in patients with ALS-bvFTD and lowest in patients with pure motor ALS, with ALSci/bi in between. This pattern was not observed for microglial activation. Associations between white matter pTDP-43 severity and cognitive/behavioural impairment were less robust than those in grey matter. Severity of cognitive and/or behavioural impairment in ALS is related to severity of pTDP-43 pathology, in particular in the grey matter of extra-motor brain regions; we did not detect a clear association with microglial activation.
To optimize surveillance in individuals with a family history (FH) of colorectal cancer (CRC), knowledge on the incidence rate of non-advanced adenomas (nAAs) and their progression rate to advanced neoplasia (AN) is crucial. We jointly estimated personalized adenoma incidence and progression rates using a novel statistical approach. We used data of individuals with ≥ 1 first-degree relative with CRC who underwent ≥ 2 colonoscopies (n = 876 individuals; n = 2384 colonoscopies). Interval-censored data on timing and yield (no adenomas/nAA/AN) of each colonoscopy were available. nAA incidence and progression time from nAA to AN were estimated using a Bayesian progressive three-state model. Over a median follow-up of 6 years (interquartile range 5–6), 60 (6.8
Problem:Quality healthcare requires effective patient communication. However, lack of personnel and increasing demands on healthcare professionals (HCPs) create a need for innovative solutions that enhance accessibility and delivery of information to patients. Goal:We propose an innovative method to convey treatment and disease information using an Artificial Intelligence (AI)-driven social robotic physical interface. The aim of this study is to develop and test the feasibility of using a social robot that can convincingly provide health information in patient dialogues within clinical practice, to support patient communication and information exchange. Methods:This paper sets out the architectural approach of an AI-reinforced social robot connected to whitelisted validated clinical sources using a Generative Pre-training Transformer (GPT)-based Large Language Model (LLM). We describe experimental results in a lab-based pilot feasibility study, and then highlight related results for user experience in clinical practice implementation for an osteoarthritis (OA) use case, in which the robot answers osteoarthritis-related questions. Results were obtained after end-user engagement using the User Experience Questionnaire (UEQ) and semi-structured interviews. Results:UEQ results were obtained in a lab-based pilot test (n = 20) and with OA patients (n = 21) and healthcare professionals (n = 7). Above average/good attractiveness, perspicuity and stimulation were reported in the pilot test; novelty was excellent, yet dependability and efficiency were reported below average. In the clinical setting, Patient UEQ score resulted in mean 2.13 with values ranging from 1.7 to 2.5, indicating a positive trend in efficiency, inventiveness and acceptability. HCPs UEQ scores reached mean 1.89, with all values above 1 except for excitement of usage, which scored 0.8 (SD 1.3). Semi-structured interviews added in-depth enrichment of the data. Conclusion:In summary, this paper demonstrates the feasibility of implementing a GPT-reinforced social robot for patient communication in clinical practice.
Accurate prediction of neurological outcome after cardiac arrest is essential for guiding intensive care decisions. Electroencephalography (EEG) supports prognostication; however, interpretation relies on expert judgment and is often subjective and delayed. We developed DeepCRI, a bedside-integrated deep learning system that produces continuously updated prognostic trajectories during the first 36 h after arrest. DeepCRI uses time-dependent decision boundaries to define good-, poor-, and gray-zone regions over time, and applies a lock-in rule that fixes classification only after sustained, concordant high-confidence evidence within a compact temporal window, thereby preventing transient threshold crossings from driving decisions. DeepCRI was developed on a multicenter EEG dataset of 522 comatose patients after cardiac arrest and subsequently evaluated in independent internal (n = 219) and external validation cohorts (n = 167). In the internal validation cohort, DeepCRI provided lock-in classifications in 179/219 patients (81.7%), with a sensitivity of 94.7% (95% CI 90.0-97.6%) and specificity of 81.9% (95% CI 73.5-88.1%) for good outcome, and a sensitivity of 49.5% (95% CI 40.2-58.9%) and specificity of 100.0% (95% CI 96.7-100.0%) for poor outcome; 40/219 patients (18.3%) remained in the gray zone. In the external validation cohort, DeepCRI provided lock-in classifications in 100/167 patients (59.9%), with a sensitivity of 67.2% (95% CI 54.7-77.7%) and specificity of 82.1% (95% CI 73.7-88.2%) for good outcome, and a sensitivity of 36.8% (95% CI 28.2-46.3%) and specificity of 98.4% (95% CI 91.3-99.7%) for poor outcome. Post hoc analysis indicated residual EMG artifacts contributed to this false poor-outcome prediction. By embedding DeepCRI into routine ICU EEG infrastructure, we demonstrate the technical feasibility and clinical promise of continuous, real-time AI-driven prognostication for comatose patients after cardiac arrest.
