Alfredo Cesario

AI-enabled digital medicine platforms in Europe are developed within overlapping rules for personal data, medical devices, AI, clinical research, health data access, and health technology assessment. This viewpoint represents the authors’ experience and interpretation from the preimplementation phase of INTERCEPT, an Innovative Health Initiative consortium that plans to validate biomarkers and develop a blood-based risk score for early Crohn disease interception. It does not report original empirical research, a systematic regulatory review, a deployed platform, or a validated AI model. Our aim is to explain how compliance-by-design can be translated into an auditable governance architecture and to identify what still requires scientific evaluation. We advance 3 messages. First, governance must begin with intended purpose and boundary mapping: a digital platform may contain research, operational, and medical device software functions, and conformity of one module cannot be inferred for the platform as a whole. Second, overlapping obligations can be operationalized through 5 linked controls: modular responsibility and change control, federated data governance, traceable AI life cycle management, consent and participant rights orchestration, and a common cross-border governance core with local adaptations. Third, compliance is not a surrogate for model validity, clinical benefit, trust, scalability, or market access. Before clinical deployment, planned AI components require prespecified labels and cohorts, leakage prevention, missing data and imbalance strategies, calibration, subgroup assessment, external validation, human factors evaluation, and postdeployment algorithmovigilance. A hypothetical biomarker risk score update illustrates how the 5 controls interact. INTERCEPT is informative because it combines cross-border biomarker research, planned predictive analytics, and heterogeneous institutional settings while key design decisions remain modifiable. Its contribution is nevertheless conceptual: the proposed framework is a testable governance hypothesis, not evidence of implementation success. Prospective studies should compare it with alternative governance models and assess safety, workload, participant experience, evidence generation, and regulatory performance.

Background: Alzheimer's disease (AD) diagnosis has shifted from a purely clinical framework to a clinical-biological paradigm, driven by biomarker integration. This evolution is motivated by the wider availability of reliable biomarkers and the advent of disease-modifying treatments. Objective: To assess changes over time in clinical characteristics, diagnostic pathways, and healthcare resource utilization in a real-world cohort of individuals with cognitive impairment attending a Memory Clinic. Methods: This secondary data retrospective observational study analyzed two patient cohorts with newly diagnosed cognitive impairment: one from 2017–2019 and another from 2021–2023. Anonymized medical records and structured hospital data were examined using natural language processing to extract demographic and clinical information, diagnostic pathways, treatment patterns and comorbidities. Results: The 2021–2023 cohort was significantly younger, exhibited higher baseline Mini-Mental State Examination scores, and underwent more instrumental assessments than the 2017–2019 cohort. These findings likely reflect a shift in public awareness and attitudes toward cognitive health. AD diagnoses increased in both cohorts over time, while mild cognitive impairment diagnoses declined. The use of diagnostic combinations was more frequent in the recent cohort, in which clinical-biological diagnoses were significantly more prevalent. Conclusions: This study provides real-world insights into the evolving landscape of cognitive impairment diagnostics and care, underscoring a shift toward earlier, biologically grounded diagnosis, supporting precision medicine in AD care. The expanded use of biomarkers reflects evolving practice standards and prepares the ground for disease-modifying therapies in AD.

Aims: To assess for the first time the impact of socio-demographic variables on prescription of guideline-directed medical therapy (GDMT) after an episode of heart failure (HF) decompensation in the Italian healthcare system. Methods and results: Utilizing ‘GENERATOR-HF DataMart’, a cross-sectional analysis was performed. We included patients with HF and reduced ejection fraction discharged between January 2019 and July 2024. The degree of GDMT implementation across the different socio-demographic variables (i.e. patient's age, sex, marital status, nationality, place of residence, and educational level) was evaluated through the modified optimal medical therapy (mOMT) score (i.e. a ratio between the number of pillars actually prescribed and the number of pillars that could be prescribed on the basis of each specific contraindication). A multivariable logistic regression model was also fitted to assess the association between the socio-demographic variables and the prescription of each pillar and loop diuretics. 1730 patients (median age: 72 years; 24% females) were included. The mOMT score was significantly lower in elderly patients, but comparable across other pre-specified socio-demographic categories. In multivariable regression analysis, older age was the only independent socio-demographic predictor of under-prescription both overall and for ACEi/ARB/ARNI (OR0.70; 95% CI 0.55–0.89), beta-blockers (OR0.59; 95% CI 0.41–0.84) and SGLT2i (OR0.66, 95% CI 0.47–0.93), while also associated with a loop diuretics use (OR1.56; 95% CI 1.13–2.17). A higher mOMT score was significantly associated with a reduced incidence of early adverse events (i.e. 30-day all-cause death and urgent re-admissions) (4.1% vs. 8.5%; P = 0.001). Conclusion: Older age was the only independent predictor of under-prescription of GDMT and enhanced use of loop diuretics, whereas no discrepancies were found across the other socio-demographic subgroups.

Background: Systemic lupus erythematosus (SLE) is a chronic disease characterized by a broad spectrum of involved organs, including neurological, renal, and vascular domains, with disease activity manifesting through unpredictable patterns that vary across individuals and over time, making the prediction of activity events particularly challenging. Objective: This paper proposes a hierarchical machine learning model to predict a 12-month SLE activity, defined as the occurrence of at least one event among SLE hospitalization, new organ-involved domain, and neurological, renal, or vascular manifestation within the following year. At each patient’s visit, the model considers all the features at the current time point, the information about the patient’s clinical history, and about its last 12 months, to predict the outcome for the next 12 months. Methods: The study cohort consists of 262 patients with at least an outpatient visit and an SLE admission from 2012 to 2020, at the Italian Gemelli Hospital, comprising a retrospective longitudinal dataset of 5962 contacts. The data include demographics, laboratory, clinical features (eg, domain involvements and manifestations), treatments, and pathways (eg, contact types as outpatients, hospitalizations, day hospitals, and visit frequency). The variables consider 3 time ranges: features about the current contact and the last 12 months, and the previous patient’s clinical history. The main model was developed by testing different machine learning approaches within a cross-validation setup. The predicted probability outputs were used in a risk stratification analysis, identifying 3 groups of predictions: strong, moderate, and mild. Mild samples were then passed through a second cascade model. The integration of the main model (applied to strong and moderate samples) with the cascade model (applied to mild contacts) forms our final hierarchical model. Results: The hierarchical model, resulting from the ensemble of the main random forest and cascade decision tree, demonstrated enhanced performance, increasing the area under the receiver operating characteristic curve from 0.696 (95% CI 0.672‐0.719) in the original main model to 0.743 (95% CI 0.717-0.769), particularly for specific patient characteristics. Through the application of explainable artificial intelligence methods, we also identified the key features that significantly influence the model’s predictions. Among the 185 collected features, 15 emerged as the most impactful, including age at contact, response to therapy modifications, abnormal laboratory tests, and clinical manifestations. This analysis plays a crucial role in enhancing model transparency, which is essential for fostering the adoption of artificial intelligence in health care settings. Conclusions: Our study introduces an explainable and reliable tool for predicting 1-year SLE activity, supporting physicians with an advanced decision-support system to improve patient management. The model identifies key features that may help characterize patient phenotypes, enabling personalized treatment plans and better outcomes. In addition, the methodology can be generalized for predictive analytics in other chronic autoimmune diseases.

The diverse data sources in hospitals are crucial for real-world evidence (RWE) analysis and for developing decision support systems. However, integrating this variety of sources into standardized data collections for a specific disease, requires the development of highly personalized frameworks. In this study, we propose a RWE approach for the development of a Data Mart for the Systemic Lupus Erythematosus (SLE) at the Gemelli hospital of Rome. Our approach combines natural language processing and data mining procedures to capture information about organ involvements, activity episodes and treatments, under rules defined by physicians. The Gemelli SLE Data Mart includes 262 SLE patients with at least a hospitalization from 2012 to 2020, and an outpatient visit, with a total amount of 5962 contacts. We also developed a visualization tool to display each patient longitudinally and we performed a data-driven analysis to stratify patients into progression groups, based on their involved organs and flares. The Gemelli SLE Data Mart and its framework were useful to build up a standardized data collection, to be used for further personalized medicine applications.

Aims: Despite advancements in disease-modifying therapies, the rate of hospitalizations in patients with heart failure (HF) remains high, with an increased risk of future adverse events and healthcare costs. In this context, the AZIMUTH study aims to evaluate the large-scale applicability of a smartphone app-based model of care to improve the quality of care and clinical outcomes of HF patients. Methods and results: The AZIMUTH trial is a multicentre, prospective, pragmatic, interventional, single-cohort study enrolling HF patients. Three hundred patients will be recruited from four different sites. For comparative analyses, historical data from participating hospitals for the 6 months before enrolment and propensity-matching score analyses from GENERATOR HF DataMart, will be used. The estimated duration of the study is 6 months. During the whole observational period, the patients are asked to provide information regarding their clinical status, transmit remote clinical parameters, and periodically answer validated questionnaires, the Kansas City Cardiomyopathy Questionnaire Health and Morisky Medication Adherence Scale 8-item, on a mobile application, through which healthcare providers implement therapeutic adjustments and remote clinical assessments. The primary objective of this study is to evaluate the feasibility, usability, and perceived benefits for key stakeholders (patients and clinical staff) of the AZIMUTH digital platform in the enrolled patients when compared to standard of care. Secondary endpoints will be the description of the rate of hospital readmissions, ambulatory visits and prescribed therapy in the 6 months following enrolment in the experimental group compared to both the historical and propensity-matched cohorts. Conclusion: The AZIMUTH aims to enhance HF management by leveraging digital technologies to support the care process and enhance monitoring, engagement, and personalized treatment for HF patients.

Electronic Health Records (EHRs) contain a wealth of unstructured patient data, making it challenging for physicians to do informed decisions. In this paper, we introduce a Natural Language Processing (NLP) approach for the extraction of therapies, diagnosis, and symptoms from ambulatory EHRs of patients with chronic Lupus disease. We aim to demonstrate the effort of a comprehensive pipeline where a rule-based system is combined with text segmentation, transformer-based topic analysis and clinical ontology, in order to enhance text preprocessing and automate rules’ identification. Our approach is applied on a sub-cohort of 56 patients, with a total of 750 EHRs written in Italian language, achieving an Accuracy and an F-score over 97% and 90% respectively, in the three extracted domains. This work has the potential to be integrated with EHR systems to automate information extraction, minimizing the human intervention, and providing personalized digital solutions in the chronic Lupus disease domain.