Development and external validation of a logistic regression derived formula based on repeated routine hematological measurements predicting survival of hospitalized Covid-19 patients

Heber, Stefan; Pereyra, David; Schrottmaier, Waltraud C.; Kammerer, Kerstin; Santol, Jonas; Pawelka, Erich; Hana, Markus; Scholz, Alexander; Liu, Markus; Hell, Agnes; Heiplik, Klara; Lickefett, Benno; Haverall, Sebastian; Traugott, Marianna; Neuboeck, Matthias; Schoergenhofer, Christian; Seitz, Tamara; Firbas, Christa; Karolyi, Mario; Weiss, Guenter; Jilma, Bernd; Thålin, Charlotte; Bellmann‐Weiler, Rosa; Salzer, H.; Fischer, Michael J.M.; Zoufaly, Alexander; Assinger, Alice

doi:10.1101/2020.12.20.20248563

Cited by 5 publications

(5 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Different assessment procedures within our study also yield highly variable performance estimates. Some studies suggested logistic regression models for COVID-19 and mortality prediction 39,53 , however, most identified (X)GB or RF as the best model classes 18,20,25,31,38 . We confirm these findings and suggest to use XGB or RF for COVID-19 diagnosis and RF for mortality prediction, as these exhibit the highest performance in our experiments.…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Machine Learning based COVID-19 Diagnosis from Blood Tests with Robustness to Domain Shifts

Roland

Böck

Tschoellitsch

et al. 2021

Preprint

View full text Add to dashboard Cite

We investigate machine learning models that identify COVID-19 positive patients and estimate the mortality risk based on routinely acquired blood tests in a hospital setting. However, during pandemics or new outbreaks, disease and testing characteristics change, thus we face domain shifts. Domain shifts can be caused, e.g., by changes in the disease prevalence (spreading or tested population), by refined RT-PCR testing procedures (taking samples, laboratory), or by virus mutations. Therefore, machine learning models for diagnosing COVID-19 or other diseases may not be reliable and degrade in performance over time. To countermand this effect, we propose methods that first identify domain shifts and then reverse their negative effects on the model performance. Frequent re-training and re-assessment, as well as stronger weighting of more recent samples, keeps model performance and credibility at a high level over time. Our diagnosis models are constructed and tested on large-scale data sets, steadily adapt to observed domain shifts, and maintain high ROC AUC values along pandemics.

show abstract

Section: Discussionmentioning

confidence: 99%

“…We additionally report the predictive ability for mortality risk of the COVID-19 positive samples on the basis of the blood tests only, again with no additional expensive features 33–35,38,40,41,53,54 . Compared to previous studies 33,36,37,39 , our mortality models are trained on a large number of COVID-19 positive patients.…”

Section: Introductionmentioning

confidence: 99%

Machine Learning based COVID-19 Diagnosis from Blood Tests with Robustness to Domain Shifts

Roland

Böck

Tschoellitsch

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Mortality risk scores have been developed, by different means, analyzing clinical data and laboratory parameters derived from blood and sometimes urine samples taken on admission to hospital, mostly under the conditions mentioned above [ 85 , 86 , 87 , 88 , 89 , 90 , 91 , 92 ].…”

Section: Mortality Risk Scores In Symptomatic Hospitalized Sars-cov-2...mentioning

confidence: 99%

SARS-CoV-2-Infection (COVID-19): Clinical Course, Viral Acute Respiratory Distress Syndrome (ARDS) and Cause(s) of Death

Ramadori

2022

Medical Sciences

View full text Add to dashboard Cite

SARS-CoV-2-infected symptomatic patients often suffer from high fever and loss of appetite which are responsible for the deficit of fluids and of protein intake. Many patients admitted to the emergency room are, therefore, hypovolemic and hypoproteinemic and often suffer from respiratory distress accompanied by ground glass opacities in the CT scan of the lungs. Ischemic damage in the lung capillaries is responsible for the microscopic hallmark, diffuse alveolar damage (DAD) characterized by hyaline membrane formation, fluid invasion of the alveoli, and progressive arrest of blood flow in the pulmonary vessels. The consequences are progressive congestion, increase in lung weight, and progressive hypoxia (progressive severity of ARDS). Sequestration of blood in the lungs worsens hypovolemia and ischemia in different organs. This is most probably responsible for the recruitment of inflammatory cells into the ischemic peripheral tissues, the release of acute-phase mediators, and for the persistence of elevated serum levels of positive acute-phase markers and of hypoalbuminemia. Autopsy studies have been performed mostly in patients who died in the ICU after SARS-CoV-2 infection because of progressive acute respiratory distress syndrome (ARDS). In the death certification charts, after respiratory insufficiency, hypovolemic heart failure should be mentioned as the main cause of death.

show abstract

“…However, little is known of the effect of the risk factors identified on the course of the disease. There is a wide range of various models for diagnosing COVID-19, predicting the prognosis of patients with COVID-19 or being admitted to hospital for COVID-19 [5,6] as well as models to support risk stratification scores [7]. In Germany, the indication of a hospital admission is decided by the outpatient doctor dependent upon age, comorbidities, breathing rate and oxygen saturation [8].…”

Section: Introductionmentioning

confidence: 99%

Surveillance and care for confirmed and suspected patients with COVID-19 in general practice (CovidCare): study protocol for an observational trial

et al. 2021

View full text Add to dashboard Cite

Background A SARS-CoV-2 infection can lead from asymptomatic through to critical disease in a dynamic and unpredictable course within a few days. The challenge in outpatient monitoring the highly contagious COVID-19 disease during the ongoing pandemic is to filter severe courses followed by admission to hospital with the aim of preventing an overburdening of clinics. However, little is known of the effect of risk factors on the course of the infection of outpatient patients. To support general practices in managing high risk patients, we designed a COVID-19 surveillance and care tool (CovidCare). It includes an initial assessment of yet known risk factors and symptoms and a continuous telephone monitoring of signs and symptoms. This study aims to investigate the effects of different risk factors on the course of the COVID-19 disease, utilisation of different health care services and to gain insights into the utilisation of CovidCare in general practices. Methods We will conduct a multi-centered prospective, longitudinal non-controlled observational trial of COVID-19 patients in general practices. Overall, 700 GPs who participate in general-practice centered care by the AOK Baden-Württemberg (large German sickness fund) are eligible and will be invited for study participation, including adult, outpatient COVID-19 patients (or urgent suspicion and ≥ 50 years) with at least one additional known risk factor, who participate in general-practice centered care. The primary outcome is hospitalisation due to COVID-19. Secondary outcomes are diagnosis of pneumonia, utilisation of palliative care, mortality rate, anxiety and identification of predictive risk factors. Quantitative data analysis will focus on valid descriptive figures and mixed regression models. The accompanying process evaluation is based on interviews and questionnaires from general practice staff and patients. The analysis of the process evaluation is descriptive and explorative. Discussion The use of the COVID-19 surveillance and care tool is expected to encourage the provision of structured quality of care during the ongoing pandemic. This trial will provide an understanding of the COVID-19-disease and the effect of several risk factors on the course of the disease and health care utilisation. The results can be used for a better management of the COVID-19 pandemic and its consequences. Trial registration German Clinical Trials Register DRKS00022054.

show abstract

Development and external validation of a logistic regression derived formula based on repeated routine hematological measurements predicting survival of hospitalized Covid-19 patients

Cited by 5 publications

References 25 publications

Machine Learning based COVID-19 Diagnosis from Blood Tests with Robustness to Domain Shifts

Machine Learning based COVID-19 Diagnosis from Blood Tests with Robustness to Domain Shifts

SARS-CoV-2-Infection (COVID-19): Clinical Course, Viral Acute Respiratory Distress Syndrome (ARDS) and Cause(s) of Death

Surveillance and care for confirmed and suspected patients with COVID-19 in general practice (CovidCare): study protocol for an observational trial

Contact Info

Product

Resources

About