Interest of the cellular population data analysis as an aid in the early diagnosis of SARS‐CoV‐2 infection

Abstract

Introduction Coronavirus disease 2019 (COVID‐19) is characterized by a high contagiousness requiring isolation measures. At this time, diagnosis is based on the positivity of specific RT‐PCR and/or chest computed tomography scan, which are time‐consuming and may delay diagnosis. Complete blood count (CBC) can potentially contribute to the diagnosis of COVID‐19. We studied whether the analysis of cellular population data (CPD), provided as part of CBC‐Diff analysis by the DxH 800 analyzers (Beckman Coulter), can help to identify SARS‐CoV‐2 infection. Methods Cellular population data of the different leukocyte subpopulations were analyzed in 137 controls, 322 patients with proven COVID‐19 (COVID+), and 285 patients for whom investigations were negative for SARS‐CoV‐2 infection (COVID−). When CPD of COVID+ were different from controls and COVID− patients, we used receiver operating characteristic analysis to test the discriminating capacity of the individual parameters. Using a random forest classifier, we developed the algorithm based on the combination of 4 monocyte CPD to discriminate COVID+ from COVID− patients. This algorithm was tested prospectively in a series of 222 patients referred to the emergency unit. Results Among the 222 patients, 86 were diagnosed as COVID‐19 and 60.5% were correctly identified using the discriminating protocol. Among the 136 COVID− patients, 10.3% were misclassified (specificity 89.7%, sensitivity 60.5%). False negatives were observed mainly in patients with a low inflammatory state whereas false positives were mainly seen in patients with sepsis. Conclusion Consideration of CPD could constitute a first step and potentially aid in the early diagnosis of COVID‐19.

Artificial Intelligence Applications for COVID-19 in Intensive Care and Emergency Settings: A Systematic Review

“…Table 1 presents the main characteristics of the study. Eleven of the 14 studies investigated predictive models and were assessed according to PROBAST and TRIPOD: eight studies developed prognostic models [ 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 ] and three studies developed diagnostic models [ 38 , 39 , 40 ]. Of the remaining three studies, two evaluated the prognostic potential of existing AI-based lung segmentation software (without integration into a multivariate predictive model) [ 41 , 42 ] and one investigated an AI-based system for resource optimisation in the ICU [ 43 ].…”

“…Three studies investigated diagnostic AI predictive models; two studies developed models to predict the outcome of COVID-19 status at admission to the ED. Only one study was externally validated: Vasse, et al [ 40 ] developed a decision tree based on cellular population data using random forest for feature selection (accuracy = 60.5%). Brinati, et al’s [ 38 ] random forest model (C-index = 0.84, accuracy = 82%) and three-way random forest model (accuracy = 86%) achieved better performance, but were validated using weaker k -fold cross-validation.…”

Artificial Intelligence Applications for COVID-19 in Intensive Care and Emergency Settings: A Systematic Review

“…Three studies investigated diagnostic AI predictive models; two studies developed models to predict the outcome of COVID-19 status at admission to the ED. Only one study was externally validated: Vasse et al 37 developed a decision tree based on cellular population data using Random Forest for feature selection (accuracy=60.5%). Brinati et al’s 35 Random Forest model (C-index=0.84, accuracy=82%) and Three-Way Random Forest model (accuracy=86%) achieved better performance but was validated using weaker k -fold cross-validation.…”

“…Notwithstanding the high risk of bias and poor reporting of the reviewed AI models, AI algorithms tend to produce uninterpretable "black box" predictive models, which may lead to decreased acceptability of both diagnostic and prognostic AI applications amongst clinicians and hospital administrators. Some studies 32,36,37 have attempted to overcome this by using AI techniques for feature selection and presenting the final model as a decision tree or scoring system with clearly defined input variables. However, such simplifications of AI models curtail performance and limit the utility of the final model.…”

“…Table 1 presents the main characteristics of the study. 11 of the 14 studies investigated predictive models and were assessed according to PROBAST and TRIPOD: eight studies developed prognostic models [27][28][29][30][31][32][33][34] and three studies developed diagnostic models [35][36][37] . Of the remaining three studies, two evaluated the prognostic potential of existing AI-based lung segmentation software (without integration into a multivariate predictive model) 38,39 and one investigated an AI-based system for resource optimisation in the ICU 40 .…”