Use of an algorithm based on routine blood laboratory tests to exclude COVID-19 in a screening-setting of healthcare workers

Abstract

Background: COVID-19 is an ongoing pandemic leading to exhaustion of the hospital care system. Our health care system has to deal with a high level of sick leave of health care workers (HCWs) with COVID-19 related complaints, in whom an infection with SARS-CoV-2 has to be ruled out before they can return back to work. The aim of the present study is to investigate if the recently described CoLab-algorithm can be used to exclude COVID-19 in a screening setting of HCWs.

Methods: In the period from January 2021 till March 2021, HCWs with COVID-19-related complaints were prospectively collected and included in this study. Next to the routinely performed SARS-CoV-2 RT-PCR, using a set of naso- and oropharyngeal swab samples, two blood tubes (one EDTA- and one heparin-tube) were drawn for analysing the 10 laboratory parameters required for running the CoLab-algorithm.

Results: In total, 726 HCWs with a complete CoLab-laboratory panel were included in this study. In this group, 684 HCWs were tested SARS-CoV-2 RT-PCR negative and 42 cases RT-PCR positive. ROC curve analysis showed an area under the curve (AUC) of 0.853 (95% CI: 0.801-0.904). At a safe cut-off value for excluding COVID-19 of -6.525, the sensitivity was 100% with a specificity of 34% (95% CI: 21 to 49%). No SARS-CoV-2 RT-PCR cases were missed with this cut-off and COVID-19 could be safely ruled out in more than one third of HCWs.

Conclusion: The CoLab-score is an easy and reliable algorithm that can be used for screening HCWs with COVID-19 related complaints. A major advantage of this approach is that the results of the score are available within 1 hour after collecting the samples. This results in a faster return to labour process of a large part of the COVID-19 negative HCWs (34%), next to a reduction in RT-PCR tests (reagents and labour costs) that can be saved.

Fig 1. Inclusion flow.

Fig 2. ROC curve of the CoLab-linear predictor.

The area under the ROC curve is shown with the 95% DeLong confidence interval in round brackets. The displayed threshold of -6.241 corresponds to a sensitivity of 100%, i.e. no HCWs below this linear predictor were RT-PCR positive.

Fig 3. Calibration plot.

A. In the calibration plot the proportion of observed COVID-19 positives versus expected proportion of positives are plotted. Observations are grouped with an average of 50 observations per group. The expected probabilities follow from applying the inverse logit function to the CoLab-linear predictor. If the observed proportion in an external dataset is lower than the expected proportion, this means risks are over-estimated, if the observed fraction is higher, risks are under-estimated. Ideally, observed proportions are equal to expected proportions, this ideal-calibration-line is shown as a straight line through the origin with a slope of 1. The logistic calibration line is a logistic regression fit of the predicted probabilities. B. Using the intercept and/or slope from the logistic regression model, recalibrated probabilities were obtained and plotted in a second calibration plot.

Fig 4. Histograms and fitted Gaussian distribution of the CoLab-linear predictor split by RT-PCR result.

A normal distribution was fitted to the RT-PCR negative group (mean: -6.04, SD: 1.73), the dashed lines represent the 95% CI. The 5^th percentile of the Gaussian distribution is shown in red and dashed lines represent the 95% CI. Linear predictor values below this 5^th percentile are regarded as non-COVID-19.

Fig 5. CoLab-linear predictor versus RT-PCR CT value.

The CoLab-linear predictor is plotted versus the RT-PCR CT value. The red line is the CoLab-linear predictor cut-off below which HCWs are regarded as non-COVDI 19, the dashed red lines represent the 95% CI of the cut-off. The dashed line is a LOESS smooth where the 95% confidence interval is shown in gray.