Risk factors associated with bacteremia in COVID-19 patients admitted to intensive care unit: a retrospective multicenter cohort study.

PURPOSE: This multicenter observational study was done to evaluate risk factors related to the development of BSI in patients admitted to ICU for COVID-19. METHODS: All patients with COVID-19 admitted in two COVID-19 dedicated ICUs in two different hospital between 02-2020 and 02-2021 were recruited. RESULT: 537 patients were included of whom 265 (49.3%) experienced at least one BSI. Patients who developed bacteremia had a higher SOFA score [10 (8-12) vs 9 (7-10), p < 0.001], had been intubated more frequently [95.8% vs 75%, p < 0.001] and for a median longer time [16 days (9-25) vs 8 days (5-14), p < 0.001]. Patients with BSI had a median longer ICU stay [18 days (12-31.5) vs 9 days (5-15), p < 0.001] and higher mortality [54% vs 42.3%, p < 0.001] than those who did not develop it. Development of BSI resulted in a higher SOFA score [aHR 1.08 (95% CI 1.03-1.12)] and a higher Charlson score [csAHR 1.15 (95% CI 1.05-1.25)]. CONCLUSION: A high SOFA score and a high Charlson score resulted associated with BSI's development. Conversely, immunosuppressive therapy like steroids and tocilizumab, has no role in increasing the risk of bacteremia.

Predictive model for bacterial co-infection in patients hospitalized for COVID-19: a multicenter observational cohort study.

OBJECTIVE: The aim of our study was to build a predictive model able to stratify the risk of bacterial co-infection at hospitalization in patients with COVID-19. METHODS: Multicenter observational study of adult patients hospitalized from February to December 2020 with confirmed COVID-19 diagnosis. Endpoint was microbiologically documented bacterial co-infection diagnosed within 72 h from hospitalization. The cohort was randomly split into derivation and validation cohort. To investigate risk factors for co-infection univariable and multivariable logistic regression analyses were performed. Predictive risk score was obtained assigning a point value corresponding to ß-coefficients to the variables in the multivariable model. ROC analysis in the validation cohort was used to estimate prediction accuracy. RESULTS: Overall, 1733 patients were analyzed: 61.4% males, median age 69 years (IQR 57-80), median Charlson 3 (IQR 2-6). Co-infection was diagnosed in 110 (6.3%) patients. Empirical antibiotics were started in 64.2 and 59.5% of patients with and without co-infection (p = 0.35). At multivariable analysis in the derivation cohort: WBC ≥ 7.7/mm3, PCT ≥ 0.2 ng/mL, and Charlson index ≥ 5 were risk factors for bacterial co-infection. A point was assigned to each variable obtaining a predictive score ranging from 0 to 5. In the validation cohort, ROC analysis showed AUC of 0.83 (95%CI 0.75-0.90). The optimal cut-point was ≥2 with sensitivity 70.0%, specificity 75.9%, positive predictive value 16.0% and negative predictive value 97.5%. According to individual risk score, patients were classified at low (point 0), intermediate (point 1), and high risk (point ≥ 2). CURB-65 ≥ 2 was further proposed to identify patients at intermediate risk who would benefit from early antibiotic coverage. CONCLUSIONS: Our score may be useful in stratifying bacterial co-infection risk in COVID-19 hospitalized patients, optimizing diagnostic testing and antibiotic use.

Epidemiology of Invasive Pulmonary Aspergillosis Among Intubated Patients With COVID-19: A Prospective Study.

BACKGROUND: We evaluated the incidence of invasive pulmonary aspergillosis among intubated patients with critical COVID-19 and evaluated different case definitions of invasive aspergillosis. METHODS: Prospective, multicenter study in adult patients with microbiologically confirmed COVID-19 receiving mechanical ventilation. All included participants underwent a screening protocol for invasive pulmonary aspergillosis with bronchoalveolar lavage galactomannan and cultures performed on admission at 7 days and in case of clinical deterioration. Cases were classified as coronavirus-associated pulmonary aspergillosis (CAPA) according to previous consensus definitions. The new definition was compared with putative invasive pulmonary aspergillosis (PIPA). RESULTS: 108 patients were enrolled. Probable CAPA was diagnosed in 30 (27.7%) patients after a median of 4 (2-8) days from intensive care unit (ICU) admission. Kaplan-Meier curves showed a significantly higher 30-day mortality rate from ICU admission among patients with either CAPA (44% vs 19%, P = .002) or PIPA (74% vs 26%, P < .001) when compared with patients not fulfilling criteria for aspergillosis. The association between CAPA (OR, 3.53; 95% CI, 1.29-9.67; P = .014) or PIPA (OR, 11.60; 95% CI, 3.24-41.29; P < .001) with 30-day mortality from ICU admission was confirmed, even after adjustment for confounders with a logistic regression model. Among patients with CAPA receiving voriconazole treatment (13 patients; 43%) a trend toward lower mortality (46% vs 59%; P = .30) and reduction in galactomannan index in consecutive samples were observed. CONCLUSIONS: We found a high incidence of CAPA among critically ill COVID-19 patients and its occurrence seems to change the natural course of disease.

Automated orofacial virtual patient creation: A proof of concept.

OBJECTIVES: To (1) construct a virtual patient (VP) using facial scan, intraoral scan, and low-dose computed tomography scab based on an Artificial intelligence (AI)-approach, (2) quantitatively compare it with AI-refined and semi-automatic registration, and (3) qualitatively evaluate user satisfaction when using virtual patient as a communication tool in clinical practice. MATERIALS AND METHODS: A dataset of 20 facial scans, intraoral scans, and low-dose computed tomography scans was imported into the Virtual Patient Creator platform to create an automated virtual patient. The accuracy of the virtual patients created using different approaches was further analyzed in the Mimics software. The accuracy (% of corrections required), consistency, and time efficiency of the AI-driven virtual patient registration were then compared with the AI-refined and semi-automatic registration (clinical reference). User satisfaction was assessed through a survey of 35 dentists and 25 laypersons who rated the virtual patient's realism and usefulness for treatment planning and communication on a 5-point scale. RESULTS: The accuracy for AI-driven, AI-refined, and semi-automatic registration virtual patient was 85 %, 85 %, and 100 % for the upper and middle thirds of the face, and 30 %, 30 %, and 35 % for the lower third. Registration consistency was 1, 1 and 0.99, and the average time was 26.5, 30.8, and 385 s, respectively (18-fold time reduction with AI). The inferior facial third exhibited the highest registration mismatch between facial scan and computed tomography. User satisfaction with the virtual patient was consistently high among both dentists and laypersons, with most responses indicating very high satisfaction regarding realism and usefulness as a communication tool. CONCLUSION: The AI-driven registration can provide clinically accurate, fast, and consistent virtual patient creation using facial scans, intraoral scans, and low-dose computed tomography scans, enabling interpersonal communication. CLINICAL SIGNIFICANCE: Using AI for automated segmentation and registration of maxillofacial structures leads to clinically efficient and accurate VP creation, opening the doors for its widespread use in diagnosis, treatment planning, and interprofessional and professional-patient communication.

Content based image retrieval using local binary pattern operator and data mining techniques.

Content based image retrieval (CBIR) concerns the retrieval of similar images from image databases, using feature vectors extracted from images. These feature vectors globally define the visual content present in an image, defined by e.g., texture, colour, shape, and spatial relations between vectors. Herein, we propose the definition of feature vectors using the Local Binary Pattern (LBP) operator. A study was performed in order to determine the optimum LBP variant for the general definition of image feature vectors. The chosen LBP variant is then subsequently used to build an ultrasound image database, and a database with images obtained from Wireless Capsule Endoscopy. The image indexing process is optimized using data clustering techniques for images belonging to the same class. Finally, the proposed indexing method is compared to the classical indexing technique, which is nowadays widely used.

Analysis and classification of ultrasound medical images using the Local Binary Pattern operator.

This paper aims to present a classification and retrieval technique applied to ultrasound medical images, based on different variations of Local Binary Pattern (LBP) algorithm. Using this technique, a dedicated application builds an ultrasound image database, determining the optimum variation of LBP algorithm. These techniques can be applied to an image or to a group of images. Characterization is done through an array of values extracted by the algorithm. The application allows the characterization of an image, a set of images, determining the similarity between different images and the degree of belonging to a particular group. There are also presented several comparisons between existent variations of this algorithm, applied on the same set of ultrasound images.