A Nonadaptive Combinatorial Group Testing Strategy to Facilitate Health Care Worker Screening during the Severe Acute Respiratory Syndrome Coronavirus-2 (SARS-CoV-2) Outbreak

Abstract

Routine testing for severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) in health care workers (HCWs) is critical. Group testing strategies to increase capacity facilitate mass population testing but do not prioritize turnaround time, an important consideration for HCW screening. We propose a nonadaptive combinatorial (NAC) group testing strategy to increase throughput while facilitating rapid turnaround. NAC matrices were constructed for sample sizes of 700, 350, and 250. Matrix performance was tested by simulation under different SARS-CoV-2 prevalence scenarios of 0.1% to 10%. NAC matrices were compared versus Dorfman sequential (DS) group testing approaches. NAC matrices performed well at low prevalence levels, with an average of 97% of samples resolved after a single round of testing via the n = 700 matrix at a prevalence of 1%. In simulations of low to medium (0.1% to 3%) prevalence, all NAC matrices were superior to the DS strategy, measured by fewer repeated tests required. At very high prevalence levels (10%), the DS matrix was marginally superior, although both group testing approaches performed poorly at high prevalence levels. This strategy maximizes the proportion of samples resolved after a single round of testing, allowing prompt return of results to HCWs. This methodology may allow laboratories to adapt their testing scheme based on required throughput and the current population prevalence, facilitating a data-driven testing strategy.

Figure 1

Severe acute respiratory syndrome coronavirus-2 testing approaches. Three approaches are outlined with six samples in which sample 3 is positive. A: Where there is no pooling (monoplex testing), each well contains one sample; therefore, the result from that well corresponds directly to that sample. B: Using Dorfman sequential pooling, wells contain more than one sample. In this example, well 2 is positive, which, according to the matrix, contains samples 3 and 4. Retesting of samples 3 and 4 is required to determine that sample 3 is positive. C: Using nonadaptive combinatorial strategies, sample 3 is in both wells 3 and 4, which appear positive. Given that sample 3 is the only sample in both wells 3 and 4, the matrix can be decoded to indicate that sample 3 is positive without a requirement for retesting.

Figure 2

Proposed testing approach for health care worker screening. A single round of group testing via a nonadaptive combinatorial scheme is used as an initial high-throughput screen. In this example, this screen can confirm the status of 335 samples (96%). The 15 indeterminate samples are then determined via monoplex testing rather than a second group testing round being implemented.

Figure 3

Proportion of samples requiring retesting. A series of density plots showing the number of retests required (ie, where the matrix was unable to differentiate the true status of sample) after a single round of testing at each prevalence level. Prevalence levels of 0.1% (A), 1% (B), 3% (C), 7% (D), and 10% (E) are shown. At very low prevalence levels (0.1 to 1%) (A and B), there are only a small number of possible outcomes, and thus the density plots appear multimodal.

Figure 4

Proportion of true negatives identified. A series of density plots showing the proportion of samples confirmed negative as a percentage of the total number of true-negative findings. Prevalence levels of 0.1% (A), 1% (B), 3% (C), 7% (D), and 10% (E) are shown.

Figure 5

Nonadaptive combinatorial versus Dorfman sequential pooling approaches. Matrices were tested at population prevalences of 0.1%, 1%, 3%, 7%, and 10%. Performance statistics for the total number of tests saved (compared with the monoplex approach) and the number of retests required are displayed.

Supplemental Figure S1

Nonadaptive pooling versus Dorfman sequential pooling. Density plot showing the performance of each nonadaptive combinatorial matrix against a sequential matrix at the same population prevalence and sample size. Here, the performance is shown by the total number of tests saved compared with the results of a monoplex approach.