key: cord-0928385-dcg5w0bw
authors: Hirschfeld, Gerrit; von Glischinski, Michael; Thiele, Christian
title: Optimal cycle thresholds for COVID-19 screening – ROC-based methods highlight between-study differences
date: 2020-12-23
journal: Clin Infect Dis
DOI: 10.1093/cid/ciaa1883
sha: 6adc45ec188d888d6d53762324674ddeadcb626a
doc_id: 928385
cord_uid: dcg5w0bw

nan

In order to demonstrate the utility of this approach, we contacted the corresponding authors of the studies identified by Jefferson and colleagues [6] as reporting on the association between cycle thresholds and viral culture positivity. Of the eight authors contacted two responded and provided the necessary data [5, 6] . For two additional studies [7, 8] we were able to extract this data from the figures which showed the cycle threshold and viral culture positivity in the published articles. We then analyzed this data using the cutpointr [5] package for the open-source software R. Specifically, we plotted the distribution of cycle thresholds in culture positive and culture negative patients across studies, the ROC-curve for the four studies, the cut points identified as optimal (criterion minimum 95% detection of viruspositive culture) and the out-of-bag estimation for the AUC.

As can be seen in figure 1 there are marked differences between the studies. Most importantly the cycle threshold scores that are identified as optimal range from 26 [95%-CI: 22-32] [8] to 37 [95%-CI:34-39] [7] , while the other two studies provide optimal cut points of 29 [95%-CI:26-29] [3] and 31 [95%-CI:31-31] [2] . The confidence intervals indicate that estimation of optimal cut points is prone to random errors and that the differences between the studies are larger than can simply be attributed to chance.

While our analysis is limited by the poor data availability our results already provide evidence for systematic differences in the optimal cycle thresholds. Therefore, great care is required when deciding which threshold should be used to determine whether a person is COVID-19 positive or negative. In addition, the width of the confidence intervals demonstrates that estimates of optimal cut points need to be based on very large samples. We believe that ROC-based methods are a valuable addition to the methodological toolkit because they allow formulating explicit criteria for what constitutes optimal cycle thresholds. Furthermore, while others have speculated before that it might not be possible to determine an universally applicable threshold [9] , the methods sketched above [5] allow a more precise answer to the question which PCR-tests may use similar cycle thresholds.

The authors declare no conflicts of interest.

To Interpret the SARS-CoV-2 Test, Consider the Cycle Threshold Value

Correlation Between 3790 Quantitative Polymerase Chain Reaction-Positives Samples and Positive Cell Cultures, Including 1941 Severe Acute Respiratory Syndrome Coronavirus 2 Isolates

Clinical infectious diseases: an official publication of the Infectious Diseases Society

Measuring the accuracy of diagnostic systems

cutpointr: Improved estimation and validation of optimal cutpoints in R

Viral cultures for COVID-19 infectivity assessment. Systematic review

Duration of infectiousness and correlation with RT-PCR cycle threshold values in cases of COVID-19

Repeat COVID-19 Molecular Testing: Correlation with Recovery of Infectious Virus, Molecular Assay Cycle Thresholds, and Analytical Sensitivity

Operation Moonshot proposals are scientifically unsound

A c c e p t e d M a n u s c r i p t A c c e p t e d M a n u s c r i p t A c c e p t e d M a n u s c r i p t