Differentially Private Graph Neural Networks for Whole-Graph Classification.

Graph Neural Networks (GNNs) have established themselves as state-of-the-art for many machine learning applications such as the analysis of social and medical networks. Several among these datasets contain privacy-sensitive data. Machine learning with differential privacy is a promising technique to allow deriving insight from sensitive data while offering formal guarantees of privacy protection. However, the differentially private training of GNNs has so far remained under-explored due to the challenges presented by the intrinsic structural connectivity of graphs. In this work, we introduce a framework for differential private graph-level classification. Our method is applicable to graph deep learning on multi-graph datasets and relies on differentially private stochastic gradient descent (DP-SGD). We show results on a variety of datasets and evaluate the impact of different GNN architectures and training hyperparameters on model performance for differentially private graph classification, as well as the scalability of the method on a large medical dataset. Our experiments show that DP-SGD can be applied to graph classification tasks with reasonable utility losses. Furthermore, we apply explainability techniques to assess whether similar representations are learned in the private and non-private settings. Our results can also function as robust baselines for future work in this area.

Multi-omics data integration methods and their applications in psychiatric disorders.

To study mental illness and health, in the past researchers have often broken down their complexity into individual subsystems (e.g., genomics, transcriptomics, proteomics, clinical data) and explored the components independently. Technological advancements and decreasing costs of high throughput sequencing has led to an unprecedented increase in data generation. Furthermore, over the years it has become increasingly clear that these subsystems do not act in isolation but instead interact with each other to drive mental illness and health. Consequently, individual subsystems are now analysed jointly to promote a holistic understanding of the underlying biological complexity of health and disease. Complementing the increasing data availability, current research is geared towards developing novel methods that can efficiently combine the information rich multi-omics data to discover biologically meaningful biomarkers for diagnosis, treatment, and prognosis. However, clinical translation of the research is still challenging. In this review, we summarise conventional and state-of-the-art statistical and machine learning approaches for discovery of biomarker, diagnosis, as well as outcome and treatment response prediction through integrating multi-omics and clinical data. In addition, we describe the role of biological model systems and in silico multi-omics model designs in clinical translation of psychiatric research from bench to bedside. Finally, we discuss the current challenges and explore the application of multi-omics integration in future psychiatric research. The review provides a structured overview and latest updates in the field of multi-omics in psychiatry.

Privacy: An Axiomatic Approach.

The increasing prevalence of large-scale data collection in modern society represents a potential threat to individual privacy. Addressing this threat, for example through privacy-enhancing technologies (PETs), requires a rigorous definition of what exactly is being protected, that is, of privacy itself. In this work, we formulate an axiomatic definition of privacy based on quantifiable and irreducible information flows. Our definition synthesizes prior work from the domain of social science with a contemporary understanding of PETs such as differential privacy (DP). Our work highlights the fact that the inevitable difficulties of protecting privacy in practice are fundamentally information-theoretic. Moreover, it enables quantitative reasoning about PETs based on what they are protecting, thus fostering objective policy discourse about their societal implementation.

Amantadine treatment is associated with improved consciousness in patients with non-traumatic brain injury.

OBJECTIVE: This study determined the effect of amantadine treatment on consciousness in patients with non-traumatic brain injury. METHODS: We pooled individual patient data of five single-centre observational studies to determine the effect of amantadine treatment among patients with ischaemic stroke, intracerebral haemorrhage, subarachnoid haemorrhage, community-acquired bacterial meningitis and status epilepticus, admitted between January 2012 and December 2015 and ventilated ≥7 days. Patient selection and multivariable regression modelling were used to adjust for differences in intergroup comparison and for parameters associated with consciousness. Improvement of consciousness 5 days after treatment initiation was defined as primary outcome. Secondary outcomes included Glasgow Coma Scale (GCS) at day 5 and GCS at day 10, rate of ICU delirium, epileptic seizures and all-cause mortality at 90 days. RESULTS: Overall, 84 of 294 (28.6%) eligible patients received amantadine. Amantadine treatment was associated with improvement of consciousness at day 5 (amantadine: 86.9% vs control: 54.0%; absolute difference: 32.9 (20.0-44.2); adjusted OR (aOR): 5.71 (2.50-13.05), p<0.001). Secondary outcomes showed differences in GCS 5 days (9 (8-11) vs 6 (3-9), p<0.001) and GCS 10 days (10(8-11) vs 9(6-11),p=0.003) after treatment initiation. There were no significant differences regarding all-cause mortality (aOR: 0.89 (0.44-1.82), p=0.758) and ICU delirium (aOR: 1.39 (0.58-3.31), p=0.462). Rate of epileptic seizures after initiation of amantadine treatment was numerically higher in the amantadine group (amantadine: 10.7% vs control: 3.0%; absolute difference: 7.7 (0.3-16.4); aOR: 3.68 (0.86-15.71), p=0.079). CONCLUSIONS: Amantadine treatment is associated with improved consciousness among patients with different types of non-traumatic brain injury in this observational cohort analysis. Epileptic seizures should be considered as potential side effects and randomised controlled trials are needed to confirm these findings.

Alpha power decrease in quantitative EEG detects development of cerebral infarction after subarachnoid hemorrhage early.

OBJECTIVE: In subarachnoid hemorrhage (SAH), transcranial Doppler/color-coded-duplex sonography (TCD/TCCS) is used to detect delayed cerebral ischemia (DCI). In previous studies, quantitative electroencephalography (qEEG) also predicted imminent DCI. This study aimed to compare and analyse the ability of qEEG and TCD/TCCS to early identify patients who will develop later manifest cerebral infarction. METHODS: We analysed cohorts of two previous qEEG studies. Continuous six-channel-EEG with artefact rejection and a detrending procedure was applied. Alpha power decline of ≥ 40% for ≥ 5 hours compared to a 6-hour-baseline was defined as significant EEG event. Median reduction and duration of alpha power decrease in each channel was determined. Vasospasm was diagnosed by TCD/TCCS, identifying the maximum frequency and days of vasospasm in each territory. RESULTS: 34 patients were included (17 male, mean age 56 ± 11 years, Hunt and Hess grade: I-V, cerebral infarction: 9). Maximum frequencies in TCD/TCCS and alpha power reduction in qEEG were correlated (r = 0.43; p = 0.015). Patients with and without infarction significantly differed in qEEG parameters (maximum alpha power decrease: 78% vs 64%, p = 0.019; summed hours of alpha power decline: 236 hours vs 39 hours, p = 0.006) but showed no significant differences in TCD/TCCS parameters. CONCLUSIONS: There was a moderate correlation of TCD/TCCS frequencies and qEEG alpha power reduction but only qEEG differentiated between patients with and without cerebral infarction. SIGNIFICANCE: qEEG represents a non-invasive, continuous tool to identify patients at risk of cerebral infarction.

The impact of the coronavirus disease (COVID-19) pandemic on outpatient epilepsy care: An analysis of physician practices in Germany.

OBJECTIVE: To gain insight into epilepsy care during coronavirus disease (COVID-19) pandemic, we analyzed prescription data of a large cohort of persons with epilepsy (PWE) during lockdown in Germany. METHODS: Information was obtained from the Disease Analyzer database, which collects anonymous demographic and medical data from practice computer systems of general practitioners (GP) and neurologists (NL) throughout Germany. We retrospectively compared prescription data for anti-seizure medication (ASM) and physicians' notes of "known" and "new" PWE from January 2020 until May 2020 with the corresponding months in the three preceding years 2017-2019. Adherence was estimated by calculating the proportion of patients with follow-up prescriptions within 90 days after initial prescriptions in January or February. We additionally analyzed hospital referrals of PWE. The significance level was set to 0.01 to adjust for multiple comparisons. RESULTS: A total of 52,844 PWE were included. Anti-seizure medication prescriptions for known PWE increased in March 2020 (GP + 36%, NL + 29%; P < 0.01). By contrast, a decrease in prescriptions to known and new PWE was observed in April and significantly in May 2020 ranging from -16% to -29% (P < 0.01). The proportion of PWE receiving follow-up prescriptions was slightly higher in 2020 (73.5%) than in 2017-2019 (70.7%, P = 0.001). General practitioners and NL referred fewer PWE to hospitals in March 2020 (GP: -30%, P < 0.01; NL: -12%), April 2020 (GP: -29%, P < 0.01; NL: -37%), and May 2020 (GP: -24%, P < 0.01; NL: -16%). CONCLUSION: Adherence of known PWE to ASM treatment appeared to remain stable during lockdown in Germany. However, this study revealed findings which point to reduced care for newly diagnosed PWE as well as fewer hospital admissions. These elements may warrant consideration during future lockdown situations.

Manufacturer switch of anti-seizure drugs may not increase the risk of seizure recurrence in Children: A nationwide study of prescription data in Germany.

OBJECTIVE: Several publications on the exchangeability of antiepileptic drugs in clinical settings revealed an increased risk for seizure recurrence after changing the manufacturer of anti-seizure drugs (ASD) in adults, possibly due to a decline of adherence. It is unclear whether this holds true in children and adolescents. METHODS: Patient data of children and adolescents (<18 years) were collected anonymously from 236 German pediatricians and pediatric neurologists between January 2011 and December 2018 using the IMS® Disease Analyzer database (IQVIA, Frankfurt, Germany). Patients with epilepsy were included if at least 2 prescriptions within 360 days and 1 within 180 days prior to the index date were available. The cohort was separated into a seizure group and seizure-free controls. Both groups were matched 1:1 according to age, gender, insurance status, and treating pediatrician. The risk for seizure recurrence after a manufacturer switch of the same ASD at the last prescription before the index date was analyzed using a multivariate regression model. RESULTS: A total of 678 children and adolescents with epilepsy were included (each group: n = 339; age: 9.6 ±â€¯4.4 years). Comparing both groups, the risk for seizures recurrence was not increased after a manufacturer switch had occurred. Albeit changes during the last prescription before the index date had occurred more often in the seizure-free group, neither change of branded and generic products nor substances reached significance. Only change of ASD strength showed a significantly reduced odds ratio for seizures (OR 0.40, 95% CI 0.24-0.65, p < 0.001). SIGNIFICANCE: In contrast to the available evidence in adults, changing the manufacturer did not appear to increase the risk for seizure recurrence in previously seizure-free children and adolescents with epilepsy.

Anti-seizure medication is not associated with an increased risk to develop cancer in epilepsy patients.

OBJECTIVE: Whether anti-seizure medication (ASM) increases the risk for cancer has been debated for decades. While for some ASM, a carcinoma-promoting effect has been suspected, carcinoma-protective effects have been shown for other ASM. However, the issue remains unresolved as data from preclinical and clinical studies have been inconsistent and contradictory. METHODS: We collected anonymous patient data from practice neurologists throughout Germany between 2009 and 2018 using the IMS Disease Analyzer database (QuintilesIMS, Frankfurt, Germany). People with epilepsy (PWE) with an initial cancer diagnosis and antiepileptic therapy prior to the index date were 1:1 matched with a control group of PWE without cancer according to age, gender, index year, Charlson Comorbidity Index, and treating physician. For both groups, the risk to develop cancer under treatment with different ASMs was analyzed using three different models (ever use vs. never use (I), effect per one (II) and per five therapy years (III). RESULTS: A total of 3152 PWE were included (each group, n = 1,576; age = 67.3 ± 14.0 years). The risk to develop cancer was not significantly elevated for any ASM. Carbamazepine was associated with a decreased cancer risk (OR Model I: 0.699, p < .0001, OR Model II: 0.952, p = .4878, OR Model III: 0.758, p < .0004). SIGNIFICANCE: Our findings suggest that ASM use does not increase the risk of cancer in epilepsy patients.

Psychosocial long-term outcome in patients with psychogenic non-epileptic seizures.

PURPOSE: To evaluate psychosocial long-term outcome in patients diagnosed with psychogenic nonepileptic seizures (PNES) and to predict outcome of PNES, economic status, and quality of life (QoL) at follow-up. METHODS: Patients diagnosed with PNES in the video-EEG-monitoring unit at our Epilepsy center between 2002-2016 were contacted by phone 1-16 years after communicating the diagnosis. Patients underwent a structured interview asking for current PNES status, psychosocial situation (economic status, marital status, setting of living, driving), depression, and QoL. RESULTS: Of 70 PNES patients without comorbid epilepsy (age: 41.1 ± 13.5 years; 74 % female, follow-up: 5.2 ± 4.2 years), 23 patients (33 %) reported to be free of PNES during the last 12 months. Patients with cessation of PNES were younger at PNES onset (p < .01) and diagnosis (p < .01) and had a higher education (p < .05). At follow-up, the proportion of economically active patients only increased in individuals with cessation of PNES (p < .001) while an increased number of patients with persisting PNES relied on governmental support (p < .001). Cessation of PNES was associated with better mood (p < .01) and QoL (p < .001). In multiple regression models, cessation of PNES was only predicted by younger age at onset, while good economic outcome was determined by younger age and good economic status at diagnosis and cessation of PNES at follow-up. Good QoL at follow-up was predicted by low depressive symptoms, freedom of PNES, and economic activity at follow-up. CONCLUSION: Long-term outcome in patients with PNES remains to be poor and the majority of patients continue to have PNES. Cessation of PNES was associated with good economic outcome, mood, and QoL.

Ex vivo radiosensitivity is increased in non-cancer patients taking valproate.

BACKGROUND: Valproate (VPA) is a commonly prescribed antiepileptic drug for patients experiencing epileptic seizures due to brain tumors. VPA increases radiation sensitivity in various tumor cells in vitro due to complex mechanisms. This could make tumors more vulnerable to ionizing radiation or overcome radioresistance. Yet, clinical data on possible improvement of tumor control by adding VPA to tumor therapy is controversial. Potentially radiosensitizing effects of VPA on healthy tissue remain unclear. To determine individual radiosensitivity, we analyzed blood samples of individuals taking VPA. METHODS: Ex vivo irradiated blood samples of 31 adult individuals with epilepsy were studied using 3-color fluorescence in situ hybridization. Aberrations in chromosomes 1, 2 and 4 were analyzed. Radiosensitivity was determined by the mean breaks per metaphase (B/M) and compared to age-matched (2:1) healthy donors. RESULTS: The patient cohort (n = 31; female: 38.7%) showed an increase of their average B/M value compared to healthy individuals (n = 61; female: 56.9%; B/M: 0.480 ± 0.09 vs. 0.415 ± 0.07; p = .001). The portion of radiosensitive (B/M >  0.500) and distinctly radiosensitive individuals (B/M >  0.600) was increased in the VPA group (54.9% vs. 11.3 and 9.7% vs. 0.0%; p < .001). In 3/31 patients, radiosensitivity was determined prior to and after VPA treatment and radiosensitivity was increased by VPA-treatment. CONCLUSIONS: In our study, we confirmed that patients treated with VPA had an increased radiosensitivity compared to the control group. This could be considered in patients taking VPA prior to the beginning of radiotherapy to avoid toxic side effects of VPA-treatment.

Influence of new versus traditional antiepileptic drugs on course and outcome of status epilepticus.

PURPOSE: New antiepileptic drugs (AEDs) are increasingly applied in second-line therapy of status epilepticus (SE). In our study, we analyzed the impact of the choice of second-line AEDs on the course and prognosis of SE. METHODS: This retrospective single- center study used data of an 8 year cohort of SE in adults from 2007 to 2014. Based on the year of market introduction with a cutoff at 1990, we classified AEDs as traditional or new. Prescription pattern associated differences in prognosis were measured through univariate and multivariable analysis of 3 endpoints: occurrence of refractory SE (RSE), functional outcome in survivors to discharge (good: mRS at discharge <3 or identical to admission mRS; otherwise poor), and in-hospital mortality. RESULTS: From 362 SE episodes during the study period, 222 episodes were included into the study, among those 150 episodes treated with new and 72 with traditional AEDs. Use of new AEDs increased during the study period. After adjustment for confounders, treatment with new AEDs was on the one hand associated with higher rate of RSE occurrence (OR 1.95, 95 % CI 1.05-3.62, p = 0.03), but, on the other hand, also with better functional outcome at discharge (OR 2.64, 95 % CI 1.16-6.00, p = 0.02), while it was not an independent predictor of in- hospital mortality (OR 0.88, 95 % CI 0.33-2.33, p = 0.80). CONCLUSION: Our observation that new AEDs may be associated with a higher rate of RSE development and relatively better functional outcome when adjusted for the premorbid mRS needs confirmation in prospective studies.

First clinical postmarketing experiences in the treatment of epilepsies with brivaracetam: a retrospective observational multicentre study.

OBJECTIVES: Brivaracetam (BRV) is the latest approved antiepileptic drug and acts as a synaptic vesicle protein 2A ligand. The aim of the present study was to evaluate the efficacy and tolerability of BRV in the clinical setting. DESIGN: Retrospective, observational multicentre study. SETTING: We retrospectively collected clinical data of patients who received BRV in 10 epilepsy centres using a questionnaire that was answered by the reporting neurologist. PARTICIPANTS: Data of 615 epilepsy patients treated with BRV were included in the study. PRIMARY AND SECONDARY OUTCOME MEASURES: Efficacy regarding seizure frequency and tolerability of BRV were evaluated. Descriptive statistics complemented by X2 contingency tests and effect sizes were performed. RESULTS: Overall, 44% of the patients had a decreased, 38% a stable and 18% an increased seizure frequency. 17% of patients achieved seizure freedom after initiation of BRV. The seizure frequency decreased in 63% of 19 patients with BRV monotherapy. 27% reported adverse effects, but only 10% of patients with monotherapy. Brivaracetam was significantly more often associated with decreased seizure frequency in levetiracetam (LEV) naïve patients (p=0.012), but BRV also led to a decreased seizure frequency in 42% of patients who had been treated with LEV before, including 17% of patients who were completely seizure free. Adverse effects under LEV improved in 62% and deteriorated in 2% of patients after the switch to BRV. At latest follow-up (mean±SD = 26.3±6.5 months), 68% were still on BRV. CONCLUSIONS: The present study shows that results of the phase III studies on BRV match data from real life clinical settings. Brivaracetam seems to be a useful alternative in patients who have suffered adverse effects while taking LEV.

Comparison of scoring tools for the prediction of in-hospital mortality in status epilepticus.

PURPOSE: Several scoring tools have been developed for the prognostication of outcome after status epilepticus (SE). In this study, we compared the performances of STESS (Status Epilepticus Severity Score), mSTESS (modified STESS), EMSE-EAL (Epidemiology-based Mortality Score in Status Epilepticus- Etiology, Age, Level of Consciousness) and END-IT (Encephalitis-NCSE-Diazepam resistance-Image abnormalities-Tracheal intubation) in predicting in-hospital mortality after SE. METHOD: Data collected retrospectively from a cohort of 287 patients with SE were used to calculate STESS, mSTESS, EMSE-EAL, and END-IT scores. The differences between the scores' performances were determined by means of area under the ROC curve (AUC) comparisons and McNemar testing. RESULTS: The in-hospital mortality rate was 11.8%. The AUC of STESS (0.628; 95% confidence interval (CI), 0.529-0.727) was similar to that of mSTESS (0.620; 95% CI, 0.510-0.731), EMSE-EAL (0.556; 95% CI, 0.446-0.665), and END-IT (0.659; 95% CI, 0.550-0.768; p > .05 for each comparison) in predicting in-hospital mortality. STESS with a cutoff of 3 was found to have lowest specificity and number of correctly classified episodes. EMSE-EAL with a cutoff at 40 had highest specificity and showed a trend towards more correctly classified episodes while sensitivity tended to be low. END-IT with a cutoff of 3 had the most balanced sensitivity-specificity ratio. CONCLUSIONS: EMSE-EAL is as easy to calculate as STESS and tended towards higher diagnostic accuracy. Adding information on premorbid functional status to STESS did not enhance outcome prediction. END-IT was not superior to other scores in prediction of in-hospital mortality despite including information of diagnostic work-up and response to initial treatment.