In JMIR formative research
BACKGROUND : Machine learning (ML) is a part of the Artificial Intelligence strategy. Its algorithms are imputed on Big Data sets to see patterns, learn from their results, and perform tasks autonomously without being instructed on how to address the problem. New diseases like Sars-Cov2 are important data stores for machine learning. Therefore, all relevant parameters should be explicitly quantified and modeled.
OBJECTIVE : The purpose of the study was to determine (a) the overall preclinical character; (b) the cumulative cutoff values and the risk ratio, and (c) the factors associated with severity by a unidimensional and multidimensional analysis on 2173 Sars-Cov2 patients.
METHODS : The machine learning study population consisted of 2173 patients (1587 mild and non-symptoms patients, 377 moderate patients, 209 severe patients). The status of the patients was recorded from September 2021 to March 2022. Two correlation test, relative risk and risk ration were used to eliminate the unbalance parameter and select also the most remarkable ones. HCA, K-means are two independent methods to classify the parameters following their R scores. Finally, Network analysis step give the view in three dimension, more complete of the results above.
RESULTS : The Covid19 Severity directly links with a significant correlation to Age, Score index of the chest X-ray, percentage and quantity of neutrophils, Albumin, C reactive protein, and ratio of Lymphocytes. Their significant risk ratio (P<.00001) from the meta-analysis, respectively, are: 4.19 [3.58-4.95], 3.29 [2.76-3.92,] and 3.03 [2.4023;3.8314], 3.18 [2.73-3.70] and 3.32 [2.6480;4.1529], 3.15 [2.6153;3.8025], 3.4[2.91-3.97], 0.46 [0.3650;0.5752] (P<.00001), 0.34 [0.2743;0.4210]. The significant inversion of correlation between the group of severity shows the important remark. ALT - Leucocytes show the strong negative link (R=-1, P<.00001) in the mild group to the significant positive correlation in the moderate group (R=1, P<.00001). Transferrin-anion Chloride has an positive association (R=1, P<.00001) in the mild group with a significant negative correlation in the moderate group (R=-0.59, P<.00001). The clustering and network analysis visualize that the mild-moderate group, the closest neighbors with the Covid19 severity are ferritins, Age. Then there is C-reactive protein, SI of X-ray, Albumin, and Lactate dehydrogenase, which are the next close neighbors of these three factors. In the moderate-severe group, the closest neighbors with the Covid19 severity are Ferritin, Fibrinogen, Albumin, the quantity of Lymphocytes, SI of X-ray, white blood cells count, Lactate dehydrogenase, and quantity of neutrophils.
CONCLUSIONS : Complete multidimensional study in 2173 Covid19 patients in Vietnam shows the part of the related preclinical factors, which may become the clinical reference marker for surveillance and diagnostic management.
Nguyen Tue Trong, Ho Tu Cam, Bui Huong Thi Thu, Ho Lam Khanh, Ta Van Thanh