Riviste UNIMI
Not a member yet
21278 research outputs found
Sort by
The consistory of the Church of Cadiz (14th-16th centuries): provisors, notaires and do- cumental praxis
No full textEn la Baja Edad Media, al igual que el resto de diócesis hispanas y europeas, la Iglesia de Cádiz contó con una oficina de administración de justicia eclesiástica: el consistorio. Aún así, a pesar de que la diócesis gadicense hunde sus raíces en la conquista cristiana de la ciudad en el siglo XIII bajo el reinado de Alfonso X el Sabio, no es hasta finales del siglo XIV cuando surgen los primeros testimonios documentales de su funcionamiento. En este artículo, por tanto, nos centraremos en los provisores que dirigieron esta oficina, así como en los notarios en que se apoyaron para formalizar sus necesidades escriturarias.In the Late Middle Ages, like the rest of the Hispanic and European dioceses, the Church of Cadiz had an office for the administration of ecclesiastical justice: the consistory. However, even though the Cadiz diocese traces its roots back to the Christian conquest of the city in the 13th century under the reign of Alfonso X the Wise, it wasn\u27t until the end of the 14th century that the first documentary evidence of its functioning emerged. In this article, therefore, we will focus on the provisors who directed this office, as well as the notaries on whom they relied to formalize their scriptural needs.En la Baja Edad Media, al igual que el resto de diócesis hispanas y europeas, la Iglesia de Cádiz contó con una oficina de administración de justicia eclesiástica: el consistorio. Aún así, a pesar de que la diócesis gadicense hunde sus raíces en la conquista cristiana de la ciudad en el siglo XIII bajo el reinado de Alfonso X el Sabio, no es hasta finales del siglo XIV cuando surgen los primeros testimonios documentales de su funcionamiento. En este artículo, por tanto, nos centraremos en los provisores que dirigieron esta oficina, así como en los notarios en que se apoyaron para formalizar sus necesidades escriturarias.Au Bas Moyen Âge, à l\u27instar des autres diocèses hispaniques et européens, l\u27Église de Cadix disposait d\u27un office d\u27administration de la justice ecclésiastique : le consistoire. Cependant, bien que la diocèse de Cadix trouve ses racines dans la conquête chrétienne de la ville au XIIIe siècle sous le règne d\u27Alphonse X le Sage, ce n\u27est qu\u27à la fin du XIVe siècle qu\u27apparaissent les premiers témoignages documentaires de son fonctionnement. Dans cet article, nous nous concentrerons donc sur les proviseurs qui ont dirigé cet office, ainsi que sur les notaires sur lesquels ils se sont appuyés pour formaliser leurs besoins scripturaux
Cozroh’s Liber Traditionum and the Memory of the Agilolfing Era: the Case of Traditions nos. 49 and 50 and the References to Hagiography and Legislative Texts
No full textThe traditiones of the Cozroh Codex, the liber traditionum of the Church of Freising, produced between 824 and 848 approximately, were edited by Theodor Bitterauf in 1905. In the footsteps of Redlich\u27s theories on the libri traditionum, this edition considers the traditiones as individual acts that are rearranged chronologically and in continuity with documents attested in other codices of the Freising archive. More recent research has however urged us to bring attention back to the codex itself and to the original arrangement of the documents. This article proposes to analyse the first two traditiones in the part of the codex dedicated to the episcopate of Arbeo of Freising (764-783) in the light of some lexical parallels with the Vitae Corbiniani and Haimhrammi and with the legislative texts of the Agilolfing era (primarily the lex Baiuvariorum and the Synod of Neuching): these clues allow us to hypothesise that the original arrangement of the documents in the codex may derive from Cozroh\u27s desire to provide his audience with a precise interpretation of Arbeo’s episcopate, enhancing the role of the bishop both as the author of the hagiographical texts and as collaborator of duke Tassilo.Le traditiones del Codice di Cozroh, il liber traditionum della Chiesa di Frisinga realizzato tra l’824 e l’848 circa, sono state oggetto dell’edizione di Theodor Bitterauf nel 1905. Seguendo le teorie di Redlich sui libri traditionum, tale edizione considera le traditiones come atti singoli, che vengono quindi riordinati cronologicamente e disposti in continuità con documenti presenti in altri codici dell’archivio di Frisinga. La ricerca più recente ha però esortato a ricondurre l’attenzione sul codice stesso e sulla disposizione originale dei documenti. L’articolo propone di analizzare le prime due traditiones nella parte del codice dedicata all’episcopato di Arbeone di Frisinga (764-783) alla luce di alcuni parallelismi lessicali con le Vitae Corbiniani e Haimhrammi e con i testi legislativi di epoca agilolfingia (in primis la lex Baiuvariorum e le disposizioni del sinodo di Neuching): questi indizi permettono di ipotizzare che la disposizione originale dei documenti nel codice possa derivare dalla volontà di Cozroh di fornire al proprio pubblico una precisa interpretazione dell’episcopato di Arbeone, intesa a valorizzare il ruolo del vescovo come autore dei testi agiografici e come collaboratore del duca Tassilone
«A new and broader investigation». Axel Goria between the Corpus Astense and 20th century medieval studies
No full text«A new and broader investigation». Axel Goria between the Corpus Astense and 20th century medieval studies
The critical edition of the Asti chroniclers and the essays completing it are the legacy of one of Turin’s leading medievalists of the 20th century, Axel Goria, a disciple of Giorgio Falco. This work was entrusted to him in 1937 by the Italian Historical Institute for the Middle Ages in Rome and accompanied Goria’s studies without, however, being published. It constitutes a synthesis of the itinerary of medieval studies in the first decades of the twentieth century and an opening to the developments that the University of Turin and the Roman Institute would later bring about. The sources edited by Goria testify to the political dynamics between the 13th and 15th centuries in the process of institutional experimentation within which the communal experience encounters supra-regional networks of relations. This crucial phase is investigated through the chroniclers’ lexicon without adhering to preconstituted models: the study does not actually adopt the dichotomy concord - discord, freedom - tyranny, but analyzes the relations between the Partes, the People and broader powers – Anjou, Church and Empire – without neglecting the socio-economic dimension. This encounter between diplomatics and historical research is at the origin of the enduring validity of Goria’s work.«Una trattazione nuova e più ampia». Axel Goria tra Corpus Astense e medievistica del Novecento.
L’edizione critica dei cronisti astesi – con i saggi che la completano – costituisce il lascito inedito di uno dei principali medievisti di formazione torinese del Novecento, Axel Goria, allievo di Giorgio Falco: affidatagli dall’Istituto storico italiano per il Medio Evo nel 1937, accompagnò gli studi di Goria senza tuttavia vedere la luce. Rappresenta una sintesi dell’itinerario della medievistica dei primi decenni del Novecento e nello stesso tempo un’originale apertura agli sviluppi impressi in seguito dalla scuola universitaria torinese e dall’Istituto romano. Le fonti edite da Goria, infatti, testimoniano le dinamiche politiche tra Duecento e Quattrocento nel processo di sperimentazione istituzionale entro il quale l’esperienza comunale incontra reti di relazione sovraregionali. Questa fase cruciale è indagata restituendo voce ai lessici dei cronisti senza aderire a modelli prefissati: lo studio non si appiattisce sulla dicotomia concordia - discordia, libertà - tirannide ma analizza i rapporti tra le Partes, il Popolo e forze di più ampia portata – gli Angiò, la Chiesa e l’Impero – senza trascurare la dimensione socio-economica. Questo appare l’esito dell’incontro tra diplomatica e ricerca storica, all’origine della persistente validità del lavoro di Axel Goria
Suoni e musica nei videogiochi: dal ruolo passivo al ruolo inter-attivo nel gameplay
No full textThis contribution analyses the evolution of the role of sound and music in video games, focusing on recent developments in sound design in the field of game studies. Through a brief historical overview, it highlights how sound, initially with a passive and accessory function, has taken on an increasingly active and interactive role. From Space Invaders in the 1980s to contemporary titles, sound and music have become central elements of the gaming experience. Orchestral tracks, songs and jingles now accompany both cutscenes and gameplay, contributing to a deeper sensory immersion and dynamically signalling the player\u27s actions. Two main interactive dynamics have been identified: the use of binaural audio, which amplifies orientation and immersion by creating a sound space parallel to the visual one; and the use of sound as a narrative tool and game mechanic, especially in contexts where the image is reduced or absent. In this context, sound is conceived as “pure action”, generated by the player\u27s inputs. The case studies analysed include Horizon Zero Dawn for the role of jingles in gameplay; Hellblade: Senua\u27s Sacrifice and Resident Evil 2 Remake for binaural audio; Perception and Three Monkeys for sound as an interactive driver of narrative.Questo contributo analizza l’evoluzione del ruolo di suoni e musica nel videogioco, focalizzandosi sugli sviluppi recenti del sound design nell’ambito dei game studies. Attraverso un breve excursus storico, si evidenzia come il sonoro, inizialmente con funzione passiva e accessoria, abbia assunto un ruolo sempre più attivo e interattivo. A partire da Space Invaders negli anni Ottanta, fino ai titoli contemporanei, suoni e musiche sono diventati elementi centrali per l’esperienza videoludica. Tracce orchestrali, canzoni e jingle accompagnano oggi sia cutscene sia gameplay, contribuendo a un’immersione sensoriale più profonda e segnalando in modo dinamico le azioni del giocatore. Due le principali dinamiche interattive individuate: l’impiego dell’audio binaurale, che amplifica l’orientamento e l’immersività creando uno spazio sonoro parallelo a quello visivo; e l’utilizzo del suono come strumento narrativo e meccanica di gioco, specialmente in contesti in cui l’immagine è ridotta o assente. In quest’ottica, il suono viene concepito come “azione pura”, generata dagli input del giocatore. I case studies analizzati includono Horizon Zero Dawn per la funzione dei jingle nel gameplay; Hellblade: Senua’s Sacrifice e Resident Evil 2 Remake per l’audio binaurale; Perception e Three Monkeys per il suono come motore interattivo della narrazione
It’s a Kind of Magic (or not)? Ai’s technological imaginary and the symbolic structures of fashion narration.
No full textThis paper delves into the dynamic relationship between imaginary and technology in the era of Artificial Intelligence, illustrating how AI’s technological imaginary is not an independent entity but rather a product of the creative process inherent in the social imaginary. Additionally, it explores how the symbolic structures of socio-cultural imaginary contribute to shaping the imaginary realms of AI. The widespread integration of generative AI in fashion has introduced visual AI, offering new avenues for creative expression while challenging the construction of the sector’s imaginary landscape (Banerjee et al. 2022). This prompts critical inquiries: to which extent fashion narratives have become the nourishment of AI? What innovative narrations does AI introduce? How is AI influencing traditional perceptions of the body and beauty? Comparing 882 covers from Vogue Italia Digital Archive (1964-2024) with visuals from the AI-produced fashion magazine Copy (August 2023 onwards), this research analyzes representations and narratives to determine if AI perpetuates or subverts beauty and gender stereotypes, and how the normativity of fashion influences AI\u27s visual output.Questo articolo approfondisce la relazione dinamica tra immaginario e tecnologia nell\u27era dell\u27Intelligenza Artificiale, illustrando come l\u27immaginario tecnologico dell\u27IA non sia un\u27entità indipendente, ma piuttosto un prodotto del processo creativo insito nell\u27immaginario sociale. Inoltre, l\u27intento è di esplorare come le strutture simboliche dell\u27immaginario socio-culturale contribuiscano a plasmare i regni immaginari dell\u27IA. La diffusa integrazione dell\u27IA generativa nella moda ha introdotto l\u27IA visuale, offrendo nuove vie per l\u27espressione creativa e sfidando al contempo la costruzione del panorama immaginario del settore (Banerjee et al. 2022). Ciò solleva alcune domande critiche: in che misura le narrazioni della moda sono diventate materiale su cui l\u27IA si forma e si sviluppa? Quali narrazioni innovative introduce l\u27IA? In che modo l\u27IA sta influenzando le percezioni tradizionali del corpo e della bellezza? Confrontando 882 copertine dell\u27archivio digitale di Vogue Italia (1964-2024) con le immagini della rivista di moda prodotta dall\u27IA Copy Magazine (da agosto 2023 in poi), questa ricerca analizza le rappresentazioni e le narrazioni per determinare se l\u27IA perpetua o sovverta gli stereotipi di bellezza e di genere, e in quale modo la normatività della moda influenza la produzione visiva dell\u27IA
Potenzialità e disavventure della drammaturgia multimediale in Italia. Dialogo tra Anna Maria Monteverdi e Andrea Balzola
No full textDialogo tra Anna Maria Monteverdi e Andrea BalzolaConversation between Anna Maria Monteverdi and Andrea Balzol
Trajectories of Adherence to Biologics in Patients with Rheumatoid Arthritis and Risk of a Secondary Immuno-Mediated Inflammatory Disease: A Large Multi-Database Italian Study
No full textBackground
Rheumatoid arthritis (RA) is an immune-mediated inflammatory disease (IMID) with a global prevalence of approximately 1%. The therapeutic strategy, aiming at achieving low disease activity, includes both conventional and biologic disease-modifying anti-rheumatic drugs (bDMARDs). Evidence from the literature suggests that patients with one IMID are at higher risk of developing another. However, data are lacking on the association between the occurrence of secondary IMIDs and longitudinal adherence to bDMARDs.
Aim
To evaluate the association between adherence trajectories to bDMARDs and the occurrence of secondary IMIDs in patients with RA.
Methods
We conducted a population-based retrospective observational cohort study using administrative data [1]. We included residents of the participating regions from 2010 to 2023 who had at least one biologic dispensing approved for RA, and a diagnosis of RA identified though a validated algorithm. We excluded individuals younger than 18 years, had less than one year of continuous enrolment (look-back), were prevalent users of RA bDMARDs, were treated with rituximab as index drug, or had a diagnosis of any IMID other than RA during the look-back period. The first dispensing date of a bDMARD was defined as the index date. Patients were observed over a two-year period: the first year (exposure period) to assess adherence, and the second year (follow-up period) to monitor the occurrence of secondary IMIDs, death, or end of the study period—whichever came first. During the exposure period, patients were censored if they developed cancer, became pregnant, or died. Treatment coverage was estimated assuming a daily intake of one Defined Daily Dose. We assessed adherence to bDMARDS monthly over the exposure period using the Medication Possession Ratio. Adherence trajectories were identified using Group-Based Trajectory Modeling (GBTM) [2,3]. We used a Cox proportional hazards model to estimate the hazard ratio, and corresponding 95% confidence interval, for developing a secondary IMIDs.
Results
We identified a cohort of 35,600 individuals, with a higher proportion of females (78%), and a mean age of 56.5 (standard deviation: 14.0). We identified four distinct adherence trajectories over a 12-month period. Group 3, labeled as the High Adherent group and comprising 49.7% of participants, maintains nearly 100% adherence consistently throughout the entire follow-up period, indicating stable and optimal adherence. Group 1, referred to as the Declining Adherent group and accounting for 11.0% of the sample, starts with high adherence but shows a gradual and marked decline, particularly after the fifth month, reaching levels around 0.2 by month 12. Group 2, named the Moderate Adherent group and comprising 19.5% of the sample, experiences an initial drop in adherence in the early months, followed by an increase and stabilization around 0.6 to 0.7. Group 4, described as the Low Adherent group and representing 19.8% of participants, demonstrates a steep and continuous decline in adherence from the beginning, falling below 0.2 within the first few months and remaining low for the rest of the follow-up period, highlighting significant variability in adherence behaviors across the groups.
During the one-year of follow-up, 205 events of secondary IMIDs were observed. The Cox proportional hazards model did not reveal statistically significant differences in the risk of developing a secondary IMID across the adherence trajectory groups (overall p = 0.20). Compared to the High Adherent group (Group 3, reference), the Declining Adherent group (Group 1) had a hazard ratio (HR) of 1.28 (95% CI: 0.83–1.99), the Moderate Adherent group (Group 2) had an HR of 0.67 (95% CI: 0.43–1.04), and the Low Adherent group (Group 4) showed a non-significant increase in risk with an HR of 1.11 (95% CI: 0.78–1.59).
Conclusion
The study identified four distinct trajectories of bDMARD adherence, showing marked heterogeneity in patients\u27 behaviors. However, there were no statistically significant differences in the risk of developing a second IMID between the groups. These findings suggest that factors other than adherence may influence the occurrence of autoimmune comorbidities
A Study of Tremor Classification in Parkinson\u27s Disease using Unsupervised Learning Methods and Wearable Sensor Signal Processing
No full textIntroduction: Parkinson\u27s disease (PD) is a progressive neurodegenerative disorder characterized primarily by motor-related symptoms as tremor, slowness of movement, rigidity and difficulty with balance [1].
Although symptoms may vary from person to person, resting tremor is usually the most common symptom [2]. At the onset of the disease, it may be mild and unrecognized, and may only be a barely perceptible tremor in a hand, or sometimes in a foot or jaw. It often starts on one side of the body and then affects both sides, but usually one side remains the more affected than the other.
A diagnosis of PD is made based on neurological and physical examinations. The Movement Disorder Society Unified Parkinson\u27s Disease Rating Scale (MDS-UPDRS) [3] is the most common clinical scale used to track the longitudinal progression of PD. The assessment is based on disease severity, as determined through interview and clinical observation. Therefore, the evaluation may be subjective and affected by variability, reflecting the need for more objective measures for tremor classification.
Machine learning algorithms have recently been used to process data collected by wearable sensors [4].
Aims: Explore the use of an unsupervised learning model (k-means) to solve two classification problems: CP1: distinguish patients from controls (i.e. tremor vs. non-tremor); CP2: classify different tremor severities, where k=n. We also consider a third problem, CP3, which aims to distinguish between severe and mild tremor (i.e. k = 2), with the aim of simplifying CP2.
Methods: We used a publicly available dataset accessible via the website https://doi.org/10.21227/g2g8-1503 [5]. This dataset includes activity, gait, and tremor measures from 17 individuals diagnosed with PD and 17 healthy control (HC) subjects who were matched for age. These measures were collected using five adhesive sensors (one on each limb and one on the trunk) which captured triaxial accelerometer data during a clinic visit. Annotation files were also collected during this visit, when subjects underwent an evaluation using the MDS-UPDRS. We only considered data from sensors placed on the upper limbs of people with PD, and we excluded four subjects due to missing clinical data. Thus, our population included 13 PD patients (mean age ± SD: 66.1 ± 11.8 years; 38.5% female) and 11 HCs (66.0 ± 8.4 years; 90.1% female).
Before to apply k-means we performed a data segmentation process with the aim of extracting time intervals in which subjects were seated in a resting state. We used the start and end timestamps of the resting periods performed during the clinical assessment according to file tasks annotations. Then, the extracted segments were concatenated into a single string, i.e. a recording instance. We analyzed the tremor-predominant arm recordings for each PD patient, except for one patient for whom we analyzed both arms, for a total of 25 recording instances.
We pre-processed the raw accelerometer data performing the mean-centering and computing modulus. This transformation was necessary because of the effects of sensor orientation and individual bias. We chose the Euclidean distance as the distance metric because of its effectiveness in measuring the similarity of the movement intensities represented by the modulus values [6].
To ensure a balanced distribution of tremor events across subjects and enhance the sensitivity of the clustering algorithm, we only considered signal periods within the 95th percentile of movement intensity. We then identified the \u27dominant cluster\u27 as the cluster label that occurred most frequently in the top 5% of modulus instances.
In order to evaluate the accuracy of the K-means algorithm, we compared its assignments with clinician diagnoses. Specifically, for CP2, we compared cluster assignments with tremor labels assigned by the neurologist according to item 3.17 of the MDS-UPDRS, which evaluates resting tremor amplitude using a scale of 0–4. HCs had an at-rest tremor score of 0. Finally, we used a best matching approach to align the cluster labels with the clinical labels, selecting the mapping that resulted in the highest accuracy percentage among all possible permutations.
Results: CP1: the algorithm achieved an accuracy of 76.0%. Specifically, most PD patients were correctly assigned to the tremor cluster, while the majority of HCs were assigned to the non-tremor cluster. CP2: The algorithm achieved an accuracy of 57.1%, with significant overlap in cluster assignments. CP3: The algorithm achieved an accuracy of 71.4%. Figure 1 shows the k-means performance in all CPs.
Conclusions: The results emphasize that raw motion data can provide valuable information independently of predefined clinical labels, achieving a high level of accuracy in distinguishing tremor states from non-tremor states. Although the results of tremor severity classification, especially in multiclass scenarios, demonstrate the complexity of subtle tremor differentiation, highlighting the importance of improving feature extraction to achieve greater accuracy, the usefulness of unsupervised learning to enable scalable and objective tremor analysis is clear. Integrating such models into wearable systems could improve continuous monitoring, enhance rehabilitation strategies, and support standardized clinical assessments. Future work should focus on developing advanced algorithms, enriched feature sets and larger datasets to enhance the robustness and generalizability of these models
Use of Network Analysis for Identifying Drug Combinations to Prevent ADRs
No full textINTRODUCTION
The identification of drug combinations is critical to prevent adverse events, such as bleeding, often associated with the concomitant use of different active ingredients. Analysis of interactions using graphs is a useful tool for intuitively and quantitatively representing drug-drug relationships in real-world clinical settings.
OBJECTIVES
This study aimed to estimate the prevalence of potentially clinically important drug-drug interactions (DDIs) and the average causal effect of DDI exposure on hospitalization for bleeding related to adverse drug reactions (ADRs). In addition, identify the type of DDI that could lead to bleeding and the most common co-prescribed therapies responsible for ADRs through a network analysis approach.
METHODS
We performed a retrospective cohort single-center study including all consecutive patients admitted to the Internal Medicine Units of the Niguarda Hospital in Milan for bleeding-related complications from 1 January 2015 to 31 December 2018. Clinical characteristics, comorbidities, and pharmacological treatments were collected for each patient. Medication exposure was defined as the therapy assumed by the patient at the moment of admission. Polytherapy was defined by concomitant chronic use of ≥5 drugs. DDIs were identified using the LexiDrug database. Network analysis was applied to go to identify drug-drug interaction. Networks are a widely used tool for describing and analyzing complex systems, such as biological systems, in which relationships between entities-such as molecules, genes or drugs-play a central role [1,2,3]. In the biomedical context, graphs make it possible to visualize and study molecular interactions, such as those between proteins, genes, or drugs. A graph is a structure composed of a set of nodes (also called vertices) and a set of arcs (or edges) connecting pairs of nodes. Each arc can be characterized by a weight, which represents the strength or frequency of the interaction. This approach allows highlighting which drug combinations are most common in the analyzed dataset. There are several types of graphs, but in this analysis we focus on an undirected, weighted graph. This type of representation is particularly suitable when the relationship between two elements is bidirectional or symmetric, as is the case when two drugs are simply taken together by a patient, without implying a direction of effect [1]. One of the main questions that network analysis seeks to address concerns the identification of the most relevant or central nodes. A frequently used metric for this purpose is the degree of a node, which represents the number of connections (i.e., that the number of arcs) it has with other nodes in the network. Nodes with a high degree can be considered potential hubs, that is, central elements that contribute to the connectivity and robustness of the network. Analyses were conducted in R, using packages for data manipulation, graphs and visualization.
RESULTS
Overall, 604 patients, 242 women, and 363 men, were admitted for bleeding: 215 clinically relevant non-major bleeding, 389 major bleeding. Among major bleeding 209 in >80 elderly, 62 in patients between 75-80, 67 between 65-75, and 51 in under 65 patients. Patients using more than 2 drugs were included and they were 87.15% in case of major bleeding and 84.65 with minor bleeding. The most used drugs are proton pump inhibitors, followed by platelet aggregation inhibitors excl heparin, and beta-blocking agents. The dataset contains 392 ddis associated with the risk of bleeding. These associations represent specific combinations of drugs that could be linked to bleeding incidents, highlighting the importance of monitoring these combinations in clinical settings. The post frequent ddi associated with bleeding are co-somministrations of cns depressants and agents with antiplatelet properties, but also vitamin k antagonists with omeprazole/ pantoprazole, corticosteroids (systemic) / salicylates, aspirin / selective serotonin reuptake inhibitors, enoxaparin / agents with antiplatelet properties.
A drug x drug adjacency matrix was then constructed, in which each cell represents the absolute frequency with which two drugs were taken in combination by at least one patient. An undirected graph was derived from this matrix, in which the nodes represent the drugs and the arcs represent the observed interactions, with weight proportional to frequency. The degree of each node (number of interactions) was calculated and then the top 10% of the most connected nodes were selected to construct a filtered graph. A highly connected node appears frequently in combination with other active ingredients, while a less connected node is present in only a few associations. The degree of a node corresponds to the number of connections it has with other nodes. In addition, the weight of the arcs is proportional to the number of patients sharing the same drug combination: a thicker arc indicates a recurrent combination, while thinner arcs signal less frequent combinations.
The dataset included 604 patients, with 348 total drugs and 2542 interactions. Limiting the analysis to bleeding-associated interactions, the number of drugs considered drops to 121, with 392 interactions. In this subgraph, warfarin emerges as the most connected node (rank = 46), followed by amiodarone, clopidogrel, enoxaparin sodium, and acetylsalicylic acid. The most frequent interaction is warfarin - omeprazole (17 times). The filtered graph, showing only the 10% most connected drugs in the bleeding risk subgroup, is shown in Figure 1, where the intensity of the arc reflects the frequency of the observed interaction.
CONCLUSIONS
Potentially clinically important DDIs carry an increased average causal effect on ADR-related admission. Especially by exposure to DDIs that increase bleeding risk, which should be targeted for medicine optimization. The analysis highlighted key drugs in the network of interactions, particularly warfarin, which confirms its clinical relevance as a critical node in high-risk bleeding settings. This approach is a valuable tool for surveillance of drug interactions and can support clinical decisions, especially in polytreatment settings
Random Forest Regression for Predicting Healthcare Costs Using Administrative Databases from a Health Protection Agency in Northern Italy
No full textINTRODUCTION
Longer life expectancies and increasing prevalence of chronic diseases drive up demand for healthcare services and related costs. In Italy, 32% of people aged 65 and over, and 48% of those over 85, have major chronic conditions and multimorbidity [1]. In 2019, individuals aged 65 and over accounted for 46% of hospital admissions and 60% of pharmaceutical expenditures, highlighting the significant burden of aging on the healthcare system [2]. In terms of costs, population’s segments with high prevalence of chronic conditions account for a large portion of healthcare spending [3,4,5]. Accurate predictions of future costs for the whole population and for key segments is crucial for healthcare planning.
AIMS
To predict yearly direct healthcare costs based on data of past National Health Service (NHS) resources utilization for the whole population and for high impacting segments. As a motivating example, we applied our approach to the dialysis patients’ segment.
METHODS
Using administrative healthcare databases, we traced NHS resource utilization (i.e., access to inpatient and outpatient services, drug dispensations) and associated costs for each individual aged ≥18 assisted by the Health Protection Agency of Bergamo (Northern Italy) between 2011 and 2023. We analyzed total cost (TC) as the sum of all services and dispensations costs, total scheduled cost (TSC) as the sum of scheduled inpatient visits, all outpatient visits and dispensations costs, and scheduled services cost (SSC) as the sum of scheduled inpatient visits and all outpatient visits costs. In the present abstract we focused on TC prediction.
We used a supervised machine learning approach, namely random forest (RF) algorithm with 500 trees, to address the prediction problem [6,7]. We trained the algorithm on the 70% of individuals’ data from 2011 to 2015 (n=815,553) with their TC in 2016 as outcome. The 373 input variables included demographic features (such as age and sex) and NHS utilization data over the 4-years period 2011-2014 and in 2015 alone, in order to assess if 2016 cost was more associated with subjects’ behavior over the preceding year or with their historic behavior. As test sets, we used the remaining 30% of the dataset (hereafter 2011-16 set) and the subsequent years’ datasets (2012-17, 2013-18, 2014-19, 2015-20, 2016-21, 2017-22, and 2018-23 sets). We considered variable importance, measured as the percent increase in mean squared error (MSE) when a given variable is permuted, as a measure of each predictor’s impact on the outcome.
For each test set, actual and predicted TCs for the whole population were calculated as the sum of all individuals’ actual and predicted TCs, respectively. The ratio of the difference between predicted and actual population TCs to actual population TCs was used as measure of the prediction error (PE). PE=0% indicates a perfect prediction, PE >0% or <0% suggests overestimation or underestimation of the actual TC.
Finally, we defined dialysis patients as those who had at least one access to outpatient dialysis services. For this segment, we calculated the mean and sum of predicted and actual TCs, and PE. Also, we derived a variability interval for the mean predicted TC based on the 2.5 and 97.5 quantiles of the distribution of the mean TCs predicted by each tree for subjects included in the segment.
RESULTS
The mean actual annual population TC in the period from 2011 to 2023 was €1,023,636,867 (range: 944,632,707 – 1,111,657,382). High-cost subjects (>€15,000 yearly), accounting for less than 1% of the annual population, absorbed more than 27% of annual TC.
Top 3 most important variables in the RF were the number of outpatient accesses to dialysis over the preceding year, and the frequency of laboratory tests and outpatient services over the 4 preceding years.
Figure 1 shows the PEs calculated across all test sets, overall and in the dialysis patients’ segment. Overall, PEs ranged from -3.1 to -1.9 across 2011-16 to 2014-19 sets (for 2014-19 set, actual annual population TC: €1,031,200,509; predicted annual population TC: €1,011,869,922), and widely increased from 2015-20 (range from -6.9 to 8.7; for 2015-20 set, actual annual population TC: €944,632,707; predicted annual population TC: €1,026,878,752)
For the dialysis patients’ segment, the lowest PE (-0.7%) was observed in the 2011-16 set (actual mean TC: €38,536; predicted mean TC [variability interval]: €38,259 [35,542 – 41,112]), while the highest was -5.4% in the 2016-21 set (actual mean TC: €38,883; predicted mean TC [variability interval]: €36,785 [33,967 – 39,342]).
CONCLUSIONS
Using a machine learning approach, we predicted healthcare TCs based on individual data of past utilization of NHS for the whole population and a high impacting segment. Predictions based on the algorithm trained on data from 2011 to 2015 were consistent until 2019, understandable given the COVID-19 pandemic in 2020. Results highlight the pandemic’s impact on the model performance, leading to overestimation of the actual TC in 2020 and underestimations thereafter. Future steps include the identification of key segments and the update of the training algorithm on the subsequent years’ datasets. This is a useful tool to assist HPA in resource allocation, e.g. as an integration to the monitoring of chronic diseases in the population