Testing, as a mainstay of an effective response to COVID-19, requires sufficient supply to meet the demand. However, since the pandemic hit the US, this has not been the case. While governments have focused on procuring and producing additional tests, researchers have focused on finding ways to make the tests we do have more efficient.

Enter Dorfman Screening, a technique used in WWII to efficiently identify syphilis cases. In this scheme, samples from multiple persons are tested together – if the test on this pooled sample is negative, it is assumed all persons in the sample are uninfected. If the test returns positive, each individual in the pooled sample is retested separately to find which person(s) led to the positive result. This scheme leads to many fewer tests being needed than the standard procedure of testing everyone individually – when samples of five individuals are pooled, and the population prevalence is 5%, only 40% as many tests are used. Researchers and producers of tests have caught on to this, and as of 18 July, the FDA approved the first test for use with pooled samples of four individuals, and this likely signals the beginning of increased use of pooled samples for testing.

One of the main downsides of pooling samples is that as the prevalence goes up, the number of tests needed also goes up – as more of the samples test positive, more tests are needed to then individually test everyone in the sample. Experts generally agree that once prevalence is greater than 10%, pooling samples loses effectiveness.

However, in infectious disease outbreaks, infections often cluster, and those clusters may be tested together, leading to a concentration of infected persons in relatively-few pooled samples, potentially increasing the effectiveness of screening in this way. In a recent research letter, I simulated a series of people queueing at a testing site with or without clusters of infection. As expected, when there was no clustering (i.e., every person’s infection status was independent of everyone else’s), Dorfman screening became ineffective when prevalence reached ~10%.

When there is clustering, batch testing remains effective even at higher levels of prevalence. For example, when the prevalence is 15%, pooled testing would be ineffective if there were no clustering. But when a person is 15% more likely to be infected if the person preceding them was infected, Dorfman screening uses only half as many tests as would be used if everyone were tested individually. As the clustering increases, so too does the efficiency of batch testing.

One may ask whether this is of concern. Given that even the worst-hit states only have a prevalence of ~2%, it may appear that batch testing will be effective even without case clustering. However, testing rates rarely reflect the overall population – those tested are usually high-risk persons, among whom the prevalence is much higher. One study of testing sites found that 1-in-6 tested were infected, and at that rate, batch sampling is only effective if there is significant clustering of cases.

The recently published study therefore highlights a few key considerations as policy is made regarding pooled samples. First, pooling should not a priori be restricted in areas where the prevalence, even among high-risk persons, is high. This work shows that pooling may be surprisingly effective even in these settings. Second, the surprising aspect should be eliminated by studying whether clustering does indeed exist at testing sites. Because the study in question was based on simulated data, it is important to estimate to what extent clustering occurs in practice. This can be done by examining the series of test results completed at a testing site. These results can then inform decisions about pooled screening at a site with a given prevalence and level of clustering. In this manner, we can improve the efficiency of testing throughout this and future pandemics and outbreaks when tests are in short supply.

See related op-ed in USA Today.