Costanza Vicentini

Background: Healthcare-associated infection (HAI) surveillance remains heavily dependent on manual review of electronic medical records (EMRs), limiting scalability, reproducibility, and between-hospital comparability. Large language models (LLMs) may support extraction of clinically relevant information from unstructured text, but implementation is constrained by data protection, governance, and workflow integration challenges. Within an ECDC-funded feasibility project on healthcare-associated bloodstream infection (HA-BSI) surveillance in Italian hospitals, we developed an implementation framework to address these barriers. Methods: The framework combines a Data Protection Impact Assessment used as a design instrument, categorization of data into raw, derived, and aggregated tiers, an on-site assisted de-identification workflow, and a tiered LLM deployment model. LLMs are used for structured clinical information extraction only, while final case classification is performed through a transparent, deterministic, rule-based algorithm based on ECDC definitions. Results: The project produced an operational workflow in which identifiable clinical records remain under hospital control and only locally reviewed de-identified outputs are transferred for downstream research processing. It also led to development of an open-source human-in-the-loop redaction prototype. Key implementation barriers included variable EMR maturity, document export constraints, and lengthy ethics, institutional, and data protection approval cycles. Final LLM benchmarking and surveillance-performance evaluation remain ongoing. Conclusions: Trustworthy LLM-assisted surveillance may depend on governance and implementation architecture as much as on model capability. The proposed framework provides a feasibility model requiring further validation across hospitals, languages, record systems, and LLM versions. This study describes a privacy-by-design workflow for using large language models to extract clinical information from Italian hospital records while preserving local data control, enabling deterministic ECDC case-definition matching for healthcare-associated bloodstream infection surveillance through on-site de-identification and human oversight.

Background: Infection prevention and control (IPC) in long-term care facilities (LTCFs) is essential to reduce the burden of healthcare-associated infections (HAI) and the spread of antimicrobial resistance (AMR). This paper describes the development of standardized IPC procedures through a stakeholder-informed approach, and integrates results from two regional surveys conducted in Piedmont (first survey 2021–2022; second survey 2024) which assessed the reported availability of written procedure documents in nursing homes and, subsequently, the uptake and perceived utility of the developed standardized procedures. Methods: Two anonymous cross-sectional surveys targeted all nursing homes registered in the Piedmont regional healthcare census. The first survey (December 2021-April 2022) assessed the reported inclusion of 16 IPC procedure domains in facility manuals and the date of last revision. Based on these findings, standardized IPC procedure documents were developed by the University of Turin in collaboration with DIRMEI, regional public health authorities and IPC stakeholders, and were made available online in 2023. A second survey (August 2024) assessed reported uptake, perceived adequacy, missing but useful procedures and overall satisfaction. Descriptive statistics were used. Results: 436/780 facilities responded to the first survey (55.9%). High availability of written procedure documents was reported for hand hygiene (99.1%), isolation for COVID-19 (99.8%) and cleaning/sanitation (98.9%). Lower availability was reported for sterilization (46.2%), bloodstream infection prevention (45.9%), and management of colonized/infected patients for alert organisms (62.0%). Revision dates varied, with many critical procedures updated predominantly in 2020–2022. After dissemination of standardized procedures, 305 facilities (39.1%) participated in the second survey: 99.7% found them adequate, 31.1% reported possessing all procedures already, and 89% provided an overall positive judgment. The most frequently requested/adopted procedures were vaccination prophylaxis (33.8%), laboratory sample management (31.8%), TB control (26.2%) and catheter management (22.0%). Conclusions: This regional initiative identified heterogeneity in IPC procedural documentation in nursing homes and documented high perceived adequacy and usefulness of standardized procedure documents after dissemination. These findings support harmonized regional resources as a structural component of IPC governance in LTCFs. However, the surveys did not assess implementation fidelity, staff adherence, audits, HAI rates, antimicrobial use or AMR outcomes; future work should incorporate objective implementation and outcome indicators.

Background: Antimicrobial resistance (AMR) represents a major public health threat. Italian AMR rates are among the highest in Europe. The Italian National action plan to combat AMR was launched in 2017 and updated in 2022. To monitor its implementation, the self-assessment tool SPiNCAR (Supporto al Piano nazionale per il contrasto all’antimicrobico resistenza) was developed. Methods: We conducted the first national data collection through SPiNCAR in 2023, assessing the level of implementation across Italian regions and autonomous provinces for the year 2022. Regional SPiNCAR scores were correlated with relevant indicators: healthcare-associated infection (HAI) and antimicrobial use prevalence, as well as antibiotic consumption. Results: Thirteen regions and autonomous provinces participated. High implementation levels were observed in the SPiNCAR areas of governance, surveillance, HAI prevention and control, and antimicrobial stewardship, while education, stakeholder engagement, and performance assessment showed the lowest scores. Moderate inverse correlations emerged between governance scores and HAI prevalence, and between surveillance/appropriate use scores and antimicrobial use prevalence. A strong inverse correlation was found between performance assessment scores and antimicrobial use prevalence. Conclusions: This study provides the first national baseline assessment of the level of implementation of the Italian National action plan to combat AMR. Results suggest SPiNCAR could be a useful tool to monitor AMR policies in decentralized health systems. Findings highlight both achievements and gaps; repeated assessments are necessary to guide targeted interventions and reduce regional inequalities.

This multicenter study focuses on the latest (2020-2024) trends in candidemia in 155 Italian ICUs, stressing its frequency and lethality, with a shift toward non-albicans, fluconazole-resistant species, namely C. parapsilosis. In this study, resistance to fluconazole prolonged ICU stays without increasing mortality, while non-catheter-related episodes had a more severe prognosis. Candidemia is the leading fungal bloodstream infection (BSI) in the ICU and carries high mortality. After the COVID-19 pandemic, non-albicans, fluconazole-resistant species increased in Europe, yet contemporary ICU-acquired data are scarce. The primary outcome of this study was to evaluate the incidence of ICU-acquired candidemia per 1,000 “at risk” ICU admissions (ICU length of stay [LOS] ≥ 48 h). Secondary analyses compared ICU-acquired candidemia with ICU-acquired bacterial BSI and assessed species distribution, fluconazole susceptibility, and outcomes by fluconazole resistance and by source of infection. Data were drawn from the PROSAFE network, a prospective multicentric observational project conducted across 192 Italian ICUs (2014–2023). Among 179,590 ICU admissions, 95,662 were at risk. ICU-acquired candidemia occurred in 373 patients (3.9‰ of at-risk admissions). Compared with ICU-acquired bacterial BSI, candidemia patients had longer intra-ICU and hospital LOS (32 vs 24 days and 44 vs 36 days, respectively; p < 0.001) and higher mortality (intra-ICU 41.3% vs 28.7%; intra-hospital 50.3% vs 34.1%; p < 0.001). Candida albicans was isolated in 47.2% of episodes, followed by Candida parapsilosis at 33.0%; the latter is often fluconazole-resistant (53.7%). Mortality did not increase in fluconazole-resistant episodes, but ICU LOS did (36 vs 31 days; p = 0.016). Non-catheter-related infections (197/373, 52.8%) had higher intra-hospital mortality rates (57.0% vs 42.9%; p = 0.007). ICU-acquired candidemia in Italy remains frequent and lethal, with a worrisome shift in epidemiology and resistance; the latter prolongs ICU stay without increasing mortality. Non-catheter-related episodes had a poorer prognosis. Findings support enhanced surveillance, catheter care, and earlier antifungal therapy. The online version contains supplementary material available at 10.1186/s13054-026-05886-1.

Objectives: To estimate the incidence of healthcare-associated infections (HAIs) in Italian long-term care facilities (LTCFs) and to evaluate whether an artificial intelligence (AI) approach, through unsupervised machine learning (ML), could stratify residents into clinically distinct groups with differing susceptibility to HAIs. Design: Prospective cohort study with 12-month follow-up. Setting: 24 LTCFs in Italy, participating in the European Centre for Disease Prevention and Control 12-month longitudinal study on HAIs in LTCFs, 2022–2023. Participants: 395 residents enrolled across the participating LTCFs. Methods: Incidence measures of HAIs (rate and ratio) were estimated, using generalized estimating equations. A hierarchical cluster analysis based on residents’ clinical and demographic characteristics was implemented as an unsupervised ML approach. Results: Overall, 75 HAIs per 100 residents (95% CI, 70.3–78.3) and 0.23 HAIs per 1,000 resident-days (95% CI, 0.11–0.76) were estimated. Respiratory tract infections (29.5%, 95% CI 24.2–31.1), COVID-19 (26.3%, 95% CI 22.1–28.4), and urinary tract infections (15%, 95% CI 11.0–35.4) were the most frequent. Clustering identified two reproducible resident groups: Group 1 (39%), more independent and cognitively preserved, with fewer comorbidities and lower infection incidence; and Group 2 (61%), more dependent and clinically complex, with higher incidence of HAIs. Cluster stability was high (mean ARI = 0.83). Conclusions: This study confirms the high burden of HAIs in Italian LTCFs and provides exploratory evidence that AI-based clustering can identify reproducible HAI susceptibility profiles in a setting where such approaches have been scarcely applied.

Background: Accurate burden estimates are necessary to inform priority setting and rational resource allocation. Weighting prevalence inversely proportional to time-at-risk has been proposed as a solution for length-biased sampling, an important limitation affecting prevalence to incidence conversion for healthcare-associated infections (HAIs). Aim: This study aimed to update Italian burden estimates by calculating HAI incidence, attributable mortality and disability-adjusted life years (DALYs). Further, we describe an adapted methodology for burden estimations. Methods: We used data from the latest European Centre for Disease Prevention and Control (ECDC) point prevalence survey (PPS) of HAIs, conducted in Italy in November 2022, to calculate the burden of five major HAIs at national level. We adapted the Burden of Communicable Diseases in Europe (BCoDE) methodology to include inverse probability weighting and compared results of naïve and weighted calculations. Results: The national sample included 18,397 patients. Overall, 564.8 DALYs per 100,000 general population resulted from weighted calculations (95% uncertainty interval (UI): 450.04–694.38), with an annual incidence of 685.42 cases per 100,000 general population (95% UI: 611.09–760.86) and 33.23 deaths per 100,000 general population per year (95% UI: 28.62–38.33). Concerning naïve estimates, overall 1,017.81 DALYs per 100,000 general population were calculated (95% UI: 855.16–1,190.59). In both calculations, healthcare-acquired bloodstream infections had the highest burden in terms of DALYs per 100,000 hospitalised and general population. Conclusion: Our study confirmed the substantial burden of HAIs in Italy and renews the need to prioritise resources for infection prevention and control interventions.