Riviste UNIMI
Not a member yet
21278 research outputs found
Sort by
Time Windows Used when Identifying Current Drug Use and Polypharmacy
No full textBackground
There is no consensus on the definition of the prevalence of drug use, including polypharmacy, regarding the length of the time window and the number of required concomitant medications.
Objectives
We aim to explore how the estimated prevalence of drug use in general, and of polypharmacy in particular, is affected by the applied definition.
Methods
We conducted a drug-utilization study divided into two parts: in the first part, we focused on estimation of current use, corresponding to ‘baseline drug use’ in a cohort study. Using population-based registries from Denmark, we identified a cohort of individuals aged ≥18 years during 2020-2022, assigned them a random index date and considered ‘current use’ of the following drugs: statins, glucose-lowering drugs (GLDs), selective serotonin reuptake inhibitors (SSRIs), opioids, and non-steroidal anti-inflammatory drugs (NSAIDs). The second part of the study focused on polypharmacy, defined according to five different definitions, with estimations of its prevalence using population-based registries from Denmark. We identified a cohort of individuals older than 65 years in 2022, and we considered all drugs available in the registries except for anti-infectives for systemic use. We also evaluated the accuracy of different criteria for predicting polypharmacy using simulations.
Results
Evaluating baseline drug use, we observed that the proportion of individuals classified as exposed increased with use of time-windows up to the first 90 days before the index date, reaching a plateau using windows around 120-150 days for statins, GLDs, and SSRIs, and around 180-300 days for opioids, whereas it was not reached for NSAIDs within 360 days. The prevalence of polypharmacy ranged from 21.1% (10 different 4th level Anatomical Therapeutic Chemical (ATC) groups in one year) to 92.3% (two different 4th level ATC groups in one year) depending on the applied definition, varying with the number of different ATC groups and time periods. In the simulation, the best criterion for identifying polypharmacy required at least two dispensations for each of at least five drugs, with sensitivity ranging between 0.93 and 1.00, and specificity between 0.72 and 1.00.
Conclusions
Time windows up to 90 days are too short to identify baseline drug use in the Danish setting. How polypharmacy is defined significantly influences its estimate, suggesting a need to use multiple definitions in each study
Adherence to Diabetes Care Pathways and Risk of Diabetes-Related Complications by Citizenship: A Population-Based Study from The Mighty Project (Cup P2022ASXKR)
No full textIntroduction
National and international guidelines on diabetes recommend continuous monitoring of this condition to prevent short- and long-term complications [1–3]. Adherence to the Diagnostic Therapeutic Care Pathway (DTCP) is evaluated by the Italian Ministry of Health using annual indicators based on five key recommendations [4], to ensure uniformity of care across all regions. However, equitable access to these services remains a public health challenge, particularly for migrant populations, who may face cultural, linguistic, or systemic barriers.
Objective
Within the MIGrants’ HealTh and healthcare access in ItalY (MIGHTY) project, this population-based study aimed to compare adherence to DTCP in subjects with new diagnosis of diabetes between migrant and Italian populations, and to evaluate the association between citizenship and diabetes complications in the Marche Region, between 2013–2023.
Methods
A population-based cohort study was conducted using Healthcare Utilization Databases from the Marche Region, including Regional Beneficiary, Hospital Discharge, Drug Prescription, Outpatient Care, and Exemption databases.
The cohort of new cases of diabetes included adults (≥18 years) with a first diabetes-related event between 2013 and 2017, defined as: ≥2 prescriptions of glucose-lowering drugs (ATC: A10*) within one year, ≥1 hospitalization with diabetes as primary/secondary diagnosis (ICD-9-CM: 250.*), or ≥1 diabetes exemption (code: 013*); the date of the first event was defined as the index date. Exclusion criteria were being resident in Marche Region for less than two years prior to index date, any diabetes-related events in the two preceding years, and childbirth-related hospitalizations (Major Diagnostic Category 14) during the year of inclusion in the cohort or the two prior years.
Subjects were classified as Italian or migrants from High Migratory Pressure Countries (HMPC) based on citizenship [5].
The five recommendations monitored by the Ministry of Health [4] were assessed: ≥2 HbA1c tests (PDTA-05.1), ≥1 lipid profile (PDTA-05.2), ≥1 microalbuminuria test (PDTA-05.3), ≥1 renal function test (PDTA-05.4), and ≥1 eye exam (PDTA-05.5) each per year. The adherence to each recommendation and the overall adherence, i.e. meeting at least 4 out of 5 recommendations (PDTA-05), was annually evaluated over six years from the index date for each subject.
Mixed-effects logistic models for repeated measures were used to evaluate adherence to recommendations considering as independent variables citizenship (Italian vs. HMPC), year of evaluation, sex, age groups (65-74 vs. 18-44, 45-54, 55-64, 75+ years), and Multisource Comorbidity Score (MCS) [6] classes (≤4 vs. 5-9, 10-14, 15-19, ≥20).
Diabetes-related complications were defined as a hospitalization with a primary diagnosis of short- or long-term diabetes complications, or uncontrolled diabetes, or non-traumatic lower limb amputation [7]. Subjects with at least one year of follow-up, were followed up from index date to a diabetes-related complication, all-cause death, emigration out of the region, or December 31, 2023, whichever came first. Cox proportional hazards model was used to estimate the association between citizenship and risk of complications, adjusting for age groups, sex, MCS classes, and including annual adherence to each DTCP’s recommendation as time-dependent covariates. Results are reported with 95% Confidence Interval (95% CI). All data were processed in compliance with the European (GDPR, EU 2016/679) and national privacy laws (D.lgs. 196/2003 and subsequent amendments).
Results
The study cohort comprised 28,674 adults with newly diagnosed diabetes, of whom 1,529 (5.3%) were migrants from HMPC. At index date, migrants from HMPC were younger (mean age 50 vs. 66 years), more frequently female (59% vs. 48%), and had lower comorbidity scores (MCS median: 2 vs. 6) compared to Italians.
Migrants from HMPC compared to Italians showed lower observed adherence to all DTCP recommendations: HbA1c (26.1%, 95%CI 25.1-27.1 vs. 36.0%, 95%CI 35.7-36.2), lipids (43.4%, 95%CI 42.2-44.5 vs. 61.3%, 95%CI 61.1-61.6), microalbuminuria (27.0%, 95%CI 25.9-28.0 vs. 31.3%, 95%CI 31.0-31.5), renal function (46.8%, 95%CI 45.7-47.9 vs. 66.2%, 95%CI 66.0-66.5), eye exam (6.7%, 95%CI 6.1-7.3 vs. 8.4%, 95%CI 8.3-8.6), and overall adherence (15.6%, 95%CI 14.7-16.4 vs. 20.3%, 95%CI 20.1-20.6). Mixed-effect models confirmed lower adherence among migrants, except for eye exams (Figure 1).
Of 24,992 with ≥1 year follow-up, 3,229 experienced complications (3,161 Italians, 68 migrants). Six-year complication-free survival was 85.7% (Italians) vs. 94.1% (migrants). In the adjusted Cox regression model, citizenship was not significantly associated with the risk of developing diabetes complications (HR=0.98; 95% CI: 0.77–1.26; p=0.896).
Conclusions
This study highlights differences in adherence to recommended diabetes care pathways between Italians and migrant populations. However, these differences did not translate into a higher risk of developing diabetes-related complications. Longer period of observation might be required to evaluate the impact of citizenship on adverse diabetes- related outcomes. Sustained monitoring and culturally tailored interventions remain essential to ensure equitable access and prevent future health inequalities.
 
Climate-Driven Patterns of West Nile Virus in Lombardy: A Spatio-Temporal Analysis (2013-2022)
No full textIntroduction
Climate change is one of the most pressing global health threats of the 21st century, driving shifts in infectious disease patterns and facilitating the spread of mosquito-borne viruses like West Nile Virus (WNV). Lombardy, in northern Italy, is a high-risk area for WNV circulation and now represents an endemic zone in the Po River Valley. Despite the well-established impact of temperature and precipitation on mosquito population dynamics and WNV transmission, current surveillance does not incorporate climatic parameters to monitor WNV epidemiology.
Objective
This study aims to describe the epidemiology of WNV in Lombardy Po Valley from 2013 to 2022 and to evaluate the association between climatic factors – specifically temperature and precipitations – and the occurrence and distribution of WNV cases, in order to determine whether climate data can support more effective surveillance and warning systems.
Methods
A retrospective observational study using human WNV case data collected through the national surveillance system in Lombardy from 2013 to 2022 was conducted. Weekly provincial case counts were matched with meteorological data from ARPA Lombardy, including weekly average temperature and total precipitation. In the descriptive phase, we created choropleth maps showing the geographical distribution of WNV cases across across the period in Lombardy provinces. Moreover we plotted weekly time series of case prevalence per 10000 inhabitants, temperature, and precipitation to assess their temporal co-occurrence and seasonal patterns. For inferential analysis, we fitted a hurdle model with two components: (1) a logistic regression modeling the probability of WNV case occurrence, and (2) a zero-truncated Poisson regression for the count of cases, conditional on occurrence. Models included a two-week lag for temperature and a one-week lag for precipitation and fixed effects for province and year, and offset terms for population size.
Results
Between 2013 and 2022, 311 WNV cases were recorded in six Lombardy provinces, with a prevalence in males (74%) and individuals over 65 years (58%). Among the 164 cases with clinical classification (available from 2019 onward), 53% were neuroinvasive (WNND), while 25% were identified through blood/tissue donation. The majority of cases occurred between July and October, peaking in August (52.4%). The most affected years were 2018, 2020 and 2022. Spatially, the highest prevalence was observed in southern provinces (Cremona, Mantua, Pavia, Lodi), forming a south-north gradient. Time-series plots highlighted a recurrent seasonal alignment between rising temperatures and WNV prevalence, while precipitation showed less consistent patterns. The hurdle model confirmed a significant role of temperature: a 1°C increase (lagged two weeks) raised the odds of case occurrence by 27% (OR=1.27) and the number of cases by 11% (IRR=1.11). Precipitation (lagged one week) had no effect on outbreak onset but increased case counts by 13% (IRR=1.13) once circulation began. Provincial and inter-annual differences were significant in predicting virus occurrence but not outbreak intensity.
Conclusions
Temperature is the most influential factor for both triggering and amplifying WNV transmission, while precipitation acts as a secondary amplifier. This study demonstrates the critical role of short-term climatic variables, particularly temperature, in shaping WNV dynamics in Lombardy. Rising temperatures significantly increase both the probability of outbreak initiation and the intensity of transmission. While precipitation alone does not appear to initiate outbreaks, it contributes to amplifying transmission once WNV is circulating. These findings reinforce the importance of integrating climate indicators into surveillance systems to enhance early warning capacity and timely public health responses. The identification of provincial-level risk differentials and temperature thresholds can inform geographically tailored interventions. In the context of ongoing climate change, such predictive models could become essential tools to mitigate the impact of future arboviral epidemics
AI-Based Tool for Early Diagnosis and Progression Prediction in Alzheimer’s Disease: A Multicenter Validation Study
No full textIntroductionAlzheimer’s disease (AD) is the most common cause of neurodegenerative dementia and poses a major healthcare challenge worldwide. Despite the availability of biological biomarkers, their application in routine clinical settings remains limited. Recent recommendations from eleven European scientific societies and Alzheimer Europe propose a patient-centered diagnostic workflow for memory clinics [1]. Within this context, artificial intelligence (AI) may offer valuable support for clinical staging and diagnosis based on widely available neuropsychological and MRI data [2-3].
ObjectivesThis study aimed to evaluate the clinical performance of TRACE4AD™, a CE-marked AI-based medical device, in supporting memory clinics during key diagnostic steps, specifically by assessing its ability to correctly stage cognitive decline, to classify clinical syndromes and formulate causal hypotheses (distinguishing AD from non-AD profiles), and to predict conversion to AD dementia within 24 months.
MethodsA total of 797 subjects were enrolled from 66 centers (Italy, US, Canada). All underwent 3D T1-weighted MRI and a detailed neuropsychological battery assessing multiple cognitive domains [4]. In 482 cases, CSF biomarkers (Aβ42, t-tau, p-tau) and/or [¹⁸F]FDG PET imaging were available [5]. TRACE4AD™ automatically analyzed imaging and cognitive data using an ensemble of Support Vector Machines (SVMs), with feature selection via Principal Component Analysis (PCA) and Fisher Discriminant Ratio (FDR) [6-7]. Clinical performance was assessed in terms of agreement with expert clinical staging (Cohen’s kappa), diagnostic accuracy against biomarker-based classification for syndrome identification, and predictive accuracy of conversion to AD dementia at 24 months using clinical follow-up as reference.
ResultsTRACE4AD™ showed substantial to almost perfect agreement with clinical staging (κ=0.81 for HS/SCI/WW, κ=0.70 for MCI/MD, κ=0.90 for moderate/severe dementia). In the subset of subjects with biomarker data (n=130), the tool correctly classified AD-related syndromes with 91% accuracy, achieving a positive predictive value of 91% and a negative predictive value of 100%. For prediction of conversion to AD-dementia at 24 months (n=341), TRACE4AD™ reached 89% sensitivity, 82% specificity, 85% overall accuracy, and an AUC of 83%. Furthermore, AI-derived brain volumetric features significantly correlated with CSF biomarkers, particularly in medial temporal regions, and cognitive performance, supporting the tool’s biological validity and interpretability.
ConclusionsTRACE4AD™ demonstrated high performance in staging, syndrome classification, and prediction of AD conversion, supporting its utility as a statistical and clinical decision-support tool. Its ability to integrate multimodal data in a reproducible, interpretable manner aligns with current intersocietal recommendations [1], providing an innovative and practical solution to enhance early diagnosis and personalized care in memory clinics
Statistical Strategies for Olink Proteomics Data: A Comparative Approach and Future Directions
No full textINTRODUCTIONOlink® proteomics platforms offer a powerful tool for high-throughput biomarker discovery through multiplexed protein quantification. Their application in cardiovascular research provides novel opportunities to identify predictive biomarkers, but the complexity and dimensionality of the resulting Omics data require tailored statistical methodologies for robust analysis and interpretation.
OBJECTIVES
This study aimed to compare multiple statistical techniques to analyze Olink data from coronary artery disease patients, with the goal of identifying plasma biomarkers associated with cardiovascular mortality.
METHODS We analyzed 69 plasma samples from patients with coronary artery disease, of whom 17 (24.6%) experienced cardiovascular mortality. Protein expression was assessed using four Olink Target 96 panels (cardiometabolic, cardiovascular II and III, inflammation), yielding 333 Normalized Protein eXpression (NPX) values. A multi-method analytical pipeline was employed, including univariate t-tests, principal component analysis (PCA), Gene Set Enrichment Analysis (GSEA), heatmap visualization, Boruta feature selection, and multivariate logistic regression with stepwise variable selection. Analyses were conducted using SAS v9.4 and R v4.3.1, including the OlinkAnalyze R package [1].
RESULTSInitial univariate analyses did not identify statistically significant differences between outcome groups after multiple testing correction. Volcano plots of adjusted p-values confirmed this lack of significance. PCA revealed low explanatory power of the first two components, suggesting limited separation between cases and controls based on the protein profiles. GSEA and heatmap analyses failed to detect any significant enrichment patterns. In contrast, the Boruta algorithm identified several relevant features, which were further evaluated in a multivariate logistic regression model. Stepwise selection based on unadjusted p-values led to the development of a predictive model with good performance (AUC = 0.89, 95% CI: 0.81–0.97). Clinical collaboration played a key role in contextualizing these findings.
CONCLUSIONSThis study highlights the importance of integrating diverse statistical methodologies for the analysis of high-dimensional Olink proteomics data. While no single approach yielded definitive results, the combination of techniques allowed for the identification of promising biomarkers and construction of a performant predictive model. However, the small sample size remains a major limitation, affecting the robustness and reproducibility of the findings. Future research should explore the integration of synthetic data generation techniques to simulate larger datasets. This could enhance the stability of statistical inferences and allow more confident identification of clinically relevant biomarkers in small-scale Omics studies
Implementation of a Bundle for Gram-Negative Bloodstream Infection: Impact on 30-Day Mortality Analyzed with a Path Model
No full textINTRODUCTION
Gram-negative bacteria are the first cause of community bacteremia and the second cause of nosocomial bacteremia with increasing incidence, relevant morbidity and mortality. The current lack of international guidelines and/or evidence-based care bundles focused on Gram-negative bloodstream infection (GN-BSI) led to challenging management.
OBJECTIVES
Our aim was to verify whether implementing a bundle in the management of GN-BSI which comprised follow-up blood culture (FUBC), diagnosis imaging (DI) and source control (SC), optimized antibiotic administration (OAA) and shortened treatment duration for uncomplicated BSIs reduced mortality after a GN-BSI event.
METHODS
The study population included patients with monomicrobial GN-BSI aged ³18 years enrolled at IRCCS Azienda Ospedaliero-Universitaria di Bologna. From January 2017 to December 2019 patients were administered standard clinical practice (pre-phase); from January 2022 to December 2023 (post-phase) all patients with monomicrobial GN-BSI were managed according to the predefined bundle designed to provide a structured, standardized approach to GN-BSI management, aimed to reduce mortality and improve patient outcomes. To verify the objective of the study a path model was developed, in which the indicator of the study phase was the exposure of interest, 30-day all-cause mortality was the dependent variable and the indicators of FUBC, DI, OAA and SC were set as mediators. FUBC, DI and OAA were posited as first-level mediators and SC as a subsequent mediator, because the decision to undertake FUBC, DI and OAA stems from the patients’ diagnosis, while SC can be undertaken according to FUBC and DI results. Treatment duration was not included in the model because it could be either cause or effect of mortality. The phase indicator was hypothesized as influencing all mediators and the outcome. Several clinical characteristics along with age and gender were included as exogenous variables. From an initial theoretical model, a final model was obtained by discarding non-significant paths and adding paths suggested by modification indices and deemed clinically relevant. Finally, a comparison with other modelling techniques was carried out to verify whether the path model actually added critical information and fitted best to the data.
RESULTS
Overall 3,355 patients were included, 2,092 were managed in the pre-phase and 1,263 with the bundle. Median age was similar (70.4 ± 16.4 vs. 71.2 ± 16.2 years), as the proportion of male patients (55.4% vs. 58.3%). No significant differences were observed in SOFA score, immunosuppression, septic shock incidence, rates of uncomplicated BSI, while the Charlson Comorbidity Index (CCI) was slightly lower in the post-phase (5.57 ± 2.74 vs. 5.79 ± 3.04; p = 0.011). In post-phase patients the BSI acquisition site showed fewer nosocomial or healthcare-associated cases (64.9% vs. 68.3%, p=0.045) and more community-acquired infections. Escherichia coli was the most common pathogen and was less frequent among post-phase patients (47.8% vs. 53.3%; p = 0.002). Among these patients, higher proportions of appropriate empirical therapy (77.0% vs. 69.2%; p <0.001), execution of FUBC within 7 days (55.8% vs. 30.2%; p <0.001), use of DI (90.8% vs. 77.5%; p <0.001) as well as SC (26.3% vs. 23.6%, p=0.083) were found, while OAA was administered in similar proportions. Mortality rate at 30 days was slightly higher in patients of the pre-phase (14.1% vs. 12.4%) but non-significant at c2 test (p =0.150). The final path analysis model was performed on 3322 patients and used 12 independent and 4 dependent variables. It obtained a very satisfactory fit to the data (RMSEA=0.015, CFI=0.973, TLI=0.953) and explained 32.7% variance of the 30-day mortality variable. The model confirmed that bundle administration was not directly associated with 30-day mortality, however an indirect negative association was found, through imaging: patients of the post-phase were more likely to have DI performed, which in turn was associated to a lower risk of 30-day mortality. The total indirect effect on mortality of being in the post-phase, expressed by standardized coefficient, was -0.050 (p<0.001). DI was positively correlated with FUBC (r=0.285). Other variables significantly affecting mortality (std. effects) were SOFA score (0.275, p<0.001), age (0.234, p<0.001), UTI (-0.172, p<0.001), CCI (0.167, p<0.001), stay in surgery ward at the time of infection (-0.130, p<0.001), BSI nosocomial infection (0.034, p<0.001), carbapenem resistance (0.063, p=0.009) and Pseudomonas spp pathogen (0.045, p<0.001). The relationship between study phase and mortality was not found by a multiple logistic regression model including the same variables used in the path model (std. coefficient 0.004, p=0.882); a treatment-effect lasso logistic model run with 75 variables on 2791 patients of this population obtained an ATE of -0.004 (p=0.798) with a PO mean of 0.137 (p<0.001).
CONCLUSIONS
Interventions like the bundle described in this study are not inherently affecting mortality directly, because they act by deploying several components that may take place with a predetermined priority and sequence, actually one affecting the other. Rather, they are expected to affect the outcome indirectly by triggering activities that can actually be related to the outcome. We have demonstrated that the bundle activation did decrease the mortality rate of patients with monomicrobial GN-BSI by increasing the use of imaging and of its correlated follow-up blood culture, which in turn were related to lower mortality. Indirect effects cannot be estimated by traditional modelling techniques like multiple regression, therefore we advocate the use of path modeling in clinical settings involving temporally related activities
Predicting Acute Biliary Pancreatitis Relapse Using CNN: The Minerva Multicentric Study
No full textIntroduction
Acute pancreatitis (AP) is the main pancreatic disease diagnosed in the world [1]. The etiology of AP is commonly alcoholic or related to biliary events [2,3]. Current guidelines recommend performing early cholecystectomy (EC), as surgery significantly reduces the risks of subsequent recurrence [4,5,6,7,8]. Recurrence of acute biliary pancreatitis (RBAP) is defined as a syndrome of multiple distinct acute inflammatory responses originating in individuals with genetic, environmental, traumatic, metabolic, who experienced a second episode of AP after at least 3 months [9]. To date, RBAP is a dangerous clinical complication of the pancreas, requiring emergency surgery and it can cause death if not operated on within 24 hours of onset [9]. However, due to particular patient frailties, medical-surgical conditions, or logistic problems, EC is not always performed [10]. The early identification of patients at high risk of recurrence could lead to better clinical and logistics management and provide practical recommendations for cholecystectomy priority [11,12,13,14]. Predicting and preventing RBAP can reduce costs of hospitalization and medical care and, more importantly, promote the management and prioritization of cases hospitalized with AP and potentially subject to relapse. Our recent systematic review [15] confirmed that there are no prospective studies that tried to model the prediction of RBAP. All evidence emerged through monocenter, retrospective data, was inconclusive and contradictory. The aim of the MINERVA study is 2-fold: on one hand, it aims to gather prospective data about RBAP from XX centers in Italy; on the other, it aims to develop and validate the first machine learning-based predictive model to identify patients at risk of RBAP [16].
Objective
The MINERVA (Machine learnINg for the rElapse Risk eValuation in Acute Biliary Pancreatitis) project is the first observational multicenter prospective trial designed to investigate the predictive factors of relapse in acute biliary pancreatitis using artificial intelligence (AI) and by collecting outcomes at 3 months, 6 months, and 1 year follow-up. The aim of this project is to develop a predictive model of acute biliary pancreatitis recurrence based on a convolutional neural network (CNN), using images generated from tabular clinical data from both prospective and retrospective multi-centric sources.
Methods
Clinical tabular data from the retrospective MANCTRA [17] and prospective MINERVA [16] datasets were merged to obtain 2413 instances appropriately preprocessed to manage missing values, normalizations and prevent errors. A strong imbalance was regulated applying the adasyn algorithm [18], to increase the minority class (initially 8% of the total) with artificial instances. The final dataset was made by 3630 subjects divided in 1441 with relapse and 2189 without it. The selected predictors were divided in numerical: the patient’s age, BMI, white blood cell (per mm3), neutrophil (per mm3), platelet (per mm3), international normalized ratio, protein c-valuesreactive, aspartate aminostrasferase (units/liter), alanine aminostrasferase (unit/liter), total birylubin (mg/dl), direct conjugated birylubin, gamma glutamyl transpeptidase (units/liter), serum amylase (units/liter), lipasis (units/liter), the lactate dehydrogenase (units/liter); and categorical: sex of the patient, previous episodes of pancreatitis, clinical history of diabetes, clinical history of chronic lung disease, hypertension, atrial fibrillation, chronic kidney disease, disease of the hematopoietic system, immunosuppressive drugs at the time of admission, cholodulelithiasis, acute cholangitis, department of hospitalization and Endoscopic Retrograde Colangiography-Pancreatography. Then, using the PCA-based Deepinsight algorithm [19], the features have been mapped to pixels with a grid size equal to 128; To process the images, we designed a CNN consisting of 3 reduction blocks with ReLU and MaxPooling activation, followed by 2 fully connected layers: the first with ReLU and Dropout, the second with sigmoid activation for binary classification (RBAP: yes/no). we used Adam optimizer (L2 adjustment) and binary cross-entropy loss. Data were divided into training (70%), validation (15%) and test sets (15%), and AUC, F1-score and accuracy were used for evaluation. We used an adaptive variable learning rate (LR) starting from 0.001, a batch size(BS) between 16 and 32, the number of periods incrementally validated according to a trial and error approach, and then fixed by early stopping (ES) technique. A classical non-parametric bootstrap approach, based on 50-iterations, was adopted to estimate the variability of performance metrics by evaluating greater robustness and reliability of the predictive capabilities compared to fluctuations in baseline data. The python code took about 30 minutes to run on a PC with Processor 12th Gen Intel(R) Core(TM) i5-12400F (12CPUs) 2.5 GHz, Operating System Microsoft Windows 11 Pro and RAM from 32GB, equipped with NVIDIA GeForce RTX 4060.
Results
The model showed good predictive performance in validation phases with an AUC of (84.38 ± 1.76)%, while on test AUC is 82.25% . Parameters as follows: BS=32, LR=0.001, ES at 60 epochs and weight decay equal to 0.0001.
Figure 1. Structure of the implemented CNN (A), Calibration Plot (B) and Roc Curve Plot (C) of the model.
Conclusion
This study presents the first AI model developed specifically for the prediction of RBAP. The integration of clinical data from complementary sources and the application of CNN techniques demonstrate the feasibility and clinical potential of this approach in an area that has been little explored so far. The results of this work are promising and the testing of AI to support the management of cases of relapse of acute biliary pancreatitis can prove a beneficial factor in supporting the medical clinic
Treatment Persistence of Lacosamide, Perampanel and Brivaracetam: An Extended Real-World Analysis from the COMPARE Italian Cohort
No full textObjectivesAntiseizure medication (ASM) discontinuation is a frequent and clinically meaningful outcome in the treatment of epilepsy, often reflecting inadequate tolerability, limited effectiveness, or both[1, 2]. In this study, we reanalyzed data from the COMPARE Italian multicenter cohort previously published by Roberti et al[3].to investigate real-world outcomes in patients treated with the most recent add-on ASMs. Our aim was to compare treatment persistence among lacosamide, brivaracetam, and perampanel over time, and to assess how adverse events and clinical response influence the discontinuation process. In contrast to the original multivariable regression approach, we applied a propensity score weighting framework to better approximate causal effects and address residual confounding.
MethodsWe estimated stabilized inverse probability of treatment weights (IPTW) using a multinomial propensity score model based on age, sex, epilepsy duration, etiology, seizure type, prior ASM exposure, epilepsy surgery, and baseline polytherapy[4, 5]. Time to treatment discontinuation was analyzed through log-logistic accelerated failure time (AFT) models, which allow direct estimation of time ratios (TRs)[6]. Clinical response and adverse events were added as covariates to explore their potential mediating role. Interaction terms with log-transformed time were introduced to assess how treatment effects evolve longitudinally. Robustness was evaluated through sensitivity analyses using cluster-robust standard errors and symmetric weight trimming to exclude individuals with extreme propensity scores.
ResultsThe analysis included 828 patients (250 lacosamide, 234 brivaracetam, 344 perampanel), with a maximum follow-up of 36 months. In unadjusted AFT models, brivaracetam (TR = 0.48, 95% CI: 0.27–0.84) and perampanel (TR = 0.46, 95% CI: 0.27–0.76) were associated with shorter persistence than lacosamide. After adjusting for adverse events and response, the association was attenuated for brivaracetam (TR = 0.68, 95% CI: 0.42–1.14), while perampanel remained significant (TR = 0.61, 95% CI: 0.38–0.97). Both adverse events (TR = 0.44, 95% CI: 0.30–0.66) and clinical response (TR = 4.25, 95% CI: 2.97–6.07) were independently associated with time to discontinuation, supporting their potential mediating role. Time-dependent models revealed that the initial disadvantage of brivaracetam (interaction TR = 2.50, 95% CI: 2.21–2.83) and perampanel (TR = 2.60, 95% CI: 2.34–2.88) diminished over time. Clinical response became increasingly protective, while the impact of adverse events waned. Sensitivity analyses confirmed the robustness of these findings.
ConclusionsThe risk of treatment discontinuation evolves dynamically and differs among ASMs. Lacosamide showed greater early persistence, whereas brivaracetam and perampanel, despite being associated with earlier dropout, demonstrated improved retention among patients who tolerated them beyond the initial period. These findings suggest that the first weeks of treatment may be critical for identifying and managing tolerability issues, especially with brivaracetam and perampanel. Our results underscore the importance of adopting a dynamic, patient-centered approach to ASM selection, considering not only baseline characteristics but also longitudinal response and tolerability. Propensity score-based methods, when applied appropriately, can enhance causal inference in observational studies and provide clinically meaningful insights to guide therapeutic decisions in epilepsy care
Predicting Methodological Quality in No Profit Clinical Trial
No full textINTRODUCTION
The SPIRIT 2013 Statement [1] has long represented the international gold standard for the content of clinical trial protocols, providing a comprehensive framework to ensure transparency, methodological rigor, and ethical soundness. However, the rapid evolution of trial methodologies, regulatory landscapes, and data-sharing practices has created the need for a revised standard. Just days ago, SPIRIT 2025 [2] was released, introducing updated and expanded guidance that reflects contemporary challenges and expectations, particularly in areas such as adaptive designs, patient involvement, and statistical analysis. Despite these efforts, numerous studies continue to document suboptimal adherence to SPIRIT recommendations [3], especially concerning trial design and statistical methods. Evaluating real-world adherence to SPIRIT guidelines [4] offers valuable insights into common shortcomings, systemic barriers, and areas requiring targeted support or training [5].
OBJECTIVES
The aim of this study was to assess whether it is possible to predict which clinical trial protocols are likely to show high adherence to SPIRIT guidelines, with a specific focus on methodological items. We sought to identify study-level characteristics that may act as potential predictors of adherence, to better understand structural drivers of protocol quality and support future improvement strategies.
METHODS
We retrieved information on design and methodological features of clinical trial protocols submitted between 2021 to 2025 to the local Ethics Committee and recorded them in a centralized REDCap registry. Adherence to SPIRIT 2013 items related to methodology, study design, data collection, management, and analysis (items 9–21b) was assessed. The 2013 version was used because all studies included were submitted prior to the release of Spirit 2025. Each item was scored as fulfilled or not, and an individual adherence score was computed as the total number of satisfied items. We described the distribution of adherence scores using median and interquartile range (IQR). Since adherence scores were not normally distributed, we dichotomized the scores at the median value, classifying them into higher adherence ("good") and lower adherence ("poor") categories. To address the predictive objective, we implemented a set of machine learning algorithms (e.g. Random Forest, eXtreme Gradient Boosting, and Boosted Logistic Regression) applied to a pool of candidate predictors selected for their potential relevance to the outcome. Model performance was evaluated using accuracy, area under the ROC curve (AUC), and F1 score. Variable importance was then assessed across models, and the most influential predictors were subsequently incorporated into a multivariable logistic regression model to evaluate their independent association with the outcome. Covariates included study characteristics related to sponsorship, methodological features, submission timing, and thematic focus. Odds ratios and corresponding 95% confidence intervals will be reported to evaluate the direction and strength of the association. Analyses were conducted using Stata software, release 19 and R version 4.4.3.
RESULTS
All 132 protocols included in the analysis were no profit interventional. Of these, 28% were monocentric, 59% multicentric within Italy, and 13% involved international sites. Overall, 32.6% of the studies were promoted by Italian sponsors with structured biostatistical support, 62.9% by other Italian sponsors for whom the availability of such support is unknown, and 4.5% by international institutions, also with unknown biostatistical support. Randomization was used in 59% of protocols. Blinding was reported in 12% of protocols: 9% were double-blinded and 3% single-blinded. Median adherence to the overall SPIRIT checklist was 72% (IQR 52.3%–85.7%). In the methods section, adherence was 80% (IQR 60%–90%) for items 9–15, 100% (IQR 66%–100%) for items 16–17, and 62.5% (IQR 37.5%–87.5%) for items 18–21b. The outcome was defined as a summary score representing the number of methodological items checked and was dichotomized at the median value. The presence of a biostatistics unit was significantly associated with higher methodological quality (OR = 3.44; 95% CI: 1.08–10.99; p = 0.037). Protocols involving pediatric populations were less likely to meet high-quality criteria (OR = 0.085; 95% CI: 0.007–1.048; p = 0.054), as were those in the Oncology/Infectious Diseases area (OR = 0.20; 95% CI: 0.045–0.870; p = 0.032). The model demonstrated good discriminative ability (AUC = 0.792) and excellent calibration (p = 0.91).
CONCLUSIONS
By combining traditional statistical approaches with innovative machine learning models, we gained a clearer understanding of which protocol features are predictive of high adherence to SPIRIT guidelines. Our findings suggest that the involvement of a multidisciplinary team, including biostatisticians, is strongly associated with better methodological quality. Protocols submitted by IRCCS institutions showed higher adherence. In contrast, trials involving special populations, particularly pediatric studies, were more likely to exhibit lower adherence, highlighting a need for targeted guidance and support in these contexts. Future analyses should include a larger sample and protocols evaluated by multiple Ethics Committees to enhance the generalizability of these findings.
 
Translation and Content Validity of a Tool for Measuring Perception of Nuclear Issues among Students Visiting Pavia Research Reactor
No full textINTRODUCTIONThere is a lack of national data describing students’ perceptions of nuclear safety. Moreover, no validated instruments currently exist for investigating this topic.
OBJECTIVES
The objectives of this study were to translate and back-translate from French to Italian and from Italian to French the questionnaire used in the study "Special Eurobarometer 271 / Wave 66.2 – Europeans and Nuclear Safety"1 conducted by the European Commission; validate the questionnaire using the Content Validity Ratio (CVR); assess the content and face validity of the tool through expert panel reviews; and collect preliminary data (pilot study) to support the planning of future research strategies.
METHODSThis pilot observational study involved the administration of a questionnaire as an additional procedure during routine educational reactor activities. Phase 1 consisted of the translation and back-translation of the questionnaire. Phase 2 involved the evaluation of content and face validity by a panel of nine experts with diverse backgrounds in nuclear science. In Phase 3, pilot data were collected using the survey instrument and subsequently analyzed through descriptive statistics. Content validity was assessed using the Item Content Validity Index (I-CVI), with a threshold of 0.78 considered acceptable, and the Scale Content Validity Index/Average (S-CVI/Ave), with a threshold of 0.80. Face validity was assessed through expert panel reviews. Items scoring below 0.60 were individually reviewed by the expert board.
RESULTSContent and face validation revealed the removability of seven items, which were individually reviewed by the expert board. As a result, two of these items were retained. The final version of the questionnaire included twelve items, with a final S-CVI/Ave of 0.89. The pilot study (n = 50) showed a divided perception among adolescents regarding nuclear energy: some participants perceived it as highly risky, while others held more favorable views. Data collection is ongoing. Sixty percent of respondents believe that the benefits of nuclear energy outweigh its risks. However, 98% reported feeling not at all or only slightly informed about nuclear energy. Sixty-seven percent believe that nuclear power plants are not very risky or not risky at all for the country, and 88% are completely or somewhat in agreement that such plants can be operated safely. Finally, 67% believe that the use of nuclear energy should be increased.
CONCLUSIONThe results of this pilot study support the successful translation, back-translation, and initial validation of the questionnaire in terms of content and face validity. These findings suggest that the instrument is a promising tool for assessing perceptions of nuclear safety and can be reliably used in future research within educational and healthcare contexts