This online tool enables you to model the performance of various pool-based testing strategies for a specific area in response to the COVID-19 pandemic. It was developed based on the reference given on the background page, available as a preprint at https://arxiv.org/abs/2004.11851.
Mass testing is increasingly understood by scientists and politicians as the only viable exit strategy from COVID-19 lockdown (see our short collection of statements). However, currently every person gets tested individually - an approach that requires too much time and resources. In light of limited testing capacities, sample pooling (also known as group testing) is a highly practicable solution to increase testing throughput considerably. To that end, several samples are combined in one test, and further testing is undertaken only in case of a positive result. We present the results of the first extensive study comparing the most relevant methods for sample pooling published so far. We determine the efficiency of the different methods using a simulation approach, quantifying how many infected cases we can identify per test. Assuming an infection rate of 1%, the conventional approach (individual testing) will require an average of 100 tests to identify one infected case (i.e. 0.01 cases are detected per test). However, when pooling e.g. ten samples, there is a higher likelihood of identifying groups of ten as non-infected with a single test, thereby significantly reducing the total number of tests required. Our results show that for current infection rates, advanced sample pooling methods can identify up to ten times more cases per test than individual testing.
Disclaimer: Please note that these findings are generated using non-audited, research-grade open source software. The code can be found here: https://github.com/SC-SGS/covid19-pooling.
Instructions: Please adapt the input parameters to reflect your specific situation - you will then receive a graphical representation showing the performance of different testing strategies for infection rates ranging from 0.1% to 30%.
Please note: The results obtained via the online tool can differ slightly from those described in the paper because of limitations in available computing power for online simulations. Instead of performing the actual, time-consuming simulation this online tool uses a response surface. The response surface is created with the sparse grid toolbox SG++. It has learned the simulation from precalculated data and approximates it almost perfectly, with an error no larger than the stochastic fluctuations within the original simulation. For full-scale simulations, please use the original code available on GitHub.
Instructions: Hovering over a data point shows simulated results. Draw rectangle to zoom in. Scenarios and standard deviation bars can be switched on and off in the legend.
|Method||Time to test 10% of the population||Optimal pool size|
|Individual testing||234.2 days||1.0|
|2-level pooling||46.0 days||10.0|
|Binary splitting||27.9 days||32.0|
|Recursive binary splitting||24.7 days||32.0|
Parameters: population: 328,240,000; testing capacity p.d.: 146,000; infection rate: 1.0%