Wastewater Based Surveillance Can Be Used to Reduce Clinical Testing Intensity on a University Campus

Abstract

Clinical testing has been a vital part of the response to and suppression of the COVID-19 pandemic; however, testing imposes significant burdens on a population. College students had to contend with clinical testing while simultaneously dealing with health risks and the academic pressures brought on by quarantines, changes to virtual platforms, and other disruptions to daily life. The objective of this study was to analyze whether wastewater surveillance can be used to decrease the intensity of clinical testing while maintaining reliable measurements of diseases incidence on campus. Twelve months of human health and wastewater surveillance data for eight residential buildings on a university campus were analyzed to establish how SARS-CoV-2 levels in the wastewater can be used to minimize clinical testing burden on students. Wastewater SARS-CoV-2 levels were used to create multiple scenarios, each with differing levels of testing intensity, which were compared to the actual testing volumes implemented by the university. We found that scenarios in which testing intensity fluctuations matched rise and falls in SARS-CoV-2 wastewater levels had stronger correlations between SARS-CoV-2 levels and recorded clinical positives. In addition to stronger correlations, most scenarios resulted in overall fewer weekly clinical tests performed. We suggest the use of wastewater surveillance to guide COVID-19 testing as it can significantly increase the efficacy of COVID-19 surveillance while reducing the burden placed on college students during a pandemic. Future efforts should be made to integrate wastewater surveillance into clinical testing strategies implemented on college campuses.

Graphical Abstract

1. Introduction

COVID-19 clinical testing has been an essential component in the effort to restore society to the normalcy of pre-pandemic times. Many aspects of clinical testing (abbreviated as “testing” in this paper) have been optimized throughout the pandemic with the aim of increasing public safety. However, improvements have not removed the toll that testing places on a population. COVID-19 testing is a multifaceted experience and persons being tested can undergo psychological pressures and have been found to employ various coping methods to deal with that affiliated stress (Kathiravan et al. 2021). Hesitancy towards COVID-19 testing has been observed for several reasons including social stigmatization, fear of enforced quarantine and simply to avoid the mental strain of waiting for test results (Thappa et al. 2020). The imposed disruption to daily life that positive test results place on an individual coupled with the self-resolving course of the disease, especially among young asymptomatic individuals, can foster the false belief that testing is a waste of time and should even be avoided (Thappa et al. 2020).

Though anyone is at risk of COVID-19-mediated mental health effects, younger adults perceived more COVID-19 stressors and reported more elevated occurrences of depression and anxiety symptoms than older adults (Bruine de Buin 2020). Furthermore, specific experiences associated with the pandemic such as disruption to daily life, perceived notions that the pandemic would continue to affect daily life, and reduced confidence in the government’s response were significant predictors of depression symptoms and were associated with substance abuse in college students (Stamatis et al. 2022). Given that COVID-19 testing intensity can have a negative effect on the mental health of young adults it is clear how a structured COVID-19 testing strategy that promotes purposeful and minimal testing frequency can be beneficial for a university campus and similar communities.

Measurements of genome concentrations present within wastewater have been used to estimate the incidence rates of COVID-19 (Ahmed et al. 2020a, Haramoto et al. 2020, Lu et al. 2020, Sherchan et al. 2020, La Rosa et al. 2021, Zhan et al. 2023) and other illnesses (Sharkey et al. 2023, Babler et al. 2023b). Incidence is defined as the number of diagnosed cases within a population over a given period of time. It is chosen as the clinical measure because of the way clinical data was mostly available during the pandemic, through clinical diagnosis of individuals who chose to get tested.

The reason wastewater surveillance can be used to estimate incidence of disease is because infected individuals (whether pre-symptomatic, symptomatic, and asymptomatic individuals) who make use of the established sewer system will release SARS-CoV-2 through their urine and feces into wastewater (Sharkey et al. 2021, Jones et al. 2020, Medema et al. 2020). The more individuals infected, the higher the SARS-CoV-2 RNA levels in the wastewater. The level of infection as measured through wastewater is an indirect measure of incidence.

In addition to the inclusive nature of wastewater, wastewater sample collection is nonintrusive, does not disrupt the daily routines of its contributors, is independent of clinical testing, and can be more cost effective to estimate disease incidence (Bowes et al. 2023). Not surprisingly, wastewater surveillance monitoring programs have been initiated throughout the world to establish early detection (Ahmed et al. 2020a, Haramoto et al. 2020, Lu et al. 2020, Sherchan et al. 2020, La Rosa et al. 2021).

It has been shown that wastewater surveillance of university campuses can be combined with clinical data to establish trends between SARS-CoV-2 levels in the wastewater and incidence of disease in individual buildings (Cohen et al. 2022, Sellers et al. 2022, Gibas et al. 2021, Betancourt et al. 2021, Vo et al. 2022, Solo-Gabriele et al. 2023) with sensitivity to find incident COVID-19 cases exceeding 60% (Godinez et al. 2022). The objective of this study was to analyze whether wastewater surveillance can be used to track disease incidence while reducing the intensity of clinical testing. At the University of Miami, which is where this study was conducted, incidence tracking was guided by comprehensive clinical testing. Wastewater levels were used to validate incidence levels from clinical testing. However, given the clinical testing intensity we aimed in this paper to evaluate whether a program guided by wastewater testing could track disease incidence with less clinical testing. We analyzed a wastewater surveillance data set that was based on samples from eight different residential buildings on campus and accompanying human health data from each building to determine whether differing testing intensities affected the correlation between SARS-CoV-2 levels in the wastewater and the number of recorded positive cases which is a measure of incidence rates. Our study is unique in that we have both extensive clinical and extensive wastewater surveillance data sets, which provide the ability to conduct such an analysis.

2. Methods

2.1. Wastewater sample collection

The University of Miami (UM) began a wastewater surveillance program in September 2020; it continued through the end of September 2022. Initially, all sample collection focused on facilitating wastewater measurements for the entire University campus. However, recognizing the utility of wastewater surveillance and the need to further fine-tune testing for smaller populations with high incidence, sample collection efforts were adjusted to focus on individual residential halls starting January 2021 and continuing through May 2022. Refinements beyond the building level were not possible due to the design of the sewage collection system. Early during the pandemic sample collections were performed once a week from eight different residential buildings on campus, which collectively housed a maximum of 5,104 students. To further facilitate prevalence tracking in the dormitories, university administration advocated for additional sample collection efforts, so from September 2021 to January 2022 sampling increased to twice a week. The timeline of sample collection efforts for individual buildings therefore corresponded to weekly sampling from January 2021 through August 2021, twice a week from September 2021 to January 2022, and weekly again from February 2022 through May 2022. The cadence of wastewater sample collection focused on collecting samples during weekdays with reporting by Thursday evening of the same week, such that the administration could meet to discuss the wastewater results along with clinical data on Fridays. The eight residential buildings for which wastewater was collected included four nearly identical towers (Hm, Hp, W, R), a large residential hall (P), a one-story residential building (N), and two additional residential halls consisting of clusters of buildings (L, V). The student population size and general characteristics of each building are provided in Table 1.

Table 1:

Characteristics of residential halls for which separate wastewater samples were collected.

Residence Hall	Maximum occupancy	Bathrooms	Number of residential floors	Number of clustered buildings	General Characteristics
Hm	428	Communal	12	1	Single Tower Dormitory
Hp	428	Communal	12	1	Single Tower Dormitory
W	428	Communal	12	1	Single Tower Dormitory
R	428	Communal	12	1	Single Tower Dormitory
P	1426	Individual	14	2	Two Tower Dormitory
N	391	Individual	4	1	Single building
L	1115	Individual	5	25	25 Interconnected buildings
V	460	Individual	21	7	7 Disconnected buildings

Grab samples were collected at each of the buildings by lowering a sterile 2-liter bottle connected to a chain into the corresponding sewer holes. The 2-L bottle containing the wastewater sample was then split in the field into a 0.5 L plastic beaker and into a 0.5-L bottle. The wastewater placed in the beaker was analyzed for basic water quality measurements in the field (pH, temperature, turbidity, dissolved oxygen, and specific conductivity). The 0.5-L bottle contained pre-dispensed 0.5 mL sodium thiosulfate (100 g/L) to remove potential chlorine residual. This bottle was immediately placed in a cooler with ice packs and was taken to UM’s Sylvester Comprehensive Cancer Center Biospecimen Shared Resource (BSSR) laboratory for subsequent treatment and concentration for SARS-CoV-2 quantification. All samples were received at the laboratory within three hours of sample collection and all ice packs were still frozen upon receipt. Ambient temperatures never fell below freezing and so samples were never frozen before pre-processing. Standard practices for field safety associated with wastewater surveillance were utilized, with 99.5% isopropyl alcohol as the main chemical disinfectant in the field.

2.2. Wastewater sample analysis

At the BSSR laboratory, the 0.5-L wastewater samples collected from each residential building underwent a pre-treatment process. First a recovery control was added (recovery from filtration through qPCR analysis). The recovery control was heat-inactivated (15 min @ 56 °C) OC43, which was added to levels of 10⁶ gc/L. Once the recovery control was added the sample then underwent electronegative filtration (Ahmed et al 2020, Babler et al. 2022, 2023a). To the sample 4.7 mL of MgCl₂ (51%) was added and the initial pH was measured. Ten percent HCl was then added with a dropper to lower the pH to a designated range of 3.5–4.5. Acidified wastewater from each sample location was then filtered to clogging through 47 mm diameter, 0.45 μm pore size filters (MilliporeHAWP4700). Filters were then folded and placed into 1.5 mL of 1×DNA/RNA Shield (Zymo) and stored at 4 °C, until subsequent RNA extraction and qPCR analysis at the UM Center for AIDS Research (CFAR).

At the CFAR laboratory, RNA was extracted from 250 μL of the filter concentrate using a Quick-RNA^™ Viral Kit (Zymo Research) as described earlier (Sharkey et al. 2021, Babler et al. 2023c). An inhibition control (a spike-in of Human Immunodeficiency Virus RNA) was added to each RNA eluate. The purpose of the inhibition control was to check for possible sample interferences during qPCR amplification. A blank was also created per sample set (10 μL nuclease-free water+30 μL HIV RNA) to evaluate the HIV qPCR signal without a sample to assess the relative interference (or inhibition) from the sample. Samples were analyzed for molecular targets via Volcano 2^nd Generation (V2G)-qPCR on a Bio-Rad CFX Connect Real-Time System (Bio-Rad Laboratories Inc., USA) as described previously (Sharkey et al. 2021, Babler et al. 2022). Molecular targets chosen for analysis included SARS-CoV-2 (targeting the nucleocapsid N3 gene) (Babler et al., 2022), plus indicators of human waste, beta-2-microglobulin (B2M) – specifically the single stranded mRNA of the protein-coding gene, and pepper mild mottled virus (PMMoV). These human waste indicators are used to estimate the amount of dilution from wastewater that does not originate from bathroom sources (e.g., dishwashing water and clothes washing water). Target-specific standards with concentrations ranging from 10¹ to 10⁵ copies/μL (used to develop the standard curve) were included with each 96-well plate analyzed. All sample handling for pre-treatment occurred within a Biosafety Level 2 laminar flow hood, and standard laboratory safety practices were followed.

2.3. Student testing

The University of Miami adapted and responded to the unprecedented events of COVID-19 with a rapidly built and robust SARS-CoV-2 testing program (Nimer et al. 2020) that included a Testing, Tracing and Tracking (3-T) surveillance platform. This initiative aimed to monitor changes in infection burden that might have required changes in public health strategy for the UM community. During the Spring 2021 semester, this program included weekly screening of UM students who chose to live on campus and/or attend classes in person. This screening was the primary method to assess disease spread on-campus. Wastewater surveillance was used to confirm the level of disease spread as measured through clinical methods.

COVID-19 vaccinations were available to students starting April 2021 such that by the summer of 2021 students who were fully vaccinated were no longer required to undergo testing. Students who returned to campus after summer break regardless of vaccination status were required to be tested within 24 hours of moving in. Students who were not vaccinated were required to participate in subsequent COVID-19 testing at least once weekly through October 1 and twice a week after October 1, 2021. The on-campus COVID-19 testing program ended at the end of March 2022. Data regarding which students had been vaccinated was not provided due to privacy concerns. Only campus-wide aggregated vaccination rates were available.

All testing for COVID-19 was conducted via mid-nasal swab followed by PCR-based diagnosis. Samples were processed at the Mailman Center for Child Development, on the UM Miller School of Medicine campus, which houses the core microbiology laboratory as well as several research laboratories. Clinical samples were placed in viral transport media or universal transport media. They were processed using a Perkin Elmer (PE) assay (Perkin Elmer, 2021a; Perkin Elmer, 2021b) which included nucleic acid extraction via the PE Chemagic. A Thermo Fisher Applied Biosystems QuantStudio 7 or Applied Biosystems 7500 Fast was used for RT-qPCR analyses of clinical samples. This human surveillance program was approved by the university’s Institutional Review Board (IRB 20210164). No individually identifiable data were collected or generated. The clinical information was released to the university community through a dashboard that reported the total daily positive COVID-19 tests. The university community was also informed of the SARS-CoV-2 wastewater sampling by reports in internal news media.

2.4. Data sets analyzed

All data analysis was focused on distinct sampling days. The sampling day was used to link the wastewater SARS-CoV-2 (WW) data to the human health (HH) data. The data were analyzed for each dormitory from January 2021 to March 2022. Although wastewater data was available for a residence halls through May 2022, the on-campus clinical testing sites closed in March 2022, thereby defining the end of the period for which both robust clinical and wastewater data were available. Data from May to August 2021 was excluded as residence halls were mostly vacant during the summer months.

WW data was collected from each of the eight residence halls on a weekly basis and evaluated using the original SARS-CoV-2 RNA signal, $S_i$, corresponding to sampling day $i$. However, wastewater at the individual building scales can be variable and can have a variety of water sources, some of which would contain human waste and others that would not. To account for the possibility of multiple SARS-CoV-2 sources, human waste targets of B2M and PMMoV (Zhan et al. 2022) were used as normalizing parameters, where $B_i=S_i/B\mathit{2}{M}_i$ and $P_i=S_i/PMMo{V}_i.S_i,B_i$, and $P_i$ were calculated for every sample collected on a sampling day. Any samples with $S_i$ levels below detection limit were assigned a value of 100 gc/L. Every $S_i,B_i,$ and $P_i$ value was individually log₁₀ transformed to create $L{S}_i,L{B}_i$, and $L{P}_i$ values, respectively.

The HH data was provided by the University of Miami and consisted of the number of performed COVID-19 tests, $T{e}_n$, and the number of recorded positive cases, $P{o}_n$, on any given day $n$. The HH values were grouped into seven-day sums and defined to correspond to a wastewater sampling day, $i$. The daily values for each sampling day along with the values from the six days before were summed to create testing and weekly incidence totals, such that testing total, $Te{w}_i$ was defined as follows (equation 1).

$$Te{w}_i={\sum }_{n-6}^{n}T{e}_n$$ (eqn. 1)

Similarly, for the daily incidence rate, $P{o}_n$, weekly incidence rate, $Po{w}_i$ was computed using the above equation except $P{o}_n$ was substituted for $T{e}_n$.

From this point forward, all analysis will be explained only in terms of $L{S}_i$, but any analysis done for $L{S}_i$ was also performed for $L{B}_i$ and $L{P}_i$. Three-week moving averages were calculated according to the following equations. Wherein, each $L{S}_i$, was averaged with the two $L{S}_i$ values from the prior two sampling days, creating a $\stackrel{-}{L{S}_i}$ value (equation 2) for each sampling day.

$$\overline{L{S}_i}=\frac{{\sum }_{i-2}^{i}L{S}_i}{3}$$ (eqn. 2)

For the HH data, the corresponding ${Tew}_{,i}$ moving average, $\stackrel{-}{T_i}$, was computed as follows (equation 3).

$$\overline{T_i}=\frac{{\sum }_{i-2}^{i}Te{w}_i}{3}$$ (eqn. 3)

Similarly, for the weekly incidence rate, ${Pow}_i$, the moving average for the weekly incidence rate, $\stackrel{-}{P_i}$ was computed using the above equation except ${Pow}_i$ was substituted for ${Tew}_i$.

Every sampling day had WW moving average values $\left(\stackrel{-}{L{S}_i},\stackrel{-}{L{B}_i},\stackrel{-}{L{P}_i}\right)$, a testing moving average value $\left(\stackrel{-}{T_i}\right)$, and a weekly incidence moving average value $\left(\stackrel{-}{P_i}\right)$. A visual representation of the data can be seen in Figure S1 in the supplemental section. All the moving average values of the same type were grouped to create lists (given by the letter “c” at the end of the reference) as described below.

$$\text{“All Sampling” data set}\{\begin{array}{l}Tc=\left\{\overline{T_1},\overline{T_2},\overline{T_3},\overline{T_4},\dots ,\overline{T_{52}}\right\}\\ Pc=\left\{\overline{P_1},\overline{P_2},\overline{P_3},\overline{P_4},\dots ,\overline{P_{52}}\right\}\\ LSc=\left\{\overline{L{S}_1},\overline{L{S}_2},\overline{L{S}_3},\overline{L{S}_4},\dots ,\overline{L{S}_{52}}\right\}\\ LBc=\left\{\overline{L{B}_1},\overline{L{B}_2},\overline{L{B}_3},\overline{L{B}_4},\dots ,\overline{L{B}_{52}}\right\}\\ LPc=\left\{\overline{L{P}_1},\overline{L{P}_2},\overline{L{P}_3},\overline{L{P}_4},\dots ,\overline{L{P}_{52}}\right\}\end{array}$$

These lists were then manipulated to create distinct data sets. Five data sets were produced for each building. The first data set, called “All Sampling”, had every WW and HH moving average value from every single sampling day. The other 4 data sets were all subsets of the “All Sampling” data set. They were created by removing specific values from the “All Sampling” data set. A visual example is provided in Table 2.

Table 2.

Visualization of the data sets for building V. The “All Sampling” data set includes all sampling days and each following data set (Matching w/ High T, Matching, Testing 10%, and Testing 20%) is made up of distinct combinations of the sampling days within “All Sampling”.

One data set called “Testing 10%”, removed data points associated with sampling days that had a $\stackrel{-}{T_i}$ value within the top and bottom 10% of the range of $\stackrel{-}{T_i}$ values for each building. Another data set was called “Testing 20%” removed data points associated with sampling days that had a $\stackrel{-}{T_i}$ value within the top and bottom 20%. The intention behind these two data sets was to explore how the removal of sampling days with either very high or very low testing intensity would affect the relationship between wastewater SARS-CoV-2 levels and incidence rates.

The other two data sets focused on sampling days that had matching classifications of $\stackrel{-}{L{S}_i}$ and $\stackrel{-}{T_i}$ values. These data sets were created to analyze how the adjustments of testing intensity in response to changes in WW levels can impact the relationship between SARS-CoV-2 and incidence rates. The classifications were either high or low. To determine high from low values, a threshold, $\stackrel{-}{LSc}$, was calculated. The threshold was calculated as the average of the elements listed in $LSc$:

$$\overline{LSc}=\frac{{\sum }_{i=1}^{52}\overline{L{S}_i}}{52}$$ (eqn. 4)

Similarly, the average of the elements listed in $Tc,\stackrel{-}{Tc}$, were computed as:

$$\overline{Tc}=\frac{{\sum }_{i=1}^{52}\overline{T_i}}{52}$$ (eqn. 5)

Then any $\stackrel{-}{L{S}_i}$ that was above $\stackrel{-}{LSc}$ was classified as a high value and any value below $\stackrel{-}{LSc}$ was classified as a low value. $\stackrel{-}{T_i}$ values were classified as high or low in the same way. Every sampling day had either a high or low $\stackrel{-}{L{S}_i}$ and a high or low $\stackrel{-}{T_i}$ value.

One of the classification data sets was called “Matching With High T,” in which any sampling day that had a low $\stackrel{-}{T_i}$ value and a high $\stackrel{-}{L{S}_i}$ value was removed. This resulted in a data set that included only sampling days in which the $\stackrel{-}{T_i}$ and corresponding $\stackrel{-}{L{S}_i}$ values had either the same classification or a high $\stackrel{-}{T_i}$ and a low $\stackrel{-}{L{S}_i}$. That is, the data set included only sampling days in which the testing intensity for that week either matched that week’s SARS-CoV-2 levels or was elevated when SARS-CoV-2 levels were low. For the final data set, “Matching”, the same process was done as for “Matching With High T” except sampling days with a high $\stackrel{-}{T_i}$ and low $\stackrel{-}{L{S}_i}$ were also removed from the dataset. This data set included only sampling days in which testing intensity was low when SARS-CoV-2 levels were low and testing intensity was high when SARS-CoV-2 levels were high.

2.5. Statistical analyses

Regressions between WW values $\left(\stackrel{-}{L{S}_i},\stackrel{-}{L{B}_i},\stackrel{-}{L{P}_i}\right)$ and incidence values $\left(\stackrel{-}{P_i}\right)$ were conducted for each data set to establish the strength of the relationship between WW SARS-CoV-2 levels and incidence rates under differing testing conditions. Each data set had unique lists of WW, incidence, and testing moving average values. The regressions were performed between the lists of WW and incidence moving average values. Fifteen regressions were performed for each building. Each of the five data sets had unique lists of $LSc,LBc$, and $LPc$ which were each regressed with their corresponding unique $Pc$ list. All analyses were performed by SPSS. Spearman correlations (r_s) were performed as the data was found to be non-normally distributed by the Shapiro-Wilk test. Correlations with p < 0.05 were considered significant. Correlation coefficients greater than 0.7 were considered strong. Correlation coefficients less than 0.3 were considered weak.

All the regressions discussed here were performed for lists $LSc,LBc$, and $LPc$ but for the sake of simplicity all analysis will be explained in terms of $LSc$ only. The regressions performed between $LSc$ and $PC$ under the “All Sampling” condition established the correlations between SARS-CoV-2 levels and weekly incidence rates from the actual testing strategy implemented by the university. Then to understand how differing testing strategies could impact the relationship between SARS-CoV-2 levels and incidence rates additional regressions, also between $LSc$ and $Pc$, were run. Each additional regression was under a different data set (Testing 10%, Testing 20%, Matching With High T, and Matching). Figure S2 in the supplemental section depicts the regression pairings.

3. Results

The first three columns of Table 3 list the Spearman r_s values for the regressions between $LSc$ and $Pc,LBc$ and $Pc$, plus $LPc$ and $Pc$ respectively. Each column has data from the five data sets resulting in fifteen r_s values for each building. There is one value for each of the three regressions for all five data sets. Results show that for all eight buildings, the regressions for the “All Sampling” data set were usually weaker than the other data sets. The “Matching” data set showed the strongest correlations across all eight buildings; six of the eight buildings had a correlation stronger than 0.7. Figure 1 is a plot of the data for building V of both the “All Sampling” (panel A) and “Matching” data sets (Panel B). The plots show how the “Matching” data set has fewer peaks in incidence rates, $\stackrel{-}{P_i}$ during periods of low SARS-CoV-2 levels, $\stackrel{-}{L{S}_i}$ resulting in an overall stronger correlation between SARS-CoV-2 levels and incidence rates. The “Testing 10%” and “Testing 20%” data sets resulted in mixed results, with sometimes stronger and sometimes weaker correlations than the “All Sampling” data set.

Table 3.

Spearman correlation coefficients per residential building. The first three columns give the Spearman r_s values between $PMc$ and $LSc,LBc$, and $LPc$, respectively. The last columns correspond to the total number of tests performed, total number of sampling days, and average weekly number of tests performed per data set, respectively. Rows represent the five different data sets for each of the eight buildings.

	LSc vs Pc	LBc vs Pc	LPc vs Pc	Test Count	Sampling Days	Average Weekly Test Count
Building Hm
All Sampling	0.45	0.42	0.45	10,900	51	214
Testing 10%	0.34	0.46	0.31	6,576	41	160
Testing 20%	0.35	0.48	0.37	3,817	29	132
Matching w/ High T	0.75	0.77	0.79	10,482	33	318
Matching	0.79	0.83	0.77	3,734	19	197
Building Hp
All Sampling	0.37	0.48	0.44	9,924	50	198
Testing 10%	0.33	0.48	0.44	6,364	40	159
Testing 20%	0.35	0.55	0.42	3,666	30	122
Matching w/ High T	0.56	0.78	0.66	9,425	38	248
Matching	0.52	0.76	0.65	4,938	29	170
Building R
All Sampling	0.41	0.09	0.13	4,618	48	96
Testing 10%	0.28	0.13	0.14	2,583	39	66
Testing 20%	0.33	0.14	0.14	1,368	30	46
Matching w/ High T	0.63	0.27	0.22	4,184	29	144
Matching	0.76	0.25	0.21	2,542	20	127
Building W
All Sampling	0.55	0.28	0.42	10,393	47	221
Testing 10%	0.51	0.26	0.40	6,775	39	174
Testing 20%	0.60	0.35	0.37	3,990	27	148
Matching w/ High T	0.79	0.75	0.49	9,601	30	320
Matching	0.78	0.79	0.48	4,567	20	228
Building V
All Sampling	0.73	0.69	0.75	65,274	52	1,255
Testing 10%	0.78	0.75	0.80	40,387	41	985
Testing 20%	0.79	0.79	0.86	22,322	32	698
Matching w/ High T	0.86	0.76	0.77	63,438	38	1,669
Matching	0.91	0.89	0.91	5,007	29	173
Building L
All Sampling	0.64	0.70	0.65	13,407	52	258
Testing 10%	0.66	0.71	0.64	9,507	42	226
Testing 20%	0.70	0.75	0.64	6,207	32	194
Matching w/ High T	0.78	0.80	0.70	12,131	31	391
Matching	0.83	0.91	0.83	4,037	18	224
Building N
All Sampling	0.03*	0.13	0.03*	10,520	47	224
Testing 10%	0.02*	0.12	0.06	7,757	39	199
Testing 20%	0.05*	0.02	0.02	4,819	28	172
Matching w/ High T	0.54	0.33	0.55	10,125	31	327
Matching	0.39	0.29	0.45	3639	19	192
Building P
All Sampling	0.37	0.27	0.22	23,933	52	460
Testing 10%	0.40	0.34	0.27	16,545	42	394
Testing 20%	0.45	0.40	0.29	10,393	31	335
Matching w/ High T	0.77	0.71	0.66	21124	27	782
Matching	0.87	0.83	0.72	3,959	12	330

Notes: All values rounded to hundredths place. Values with * have $p$-value greater than 0.05

Figure 1.

Time series plot of “All Sampling” data set and “Matching” data set for building V. Upper panel A corresponds to the “All Sampling” data set and lower panel B corresponds to the “Matching” data set. Due to the large range of $\stackrel{-}{T_i}$ values, it is difficult to see the number of tests conducted when values were low. On March 30, April 6, and April 13, 2021, the values of $\stackrel{-}{T_i}$ were 27.7, 28, and 29.3, respectively.

The Test Count column in Table 3 shows the number of clinical tests performed per building and per data set. Each building has five Test Count values, one for each data set (All Sampling, Testing 10%, Testing 20%, Matching w/ High T, and Matching). The total number of tests performed for all eight buildings during the study period is 148,969 tests. The “All Sampling” Test Count values are the true number of tests performed by the university per building. The “All Sampling” test counts range from 4,618 for building R to 65,274 for building V. Within each building, the Test Count values for the other data sets will always be less than the “All Sampling” Test Count value. This is because other data sets were created by the selective removal of data points associated with specific sampling days. The Sample Count column gives the number of sampling days included in each data set. The Average Weekly Test Count column is the quotient of the Test Count column and the Sample Count column. It provides an average number of tests performed weekly under each data set for each building.

Table 4 focuses specifically on the test count differences between the “All Sampling” and “Matching” data sets for each building. By multiplying the Sample Count and Average Weekly Test Count values for the “Matching” data set we were able to estimate the total number of tests performed if the “Matching” data set could have been applied to the same number of sampling days as the “All Sampling” data set. This value is reflected in the Matching (All Sampling Days) column of Table 4. The Test Count Difference and Percent Difference columns of Table 4 show the difference between the All Sampling and Matching (All Weeks) total test count and the percent decrease or increase of that difference, respectively. Table 4 shows that for most of the buildings if the “Matching” data set could have been applied to the same time frame as the “All Sampling” data set then the total number of tests performed would have decreased. These results are described in greater detail in the discussion.

Table 4.

Comparison of testing data per residential building. Data sets compared are “All Sampling” and “Matching”, with “Matching” showing the highest correlations. Sample counts correspond to the number of wastewater sampling points included in the regression analysis. The average weekly test counts correspond to the number of clinical tests associated with each data set. The total test count corresponds to the total clinical tests corresponding to each data set plus extending the “Matching” data set to estimate the number of clinical tests needed if clinical tests were ordered consistent with the “Matching” data strategy. The last two columns list the differences in the number of clinical tests for each strategy. A negative percent difference indicates that the number of tests would be less for the “Matching” strategy in comparison to the “All Sampling” strategy.

		Sample Count		Average Weekly Test Count		Total Test Count
Building	Max Occupancy	All Sampling	Matching	All Sampling	Matching	All Sampling	Matching (Partial Sampling Days)	Matching (All Sampling Days)	Test Count Difference	Percent Difference
Hm	428	51	19	214	197	10,900	3,374	10,047	853	−7.8%
Hp	428	50	29	198	170	9,924	4,938	8,500	1,424	−14%
R	428	48	20	96	127	4,618	2,542	6,096	1,478	32%
W	428	47	20	221	228	10,393	4,567	10,716	323	3.1%
V	460	52	29	1,255	173	65,274	5,007	8,996	56,278	−86%
L	1,115	52	18	258	224	13,407	4,037	11,648	1,759	−13%
N	391	47	19	224	192	10,520	3,639	9,024	1,496	−14%
P	1,426	52	12	460	330	23,933	3,959	17,160	6,773	−28%
Overall	5,104	399	166	2,926	1,641	148,969	32,423	82,187	66,509

4. Discussion

As stated in the results, the “Matching” data set had the strongest correlations between WW SARS-CoV-2 levels and incidence rates. “Matching with High T” was a very similar data set to “Matching” except that it included sampling days with high testing during low WW levels. The difference in correlations between the two data sets indicate that surplus testing efforts, especially during periods where SARS-CoV-2 levels in the WW were low, negatively impacted the relationship between SARS-CoV-2 WW levels and incidence rates on campus. This negative impact can be due, in part, to overrepresentation of positive cases in which a positive individual is repeatedly tested in too short a time span resulting in multiple counts for the same infected person. The results support the conclusion that testing intensity guided by SARS-CoV-2 WW levels yields stronger correlations between SARS-CoV-2 levels and incidence rates on campus.

The overall decrease in total test counts between the “All Sampling” and “Matching” data sets further supports the advantage of using testing strategies that are guided by WW levels. If a testing strategy that operated under similar conditions as the “Matching” data set (increasing and decreasing testing intensity in rhythm with WW SARS-CoV-2 levels) would have been implemented during our study period, the total number of tests performed could have been reduced by over 66,000. It should be noted that most of those tests are a result of increased testing at building V which could have been due to building V housing students that were more involved in large group activities. These activities may have required a higher number of tests to allow for safe congregation. Even still, if the testing data from building V is excluded, the total number of tests performed for the remaining seven buildings could have been reduced by 7%, or more than 10,000 tests.

Implementing a WW guided testing strategy onto a university campus can maintain a strong correlation between WW SARS-CoV-2 levels and incidence rates across campus while also limiting the total number of tests performed and thereby minimizing the stress placed on students. We now expand on a hypothetical surveillance strategy that uses weekly wastewater sampling and weekly test counts to adjust testing intensity so that it mirrors WW SARS-CoV-2 levels.

4.1. Establishing and continuously updating thresholds

Our proposed strategy of using wastewater surveillance to determine weekly testing intensity adjustments to minimize testing burden on students begins with creating a $LSc$ threshold, $\overline{LSc}$. The threshold will be used to classify weekly $\overline{L{S}_i}$ values as either high or low. The threshold, $\overline{LSc}$, should be continuously updated every week. There will be an initial period of threshold instability at the beginning of this surveillance strategy until enough weeks pass for stable thresholds to be calculated. Figuring out exactly how many weeks are needed for the thresholds to stabilize will require further investigation and is a limitation present in our analysis. We performed all our analysis with a threshold established using a year’s worth of data.

4.2. Adjusting testing intensity

With an established $\overline{LSc}$ it is then possible to classify any new $\overline{L{S}_i}$ value from weekly WW sampling as either being a high or low value. If the classification is high, then for that upcoming week testing efforts should be increased. If the classification is low, then testing efforts can be reduced for the incoming week. Figure S3 in the supplemental text section illustrates the hypothetical surveillance scenario.

This is a very simplified strategy that only attempts to depict how incorporation of wastewater surveillance into testing effort decisions can potentially optimize disease tracking and reduce social burden on a student population. As such, our strategy does not provide specifics of how testing intensity should be adjusted based on weekly SARS-CoV-2 levels. Meaning we only state that testing efforts should either be increased or decreased but exactly how many tests should be performed per building based on SARS-CoV-2 wastewater levels would require more investigation.

4.3. Approaches by other institutions

Across the country, academic institutions were trying to keep their students safe while also attempting to minimize disruptions to their academic careers. (Harris-Lovett et al. 2021, Scott et al. 2021, Gibas et al. 2021, Fox et al. 2021, Wang et al. 2022, Wartell et al. 2022, Lee et al. 2022, Welling et al. 2022, Kazenelson et al. 2023). Some schools focused on clinical testing whereas others incorporated clinical testing with wastewater surveillance. According to Fox et al (2020) Indiana University (IU) used clinical testing to establish mitigation strategies to protect its students. IU switched to online classes at the start of the fall semester of 2020 and implemented a steady increase of diagnostic testing throughout the semester. Testing increased from 17.9 tests per day at the beginning of the semester to 205 tests per day by the end of August, reaching 3,981 tests by October 10^th (Fox et al. 2020). Through the temporary switch to online format, the use of their testing strategy, and other safety implementations IU was able to navigate an outbreak without delaying the academic progress of its students. IU did not incorporate wastewater surveillance into their response to COVID-19, so we are left to consider if fewer tests could have been performed had they used WW measurements as a guiding metric in their response. We now look at other institutions that used a combination of wastewater surveillance and clinical testing in their COVID-19 response.

The University of South Carolina (USC) conducted wastewater surveillance on individual buildings. It used wastewater surveillance data in conjunction with clinical testing data to confirm clinical trends. During the Fall 2020 semester USC collected samples twice a week from the ten residence halls and the on-campus student isolation and quarantine building which collectively served as the housing for 5,955 students that semester. They documented a statistically significant positive relationship between the log₁₀ viral RNA values (copies/L) acquired from wastewater samples collected and the number of positive cases reported in the seven days after each collection date (Sellers et al. 2022).

The University of Arizona (UA) collected wastewater samples twice per week from sewer holes that serviced 13 dorms throughout the Fall 2020 semester (Betancourt et al. 2020). Students were tested through two routes, Campus Health Services (CHS) or Test ALL Test Smart (TATS). CHS testing was reserved only for students that reported symptoms to health services while TATS was used to test every individual living in a dormitory that yielded positive detection of SARS-CoV-2 in its wastewater (Betancourt et al. 2021).

Virginia Tech (VT) collected wastewater samples twice a week from 17 sewer holes throughout the campus from the Fall 2020 semester until the near end of the Spring 2021 semester. In addition to clinical data, VT also acquired accurate building-specific occupancy levels through access-card data to adjust their results to population levels. Interestingly, the work at VT further emphasized that other factors, such as card swipes, can be also used to guide clinical testing. Overall, VT found a positive correlation between the number of viral copies present in the wastewater and the number of positive cases reported in the following days after sample collection (Cohen et al. 2022).

The combined clinical and wastewater surveillance approaches described above for USC, UA, and VT, were the norm during the pandemic and established the correlations between disease incidence and wastewater levels of SARS-CoV-2. However, these schools did not document whether wastewater surveillance was used to establish the testing intensity. One university, Kenyon College (KC), did document a link between wastewater surveillance results and clinical testing intensity. Kenyon College used the viral results obtained from their wastewater surveillance to broadly regulate their clinical testing protocols. KC tested the entire student population, of about 1,600, when consecutive peaks in viral levels were found in the wastewater (Barich and Slonczewski 2021).

The recommendation from this study, using the UM data, differs in that we propose more purposeful adjustments to clinical testing intensity based on viral levels in the wastewater. Our analysis was possible because of the large number of clinical tests performed. There was a total test count of 148,969 for a maximum student resident population of 5,104. The UM was the only university in the south Florida region that remained open to in-person classes during the fall 2020 semester, and given the uncertainty in knowledge about pandemic spread, rightfully relied heavily on clinical testing. The analysis of the clinical and wastewater data from the UM, however, suggests that clinical testing may have been excessive for the goal of establishing disease incidence on campus. Our results showed an improvement in correlative strength between SARS WW levels and recorded positives across the eight residential buildings. This improvement could have been obtained with a 45% decrease in the total number of tests performed, of which a major component was associated with the intense clinical testing at one building. As a result, for future pandemics or outbreaks, a combined approach is recommended where wastewater surveillance guides clinical testing. Testing reduction is not straight forward, and our analysis does not take into account special circumstances that may require intense testing. However, if the goal is to establish disease incidence on campus, the approach presented here can be further developed to reduce the amount of clinical testing.

5. Conclusion

Enduring a pandemic is a stressful experience for all and can be especially difficult for those who struggle to continue their academic career during and immediately after a pandemic. Easing the burdens on students, while maintaining their safety, deserves attention and refined data analyses. We have proposed that combining wastewater surveillance with adjusting testing efforts on a university campus can be useful in easing those burdens. A relentlessly high level of testing will be effective in monitoring the progression of any disease, but such testing imposes financial and personal costs, including the mental and social consequences of constant medical testing. These challenges were especially prominent on university campuses throughout the COVID-19 pandemic when a positive COVID-19 diagnosis could have meant isolation for weeks, removal from campus, or even a loss of an entire semester for a student. The results of this study show that increased testing does not necessarily improve correlations between wastewater SARS-CoV-2 levels and COVID-19 cases. We hypothesize that using wastewater surveillance to guide testing can reduce the number of tests performed while not compromising the ability of a university to monitor the health and safety of its students.

This study was a preliminary analysis that demonstrates that wastewater surveillance can be an effective resource to optimize COVID-19 testing strategies on a university campus. Our data was limited to 2.5 semesters in which a variety of safety procedures were in place for students across one university campus. Varying degrees of residential occupancy within different buildings, the difficulty of tracking cases throughout a commuter student population, and a limited number of wastewater sampling days all render our working data incomplete. In addition, the clinical testing, although extensive, may have not diagnosed all infected individuals, especially after mandatory testing ended for vaccinated students. An additional limitation of the study was the lack of vaccination data by residential building as only the aggregated vaccination statistics were available for the entire campus.

Given the limitations, our analysis for the data that was available supports the hypothesis that more testing alone does not necessarily improve the ability of wastewater SARS-CoV-2 measurements to predict the number of positive COVID-19 cases. Indeed, we were able to improve correlations by decreasing testing frequency. The results of this study support future efforts to develop testing strategies that can use quantifiable changes in wastewater SARS-CoV-2 levels to guide specific numerical adjustments in testing intensity. Future research is needed to better understand the relationship between wastewater SARS-CoV-2 levels and testing intensity so that a set change in SARS-CoV-2 levels in wastewater from one week to the next might trigger a specific and optimal change in testing intensity. Future studies should integrate modeling approaches inclusive of machine learning and Bayesian statistical approaches to refine the strategy of integrating wastewater measurements into a clinical testing program. Using wastewater surveillance coupled with advanced modeling approaches to regulate testing efforts on a university campus can potentially help to maximize the safety of students while also minimizing the burdens placed on them.

Highlights.

• Measured WW Sars-CoV-2 levels were compared to human health data

• Different testing strategy scenarios were created and compared

• Strategy with testing intensity that matched WW levels had strongest correlation

• Showed that fewer total tests and fewer weekly tests were more efficient

• Universities can use WW data to regulate testing intensity to reduce mental stress