Assessment of the Effective Sensitivity of SARS-CoV-2 Sample Pooling Based on a Large-Scale Screening Experience: Retrospective Analysis

Abstract

Background: The development of new large-scale saliva pooling detection strategies can significantly enhance testing capacity and frequency for asymptomatic individuals, which is crucial for containing SARS-CoV-2.

Objective: This study aims to implement and scale-up a SARS-CoV-2 screening method using pooled saliva samples to control the virus in critical areas and assess its effectiveness in detecting asymptomatic infections.

Methods: Between August 2020 and February 2022, our laboratory received a total of 928,357 samples. Participants collected at least 1 mL of saliva using a self-sampling kit and registered their samples via a smartphone app. All samples were directly processed using AutoMate 2550 for preanalytical steps and then transferred to Microlab STAR, managed with the HAMILTON Pooling software for pooling. The standard pool preset size was 20 samples but was adjusted to 5 when the prevalence exceeded 2% in any group. Real-time polymerase chain reaction (RT-PCR) was conducted using the Allplex SARS-CoV-2 Assay until July 2021, followed by the Allplex SARS-CoV-2 FluA/FluB/RSV assay for the remainder of the study period.

Results: Of the 928,357 samples received, 887,926 (95.64%) were fully processed into 56,126 pools. Of these pools, 4863 tested positive, detecting 5720 asymptomatic infections. This allowed for a comprehensive analysis of pooling's impact on RT-PCR sensitivity and false-negative rate (FNR), including data on positive samples per pool (PPP). We defined Ct_ref as the minimum cycle threshold (Ct) of each data set from a sample or pool and compared these Ct_ref results from pooled samples with those of the individual tests (ΔCt^P). We then examined their deviation from the expected offset due to dilution [ΔΔCt^P = ΔCt^P - log₂]. In this work, the ΔCt^P and ΔΔCt^P were 2.23 versus 3.33 and -0.89 versus 0.23, respectively, comparing global results with results for pools with 1 positive sample per pool. Therefore, depending on the number of genes used in the test and the size of the pool, we can evaluate the FNR and effective sensitivity (1 - FNR) of the test configuration. In our scenario, with a maximum of 20 samples per pool and 3 target genes, statistical observations indicated an effective sensitivity exceeding 99%. From an economic perspective, the focus is on pooling efficiency, measured by the effective number of persons that can be tested with 1 test, referred to as persons per test (PPT). In this study, the global PPT was 8.66, reflecting savings of over 20 million euros (US $22 million) based on our reagent prices.

Conclusions: Our results demonstrate that, as expected, pooling reduces the sensitivity of RT-PCR. However, with the appropriate pool size and the use of multiple target genes, effective sensitivity can remain above 99%. Saliva pooling may be a valuable tool for screening and surveillance in asymptomatic individuals and can aid in controlling SARS-CoV-2 transmission. Further studies are needed to assess the effectiveness of these strategies for SARS-CoV-2 and their application to other microorganisms or biomarkers detected by PCR.

Keywords: COVID-19; PCR; SARS-CoV-2; nonsymptomatic; pooling; saliva; screening; sensitivity; surveillance; transmission control.

Figure 1

Schematic diagram of the pooling process. Starting with a population sampling of m individuals (where the red individuals symbolize an infected person), a total of N pools were formed, with each composed of the selected preset number. Afterward, RT-PCR was performed, and a result was obtained in which the luminescence of each target gene was analyzed and it was determined whether the sample was positive (ie, infected). If it was positive, the pool was undone, and an individual PCR was performed on the members of the pool who were detected to be positive. RT-PCR: real-time polymerase chain reaction.

Figure 2

Pooling versus individual results (Galicia, Spain, data collected between August 2020 and February 2022; N=887,926 samples; retrospective analysis of SARS-CoV-2 infection). This compound figure includes the distributions of minimal Ct values for (A) individual tests and (C) sample pooling, and (B) breakdown of the Ct distributions for each of the genes (E, N, RdRp, and S) on the individual test (left) and within the pool (right). In such decomposition, individually, there are genes that tend to deviate to higher Ct values (such as RdRp), with gene S generally marking the positive detections.

Figure 3

ΔΔCt^P_G(i) concept. (Galicia, Spain, data collected between August 2020 and February 2022; N=887,926 samples; retrospective analysis of SARS-CoV-2 infection). Explanation of the concept of ΔΔCt^P_G(i) and its relationship with ΔΔCt^P_G(i) for gene S and pool size 20 (as an example). This differentiation is crucial for assessing the true deviation in infection detection due to the intrinsic deviation of log₂(N) inherent to the chosen pool size (N). This is evident through the distribution of ΔΔCt^P_G(i), which demonstrates a dispersion centered around 0 and illustrates instances where certain cases fall ahead of or behind the reference threshold cycle (Ct) value.

Figure 4

Global insights. The overall results of the analysis of individual and pool tests were evaluated. The results have been compartmentalized based on the samples that have undergone processing through pooling and those that have been discarded due to noncompliance with established criteria. Of the samples that were fully processed, a differentiation was made between those that were nondetected and those that were positive (ie, infected). Additionally, a representation of all the pools that were executed is provided, and they have been categorized into those that were nondetected and those that were positive and further individually analyzed.

Figure 5

Positivity evolution (Galicia, Spain, data collected between August 2020 and February 2022; N=887,926 samples; retrospective analysis of SARS-CoV-2 infection). Comparative analysis of the distribution of infected cases during the pandemic in the local area of Vigo (Spain; light blue) with the positives detected in our laboratory (dark blue) in temporal increments of 2 weeks. Furthermore, a histogram is included below, which illustrates the progression of pool sizes (ie, N) in relation to the percentage of positives for each pool size (ie, %POS/N) throughout the pandemic.

Figure 6

Evolution of the COVID-19 pandemic throughout the study period (Galicia, Spain, data collected between August 2020 and February 2022; N=887,926 samples; retrospective analysis of SARS-CoV-2 infection). (A) The number of cases analyzed is shown (solid black line); additionally, the number of PPT (solid red line) and the prevalence sampled in pharmacies (gray bar), other sources (orange bar), and globally (green bar) are shown. (B) The relationship between the PPT and the global prevalence is shown with the theoretical performance under the 20:1 and 5:1 protocols given as a reference. PPT: persons per test.

Figure 7

Individual Ct versus Ct_ref (Galicia, Spain) between August 2020 and February 2022 (N=887,926 samples; retrospective analysis of SARS-CoV-2 infection). A scatter plot showing the Ct value of each gene on the individual test versus the minimal Ct (ie, Ct_ref), colored by the vertical distance between such values, which is known as ΔCt. The violin plots shown at the top of each subfigure are those cases in which the individual gene (either E, N, RdRPS, or S) failed to detect a positive sample. Gene S shows a lower dispersion in the results in comparison with gene N, which not only has a higher dispersion of Ct but also has a higher distribution of failures reaching low values of Ct_ref.

Figure 8

Pooling minimal Ct versus Ct_ref (Galicia, Spain, data collected between August 2020 and February 2022; N=887,926 samples; retrospective analysis of SARS-CoV-2 infection). Comparison of individual Ct_ref (x-axis) and pool minimal Ct (y-axis) for all pools with (A) PPP=1 and small pool size (N<6), (B) PPP=1 and large pool size (N>15), (C) PPP>1 and small pool size (N<6), and (D) PPP>1 and large pool size (N>15). The correlation between the 2 minimum Ct values is especially apparent for those pools that contained a single positive (top) and even more so for those of smaller size (A). However, as the viral load increases in the sample pool, this correlation vanishes. PPP: positive samples per pool.

Figure 9

False negative (FN) versus Ct_ref. The probability of an FN for each test is presented, with the results for individual tests (solid line) and a reference pool size of 20 (dashed line), separated by gene (E, N, RdRp/S, and S). As anticipated, the probability of an FN is higher for pooled samples compared with individual tests and higher when it gets closer to 40, as expected.