tetano
Editor, Senior Moderator
BMC Infect Dis
. 2022 Feb 24;22(1):187.
doi: 10.1186/s12879-022-07164-1.
A new screening tool for SARS-CoV-2 infection based on self-reported patient clinical characteristics: the COV [SUB]19[/SUB]-ID score
Pablo Diaz Badial[SUP] #[/SUP][SUP] 1 [/SUP], Hugo Bothorel[SUP] #[/SUP][SUP] 2 [/SUP], Omar Kherad[SUP] 3 [/SUP], Philippe Dussoix[SUP] 1 [/SUP], Faustine Tallonneau Bory[SUP] 1 [/SUP], Majd Ramlawi[SUP] 1 [/SUP]
Affiliations
Abstract
Background: While several studies aimed to identify risk factors for severe COVID-19 cases to better anticipate intensive care unit admissions, very few have been conducted on self-reported patient symptoms and characteristics, predictive of RT-PCR test positivity. We therefore aimed to identify those predictive factors and construct a predictive score for the screening of patients at admission.
Methods: This was a monocentric retrospective analysis of clinical data from 9081 patients tested for SARS-CoV-2 infection from August 1 to November 30 2020. A multivariable logistic regression using least absolute shrinkage and selection operator (LASSO) was performed on a training dataset (60% of the data) to determine associations between self-reported patient characteristics and COVID-19 diagnosis. Regression coefficients were used to construct the Coronavirus 2019 Identification score (COV[SUB]19[/SUB]-ID) and the optimal threshold calculated on the validation dataset (20%). Its predictive performance was finally evaluated on a test dataset (20%).
Results: A total of 2084 (22.9%) patients were tested positive to SARS-CoV-2 infection. Using the LASSO model, COVID-19 was independently associated with loss of smell (Odds Ratio, 6.4), fever (OR, 2.7), history of contact with an infected person (OR, 1.7), loss of taste (OR, 1.5), muscle stiffness (OR, 1.5), cough (OR, 1.5), back pain (OR, 1.4), loss of appetite (OR, 1.3), as well as male sex (OR, 1.05). Conversely, COVID-19 was less likely associated with smoking (OR, 0.5), sore throat (OR, 0.9) and ear pain (OR, 0.9). All aforementioned variables were included in the COV[SUB]19[/SUB]-ID score, which demonstrated on the test dataset an area under the receiver-operating characteristic curve of 82.9% (95% CI 80.6%-84.9%), and an accuracy of 74.2% (95% CI 74.1%-74.3%) with a high sensitivity (80.4%, 95% CI [80.3%-80.6%]) and specificity (72.2%, 95% CI [72.2%-72.4%]).
Conclusions: The COV[SUB]19[/SUB]-ID score could be useful in early triage of patients needing RT-PCR testing thus alleviating the burden on laboratories, emergency rooms, and wards.
Keywords: Artifiicial Intelligence; COVID-19; Machine learning; Predictive score; SARS-CoV-2; Screening; Self-reported symptoms; Triage.
. 2022 Feb 24;22(1):187.
doi: 10.1186/s12879-022-07164-1.
A new screening tool for SARS-CoV-2 infection based on self-reported patient clinical characteristics: the COV [SUB]19[/SUB]-ID score
Pablo Diaz Badial[SUP] #[/SUP][SUP] 1 [/SUP], Hugo Bothorel[SUP] #[/SUP][SUP] 2 [/SUP], Omar Kherad[SUP] 3 [/SUP], Philippe Dussoix[SUP] 1 [/SUP], Faustine Tallonneau Bory[SUP] 1 [/SUP], Majd Ramlawi[SUP] 1 [/SUP]
Affiliations
- PMID: 35209872
- DOI: 10.1186/s12879-022-07164-1
Abstract
Background: While several studies aimed to identify risk factors for severe COVID-19 cases to better anticipate intensive care unit admissions, very few have been conducted on self-reported patient symptoms and characteristics, predictive of RT-PCR test positivity. We therefore aimed to identify those predictive factors and construct a predictive score for the screening of patients at admission.
Methods: This was a monocentric retrospective analysis of clinical data from 9081 patients tested for SARS-CoV-2 infection from August 1 to November 30 2020. A multivariable logistic regression using least absolute shrinkage and selection operator (LASSO) was performed on a training dataset (60% of the data) to determine associations between self-reported patient characteristics and COVID-19 diagnosis. Regression coefficients were used to construct the Coronavirus 2019 Identification score (COV[SUB]19[/SUB]-ID) and the optimal threshold calculated on the validation dataset (20%). Its predictive performance was finally evaluated on a test dataset (20%).
Results: A total of 2084 (22.9%) patients were tested positive to SARS-CoV-2 infection. Using the LASSO model, COVID-19 was independently associated with loss of smell (Odds Ratio, 6.4), fever (OR, 2.7), history of contact with an infected person (OR, 1.7), loss of taste (OR, 1.5), muscle stiffness (OR, 1.5), cough (OR, 1.5), back pain (OR, 1.4), loss of appetite (OR, 1.3), as well as male sex (OR, 1.05). Conversely, COVID-19 was less likely associated with smoking (OR, 0.5), sore throat (OR, 0.9) and ear pain (OR, 0.9). All aforementioned variables were included in the COV[SUB]19[/SUB]-ID score, which demonstrated on the test dataset an area under the receiver-operating characteristic curve of 82.9% (95% CI 80.6%-84.9%), and an accuracy of 74.2% (95% CI 74.1%-74.3%) with a high sensitivity (80.4%, 95% CI [80.3%-80.6%]) and specificity (72.2%, 95% CI [72.2%-72.4%]).
Conclusions: The COV[SUB]19[/SUB]-ID score could be useful in early triage of patients needing RT-PCR testing thus alleviating the burden on laboratories, emergency rooms, and wards.
Keywords: Artifiicial Intelligence; COVID-19; Machine learning; Predictive score; SARS-CoV-2; Screening; Self-reported symptoms; Triage.