A Transformer-Based Model Trained on Large Scale Claims Data for Prediction of Severe COVID-19 Disease Progression.

In situations like the COVID-19 pandemic, healthcare systems are under enormous pressure as they can rapidly collapse under the burden of the crisis. Machine learning (ML) based risk models could lift the burden by identifying patients with a high risk of severe disease progression. Electronic Health Records (EHRs) provide crucial sources of information to develop these models because they rely on routinely collected healthcare data. However, EHR data is challenging for training ML models because it contains irregularly timestamped diagnosis, prescription, and procedure codes. For such data, transformer-based models are promising. We extended the previously published Med-BERT model by including age, sex, medications, quantitative clinical measures, and state information. After pre-training on approximately 988 million EHRs from 3.5 million patients, we developed models to predict Acute Respiratory Manifestations (ARM) risk using the medical history of 80,211 COVID-19 patients. Compared to Random Forests, XGBoost, and RETAIN, our transformer-based models more accurately forecast the risk of developing ARM after COVID-19 infection. We used Integrated Gradients and Bayesian networks to understand the link between the essential features of our model. Finally, we evaluated adapting our model to Austrian in-patient data. Our study highlights the promise of predictive transformer-based models for precision medicine.

Common data model for COVID-19 datasets.

MOTIVATION: A global medical crisis like the coronavirus disease 2019 (COVID-19) pandemic requires interdisciplinary and highly collaborative research from all over the world. One of the key challenges for collaborative research is a lack of interoperability among various heterogeneous data sources. Interoperability, standardization and mapping of datasets are necessary for data analysis and applications in advanced algorithms such as developing personalized risk prediction modeling. RESULTS: To ensure the interoperability and compatibility among COVID-19 datasets, we present here a common data model (CDM) which has been built from 11 different COVID-19 datasets from various geographical locations. The current version of the CDM holds 4639 data variables related to COVID-19 such as basic patient information (age, biological sex and diagnosis) as well as disease-specific data variables, for example, Anosmia and Dyspnea. Each of the data variables in the data model is associated with specific data types, variable mappings, value ranges, data units and data encodings that could be used for standardizing any dataset. Moreover, the compatibility with established data standards like OMOP and FHIR makes the CDM a well-designed CDM for COVID-19 data interoperability. AVAILABILITY AND IMPLEMENTATION: The CDM is available in a public repo here: https://github.com/Fraunhofer-SCAI-Applied-Semantics/COVID-19-Global-Model. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Do P2Y12 receptor inhibitors prescribed poststroke modify the risk of cognitive disorder or dementia? Protocol for a target trial using multiple national Swedish registries.

INTRODUCTION: The target of a class of antiplatelet medicines, P2Y12R inhibitors, exists both on platelets and on brain immune cells (microglia). This protocol aims to describe a causal (based on a counterfactual model) approach for analysing whether P2Y12R inhibitors prescribed for secondary prevention poststroke may increase the risk of cognitive disorder or dementia via their actions on microglia, using real-world evidence. METHODS AND ANALYSIS: This will be a cohort study nested within the Swedish National Health and Medical Registers, including all people with incident stroke from 2006 to 2016. We developed directed acyclic graphs to operationalise the causal research question considering potential time-independent and time-dependent confounding, using input from several experts. We developed a study protocol following the components of the target trial approach described by Hernan et al and describe the data structure that would be required in order to make a causal inference. We also describe the statistical approach required to derive the causal estimand associated with this important clinical question; that is, a time-to-event analysis for the development of cognitive disorder or dementia at 1, 2 and 5-year follow-up, based on approaches for competing events to account for the risk of all-cause mortality. Causal effect estimates and the precision in these estimates will be quantified. ETHICS AND DISSEMINATION: This study has been approved by the Ethics Committee of the University of Gothenburg and Confidentiality Clearance at Statistics Sweden with Dnr 937-18, and an approved addendum with Dnr 2019-0157. The analysis and interpretation of the results will be heavily reliant on the structure, quality and potential for bias of the databases used. When we implement the protocol, we will consider and document any biases specific to the dataset and conduct appropriate sensitivity analyses. Findings will be disseminated to local stakeholders via conferences, and published in appropriate scientific journals.

Corrigendum to "Machine Learning Based Prediction of COVID-19 Mortality Suggests Repositioning of Anticancer Drug for Treating Severe Cases"[Artificial Intelligence in Life Sciences] 1(2021), 100020.

[This corrects the article DOI: 10.1016/j.ailsci.2021.100020.].

Mortality in hospitalized COVID-19 patients was associated with the COVID-19 admission rate during the first year of the pandemic in Sweden.

INTRODUCTION: Studies from the first pandemic wave found associations between COVID-19 hospital load and mortality. Here, we aimed to study if mortality of hospitalized COVID-19 patients was associated with the COVID-19 admission rate during a full year of the pandemic in Sweden. METHOD: Observational review of all patients admitted to hospital with COVID-19 in Sweden between March 2020 and February 2021 (n = 42,017). Primary outcome was 60-day all-cause mortality related to number of COVID-19 hospital admissions per month/100,000 inhabitants. Poisson regression was used to estimate the relative risk for death by month of admission, adjusting for pre-existing factors. RESULTS: The overall mortality was 17.4%. Excluding March 2020, mortality was clearly correlated to the number of COVID-19 admissions per month (coefficient of correlation ρ=.96; p<.0001). After adjustment for pre-existing factors, the correlation remained significant (ρ=.75, p=.02). Patients admitted in December (high admission rate and high mortality) had more comorbidities and longer hospital stays, and patients treated in intensive care units (ICU) had longer pre-ICU hospital stays and worse respiratory status on ICU admission than those admitted in July to September (low admission rate and low mortality). CONCLUSION: Mortality in hospitalized COVID-19 patients was clearly associated with the COVID-19 admission rate. Admission of healthier patients between pandemic waves and delayed ICU care during wave peaks could contribute to this pattern. The study supports measures to flatten-the-curve to reduce the number of COVID-19 patients admitted to hospital.

An Explainable Multimodal Neural Network Architecture for Predicting Epilepsy Comorbidities Based on Administrative Claims Data.

Epilepsy is a complex brain disorder characterized by repetitive seizure events. Epilepsy patients often suffer from various and severe physical and psychological comorbidities (e.g., anxiety, migraine, and stroke). While general comorbidity prevalences and incidences can be estimated from epidemiological data, such an approach does not take into account that actual patient-specific risks can depend on various individual factors, including medication. This motivates to develop a machine learning approach for predicting risks of future comorbidities for individual epilepsy patients. In this work, we use inpatient and outpatient administrative health claims data of around 19,500 U.S. epilepsy patients. We suggest a dedicated multimodal neural network architecture (Deep personalized LOngitudinal convolutional RIsk model-DeepLORI) to predict the time-dependent risk of six common comorbidities of epilepsy patients. We demonstrate superior performance of DeepLORI in a comparison with several existing methods. Moreover, we show that DeepLORI-based predictions can be interpreted on the level of individual patients. Using a game theoretic approach, we identify relevant features in DeepLORI models and demonstrate that model predictions are explainable in light of existing knowledge about the disease. Finally, we validate the model on independent data from around 97,000 patients, showing good generalization and stable prediction performance over time.

Mortality trends among hospitalised COVID-19 patients in Sweden: A nationwide observational cohort study.

BACKGROUND: It is important to know if mortality among hospitalised COVID-19 patients has changed as the pandemic has progressed. The aim of this study was to describe the dynamics over time of mortality among patients hospitalised for COVID-19 in Sweden, using nationwide data compiled by the Swedish National Board of Health and Welfare. METHODS: Observational cohort study where all patients hospitalised in Sweden between March 1 and September 30, 2020, with SARS-CoV-2 RNA positivity 14 days before to 5 days after admission and a discharge code for COVID-19 were included. Outcome was 60-day all-cause mortality. Patients were categorised according to month of hospital admission. Poisson regression was used to estimate the relative risk of death by month of admission, adjusting for, age, sex, comorbidities, care dependency, country of birth, healthcare region, and Simplified Acute Physiology, version 3 (patients in intensive care units; ICU). FINDINGS: A total of 17,140 patients were included, of which 2943 died within 60 days of admission. The overall 60-day mortality was thus 17·2% (95% CI, 16·6%-17·7%), and it decreased from 24·7% (95% CI, 23·0%-26·5%) in March to 10·4% (95% CI, 8·9%-12·1%) post-wave (July-September). Adjusted relative risk (RR) of death was 0·46 (95% CI, 0·39-0·54) post-wave, using March as reference. Corresponding RR for patients not admitted to ICU and those admitted to ICU were 0·49 (95% CI, 0·42-0·59) and 0·49 (95% CI, 0·33-0·72), respectively. The proportion of patients admitted to ICU decreased from 19·4% (95% CI, 17·9%-21·1%) in the March cohort to 8·9% (95% CI, 7·5%-10·6%) post-wave. INTERPRETATION: There was a gradual decline in mortality during the spring of 2020 in Swedish hospitalised COVID-19 patients, independent of baseline patient characteristics. Future research is needed to explain the reasons for this decline. The changing COVID-19 mortality should be taken into account when management and results of studies from the first pandemic wave are evaluated. FUNDING: This study was funded by Sweden's National Board of Health and Welfare.

Machine Learning Based Prediction of COVID-19 Mortality Suggests Repositioning of Anticancer Drug for Treating Severe Cases.

Despite available vaccinations COVID-19 case numbers around the world are still growing, and effective medications against severe cases are lacking. In this work, we developed a machine learning model which predicts mortality for COVID-19 patients using data from the multi-center 'Lean European Open Survey on SARS-CoV-2-infected patients' (LEOSS) observational study (>100 active sites in Europe, primarily in Germany), resulting into an AUC of almost 80%. We showed that molecular mechanisms related to dementia, one of the relevant predictors in our model, intersect with those associated to COVID-19. Most notably, among these molecules was tyrosine kinase 2 (TYK2), a protein that has been patented as drug target in Alzheimer's Disease but also genetically associated with severe COVID-19 outcomes. We experimentally verified that anti-cancer drugs Sorafenib and Regorafenib showed a clear anti-cytopathic effect in Caco2 and VERO-E6 cells and can thus be regarded as potential treatments against COVID-19. Altogether, our work demonstrates that interpretation of machine learning based risk models can point towards drug targets and new treatment options, which are strongly needed for COVID-19.

The AVERT MoCA Data: Scoring Reliability in a Large Multicenter Trial.

The Montreal Cognitive Assessment (MoCA) is a widely used cognitive screening tool in stroke. As scoring the visuospatial/executive MoCA items involves subjective judgement, reliability is important. Analyzing data on these items from A Very Early Rehabilitation Trial (AVERT), we compared the original scoring of assessors (n = 102) to blind scoring by a single, independent rater. In a sample of scoresheets from 1,119 participants, we found variable interrater reliability. The match between original assessors and the independent rater was the following: trail-making 97% (κ = 0.94), cube copy 90% (κ = 0.80), clock contour 92% (κ = 0.49), clock numbers 89% (κ = 0.67), and clock hands 72% (κ = 0.46). For all items except clock contour, the independent rater was "stricter" than the original assessors. Discrepancies were typically errors in original scoring, rather than borderline differences in subjective judgement. In trials that include the MoCA, researchers should emphasize scoring rules to assessors and implement independent data checking, especially for clock hands, to maximize accuracy.

Serum cholesterol, body mass index and smoking status do not predict long-term cognitive impairment in elderly stroke patients.

OBJECTIVES: Older stroke survivors are at risk of long-term cognitive impairment, which is associated with a number of modifiable and non-modifiable factors. We aimed to assess the association between the modifiable risk factors, serum cholesterol, low density lipoprotein, high density lipoprotein, serum triglycerides, body mass index (BMI) and smoking status on cognitive function, while controlling for the non-modifiable factors, acute functional impairment, diabetes status and age. METHODS: A cross-sectional study from a metropolitan university hospital in Sweden involving older adults (n = 149). Assessments occurred at 20 months post-stroke, using the Mini Mental State Examination and serum blood levels of cholesterol, low density lipoprotein, high density lipoprotein and serum triglycerides. RESULTS: Hierarchical linear regression showed that only acute functional impairment significantly contributed to long-term cognitive impairment in stroke survivors. Only 12% of the sample showed healthy cholesterol levels while the remaining patients showed borderline or high cholesterol levels. In terms of BMI, only 2% of the sample were underweight, 38% were within healthy range and 26% were overweight/obese. Only eight women and four men were smokers, therefore our sample of smokers was likely too small to detect any differences between smokers and non-smokers in regard to cognitive outcomes. CONCLUSION: Serum cholesterol, low density lipoprotein, high density lipoprotein, serum triglycerides, BMI or smoking status did not influence cognitive outcomes in older stroke surviving individuals. These findings suggest that modification of these factors may not influence cognitive outcomes in stroke-surviving individuals however should be interpreted as preliminary given limitations in the current study.

Longitudinal changes in activity participation in the first year post-stroke and association with depressive symptoms.

Research question: 1. Does activity participation improve over time in the first year after stroke? 2. What is the association of depressive symptoms on retained activity participation 12-months post-stroke adjusting for neurological stroke severity and age? 3. Is an improvement in activity participation associated with a decrease in depressive symptoms between 3- and 12-months post-stroke? Design: Longitudinal observational study of activity participation and depressive symptoms in ischemic stroke survivors. Participants: A total of 100 stroke survivors with mild neurological stroke severity. Methods: A total of 100 stroke survivors were recruited from five metropolitan hospitals and assessed at 3- and 12-months post-stroke using measures of activity participation (Activity Card Sort-Australia (ACS-Aus)) and depressive symptoms (Montgomery-Asberg Depression Rating Scale Structured Interview Guide (MADRS-SIGMA)). Results: There was a significant association between time (pre-stroke to 3-months post-stroke) and current activity participation (-5.2 activities 95% CI -6.8 to -3.5, p < 0.01) and time (pre-stroke to 12-months) and current activity participation (-2.1 activities 95% CI -3.7 to -0.5, p = 0.01). At 12-months post-stroke, a one-point increase in depressive symptoms was associated with a median decrease of 0.3% (95% CI -1.4% to -0.1%, p = 0.02) of retained overall activity participation, assuming similar neurological stroke severity and age. A decrease in depressive symptoms between 3- and 12-months post-stroke was associated with an improvement of 0.31 (95% CI -0.5 to -0.1, p = 0.01) in current activity participation. Conclusions: Activity participation improves during the first year of recovery post-stroke in stroke survivors with mild neurological stroke severity and is associated with depressive symptoms over time and at 12-months post-stroke. Implications for rehabilitation Improvements in participation occur in the first 3-months post-stroke and continue to a lesser degree in the first year after stroke. Depressive symptoms are associated with lower participation at 12-months. A multidimensional approach targeting depressive symptoms and increasing participation in the early months post-stroke and throughout the first-year after stroke is recommended to increase overall recovery following stroke. A focus on increasing leisure activity participation is recommended to improve depressive symptoms.

Early Mobilization After Stroke Is Not Associated With Cognitive Outcome.

Background and Purpose- We aimed to determine whether early mobilization after stroke affects subsequent cognitive function. Methods- AVERT (A Very Early Rehabilitation Trial) was an international, 56-site, phase 3 randomized controlled trial, conducted from 2006 to 2015. Participants were included if they were aged 18+, presented within 24 hours of stroke, and satisfied physiological limits for blood pressure, heart rate, and temperature. Participants were randomized to receive either usual stroke unit care or very early and more frequent mobilization in addition to usual stroke unit care. The Montreal Cognitive Assessment, scored 0 to 30, was introduced as a 3-month outcome during 2008. Results- Of the 2104 patients included in AVERT, 317 were assessed before the Montreal Cognitive Assessment's introduction. Of the remaining 1787, 1189 (66.5%) had complete Montreal Cognitive Assessment data, 456 (25.5%) had partially or completely missing data, 136 (7.6%) had died, and 6 (0.3%) were lost to follow-up. In surviving participants with complete data, adjusting for age and stroke severity, total Montreal Cognitive Assessment score was no different in the intervention (n=595; median, 23; interquartile range, 19-26; mean, 21.9; SD, 5.9) and usual care (n=594; median, 23; interquartile range, 19-26; mean, 21.8; SD, 5.9) groups ( P=0.68). Conclusions- Exposure to earlier and more frequent mobilization in the acute stage of stroke does not influence cognitive outcome at 3 months. This stands in contrast to the primary outcome from AVERT (modified Rankin Scale), where the intervention group had less favorable outcomes than controls. Clinical Trial Registration- URL: https://www.anzctr.org.au . Unique identifier: ACTRN12606000185561.

STROKOG (stroke and cognition consortium): An international consortium to examine the epidemiology, diagnosis, and treatment of neurocognitive disorders in relation to cerebrovascular disease.

INTRODUCTION: The Stroke and Cognition consortium (STROKOG) aims to facilitate a better understanding of the determinants of vascular contributions to cognitive disorders and help improve the diagnosis and treatment of vascular cognitive disorders (VCD). METHODS: Longitudinal studies with ≥75 participants who had suffered or were at risk of stroke or TIA and which evaluated cognitive function were invited to join STROKOG. The consortium will facilitate projects investigating rates and patterns of cognitive decline, risk factors for VCD, and biomarkers of vascular dementia. RESULTS: Currently, STROKOG includes 25 (21 published) studies, with 12,092 participants from five continents. The duration of follow-up ranges from 3 months to 21 years. DISCUSSION: Although data harmonization will be a key challenge, STROKOG is in a unique position to reuse and combine international cohort data and fully explore patient level characteristics and outcomes. STROKOG could potentially transform our understanding of VCD and have a worldwide impact on promoting better vascular cognitive outcomes.

Reduction in retained activity participation is associated with depressive symptoms 3 months after mild stroke: An observational cohort study.

OBJECTIVE: To quantify the association of depressive symptoms with retained activity participation 3 months post-stroke, after adjusting for neurological stroke severity and age. DESIGN: A cross-sectional observational study of retained activity participation and depressive symptoms in stroke survivors with ischaemic stroke. PARTICIPANTS: One hundred stroke survivors with mild neurological stroke severity. METHODS: One hundred stroke survivors were recruited from 5 metropolitan hospitals and reviewed at 3 months post-stroke using measures of activity participation, Activity Card Sort-Australia, and depressive symptoms, Montgomery-Asberg Depression Rating Scale Structured Interview Guide (MADRS-SIGMA). RESULTS: The median percentage of retained overall activity participation was 97%, (interquartile range 79-100%). Using multiple median regression, 1 point increase in the MADRS-SIGMA was associated with a median decrease of 0.7% (95% CI -1.4 to -0.1, p=0.02) of retained overall activity participation, assuming similar neurological stroke severity and age. CONCLUSION: The findings of this study establish the association of depressive symptoms with retained activity participation 3 months post-stroke in stroke survivors with mild neurological stroke severity. Clinical rehabilitation recommendations to enhance activity participation need to account for those with even mild depressive symptoms post-stroke.

Folate and MMA predict cognitive impairment in elderly stroke survivors: A cross sectional study.

Elderly stroke survivors are at risk of malnutrition and long-term cognitive impairment. Vitamin B-related metabolites, folate and methylmalonic acid, have been implicated in cognitive function. We conducted a study exploring the relationship between blood folate, methylmalonic acid and post-stroke cognitive impairment. This is a cross sectional study of elderly Swedish patients (n=149) 20 months post-stroke, assessed using the Mini Mental State Examination, serum blood levels of methylmalonic acid and red blood cell levels of folate. Linear modeling indicated that low levels of blood folate and elevated methylmalonic acid significantly contributed to cognitive impairment in stroke survivors. Half of the stroke survivors were shown to have folate deficiency at 20 months after stroke. Folate deficiency is common long term after stroke and both low folate and elevated methylmalonic acid appear to be associated with long term cognitive impairment, in elderly Swedish stroke survivors.