Using Machine Learning to Predict Mortality for COVID-19 Patients on Day 0 in the ICU

Rationale: Given the expanding number of COVID-19 cases and the potential for new waves of infection, there is an urgent need for early prediction of the severity of the disease in intensive care unit (ICU) patients to optimize treatment strategies.Objectives: Early prediction of mortality using machine learning based on typical laboratory results and clinical data registered on the day of ICU admission.Methods: We retrospectively studied 797 patients diagnosed with COVID-19 in Iran and the United Kingdom (U.K.). To find parameters with the highest predictive values, Kolmogorov-Smirnov and Pearson chi-squared tests were used. Several machine learning algorithms, including Random Forest (RF), logistic regression, gradient boosting classifier, support vector machine classifier, and artificial neural network algorithms were utilized to build classification models. The impact of each marker on the RF model predictions was studied by implementing the local interpretable model-agnostic explanation technique (LIME-SP).Results: Among 66 documented parameters, 15 factors with the highest predictive values were identified as follows: gender, age, blood urea nitrogen (BUN), creatinine, international normalized ratio (INR), albumin, mean corpuscular volume (MCV), white blood cell count, segmented neutrophil count, lymphocyte count, red cell distribution width (RDW), and mean cell hemoglobin (MCH) along with a history of neurological, cardiovascular, and respiratory disorders. Our RF model can predict patient outcomes with a sensitivity of 70% and a specificity of 75%. The performance of the models was confirmed by blindly testing the models in an external dataset.Conclusions: Using two independent patient datasets, we designed a machine-learning-based model that could predict the risk of mortality from severe COVID-19 with high accuracy. The most decisive variables in our model were increased levels of BUN, lowered albumin levels, increased creatinine, INR, and RDW, along with gender and age. Considering the importance of early triage decisions, this model can be a useful tool in COVID-19 ICU decision-making.

Prediction of mortality in hospitalized Egyptian patients with Coronavirus disease-2019: A multicenter retrospective study

We aimed to assess the epidemiological, clinical, and laboratory characteristics associated with mortality among hospitalized Egyptian patients with COVID-19. A multicenter, retrospective study was conducted on all polymerase chain reaction (PCR)-confirmed COVID-19 cases admitted through the period from April to July 2020. A generalized linear model was reconstructed with covariates based on predictor’s statistical significance and clinically relevance. The odds ratio (OR) was calculated by using stepwise logistic regression modeling. A total of 3712 hospitalized patients were included; of them, 900 deaths were recorded (24.2%). Compared to survived patients, non-survived patients were more likely to be older than 60 years (65.7%), males (53.6%) diabetic (37.6%), hypertensive (37.2%), and had chronic renal insufficiency (9%). Non-survived patients were less likely to receive azithromycin (p <0.001), anticoagulants (p <0.001), and steroids (p <0.001). We found that age ≥ 60 years old (OR = 2.82, 95% CI 2.05–3.86; p <0.0001), diabetes mellitus (OR = 1.58, 95% CI 1.14–2.19; p = 0.006), hypertension (OR = 1.69, 95% CI 1.22–2.36; p = 0.002), chronic renal insufficiency (OR = 3.15, 95% CI 1.84–5.38; p <0.0001), tachycardia (OR = 1.65, 95% CI 1.22–2.23; p <0.001), hypoxemia (OR = 5.69, 95% CI 4.05–7.98; p <0.0001), GCS <13 (OR 515.2, 95% CI 148.5–1786.9; p <0.0001), the use of therapeutic dose of anticoagulation (OR = 0.4, 95% CI 0.22–0.74, p = 0.003) and azithromycin (OR = 0.16, 95% CI 0.09–0.26; p <0.0001) were independent negative predictors of mortality. In conclusion, age >60 years, comorbidities, tachycardia, hypoxemia, and altered consciousness level are independent predictors of mortality among Egyptian hospitalized patients with COVID-19. On the other hand, the use of anticoagulants and azithromycin is associated with reduced mortality.

Course of lactate, pH and base excess for prediction of mortality in medical intensive care patients

Introduction As base excess had shown superiority over lactate as a prognostic parameter in intensive care unit (ICU) surgical patients we aimed to evaluate course of lactate, base excess and pH for prediction of mortality of medical ICU patients. Materials and methods For lactate, pH and base excess, values at the admission to ICU, at 24 ± 4 hours, maximum or minimum in the first 24 hours and in 24–48 hours after admission were collected from all patients admitted to the Medical ICU of the University Hospital Tübingen between January 2016 until December 2018 (N = 4067 at admission, N = 1715 with ICU treatment > 48 h) and investigated for prediction of in-hospital-mortality. Results Mortality was 22% and significantly correlated with all evaluated parameters. Strongest predictors of mortality determined by ROC were maximum lactate in 24 h (AUROC 0.74, cut off 2.7 mmol/L, hazard ratio of risk group with value > cut off 3.20) and minimum pH in 24 h (AUROC 0.71, cut off 7.31, hazard ratio for risk group 2.94). Kaplan Meier Curves stratified across these cut offs showed early and clear separation. Hazard ratios per standard deviation increase were highest for maximum lactate in 24 h (HR 1.65), minimum base excess in 24 h (HR 1.56) and minimum pH in 24 h (HR 0.75). Conclusion Lactate, pH and base excess were all suitable predictors of mortality in internal ICU patients, with maximum / minimum values in 24 and 24–48 h after admission altogether stronger predictors than values at admission. Base excess and pH were not superior to lactate for prediction of mortality.

Clinical and Laboratory Markers in Determination of Treatment Modalities and Short-Term Prognosis in Patients with Traumatic Acute Subdural Hematoma: Original Study

Objective In patients with traumatic acute subdural hematoma (ASH), it has not been yet fully elucidated which patients can benefit from surgery or from clinical follow-up. This study was constructed to predict treatment modality and short-term prognosis in patients with ASH using their clinical, radiological, and biochemical laboratory findings during admission to hospital. Methods Findings of patients with ASH determined on their CT scan between 2015 and 2018 were evaluated. Patients were grouped in terms of ASH-FOL (patients followed-up without surgery, n = 13), ASH-OP (patients treated surgically, n = 10), and ASH-INOP (patients considered as inoperable, n = 5) groups. They also were divided into “survived (n = 14)” and “nonsurvived (n = 14)” groups. Results ASH developed as a result of fall from a height in 15 patients and traffic accidents in 13 patients. In deciding for surgery, it was determined that Glasgow coma scale (GCS) scores < 8, midline shift (MLS) level > 5 mm, MLS-hematoma thickness ratio > 0.22, leukocyte count > 12730 uL, and presence of anisocoria could be used as predictive markers. It was determined that GCS scores < 8, hematoma thickness value > 8 mm, and the presence of anisocoria could be considered as biomarkers in prediction of mortality likelihood. Conclusion It could be suggested that GCS scores, MLS level, MLS-hematoma thickness ratio, presence of anisocoria, and leukocyte count value could help in determination of the treatment modality in patients with ASH. Additionally, GCS scores, hematoma thickness value, and presence of anisocoria could each be used as a marker in the prediction of early-stage prognosis and mortality likelihood of these patients.

The Use of Prognostic Prediction Models for Mortality or Clinical Deterioration among Hospitalized and Non-hospitalized Adults with COVID-19: A Systematic Review

Objective. To assess the performance of prognostic models in predicting mortality or clinical deterioration among patients with COVID-19, both hospitalized and non-hospitalized Methods. We conducted a systematic review of the literature until March 8, 2021. We included models for the prediction of mortality or clinical deterioration in COVID-19 with external validation. We used the Prediction model Risk Of Bias ASsessment Tool (PROBAST) and the GRADEpro Guideline Development Tool (GDT) to assess the evidence obtained. Results. We reviewed 33 cohort studies. Two studies had a low risk of bias, four unclear risks, and 27 with a high risk of bias due to participant selection and analysis. For the outcome of mortality, the QCOVID model had excellent prediction with high certainty of evidence but was specific for use in England. The COVID Outcome Prediction in the Emergency Department (COPE) model, the 4C Mortality Score, the Age, BUN, number of comorbidities, CRP, SpO2/FiO2 ratio, platelet count, heart rate (ABC2-SPH) risk score, the Confusion Urea Respiration Blood Pressure (CURB-65) severity score, the Rapid Emergency Medicine Score (REMS), and the Risk Stratification in the Emergency Department in Acutely Ill Older Patients (RISE UP) score had fair to good prediction of death among inpatients, while the quick Sepsis-related Organ Failure Assessment (qSOFA) score had poor to fair prediction. The certainty of evidence for these models was very low to low. For the outcome of clinical deterioration, the 4C Deterioration Score had fair prediction, the National Early Warning Score 2 (NEWS2) score poor to good, and the Modified Early Warning Score (MEWS) had poor prediction. The certainty of evidence for these three models was also very low to low. None of these models had been validated in the Philippine setting. Conclusion. The QCOVID, COPE, ABC2-SPH, 4C, CURB-65, REMS, RISE-UP models for prediction of mortality and the 4C Deterioration and NEWS2 models for prediction of clinical deterioration are potentially useful but need to be validated among patients with COVID-19 of varying severity in the Philippine setting.

Difference and efficacy of simplified motor scale compared to Glasgow coma scale

There have been many limitations reported with using the Glasgow coma scale (GCS), including complexity, and being difficult to apply among aphasic, intubated, and pediatric patients. Accordingly, many researchers exerted serious efforts to enhance and modify the scale to make it more applicable and easy to interpret in these settings. The simplified motor score (SMS) was reported in the literature in 2012 for the assessment of patients with coma in different traumatic and non-traumatic settings. In the present study, we have discussed the findings of previous studies in the literature that compared the efficacy between the SMS and GCS in the assessment of patients with traumatic brain injuries within the emergency department and out-patient settings. Our results indicate the efficacy of the SMS is similar to that of the GCS score in predicting the different outcomes, including functional performance, need to perform tracheal intubation and hospital admission. Nevertheless, evidence regarding the prediction of mortality seems to be inconsistent across the different investigations. However, the differences between the two scores is not remarkable among these studies, indicating that the SMS is an efficacious tool in this regard within an acceptable test performance results. Furthermore, the SMS score can be easily applied within these without performing complex approaches, which makes it more advantageous than the GCS. However, this evidence is based on a limited number of investigations, and more studies are required.