Riviste UNIMI
Not a member yet
21278 research outputs found
Sort by
Integrating Calibration into the Evaluation of Clinical Utility: A Proposal for a Weighted Net Benefit
No full text
INTRODUCTION
Decision Curve Analysis (DCA) is a widely used framework for evaluating diagnostic and prognostic strategies, as it explicitly incorporates the clinical consequences associated with decision-making [1]. Within this framework, Net Benefit (NB) is a key indicator of the clinical utility of a predictive model. However, the original NB formulation does not account for model calibration, despite strong evidence that poor calibration systematically reduces clinical utility [2]. In real-world settings, many predictive models, particularly those based on machine learning (ML), often show suboptimal global calibration compared to traditional statistical models, due to their greater complexity and susceptibility to overfitting.Therefore, when focusing on a specific probability threshold as a decision point, it becomes crucial to evaluate calibration in the vicinity of that threshold. This targeted approach may lead to a different comparative assessment of the clinical potential of various predictive algorithms.
OBJECTIVES
To propose a pragmatic extension of DCA based on the concept of Weighted Net Benefit (WNB), in which a model’s utility is penalized more heavily for calibration errors in the decision-making region. This avoids discarding models that, while globally less calibrated, are reliable near the clinical threshold.
METHODS
The framework involves four steps:i) the decision threshold is defined a priori through DCA and/or clinical consultation;ii) calibration is assessed around the threshold using the Bayesian Calibration Error (BCE);iii) the weighted posterior standard deviation (wSD) is computed to quantify the statistical uncertainty associated with local calibration error (LCE) estimates;iv) NB at the predetermined threshold is adjusted according to the following formula:
.
Specifically, the BCE is defined as the weighted mean of local calibration errors (LCE), calculated as the absolute difference between the average predicted probability in each bin and the Bayesian estimate of the observed event rate, computed as the posterior mean of a Beta(1 + k, 1 + n − k) distribution, where k is the number of events and n the total number of observations in the bin. Each LCE is weighted using a Gaussian function centered on the decision threshold, thus emphasizing calibration errors in clinically critical regions.The wSD is calculated as the weighted mean of the standard deviations of the posterior Beta distributions in each bin, using the same weighting function. This penalizes models not only for local miscalibration but also for greater statistical uncertainty in the estimation of calibration error near the decision threshold [3]. In this way, the WNB provides a more cautious and context-aware estimate of clinical utility, allowing recognition of models that, despite suboptimal global calibration, provide reliable predictions around the decision threshold.We applied this framework to a clinical dataset of 3,672 pregnancies with small-for-gestational-age fetuses collected in the multicenter TRUFFLE 2 study. Three predictive models—logistic regression (LR), Random Forest (RF), and XGBoost—were developed using 11 clinical variables to predict adverse perinatal outcomes. NB and WNB were calculated at the decision threshold t = 0.3.Additionally, to assess prediction instability, NB and WNB were computed as the mean and standard deviation over 500 bootstrap replicates at the clinical threshold.
RESULTS
The incidence of adverse outcomes was 13%.RF achieved the best discrimination after bootstrap optimism correction (AUROC = 0.91, 95% CI: 0.85–0.94), while LR showed the poorest performance (AUROC = 0.71, 95% CI: 0.67–0.75). XGBoost had intermediate performance (AUROC = 0.83, 95% CI: 0.74–0.88).At the threshold t = 0.3, LR showed a low NB (0.02 ± 0.001) and an even lower WNB (0.01 ± 0.001), reflecting limited clinical utility at the threshold despite near-optimal global calibration.RF yielded the highest NB (0.08 ± 0.002), though it received the strongest penalty (WNB = 0.04 ± 0.001), while XGBoost displayed intermediate behavior (NB = 0.05 ± 0.003; WNB = 0.03 ± 0.002).These results suggest that, when discrimination is high, even a model with suboptimal calibration may retain clinical utility—if the expected decision quality near the critical region compensates for calibration penalties.Low variability across bootstrap samples (SD ≤ 0.003) indicates that both NB and WNB are highly stable around the clinical threshold t = 0.3 for all models (Figure 1).
CONCLUSIONS
The WNB offers a tailored evaluation of the clinical utility of predictive models by incorporating local calibration information near the decision threshold.This approach is not intended to replace traditional DCA but rather to serve as a methodologically coherent extension—particularly relevant when comparing predictive algorithms of varying complexity in terms of calibration performance.In clinical contexts where decisions are based on well-defined thresholds, WNB can support more informed, reliable, and decision-centered evaluations
Vittorio Sereni as a “French literary specialist”. An unpublished Immagine di Guillaume Apollinaire
No full textIl contributo riporta alla luce un testo inedito di Vittorio Sereni, intitolato Immagine di Guillaume Apollinaire, ora conservato nell’Archivio Vittorio Sereni presso il Palazzo Verbania di Luino. Verosimilmente databile agli anni dell’immediato secondo dopoguerra, esso verrebbe a costituirsi, rispetto agli altri due finora noti (pubblicati, rispettivamente, ne Gli immediati dintorni, 1962, e in Letture preliminari, 1973), come il più antico saggio di Sereni dedicato al poeta francese. L’introduzione della curatrice contestualizza il testo nell’ambito delle versioni poetiche poi pubblicate da Sereni.This contribution brings to light an unpublished text written by Vittorio Sereni, titled Immagine di Guillaume Apollinaire and currently held in the Vittorio Sereni Archive at Palazzo Verbania in Luino. Hypothetically dating back to the immediate post-Second World War period, this essay is compared to the two other known works by Sereni (published in Gli immediati dintorni, 1962, and in Letture preliminari, 1973), making it potentially the earliest known piece dedicated to the French poet. The curator’s introduction provides a contextualization of the text within the framework of Sereni’s subsequent poetic translations from Apollinaire
The “burgundy one”: 5th edition of the Ringer’s Bible
No full textReview of Svensson L., 2023. Identification Guide to European Passerines. Lullula Förlag, Stockholm. 448 pp
Il valore dell’AFAM
No full textNel panorama dell’istruzione superiore italiana, il sistema dell’Alta Formazione Artistica, Musicale e Coreutica (AFAM) occupa una posizione peculiare e, in un certo senso, ancora irrisolta. Riconosciuto come presidio di eccellenza per la formazione nelle arti e nella musica, esso rimane tuttavia privo di un pieno statuto accademico compiuto, sia sul piano normativo sia su quello epistemologico. La sua collocazione istituzionale, frutto di stratificazioni storiche e di riforme parziali, riflette una tensione costante tra riconoscimento formale e marginalità sostanziale: una condizione che ne limita il potenziale, pur in presenza di risultati formativi e artistici di altissimo livello.L’AFAM costituisce a pieno titolo un segmento fondamentale del sistema dell’istruzione superiore italiano, come riconosciuto anche dalla legge n. 508 del 21 dicembre 1999, in cui la trasmissione del sapere si configura come un equilibrio originale tra prassi, riflessione teorica e ricerca applicata. In questo modello, il laboratorio, la pratica performativa e la dimensione progettuale non rappresentano un semplice supporto alla teoria, ma un luogo epistemico in cui il sapere si produce e si rigenera. In tale prospettiva, l’AFAM si distingue dalla tradizione universitaria non per minore rigore, bensì per la diversa natura della sua epistemologia: un sapere che integra dimensione embodied, competenza tecnica, riflessione critica e capacità creativa
Danza, schermi e visori. Contaminazioni coreografiche nella scena italiana — Audino, 2024
No full text« De la mer au désert, visitez l’Algérie ». La revue Algeria (1933-1962) entre publicité et propagande
No full textIn 1929, the governor of Algeria, Pierre Bordes, declared that it was necessary to make France become obsessed with Algeria. Since the 1930s, the magazine Algeria seemed to set a similar agenda throughout its three series (1933-1939, 1940-1943, 1948-1962). Algeria was the monthly magazine published by the « Office Algérien d’Action Économique et Touristique » (OFALAC), which was responsible for the commercial promotion of Algeria in France and abroad. Through « propagandistic » advertising, Algeria aimed to build a strong relationship between metropolitan France and the colony. Which strategies did it employ to popularise this « obsession with Algeria »? And what role did this sponsorship play in the broader historical context of WW2 and the independence movement? By analysing both paratextual elements such as covers, reportage images, advertisements, and the texts themselves, I will discuss Algeria’s visual and textual « branding ». Through its iconographic and descriptive bric-a-brac, can Algeria be seen as part of the « touristic invention » of the Mediterranean that follows its « scientific invention » and «literary invention» ?Le gouverneur d’Algérie Pierre Bordes déclarait en 1929 qu’il fallait « créer en France une “obsession de l’Algérie” ». À partir des années 1930, la revue Algeria semble se donner ce même objectif. Avec ses trois séries (1933-1939, 1940-1943, 1948-1962), Algeria est la revue mensuelle éditée par l’Ofalac, l’Office Algérien d’Action Économique et Touristique, qui, dès les années 1930, s’occupe de la promotion commerciale de l’Algérie en France et à l’étranger.
Par le biais d’une publicité ouvertement qualifiée de « propagande », Algeria vise à mettre en communication la métropole et la colonie. Par quelles stratégies la revue insinue-t-elle une « obsession de l’Algérie » ? Et comment cette sponsorisation s’articule-t-elle avec la grande Histoire (deuxième guerre mondiale, mouvement pour l’indépendance) ? En analysant le paratexte (couvertures, images de reportage, publicités) et les textes, on entend ouvrir une réflexion au sujet du « branding » avant l’heure qu’Algeria construit par ses mots et ses images. Par son bric-à-brac iconographique et descriptif, Algeria ne participe-t-elle pas à l’« invention touristique » d’un coin de la Méditerranée, déjà objet d’une « invention scientifique » et d’une « invention littéraire »
Nicla MERCURIO, Les variations de la langue française en Suisse romande : le cas du canton du Jura. Enquête (socio)linguistique et terminologique, Torino/Paris, L’Harmattan Italia, 2023, 202 pp.
No full textLaurent DEMOULIN, « Belgicismes et littérature belge. Confidences d’un Belge francophone édité à Paris », in Maxime DEL FIOL (dir.), Travaux de littérature : Francophonie, plurilinguisme et production littéraire transnationale en français depuis le Moyen Âge (« Travaux de Littérature »), n. 35, 2022, pp. 369-388
No full text
