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

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 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 XGBoost and Random Forests, 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.

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

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 Alzheimers 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.

Second wave mortality among patients hospitalised for COVID-19 in Sweden: a nationwide observational cohort study (preprint)

BackgroundDuring the first pandemic wave, a substantial decline in mortality was seen among hospitalised COVID-19 patients. We aimed to study if the decreased mortality continued during the second wave, using data compiled by the Swedish National Board of Health and Welfare. MethodRetrospective nationwide observational study of all patients hospitalised in Sweden between March 1st and December 31st, 2020, with SARS-CoV-2 RNA positivity 14 days before to 5 days after admission and a discharge code for COVID-19. Outcome was 60-day all-cause mortality. Poisson regression was used to estimate the relative risk (RR) for death by month of admission, adjusting for age, sex, socio-economic data, comorbidity, care dependency, and country of birth. FindingsA total of 32 452 patients were included. December had the highest number of admissions/month (n=8253) followed by April (n=6430). The 60-day crude mortality decreased from 24{middle dot}7% (95% CI, 23{middle dot}0%-26{middle dot}5%) for March to 10{middle dot}4% (95% CI, 8{middle dot}9%-12{middle dot}1%) for July-September (as reported previously), later increased to 19{middle dot}9% (95% CI, 19{middle dot}1-20{middle dot}8) for December. RR for 60-day death for December (reference) was higher than those for June to November (RR ranging from 0{middle dot}74 to 0{middle dot}89; 95% CI <1 for all months). SARS-CoV-2 variants of concern were only sporadically found in Sweden before January 2021. InterpretationThe decreased mortality of hospitalised COVID-19 patients after the first wave turned and increased during the second wave. Focused research is urgent to describe if this increase was caused by a high load of patients, management and treatment, viral properties, or other factors. Research in contextO_ST_ABSEvidence before this studyC_ST_ABSDuring the first pandemic wave, a substantial decline in mortality was seen among hospitalised COVID-19 patients in many countries. As the reason for this decline has not been clarified, no one could foresee how mortality would change during forthcoming waves. Added valueThis retrospective nationwide study of all patients hospitalised for COVID-19 in Sweden from March to December 2020 showed that the gradual decrease in mortality seen in the first pandemic wave was followed by an increased crude and adjusted 60-day all-cause mortality during the second wave. This increase in mortality occurred although the standard-of-care recommendations for hospitalised COVID-19 patients did not change in Sweden during the second half of 2020. Implications of all the available evidenceWhile improved standard-of-care was believed to be an important factor for the decrease in mortality during the first pandemic wave, the increasing mortality during the second wave has no apparent explanation. As the currently known virus variants of concern occurred only sporadically in Sweden before January 2021, they were most likely not involved. Focused research is urgent to describe if this increase in mortality was caused by a high load of patients, management and treatment factors, viral properties, or other circumstances

Decline in mortality among hospitalised covid-19 patients in Sweden: a nationwide observational study (preprint)

OBJECTIVE 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 of mortality among patients hospitalised for covid-19 in a nationwide study. DESIGN Nationwide observational cohort study of all patients hospitalised in Sweden 1 March to 30 June 2020 with SARS-CoV-2 RNA positivity 14 days before to 5 days after admission, and a discharge code for covid-19. SETTING All hospitals in Sweden. PARTICIPANTS 15 761 hospitalised patients with covid-19, with data compiled by the Swedish National Board of Health and Welfare. MAIN OUTCOME MEASURES Outcome was 60-day all-cause mortality. Patients were stratified according to month of hospital admission. Poisson regression was used to estimate the relative risk of death by month of admission, adjusting for pre-existing conditions, age, sex, care dependency, and severity of illness (Simplified Acute Physiology, version 3), for patients in intensive care units (ICU). RESULTS The overall 60-day mortality was 17.8% (95% confidence interval (CI), 17.2% to 18.4%), and it decreased from 24.7% (95% CI, 23.0% to 26.5%) in March to 13.3% (95% CI, 12.1% to 14.7%) in June. Adjusted relative risk (RR) of death was 0.56 (95% CI, 0.51 to 0.63) for June, using March as reference. Corresponding RR for patients not admitted to ICU and those admitted to ICU were 0.60 (95% CI, 0.53 to 0.67) and 0.61 (95% CI, 0.48 to 0.79), 3 respectively. The proportion of patients admitted to ICU decreased from 19.5% (95% CI, 17.9% to 21.0%) in the March cohort to 11.0% (95% CI, 9.9% to 12.2%) in the June cohort. CONCLUSIONS There was a gradual decline in mortality from March to June 2020 in Swedish hospitalised covid-19 patients, which was independent of pre-existing conditions, age, and sex. 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.