Search

The adaptation of deep learning models within safety-critical systems cannot rely only on good prediction performance but needs to provide interpretable and robust explanations for their decisions. When modeling complex sequences, attention mechanisms are regarded as the established approach to support deep neural networks with intrinsic interpretability. This paper focuses on the emerging trend of specifically designing diagnostic datasets for understanding the inner workings of attention mechanism based deep learning models for multivariate forecasting tasks. We design a novel benchmark of synthetically designed datasets with the transparent underlying generating process of multiple time series interactions with increasing complexity. The benchmark enables empirical evaluation of the performance of attention based deep neural networks in three different aspects: (i) prediction performance score, (ii) interpretability correctness, (iii) sensitivity analysis. Our analysis shows that although most models have satisfying and stable prediction performance results, they often fail to give correct interpretability. The only model with both a satisfying performance score and correct interpretability is IMV-LSTM, capturing both autocorrelations and crosscorrelations between multiple time series. Interestingly, while evaluating IMV-LSTM on simulated data from statistical and mechanistic models, the correctness of interpretability increases with more complex datasets.

Aims: To estimate prevalence of diagnosed (dDM) and undiagnosed diabetes (uDM) in Hungary and investigate determinants of uDM., Methods: Data was obtained from the nationally representative H-UNCOVER study. As laboratory measurements were available for 11/19 Hungarian counties, n = 5,974/17,787 people were eligible. After exclusions, 5,673 (representing 4,976,097 people) were included. dDM was defined by self-reporting, while uDM as negative self-reporting and elevated fasting glucose (≥7 mmol/l) and/or HbA1c (≥48 mmol/mol). Logistic regression for complex samples was used to calculate comparisons between dDM and uDM adjusted for age and BMI., Results: Diabetes prevalence was 12.0 %/11.9 % (women/men, 95 %CI:10.7-13.4 %/10.7-13.2 %), while 2.2 %/2.8 % (1.7-2.8 %/2.2-3.6 %) of women/men were uDM. While the proportion of uDM vs. dDM was similar for women ≥ 40, men in their forties had the highest odds for uDM. Neither unemployment (women/men OR:0.58 [0.14-2.45]/0.50 [0.13-1.92]), nor education level (tertiary vs. primary; women/men OR: 1.16 [0.53-2.56]/ 0.53 [0.24-1.18]) were associated with uDM. The risk of uDM was lower in both sexes with chronic morbidities., Conclusions: We report higher prevalence of diabetes and undiagnosed diabetes than previous Hungarian estimates. The finding that socioeconomic factors are not associated to uDM suggests that universal health care could provide equitable access to diabetes diagnosis., Competing Interests: Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper., (Copyright © 2024 The Authors. Published by Elsevier B.V. All rights reserved.)

Background: The increasing birthweight trend stopped and even reversed in several high income countries in the last 20 years, however the reason for these changes is not well characterized. We aimed to describe birthweight trends of term deliveries in Hungary between 1999 and 2018 and to investigate potential maternal and foetal variables that could drive these changes., Methods: We analysed data from the Hungarian Tauffer registry, a compulsory anonymized data collection of each delivery. We included all singleton term deliveries in 1999-2018 (n = 1,591,932). We modelled birthweight trends separately in 1999-2008 and 2008-2018 in hierarchical multiple linear regression models adjusted for calendar year, newborn sex, maternal age, gestational age at delivery, and other important determinants., Results: Median birthweights increased from 3250/3400 g (girl/boy) to 3300/3440 g from 1999 to 2008 and decreased to 3260/3400 g in 2018. When we adjusted for gestational age at delivery the increase in the first period became more pronounced (5.4 g/year). During the second period, similar adjustment substantially decreased the rate of decline from 2.5 to 1.4 g/year. Further adjustment for maternal age halved the rate of increase to 2.4 g/year in the first period. During the second period, adjustment for maternal age had little effect on the estimate., Conclusions: Our findings of an increasing birthweight trend (mostly related to the aging of the mothers) in 1999-2008 may forecast an increased risk of cardiometabolic diseases in offsprings born in this period. In contrast, the decreasing birthweight trends after 2008 may reflect some beneficial effects on perinatal morbidity. However, the long-term effect cannot be predicted, as the trend is mostly explained by the shorter pregnancies., (© 2024. The Author(s).)

The high mortality of patients with coronavirus disease 2019 (COVID-19) is effectively reduced by vaccination. However, the effect of vaccination on mortality among hospitalised patients is under-researched. Thus, we investigated the effect of a full primary or an additional booster vaccination on in-hospital mortality among patients hospitalised with COVID-19 during the delta wave of the pandemic. This retrospective cohort included all patients (n = 430) admitted with COVID-19 at Semmelweis University Department of Medicine and Oncology in 01/OCT/2021-15/DEC/2021. Logistic regression models were built with COVID-19-associated in-hospital/30 day-mortality as outcome with hierarchical entry of predictors of vaccination, vaccination status, measures of disease severity, and chronic comorbidities. Deceased COVID-19 patients were older and presented more frequently with cardiac complications, chronic kidney disease, and active malignancy, as well as higher levels of inflammatory markers, serum creatinine, and lower albumin compared to surviving patients (all p < 0.05). However, the rates of vaccination were similar (52-55%) in both groups. Based on the fully adjusted model, there was a linear decrease of mortality from no/incomplete vaccination (ref) through full primary (OR 0.69, 95% CI: 0.39-1.23) to booster vaccination (OR 0.31, 95% CI 0.13-0.72, p = 0.006). Although unadjusted mortality was similar among vaccinated and unvaccinated patients, this was explained by differences in comorbidities and disease severity. In adjusted models, a full primary and especially a booster vaccination improved survival of patients hospitalised with COVID-19 during the delta wave of the pandemic. Our findings may improve the quality of patient provider discussions at the time of admission., (© 2023. The Author(s).)

Aims: We compared pregnancy outcomes of untreated 'mild' GDM (GDM by WHO 2013 but not by WHO-1999) to normal glucose tolerant women (NGT)., Methods: In a universal screening program 4333 pregnant women had a 3-point 75 g OGTT in Hungary in 2009-2013. By WHO-2013 untreated NGT was diagnosed in n = 3303, 'mild' GDM in n = 336 cases., Results: 'Mild' GDM women were older (mean difference, SE: 1.4, 0.3 yrs), had higher fasting (1.0, 0.02), 60-minute (1.0, 0.09), and 120-minute (0.4, 0.06 mmol/l) blood glucose, and blood pressure (2.6, 0.5/2.0, 0.5 mmHg). Weight gain was similar in both groups (-0.3, 0.3 kg). GDM newborns were heavier (142, 50 g) and were more frequently macrosomic (>4000 g, OR 1.85, 95 %CI 1.35-2.54). Hypertension during pregnancy was more prevalent in the GDM group (OR 1.55, 95 %CI 1.05-2.28), as well as induced (OR 1.38, 95 %CI 1.10-1.74) and instrumental delivery (OR 1.34, 95 %CI 1.07-1.68), and acute caesarean section (OR 1.32, 95 %CI 1.04-1.64). Most of these differences substantially attenuated or became non-significant after adjustment for pre-pregnancy BMI., Conclusions: Pregnancy outcomes of 'mild' GDM were worse compared to normal glucose tolerant women however these differences were explained by the pre-pregnancy BMI difference between groups., Competing Interests: Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper., (Copyright © 2023 The Authors. Published by Elsevier B.V. All rights reserved.)

Introduction: In 1989, the St Vincent declaration aimed to approximate pregnancy outcomes of diabetes to that of healthy pregnancies. We aimed to compare frequency and trends of outcomes of pregnancies affected by type 1 diabetes and controls in 1996-2018., Methods: We used anonymized records of a mandatory nation-wide registry of all deliveries between gestational weeks 24 and 42 in Hungary. We included all singleton births (4,091 type 1 diabetes, 1,879,183 controls) between 1996 and 2018. We compared frequency and trends of pregnancy outcomes between type 1 diabetes and control pregnancies using hierarchical Poisson regression., Results: The frequency of stillbirth, perinatal mortality, large for gestational age, caesarean section, admission to neonatal intensive care unit (NICU), and low Appearance, Pulse, Grimace, Activity, and Respiration (APGAR) score was 2-4 times higher in type 1 diabetes compared to controls, while the risk of congenital malformations was increased by 51% and SGA was decreased by 42% (all p<0.05). These observations remained significant after adjustment for confounders except for low APGAR scores. We found decreasing rate ratios comparing cases and controls over time for caesarean sections, low APGAR scores (p<0.05), and for NICU admissions (p=0.052) in adjusted models. The difference between cases and controls became non-significant after 2009. No linear trends were observed for the other outcomes., Conclusions: Although we found that the rates of SGA, NICU care, and low APGAR score improved in pregnancies complicated by type 1 diabetes, the target of the St Vincent Declaration was only achieved for the occurrence of low APGAR scores., Competing Interests: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest., (Copyright © 2023 Fazekas-Pongor, Svébis, Major, Pártos, Dósa, Mészáros, Horváth, Domján, Zsirai and Tabák.)

Objectives: Our aim in this study was to compare the efficacy and safety of commercially available fixed-ratio combinations (FRCs) of glucagon-like peptide-1 receptor agonists (GLP-1RAs) and basal insulins by a network meta-analysis of randomized controlled trials (RCTs) of people with type 2 diabetes., Methods: We present a systematic review and network meta-analyses of RCTs of individuals with type 2 diabetes randomized to FRCs or to their components for ≥24 weeks. All reports were obtained from PubMed or ClinicalTrials.gov up to February 28, 2022. The primary outcome was glycated hemoglobin (A1C) level attained. Secondary outcomes included fasting plasma glucose, change in body weight, and incident hypoglycemia. Treatment effects were estimated as mean difference (MD) and standard error (SE), or as odds ratio (OR) with 95% confidence interval (CI) using the fixed combination of insulin glargine 100 IU/mL and lixisenatide (iGlarLixi) as reference., Results: We included 29 RCTs from among the 1,404 articles identified. No direct comparisons between FRCs were found. After excluding some insulin-capped trials to reach model consistency, both FRCs were more efficacious regarding A1C than their components, but no difference between FRCs was found (MD, -0.10%; SE, 0.10%). The effect of the fixed combination of insulin degludec and liraglutide (IDegLira) (MD, -0.47 mmol/L; SE, 0.24 mmol/L) and basal insulins was similar to that of iGlarLixi (reference) on fasting glucose, whereas GLP-1RAs had lower efficacy than iGlarLixi. Weight gain was lower with GLP-1RAs and IDegLira (MD, -0.72 kg; SE, 0.32 kg) than with iGlarLixi (reference) and higher with basal insulins. Incident hypoglycemia (based on different definitions) was least frequent with GLP-1RAs, followed by IDegLira (OR, 0.78; 95% CI, 0.39 to 1.57), iGlarLixi (reference), and basal insulins., Conclusions: For A1C, both FRCs were more efficacious over their individual components, with similar efficacies of the 2 FRCs., (Copyright © 2023 The Author(s). Published by Elsevier Inc. All rights reserved.)

Background: People with diabetic cardiovascular autonomic neuropathy (CAN) have increased cardiovascular mortality. However, the association between distal symmetric polyneuropathy (DSPN) or CAN with all-cause mortality is much less investigated. Thus, we set out to examine the effect of CAN and DSPN on all-cause mortality in a well-phenotyped cohort., Methods: All diabetes cases ( n = 1,347) from the catchment area of a secondary diabetes care centre who had medical examination including neuropathy assessment between 1997 and 2016 were followed up for all-cause mortality in the NHS Hungary reimbursement database until 2018. We investigated the association of CAN (Ewing tests) and DSPN (Neurometer) with all-cause mortality using Cox models stratified by diabetes type., Results: Altogether, n = 131/1,011 persons with type 1/type 2 diabetes were included. Of the participants, 53%/43% were male, mean age was 46 ± 12/64 ± 10 years, diabetes duration was 13 ± 10/7 ± 8 years, 42%/29% had CAN, and 39%/37% had DSPN. During the 9 ± 5/8 ± 5-year follow-up, n = 28/494 participants died. In fully adjusted models, participants with type 1 diabetes patients with versus without DSPN had an increased mortality (HR 2.99, 95% CI 1.4-8.63), while no association with CAN was observed. In type 2 diabetes, both DSPN and CAN independently increased mortality (HR 1.32, 95% CI: 1.07-1.64, and HR 1.44, 95% CI: 1.17-1.76)., Conclusions: Our results are compatible with an increased risk of mortality in people with type 1 diabetes and DSPN. Furthermore, we report a similarly strong association between DSPN and CAN and all-cause mortality in type 2 diabetes mellitus., Competing Interests: The authors declare that that they have no competing interests., (Copyright © 2021 Orsolya E. Vági et al.)

The distinction between severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2)-related and community-acquired pneumonias poses significant difficulties, as both frequently involve the elderly. This study aimed to predict the risk of SARS-CoV-2-related pneumonia based on clinical characteristics at hospital presentation. Case-control study of all patients admitted for pneumonia at Semmelweis University Emergency Department. Cases (n = 30) were patients diagnosed with SARS-CoV-2-related pneumonia (based on polymerase chain reaction test) between 26 March 2020 and 30 April 2020; controls (n = 82) were historical pneumonia cases between 1 January 2019 and 30 April 2019. Logistic models were built with SARS-CoV-2 infection as outcome using clinical characteristics at presentation. Patients with SARS-CoV-2-related pneumonia were younger (mean difference, 95% CI: 9.3, 3.2-15.5 years) and had a higher lymphocyte count, lower C-reactive protein, presented more frequently with bilateral infiltrate, less frequently with abdominal pain, diarrhoea, and nausea in age- and sex-adjusted models. A logistic model using age, sex, abdominal pain, C-reactive protein, and the presence of bilateral infiltrate as predictors had an excellent discrimination (AUC 0.88, 95% CI: 0.81-0.96) and calibration (p = 0.27-Hosmer-Lemeshow test). The clinical use of our screening prediction model could improve the discrimination of SARS-CoV-2 related from other community-acquired pneumonias and thus help patient triage based on commonly used diagnostic approaches. However, external validation in independent datasets is required before its clinical use.

Aims: To estimate and compare the prevalence of self-reported diabetes based on nationally representative surveys of the Hungarian adult population in 2002 (published data - Hungarostudy) and a survey in 2012., Methods: A cross-sectional computer-assisted telephone interview survey on a stratified representative sample of community-dwelling adults (n=1000) in 2012. To describe self-reported diabetes prevalence and its temporal changes generalized linear models were used and results were compared to figures from Hungarostudy., Results: Age standardized prevalence of self-reported type 2 diabetes was 11.7% (95%CI 10.0-13.8%) without gender or rural-urban differences in 2012. People with self-reported diabetes were older than controls (mean [SE]: 63.9 [0.9] vs. 45.9 [0.3] years, p<0.0001). The prevalence of diabetes sharply increased after 40 years of age and peaked at age 70 (27.7% [2.5], p age*age <0.0001). The prevalence of self-reported diabetes increased by 89% (OR 1.89, 95%CI 1.53-2.32) from 6.2 to 11.7% between the two surveys with the most pronounced increase in the age group 55-64 years (from 11.6 to 24.4%)., Conclusions: We reported an alarming increase in the prevalence of self-reported type 2 diabetes in the last decade that mostly affects working age people. If this trend continues, a major public health crisis in Hungary can be envisaged., (Copyright © 2016 Primary Care Diabetes Europe. Published by Elsevier Ltd. All rights reserved.)

Introduction: Oxidative-nitrative stress and poly(ADP-ribose) polymerase activation observed in gestational diabetes may play role in the increased cardiovascular risk in later life., Aim: The present study aimed to examine the influence of the severity of previous gestational diabetes (insulin need) on vascular function three years after delivery. Furthermore, the authors investigated the relation of vascular function with oxidative-nitrative stress and poly(ADP-ribose) polymerase activation., Method: Macrovascular function was measured by applanation tonometry; microvascular reactivity was assessed by provocation tests during Laser-Doppler flowmetry in 40 women who had gestational diabetes 3 years before the study. Oxidative-nitrative stress and poly(ADP-ribose) polymerase activity in blood components were determined by colorimetry and immunohistochemistry., Results: Three years after insulin treated gestational diabetes impaired microvascular function and increased oxidative stress was observed compared to mild cases., Conclusions: The severity of previous gestational diabetes affects microvascular dysfunction that is accompanied by elevated oxidative stress. Nitrative stress and poly(ADP-ribose) polymerase activity correlates with certain vascular factors not related to the severity of the disease.