Challenges and issues of SARS-CoV-2 pool testing

We read with interest the Correspondence by Stefan Lohse and colleagues,¹ who evaluated the practicability of pool testing for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). Pooling of samples yields considerable savings of test kits when the prevalence of infection is low because pools with all-negative samples can be discarded with a single test. Lohse and colleagues' findings suggest that pooling up to 30 samples is technically feasible with currently used and commercially available SARS-CoV-2 RT-PCR kits.¹

But is bigger always better? Is it really efficient to pool 30 samples? First discussed around 80 years ago by Dorfman² in the context of large-scale syphilis testing, the matter is complex and optimal pool sizes depend on the prevalence of infection in the population. Furthermore, there are more sophisticated pooling schemes than the one originally discussed by Dorfman, which use multiple stages of pooling or test samples in rows and columns of a matrix.³

We propose an adaptive approach that uses different pooling schemes depending on the estimated prevalence in a population.⁴ Our exhaustive comparison of testing schemes shows that three different schemes with initial pool sizes of 16, nine, and three samples are optimal for a prevalence of up to 3·5%, 3·5–12%, and 12–30%, respectively (appendix). The first two schemes are three-staged, meaning that if a pool tests positive it is further divided into sub-pools of four or three samples, before then testing samples individually. These schemes have a consistently higher testing efficiency than the method proposed by Lohse and colleagues, who used a three-staged scheme with initial pools of 30 samples and sub-pools of ten samples (appendix). For a prevalence of 2%, as in the population tested by Lohse and colleagues,¹ our proposed testing scheme (pool size of 16 and four sub-pools of four samples) uses around 20% fewer tests. At higher prevalence, differences become even more pronounced and smaller pool sizes are optimal. For prevalence over 18%, pools of 30 samples are even less efficient than individual testing, whereas small pool sizes of three samples still yield a considerable improvement in efficiency (appendix).

Hence bigger is not always better. Rather, it is preferable to choose one of the three proposed testing schemes based on the estimated underlying prevalence.