Improving COVID-19 Testing Efficiency using Guided Agglomerative Sampling

Fayyaz Minhas; Dimitris Grammatopoulos; Lawrence Young; Imran Amin; David Snead; Neil Anderson; Asa Ben-Hur; Nasir Rajpoot

As discussed in the methods section, if there is a way of predicting the likelihood of someone testing positive for CoV-2 (e.g., by using a machine learning method) or assigning such belief based on expert opinion, then the efficiency of the proposed scheme can be further improved by first ranking (sorting) the given samples with respect to their belief values. The concordance of the belief value and the true status can be measured by using the area under the receiver operating characteristic curve (AUC) between these values [4]: = 0.5 implies poor concordance between belief and the actual test status whereas = 1.0 implies perfect concordance. Please note that this AUC score is not between the test outcomes and the actual status but is used as a means of measuring the impact of the additive noise on the belief values for each individual. The degree of concordance is dependent upon the value of the noise factor : = 0 will result in perfect concordance ( = 1) in which case, no testing is needed as the belief is perfect whereas for large values of , the AUC value will be 0.5. Below we show the results of the proposed scheme for various values of , and . For = 1.0, we get an average AUC score of 0.75 and this leads to a moderate increase in the number of tests that can be saved in comparison to the no-belief simulation. This shows that even a weak belief assignment model coupled with the proposed scheme can significantly reduce the number of required tests. For = 0.5 (with an AUC score of 0.9), the saving is even more substantial (up to 60%). This clearly shows that the proposed testing scheme can lead to further improvements by incorporating belief through machine learning models or expert assignment. The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/2020.04.13.039792 doi: bioRxiv preprint