Federated learning for describing COVID-19 patients and hospital outcomes: An unCoVer analysis

Abstract Background Since the onset of the pandemic, the unCoVer network has been identifying real-world data from EMR of hospitalised patients with COVID-19 across countries. These heterogeneous data are integrated into a multi-user data repository operated through Opal/DataSHIELD, an interoperable open-source server application, providing privacy-preserving access to individual-level information for federated data analyses. Methods unCoVer’s federated data platform provided access to EMR collected between 02/2020 - 04/2022 from 6 hospitals in Bosnia and Herzegovina (1), Romania (2), Spain (2), and Turkey (1) for a total of 14,236 patients. Demographics, and co-morbidities at admission, length of hospital stay and intensive care (ICU) needs, are presented according to the patients’ status at discharge. Results A total of 11,248 (79.0%) of all patients reviewed recovered from COVID-19 after an average 11.5 (SD 10.8) days hospitalised, with only 4.09% of patients needing ICU. A smaller proportion of patients were transferred (5.93%), and 2143 (15.1%) were considered in-hospital deaths after an average 11.6 (SD 10.5) days in the hospital where most (81.2%) needed ICU. Recovered patients had a mean age of 57.7 (SD 16.3) years old, and gender neutral (51.2% men), in contrast to deceased patients that were 74.2 (SD 12.4) years old (59.7% men). Current smoking was infrequent for both recovered or deceased patients (3.27%, and 2.83%, respectively). Cardiometabolic conditions were less commonly reported among later recovered patients in comparison with deceased patients: obesity (10.7% vs 12.1%), diabetes (15.9% vs 27.4%), hypertension (23.2% vs 42.7%), and CVD (9.33% vs 44.9%). Chronic pulmonary disease was also more frequent among deceased patients (10.3% vs 18.1%). Conclusions Characteristics of hospitalised COVID-19 patients differ according to outcomes at discharge with more in-hospital death reported among older, chronic patients across 6 hospitals in 4 countries. Key messages • Federated analyses provide unique opportunities for robust results by privacy-preserving accessing individual-level data from heterogeneous data sources. • The unCoVer network aims to demonstrate the usability of the infrastructure to address research questions related to the COVID-19 while extending the concept to other clinical areas.


Background:
Thoracoabdominal aortic aneurysms (TAAAs) are defined as those aortic aneurysms involving renovisceral arteries. They account for around 10% of aortic aneurysms, and their treatment is burdened by considerable mortality and morbidity. Open surgical repair has been practised as the standard of care since the 1950s. In 2001 endovascular repair was introduced to reduce treatment invasiveness, and the technology is still evolving. The potential benefits of endovascular repair over open surgery should be carefully weighed in a multidimensional framework.

Methods:
We applied the Health Technology Assessment (HTA) framework (EUnetHTA core model with 9 dimensions) to conduct a report comparing the two technologies. A multidisciplinary working group was established. We derived and pooled: i) secondary data derived from systematic literature reviews, and ii) original data from IRCCS San Raffaele Hospital, Milan, a national referral centre for TAAA (qualitative and economic data). Results: Endovascular repair resulted superior to the traditional open surgery in terms of efficacy and safety, as justified by the metaanalysis we performed. Despite the higher costs, a significant impact on budget and slightly lower cost-effectiveness, the endovascular protheses' adoption could provide conspicuous benefits in terms of social and ethical dimensions without affecting long-term organisational and legal aspects.

Conclusions:
The multi-criteria decision analysis carried out from a hospital point of view shows that there is no significant difference (

Background:
Since the onset of the pandemic, the unCoVer network has been identifying real-world data from EMR of hospitalised patients with COVID-19 across countries. These heterogeneous data are integrated into a multi-user data repository operated through Opal/DataSHIELD, an interoperable opensource server application, providing privacy-preserving access to individual-level information for federated data analyses. Methods: unCoVer's federated data platform provided access to EMR collected between 02/2020 -04/2022 from 6 hospitals in Bosnia and Herzegovina (1), Romania (2), Spain (2), and Turkey (1)

Introduction:
Currently, the consequences of social networks on the selfesteem of users have been demonstrated in dissemination sources, and can generate problems in their mental and physical health. Objective: To determine the association between self-esteem and addiction to social networks in university students.

Methods:
A cross-sectional, prospective study was carried out between January and April 2022, including male and female university students, aged between 18 and 26 years, excluding participants with already diagnosed psychological disorder, such as depression, anxiety and body dysmorphia. A survey was applied through Google Forms, including the Rosenberg selfesteem scale and the social network addiction test (Cronbach's alpha 0.91). SPSS v22 software was used for data analysis, X2 test with Odds Ratio (OR) and 95% confidence interval (95%CI), assigning statistical significance with p < 0.05.

Results:
A total of 407 students were included, with moderate social network addiction in 57.4% and severe in 8.1%, with low selfesteem in 22.1% and moderate in 50.6%. Values of p < 0.05 were obtained for low self-esteem (OR/95%CI) when the most used social network is WhatsApp (0.5/0.2-0.8) and Instagram (1.8/108-3.2), as well as when having a number of followers between 1501-2000 (0.2/0.1-0.7) and 2501-3000 (10.8/1.1-106). Regarding network addiction, studying a humanities degree showed OR of 12(1.6-88.1). The rest of social networks such as Facebook, tik-tok, twitter, youtube, among others, showed values of p > 0.05 for both addiction to social networks and low self-esteem.

Conclusions:
Instagram is identified as a social network that facilitates the presence of low self-esteem in its users, while addiction to social networks is not associated with self-esteem. The number of followers is an associated factor for low self-esteem, and may be preventive or risky according to their total. Key messages: We must propose the legislation of social networks, requesting the guarantee of studies for the algorithms that compose them in order to avoid harm to users from them. Work with students to have mental health and emotional intelligence should be one of the aspects to be covered by universities, as part of the comprehensive care for students.

DM Health at work, Social security and social welfare
Background: We examined prospective associations between atypical working hours, substance use and sugar and fat consumption.

Methods:
In the French population-based CONSTANCES cohort, 47,288 men and 53,324 women currently employed included between 2012 and 2017 were annually followed for tobacco and cannabis use; among them, 35,647 men and 39,767 women included between 2012 and 2016 were also followed for alcohol and sugar and fat consumption. Three indicators of atypical working hours were self-reported at baseline: working at night, weekend work and non-fixed working hours. Generalized linear models computed odds of substance use and sugar and fat consumption at follow-up according to atypical working hours at baseline while adjusting for sociodemographic factors, depression and baseline substance use if appropriate.