Andrija Štajduhar

The uptake of new digital and AI based technologies in the healthcare setting, including digital diagnostics, has been comparatively slower than in other sectors. This is a consequence of the complex legislature, and the lack of routine procedures for the secondary use of sensitive medical data. Inadequate level of education and incentives for different stakeholders prevents overcoming strong barriers, such as clinical workflow integration, and need for comprehensive evaluation of utility of interventions which requires multi-disciplinary collaboration effort. The paper presents a re-hospitalization challenge, organised by the European digital innovation hub (EDIH) - AI4Health.Cro, a co-creation activity and an effective tool for overcoming the mentioned barriers for implementation of new AI technologies and overall innovation ecosystem maturation. Crowdsourcing challenges or datathons are R&D strategies where ad-hoc assembled competing teams collaborate on early-stage technology development or research problem solving. Our re-hospitalisation challenge has had multiple facets: as activity for construction of an AI predictive model with explanatory support and plausible business model for its deployment; it involved also extensive organisers’ effort: data anonymisation, provision of secure processing environment, and comprehensive assessment of solutions. The paper details the challenge structure and processes, from problem inception to execution and evaluation. As a co-creation effort in digital healthcare, the challenge provided results on multiple levels: (i) tailored solutions based on primary data, serving as prototypes for later deployment in practice; (ii) sandboxing EHDS principles to enable secondary data use; (iii) ecosystem building event yielding cross-disciplinary participative learning experience. Innovation challenges aren’t just competitions; they are a building tool that provides several direct and indirect impacts, from developing healthcare technologies, supporting startups, to streamlining secondary use of clinical data for research and innovation and eliciting regulatory needs. Raising the awareness of all relevant stakeholders about the translation of AI technologies into routine clinical setting. Author summary: Healthcare has been slower than other sectors to adopt new digital and AI-based technologies. This slow progress largely reflects the complexity of EU legislation and the lack of practical procedures that translate legal requirements into guidance for stakeholders using sensitive medical data in research and innovation. A crowdsourcing pilot to predict likelihood of early rehospitalisation using AI methodology provided opportunity for hands-on experience for experts from clinical, computational and business background on both synthetic and real-world medical data. Together we clarify legal responsibilities and data access conditions, and facilitate collaboration of relevant stakeholders, clinical, technical, public and research entities for ecosystem building. We provide the first Croatian European Health Data Space compliant secure processing environment, created for future use by researchers and innovators. Methodologically varied AI based prototypes were developed to predict early rehospitalization carefully balancing data overfitting, AI explainability and market potential of technology, in turn raising multidisciplinary skills, and awareness of conditions required for innovation to thrive in 28 participating teams. This type of competition, provided within European Digital Innovation Hub is an ideal platform to build regulatory sandboxes and pilot novel legislative platforms ahead of their full-scale implementation.

Perinatal hypoxia is a major contributor to neurodevelopmental disorders; however, the consequences of mild-to-moderate perinatal hypoxia (MPH) remain insufficiently characterized. Here, we investigated cortical plasticity following MPH using a multimodal approach that combines behavioral assessment, histological analysis, and in vivo magnetic resonance imaging (MRI). Fifty-six Wistar Han rats were exposed to hypoxia or normoxia at postnatal day 1 (P1). Neurodevelopmental assessment from P3 to P14 revealed impaired rooting and vibrissae-placing reflexes in hypoxic rats. Histological analysis demonstrated: altered expression of microtubule-associated protein-2, apical dendrite bundling, reduced neurofilament-H expression, and decreased dendritic arbor complexity in large pyramidal neurons, indicating disrupted maturation of excitatory circuits. Increased parvalbumin expression, higher interneuron density, and its enhanced neurite elaboration indicated precocious development of inhibitory circuits, consistent with a compensatory response. MRI at P15, combined with whole-brain voxel-wise analysis, revealed a significant increase in fractional anisotropy in the anterior cingulate cortex (ACC). Convergent behavioral, histological, and imaging findings identified the ACC as the most vulnerable region following MPH, followed by the somatosensory cortex. These findings reveal early cytoarchitectural and MRI detectable correlates of a single episode of MPH, which, together with previous findings from this model, support the neurodevelopmental origin of persistent alterations in cortical structure and circuit function, characterized by an excitatory-inhibitory imbalance. The study identifies and defines a framework for understanding region-specific vulnerability and plasticity in the immature brain, with implications for improving the early detection of subtle perinatal brain injury, as a prerequisite for timely therapeutic intervention.

Intermittent claudication (IC) is the most frequent symptomatic manifestation of lower-extremity peripheral artery disease (PAD). Supervised exercise therapy (SET) and endovascular revascularization (ER) are established treatments, but their relative and combined effects on health-related quality of life (HRQoL) remain. We conducted a systematic review and meta-analysis of randomized controlled trials (RCTs) comparing SET, ER, and ER+SET, with HRQoL as the primary outcome. Methods: Following PRISMA 2020, PubMed, Embase, and CENTRAL were used in December 2024. Eligible RCTs enrolled with IC (excluding critical limb-threatening ischemia) and reported validated HRQoL outcomes at ≥3 months. Two reviewers independently extracted data and assessed risk of bias using the Cochrane RoB 2.0 tool. Random-effects meta-analyses pooled standardized mean differences (SMDs) for HRQoL and mean differences (MDs) for walking distance. Results: Five RCTs (n = 728) were included. Compared with optimal medical therapy, both SET and ER improved HRQoL and walking distance. At 12 months, no significant effect was observed between SET and ER (SMD 0.02; 95% CI: −0.18 to 0.22). ER+SET was superior to SET alone (SMD 0.35; 95% CI: 0.12–0.57). Beyond 24 months, improvements were sustained with SET but attenuated with ER, accompanied by higher reintervention rates in ER-containing arms (approximately 20–30% by 2 years). Adverse events were rare (<1%). Conclusions: Given moderate-certainty evidence (GRADE), SET should remain the first-line therapy for intermittent claudication because it provides durable improvements in patient-centered outcomes with minimal harm. Endovascular revascularization (ER) can provide faster symptom relief, but its long-term benefits are constrained by restenosis and repeat procedures, particularly in femoropopliteal disease.

Cilj. Ispitati mogućnosti implementacije umjetne inteligencije (UI) u poslovanje visokoškolske biomedicinske knjižnice te analizirati korisničke potrebe vezane za upotrebu virtualnog asistenta (VA) u svrhu unaprjeđenja dostupnosti i kvalitete knjižničnih usluga. Rezultati. Većina studenata (92,8 %) prepoznaje potencijal UI-a za poboljšanje knjižničnih usluga. Znatni broj ispitanika (86,3 %) izrazio je interes za uporabu alata na studiju i istraživanjima, osobito kao pomoć pri pisanju radova, izvještaja, sažimanju i analizi literature. Iako su studenti uglavnom pozitivni prema UI-u, postoje zabrinutosti zbog etičkih pitanja, poput privatnosti podataka i tehničkih problema. Analiza upita VA-a otkrila je specifične korisničke potrebe koje usmjeravaju razvoj prema većoj vidljivosti Knjižnice. Ograničenja. Istraživanje ima prigodan uzorak, što ograničava mogućnost poopćavanja nalaza. Buduća istraživanja trebaju obuhvatiti veći i raznolikiji uzorak te produžiti testno razdoblje, uz rješavanje tehničkih i etičkih izazova. Praktična primjena. Rad predstavlja novi iskorak u Hrvatskoj jer je prvi put u biomedicinskoj knjižnici uvedena usluga VA-a i opisan je njegov rad u tom kontekstu. Testirana je učinkovitost i korisnost VA-s tijekom određenog razdoblja, što je dalo značenje povratne informacije za daljnje poboljšanje. Originalnost/vrijednost. Istraživanje doprinosi razumijevanju utjecaja alata UI-a na knjižnice, istražujući specifične korisničke potrebe i percepcije u kontekstu VA-a. Rezultati mogu poslužiti kao osnova za daljnja istraživanja i razvoj strategija za implementaciju UI-a te potaknuti širu primjenu u knjižničnim uslugama.

Objectives: To evaluate the quality of responses generated by GPT-4 in comparison to those written by hospital specialists and general practice physicians across multiple medical specialties. The goal was to assess whether large language models (LLMs) can support patient communication by providing accurate, useful, complete, and empathetic responses to real-world patient inquiries. Methods: We collected 100 anonymized patient questions from a public online health forum covering five specialties: cardiology, infectious diseases, neurology, gynecology, and gastroenterology. Responses were generated by GPT-4, 50 hospital-based specialists, and 50 general practice physicians. A group of 50 specialists, blinded to response source, evaluated each answer using four 7-point Likert scales: accuracy, usefulness, completeness, and empathy. The study design and reporting were informed by the CONSORT-AI Extension to promote transparency in AI evaluation. Results: The model received significantly higher ratings across all four categories compared to both physician groups. It was ranked best in 67% of evaluations, particularly outperforming physicians in completeness and empathy. While response length correlated positively with quality for physicians, model's longer responses were less useful when overly detailed. The model consistently produced longer, more comprehensive replies than human groups. Discussion: The model's strong performance in completeness and empathy highlights its potential role in enhancing patient communication. Although it matched or exceeded physicians in accuracy and usefulness, caution is warranted due to risks like hallucinations and lack of true understanding. Conclusion: This blinded evaluation suggests that AI-generated responses may support clinical practice by delivering accurate, comprehensive, and empathetic information for patient communication.

Hashimoto’s thyroiditis (HT) is an autoimmune thyroid disease with characteristic lymphocytic infiltration and fibrosis. Chronic autoimmune changes that occur in the thyroid gland in HT may also affect the lacrimal gland. Objectives: This study aimed to analyze tear biomarkers and explore correlations between these biomarkers and clinical ocular parameters in patients with HT. Methods: A total of 150 participants were divided into three groups: HT (N = 50), non-HT DED (N = 50), and healthy controls (N = 50). The participants underwent a series of diagnostic tests for DED, including the Ocular Surface Disease Index, Tear Break-Up Time, Lid-Parallel Conjunctival Folds, Schirmer test without anesthetic, lissamine green and fluorescein staining. Tear samples were analyzed for cytokine and enzyme levels (interleukin 1β, tumor necrosis factor α, interleukin 6 (IL-6), interleukin 8, interleukin 10 (IL-10), interleukin 17A, matrix metalloproteinase 9 (MMP-9)) using ELISA and multiplex immunoassay. Statistical analyses were conducted to compare groups and assess biomarker correlations. Results: Dry eye disease was observed in more than half of the study group (27/50), with severe symptoms observed in 48.15% of the DED HT subgroup. IL-6 levels were significantly elevated in the DED HT subgroup compared to the non-HT DED group (p = 0.010), suggesting specificity for HT-associated DED. MMP-9 was elevated in both the HT and non-HT DED groups (p < 0.001) but lacked specificity for HT (p = 0.059). The DED HT subgroup exhibited significantly lower IL-10 levels (p = 0.008). Lissamine green staining and LIPCOF were significantly higher in the DED HT subgroup (p < 0.001). Conclusions: Dry eye disease is common in euthyroid HT patients without signs of TAO. This study highlights the potential role of IL-6. Lissamine green staining and LIPCOF are valuable diagnostic tools for assessing the ocular surface in DED HT patients.

Introduction The aim of this study was to report the long-term follow-up results of cryoballoon (CB) ablation in patients with atrial fibrillation. Methods All consecutive patients who underwent second-generation CB ablation from February 2015 to December 2017 were included in our study. In all procedures, we used a 28-mm CB placed via a single transseptal puncture guided by intracardiac ultrasound. A 20-mm octapolar intraluminal circular catheter was used for intracardiac recordings. A single 180-s freeze strategy was employed. Repeated procedures were performed with a 3D mapping system and radiofrequency catheters. Results A total of 126 patients (69.8% male, mean age 57 ± 11 years), of which 77.0% had paroxysmal atrial fibrillation (PAF), were included in the study. After a 5-year period, 52.4% of patients were in sinus rhythm without AF recurrence, off antiarrhythmic drugs. A total of 61.9% of patients were free of AF recurrence when redo PVI procedures were performed. When accounting for redo pulmonary vein isolation and antiarrhythmic drugs, a total of 73.8% of the patients were without AF recurrence in long-term follow-up. The patients who underwent redo pulmonary vein isolation procedures had statistically significant lower rates of AF recurrence (p = 0.006). In patients with PAF, long-term success rates improved from 62.9 to 79.4% for patients who underwent the redo procedure (p = 0.020). In patients with persistent atrial fibrillation (PersAF), success rates went up from 41.4 to 55.1% for patients with single or repeated PVI procedure (p = 0.071). In the whole cohort, a total of 3 (2.4%) procedure-related major complications occurred which included persistent PNP, arterial pseudoaneurysm, and arteriovenous fistula. Conclusion Our data suggest a favorable long-term safety and efficacy profile of second-generation CB ablation. In the mixed paroxysmal and persistent population, up to 73.8% of patients remained free of AF recurrence in the 5-year follow-up, when accounting for redo procedures and AADs. Only 2.4% of patients experienced major complications of the ablation procedure, none with permanent sequelae. Graphical Abstract Central figure. K-M curves for overall population