A Comparative Study of Predictive Machine Learning Algorithms for COVID-19 Trends and Analysis

This paper attempts to conduct analysis on the WHO dataset to produce predictive analysis applying different machine learning regression approaches such as decision trees, LSTM, and CNN regressor. The primary data has 91 entries, which consists of data of various countries with respect to dates along with confirmed cases, confirmed deaths, and recovered cases. The dataset has been divided into 70:30 in which 70 percent is used for training and validation, and 30 percent is used for testing. The coronavirus disease outbreak started in 2019, arising in Wuhan, China. The key objective is to exercise different artificial intelligence approaches, we ought to predict the confirmed cases, confirmed deaths, and recovered cases, and further, various visualization techniques have been used to deduce the meaningful inferences from the model's prediction and perform specific analytics on the results concluded. The prediction models such as LSTM and CNN are evaluated on the basis of several loss functions such as R2 score and Mean Squared Error.

Neutrophil-to-lymphocyte ratio and D-dimer assay in prediction of disease severity in covid positive Indian patients in a tertiary care centre

Aims & Objectives: To evaluate role of hematological parameters in prediction of disease severity and also analyze the trends of NLR and D-Dimer during its course. Patients/Materials & Methods: A retrospective analysis of 83 patients diagnosed with COVID-19 by RT-PCR at Medanta-the Medicity hospital in June 2020 was done. The data included neutrophil- to-lymphocyte ratio (NLR), D-Dimer, PT/APTT and platelet count. The patients were divided into 20 critical patients and 63 Noncritical patients group, based on disease severity. The parameters were compared and trends analyzed. Results: The COVID positive cases had a mean age of 56.7 years (Range: 7-84 years) with a male:female ratio of 2.6:1. The critical group had mean age of 64.7 years (Range: 42-76), versus 54.1 years (Range: 7-84) in non-critical group. At admission, the mean NLR in the critical and non-critical group was 12.26 and 5.7. Further, the critical and non-critical group had NLR>3.13 in 19 cases (95%) and 49 cases (77.8%), respectively. On receiver operating characteristic curve (ROC) analysis, predictive ability of NLR for detection of critical patients was significant(p value = 0.0001;AUC:0.779) with optimal cut-off value of 6.01, having 85% sensitivity, 68.9% specificity and ∼ 93.5% negative predictable value (NPV). The mean D-Dimer value in critical and non-critical group on Day-4 of admission was 18.89 mg/L and 2.48 mg/L. Moreover, the D-Dimer>0.55 mg/L were seen in 17 critical cases (85%) in contrast with 35 non-critical cases (55.5%). On ROC analysis, the ability of D-Dimer in predicting disease severity was significant( p value = 0.0001;AUC:0.896) with optimal cut-off value of 2.27 mg/L, having 85% sensitivity, 76.2% specificity and NPV ∼ 94.1%. On trend analysis, it was observed that the D-Dimer and NLR showed a progressive upward trend in critical patients, whereas there were more of plateau/declining values in non-critical patients. PT was mildly prolonged in 14 critical patients (70%) and 13 non- critical patients (40.6%). The mean platelet counts were similar in both the groups. Discussion & Conclusion: The study shows that the severity of the disease is more in elderly (Mean age: 64.7 years). Also, at admission NLR>6.05 and Day-4 D-Dimer>2.27 mg/L are significantly predictive of disease severity and such patients should receive prompt treatment to minimize further sequel and morbidities.