key: cord-1012496-hqc3f9sx
authors: Pileggi, V.; Shurgold, J.; Sun, J.; Yang, M. I.; Edwards, E.; Peng, H.; Tehrani, A.; Gilbride, K.; Oswald, C.; Wijayarsi, S.; Al-Bargash, D.; Stuart, R.; Khansari, Z.; Raby, M.; Thomas, J.; Fletcher, T.; Simhon, A.
title: Quantitative Trend Analysis of SARS-CoV-2 RNA in Municipal Wastewater Exemplified with Sewershed-Specific COVID-19 Clinical Case Counts
date: 2022-03-13
journal: nan
DOI: 10.1101/2022.03.13.22272304
sha: 770358655b945cc727dfe719f5a840d0bf0ee4de
doc_id: 1012496
cord_uid: hqc3f9sx

We present and demonstrate a quantitative statistical linear trend analysis (QTA) approach to analyze and interpret SARS-CoV-2 RNA wastewater surveillance results concurrently with clinical case data. This demonstration is based on the work completed under the Ontario (Canada) Wastewater Surveillance Initiative (WSI) by two laboratories in four large sewersheds within the Toronto Public Health (TPH) jurisdiction. The sewersheds were sampled over a 9-month period and data were uploaded to the Ontario Wastewater Surveillance Data and Visualization Hub (Ontario Dashboard) along with clinical case counts, both on a sewershed-specific basis. The data from the last 5-months, representing a range of high and low cases, was used for this demonstration. The QTA was conducted on a sewershed specific approach using the recommendations for public health interpretation and use of wastewater surveillance data by the United States Centers for Disease Control and Prevention (US CDC). The interpretation of the QTA results was based on the integration of both clinical and wastewater virus signals using an integration matrix in an interim draft guide by the Public Health Agency of Canada (PHAC). The key steps in the QTA consisted of (i) the calculation of Pepper Mild Mottle Virus (PMMoV), flow and flow-PMMoV-normalized virus loads; (ii) computation of the linear trends including interval estimation to identify the key inflection points using a segmented linear regression method and (iii) integrated interpretations based on consideration of both the cases and wastewater signals, as well as end user actionability. This approach is considered a complementary tool to commonly used qualitative analyses of SARS-CoV-2 RNA in wastewater and is intended to directly support public health decisions using a systematic quantitative approach.

As of August 1, 2021 the wastewater contributed by over 70% of the population of the province of Ontario Ai et al, 2021 [7] from the state of Ohio in the USA investigated nine WWTPs in central Ohio and 96 considered normalization methods with PMMoV and Cross-Assembly phage (crAssphage) and also considered if wastewater surveillance can serve as a sentinel piece for detecting SARS-CoV-2 variants of concern within 98 a community. SARS-CoV-2 RNA concentrations in wastewater and COVID-19 cases were found to correlate well when they were imputed using a 5-day moving day average. Normalization using the mean concentration 100 in the community was investigated along with linear and polynomial models to improve correlations. Also, the sequencing results from wastewater samples showed agreement with the sequencing results from clinical 102 nasal swab samples during an early period of new strain emergence. It was concluded that wastewater surveillance is ideal for fast tracking variant emergence and transmission within a community. 104 An additional metric introduced by Xiao et al, 2021 [8] which when combined with clinical case data trends may provide additional insight into the potential for under-or over-estimation of disease incidence 106 in the community is the wastewater viral signal (WVS) to clinical case count (CCC) ratio or WCR. It was reported that an increase in the WCR would imply an increase in asymptomatic cases in the community; a 108 decrease in the WC ratio would imply clinical testing may be over-estimating disease incidence when counting previously infected cases as new cases and a baseline or no change in the WC ratio, would imply there is 110 sufficient public health capacity. Additionally it was recommended that the magnitude and range of the WC ratio be established for each sewershed to assess a population baseline where sufficient public health testing 112 capacity indicated a relatively constant WC ratio. Furthermore the WCR was recommended as a useful metric to better detect short-term trends of asymptomatic disease transmission compared to the wastewater 114 and clinical data independently [8] .

Despite these significant advances in demonstrating a strong relationship between wastewater-based 116 surveillance data and clinical cases for COVID there remains a gap in quantitative approaches to interpret these trends collectively. Here we report and demonstrate the utility of a novel approach for quantifying 118 and interpreting sewershed-specific trends of both wastewater viral RNA signals (WVS) and clinical case counts (CCC) with Toronto as a case study. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 13, 2022. ; https://doi.org/10.1101/2022.03.13.22272304 doi: medRxiv preprint 122 The QTA methodology is based on an implementation of the recommendations on public health interpretation and use of wastewater surveillance data from the United States Centers for Disease Control and 124 Prevention (US CDC) [9] and the recommendations on the integration of clinical and wastewater surveillance signals in the interim draft guide by the Public Health Agency of Canada (PHAC) [10, 11] . The following key 126 steps were applied:

(a) Calculation of the normalized virus loads per US CDC recommendations [9] . 128 (b) Computation of the linear trends as per US CDC recommendations [9] and using break-point linear regression [12] . 130 (c) Interpretations based on PHAC draft integration matrix and guidance [10, 11] .

The data used in this report was downloaded from the secure Ontario Dashboard. Two academic labo-132 ratories in their analysis of 24-hour wastewater composite field samples from the influent to WWTPs (post screening and prior to degritting) followed prescribed laboratory quality assessment and quality control 134 (QA/QC) measures and field recommended practices referenced and described in the MECP documents [13, 14] . The generated primary data included viral RNA quantified using RT-PCR with the N1 and N2 136 primers as well as PMMoV. Each lab collected and analyzed samples 3-5 days a week during the testing period. Key metadata included the mean daily flowrate over the 24-hour sample period provided by Toronto [15,16], a framework for wastewater surveillance, before being uploaded to the Ontario Dashboard. An additional MECP document provides details about laboratory protocols and is intended to address the issue of 142 dealing with 'trace' concentration levels or values measured at or below the method limit of detection (LOD) [17] . In our approach we considered all technical and biological replicate values as reported for integration 144 into arithmetic mean values. Further, if the arithmetic mean average resulted in a zero-value, this was replaced by one-half of the 'UJ' value [17] within the reported data set. This final data adjustment was done 146 for wastewater data only and to eliminate numerical issues associated with log-transformation of zero-values prior to break-point linear regression.

The case studies for this work focused on four large urban municipal sewersheds described in Table 1 and included the Ashbridges Bay (TAB), the Humber River (THR), the Highland Creek (THC) and the North CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted March 13, 2022. ; The overall workflow in the QTA process is given in Figure 1 with a description of the data sets used and the related sample R-code provided in the Supplement. Details of the steps used in the QTA are described 154 in the following sections.

Three different normalizations of the wastewater SARS-CoV-2 RNA signal were considered and included:

PMMoV normalized concentration with the 2020 census population per 10 5 introduced as a factor C vb in 

where:

(i) C v is the virus RNA gene concentration in gene copies per ml ( gc virus mL ).

(ii) C b is the fecal biomarker concentration in gene copies per ml ( gc biomarker mL ) which takes into account the temporal variation of the population (P ) in the sewershed [3].

(iii) Q is the mean daily flowrate in megaliters per day ( M L d ) corresponding to the 24-hour composite sample collected of the influent wastewater at the wastewater sewage treatment plant (WWTP).

(iv) P is the contributing sewershed 2020 population (cap).

(v) M b is the median biomarker gene counts per mL of ( gc biomarker mL ) in the wastewater throughout the 168 population during the complete sampling campaign [3].

(vi) The 10 3 is included for the unit conversion of L to mL. 170 6 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted March 13, 2022. ; Figure 1 . Quantitative trend analysis (QTA) flowchart showing the three major steps in the QTA process: Normalization, Trend Analysis and Integrated Interpretation with automated optimum inflection point selection (Note that segmented() refers to the main R-function used from the segmented R-package).

(vii) The 10 6 is included for the unit conversion of ML to L.

(viii) The 10 5 is included to adjust the value per 10 5 population (cap).

Eqs. 1, 2 and 3 apply equally for any virus RNA gene target including N1, N2 or N1N2 (average of N1 and N2). Further N1 and N2 represent the mean of technical replicates.

The virus biomarker normalized concentration C vb in Eq. 1 may be interpreted as the virus concentration 7 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The segmented R-package employs internal hypothesis testing about the existence of breakpoints and 198 the use of sequential hypothesis testing to select the optimum number of breakpoints from the seed values.

Various appropriate statistical methods are also integrated into the segmented routine to assess the confidence 200 intervals and error margins associated with the break-points and linear trends [12].

The above analysis was applied to each sewershed specific data set and the results of the segmented 202 routine, were extracted to determine the break-points identifying the dates of major changes in either the clinical case counts (CCC) or the wastewater signal (WVS) which were used to determine the time intervals 204 and duration between significant changes. The slope of each segmented linear trend line was determined and associated with the slope standard errors (SE) and 95% confidence intervals (CI). Because the wastewater 206 signal was log-transformed, the percent daily change (P DC) was computed per US CDC [9] using Eq. 4:

where: 208 8 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted March 13, 2022. ; https://doi.org/10.1101/2022.03.13.22272304 doi: medRxiv preprint (i) P DC is the percent daily change (%) in wastewater signal.

(ii) slope is the log 10 transformed data regression line segment slope. 

The interpretations for each time interval were determined based on a comparison and integration of 224 trend directions for both the case and wastewater signals using PHAC led collaborative recommendations derived from an interpretation integration matrix [10] . A total of 9 potential permutations were considered 226 with each of the clinical case counts and wastewater signal having three potential outcomes for each slope or trend: increasing, decreasing or baseline. Table 2 provides the general interpretations of the integrated 228 signals. Note that baseline is taken to mean that clinical case counts (CCC) or the wastewater signal (WVS) is near zero or at the limit of quantification, respectively. Additionally, general interpretations are provided 230 to enable local public health epidemiology, recognizing that each scenario requires additional context that may be influencing either signal.

For example, even a scenario with both the CCC and WVS trending up, could be a result of local factors (e.g., an outbreak among elderly residents at a long-term care facility combined with WWTP sewershed 234 maintenance operations that results in increased sediments in the collected wastewater sample which would not provide any real evidence of increased community transmission). Further, now that more than 78% of the 236 population aged 12 and older are fully vaccinated in Ontario, these interpretations might be more applicable to urban centers. However, there are still many caveats to consider. The QTA approach alone may not be 238 informative and interpretations may be confounding without a fuller consideration of the local field conditions. stratification of the case counts by vaccination status is not fully understood and may change as a function of waning immunity; and (iv) As mass clinical testing is discontinued, the clinical signal may increasingly rely 244 9 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 13, 2022. ; https://doi.org/10.1101/2022.03.13.22272304 doi: medRxiv preprint on hospitalization data as the only remaining robust and high-quality source of clinical information. Further, would a spike in the wastewater signal and an increase in hospitalizations trigger investigations for variants 246 of concern (VOC) and outbreak management decisions? [10, 11] .

The core of the interpretation message is that, in most cases, wastewater provides a source of high-quality 248 information that is independent from clinical surveillance (and associated pit-falls) that ultimately allows epidemiologists to better interpret local data in order to make appropriate decisions. Essentially, WVS that 250 are baseline or trending down contribute public health intelligence that there is less cause for concern, while WVS that are trending up signify increasing concern [10,11].

The wastewater to cases ratio (WCR) [8] was also considered and integrated as part of an extension to the base QTA report intended to demonstrate the versatility of the QTA methodology to integrate new metrics 254 for additional trend analysis and insight. Table 3 provides an integration of the PHAC and WC recommended interpretations intended to supplement the CCC and WVS trend interpretations [10, 11] . 10 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 13, 2022. ; 

The QTA results were provided in a succinct 2-page report considered a template model report for different 258 metrics under considerations and for direct use by public health officials. In addition to the trend analysis results the QTA report also includes a statistical error analysis of each break point and trend slope for the 260 end user to better assess the confidence of the reported trends. 

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 13, 2022. ; https://doi.org/10.1101/2022.03.13.22272304 doi: medRxiv preprint Table 2 and Table 3 12 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The TAB extended aggregated data set was downloaded from the Ontario Dashboard and used to generate the longer-term trends as part of the QTA under different normalization conditions (C vb , L v and L vb ). In 290 this section we focus on the TAB sewershed which has an approximate area of 25,000 ha and a population of 1.6 million which is about 52 % of the population of the City of Toronto (Table 1) Table 4 summarizes the trend results with the 95% confidence interval (CI) and Table 5 provides the interpretations based on the PHAC interpretations 306 given in Table 2 . The standard errors (SE) associated with the estimated breakpoint and trend slopes for 13 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) a DC and PDC, refer to daily change and percent daily change, respectively; CI, is the 95 % confidence interval. The CCC is trending down and the WWS is trending up. Weak evidence of escalating community transmission.

Both the CCC and WWS are trending up. Strong evidence of escalating community transmission. 15 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 13, 2022. ; 

This section introduces the WCR metric to demonstrate the versatility of the QTA approach and consti-312 tutes Part 2 of 2 of the model QTA report. Here we adopt the interpretations provided by Xiao et al, 2021

[8] within the QTA framework and this is provided in Figure 5 and Tables 8 − 11 for the TAB sewershed 16 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 13, 2022. ; The CCC is trending down and the WC Ratio is trending up. Weak evidence of escalating community transmission and increase in asymptomatic cases in the community. T3

The CCC is trending down and the WC Ratio is trending up. Weak evidence of escalating community transmission and increase in asymptomatic cases in the community. T4

Both the CCC and WC Ratio signals are trending down. No evidence of escalating community transmission and clinical testing may be over-estimating disease incidence when counting previously infected cases as new cases. T5

The CCC trending up and the WC Ratio is trending down. Weak evidence of escalating community transmission and clinical testing may be over-estimating disease incidence when counting previously infected cases as new cases.

The CCC trending up and the WC Ratio is trending down. Weak evidence of escalating community transmission and clinical testing may be over-estimating disease incidence when counting previously infected cases as new cases. T7

Both the CCC and WC Ratio are trending up. Strong evidence of escalating community transmission and increase in asymptomatic cases in the community.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 13, 2022. ; 

An assessment of the linear and non-linear correlation between C vb , L v , L vb and mN 1 (un-normalized N1) and clinical case counts by reported data were also considered ( Figure 6, Table 12 ). In terms of the 318 Linear correlation results and the R 2 )values, the PMMoV normalization (C vb ) was slightly better than the un-normalized (mN 1) which was better than the flow normalization (L v ) and the mean PMMoV-flow 320 normalization (L vb ) for the TAB WWTP, sequentially. However, when considering the non-linear correlation results, the AICcWt factor which refers to the proportion of the total predictive power that can attributed to preferred. Also considering that the adjusted R 2 (aR 2 ) is higher than the R 2 for both C vb and L vb using the PMMoV-flow normalization is preferred over the PMMoV normalization. Similar screening level analyses 328 can be conducted for N1N2 or N2 prior to settling on a specific normalization mode that can then be used consistently for a given sewershed.

18 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. a Slope ± SE, is the slope and standard error associated with the linear correlation regression; R 2 , is the coefficient of determination; p-value, is the probability that the predictor (WVS) is related to the response (CCC) compared to the null hypothesis of being unrelated; AICc, is the Akaike information criterion; AICWt, is the AICc weight; aR 2 , is the adjusted R 2 and PO, is the polynomial order of the best fit curve up to order 9.

The methodology of the QTA (Figure 1 ) provided a means to integrate the three key steps of data 332 normalization, trend analysis and interpretation in a semi-automated approach to reduce personal bias.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. the type of normalization most appropriate was found to be sewershed specific and dependent on the time scale considered, but in some cases significant similarities were observed between the different normalization 342 methods. Additional effort is required to collect and analyze the necessary metadata (flow) so this should be considered at the planning stages.

To automate the trend analysis, the minimum AIC was introduced as a means to select the best polynomialfit and to determine the seed inflection points for the segmented break-point linear regression model. Despite 346 the noisy data sets, this method was found effective using a 2 -9 order polynomial (m = 2 − 9, step 2 (iv) Figure 1 ). However, depending on the resolution required and data sets considered, the upper value of m 348 may be readily increased to serve the purpose and computed seed inflection points may be manually selected for the intended purpose. An additional consideration of the method included error propagation with the 350 standard error (SE) and 95 % confidence interval (CI), associated with the breakpoints and trends. As the data set sample size (n) is reduced the corresponding SE and CI increase and reducing the confidence level in 352 the final results and potentially compromising the interpretations. Sample size of n < 3 is not recommended to determine trends. As an example, in our QTA analysis for the current data sets we observed a trend based 354 on n = 3 ( Figure S5 , THR shorter term C vb CCC by reported date) which also included the highest SE and CI that crossed the zero value and did not prove to be significant at the 95 % confidence level.

The interpretation relied on pre-determined interpretations considering the integration of the CCC and WVS or WCR trend results provided in Table 2 

The TAB sewershed is a combined system and was the largest in land area (25,000 ha) and most populous (1.6 M) ( Table 1) CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 13, 2022. ; https://doi.org/10.1101/2022.03.13.22272304 doi: medRxiv preprint L vb ), an assessment of CCC by reported data versus episode date along with the use of the WCR compared 366 to the more direct WVS. The Supplement includes additional QTA results using L v and L vb normalizations shown in Figures S1 -S3 and Tables S1 -S12.

From Figure 4 C vb , we can see that the WVS trends up past the peak of the CCC (end of interval T1) and continues to trend up for 14-days (end of interval T2) while cases are trending down during interval T2. This 370 may be indicative that virus shedding continues to increase (additive loadings by active cases) within the community up to the beginning of interval T2. This was consistently observed in the L v , L vb normalizations 372 for CCC by reported date and also for L vb with CCC by episode date (Figures S1 -S3) .

A comparison between L vb ) using CCC by episode date versus reported date (Figures S2 and S3) shows 

The THR sewershed serves a population of 685,000 with an area of 18,000 ha having a partially combined Table S13 we can see that the WVS trends up marginally but close to a baseline and generally not responsive to the CCC increases (from 5 to 10 and from 10 to 35 cases) during intervals (T1 398 21 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 13, 2022. ; https://doi.org/10.1101/2022.03.13.22272304 doi: medRxiv preprint -T3) from July 6/21 to August 23/21. Only during interval T4 (August 23 -25/21) when CCC increased above 40 do we see a response and increase in the WVS trend. However, throughout this time period, the 400 WVS trends are not statistically significant (CI range crosses zero) and this may be related to being at the apparent limit of resolution. A closer look shows that the CCC are below 10 during T1, less than 40 during 402 interval T2 and less than 50 during T3 -T4 and assuming a factor of 10 in community cases compared to the CCC, we have a ratio in the order of less than 1/10,000 based on a population served of 685,000. From 404 Figure S5 its also observed that 16/18 (89 %) WVS data points are below the LOD (color coded green) of 2.5 gc/mL during T1 and suggests that we are at the limit of resolving the change in cases during these event 406 up to when the CCC increases to above the order of 1/10,000.

Comparing the WVS trends using flow normalization (L v ) compared to the L vb and C vb normalizations, 408 shows an increase in the number of intervals or break-points (6 versus 4) suggesting the flow normalization, at low CCC, shows higher variability. This may be attributed to the fact that the THR sewershed is a partially (Tables S26 -S29) . 430 

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 13, 2022. ; https://doi.org/10.1101/2022.03.13.22272304 doi: medRxiv preprint Figure S11 (WC ratio of $L_{vb}) shows the WC ratio and CCC within four intervals (T1 -T4). During interval T1 and T3 the WC ratio supports the up-trend of CCC suggesting that there is an increase in 432 asymptomatic cases in the community. During interval T2 the WC ratio is trending down while the CCC is trending up and the WC ratio trend results is suggesting that clinical testing may be over-estimating when 434 counting previously infected cases as new cases. During interval T4 the CCC cases are trending down while the WC ratio is trending up which is suggesting that asymptomatic cases in the community are increasing 436 despite the decreasing CCC trend indicating weak evidence of escalating community transmission. Figure S12 provides linear correlation of the CCC by reported date to the WVS by 100K population.

438 Table S34 shows a weak but significant linear data correlation (R 2 < 0.24 and p-value < 0.02, generally). The non-linear AICwt is between 0.54 -0.63 with C vb and L v normalization providing the best model correlation.

This data set differs from the THR data set by not having any data below the LOD. However the the weak linear correlation appears to have contributed to a statistically insignificant trends in 3 of the 4 time intervals 442 at the 95% confidence level and this is attributed to smaller sample size (n) when compared to the daily CCC. Figure S14 using C vb normalization, shows the CCC and WVS trends over seven time intervals (T1 -T7) and similarly to the TAB longer-term QTA results the WVS, WVS trends up past the peak of the CCC 452 (end of interval T1) and continues to trend up for 13-days (end of interval T2) while cases are trending down during interval T2. This appears to be a common phenomena most evident in the longer-term QTA results 454 and attributed to continued virus shedding in the community beyond the peak observed in CCC. Figure S14 , during interval T3 and T4, both the WVS and CCC are trending down in 456 concordance. During interval T5 the CCC trend is at baseline and the WVS is trending down but at a low level with much of the data below the LOD and this continues during interval T6 and T7. During T6 and 458 T7 the WVS does not change and does not respond to the increase and decrease in CCC while the trends in cases are less than 1 case/day. For the TNT sewershed, the QTA results are similar for both the C vb and L v 460 normalizations ( Figure S15 and Tables 39 -S42) using the WVS metric directly.

The trends ( Figure S16 and Table S43 -S45) are increasing during intervals T1 -T3 and T7 and this 462 suggests that clinical testing may be over-estimating disease incidence when counting previously infected 23 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Table S47 shows a good and significant linear data correlation (0.63 ≤ R 2 ≤ 0.73 and p-value < 0.01) as well as non-linear correlation. Based on the AICcWt value of 1.0 and 0.98 the C vb and L vb normalization modes 468 are preferred. This data set is similar to the TAB data set in correlation values (Table 12 and showing similar consistent properties in terms of significant trends. 

The primary purpose of this work was to demonstrate the utility of the QTA approach to help inform 472 public health decisions. Data sets from four urban sewersheds within the Toronto PHU were analyzed using QTA both over a longer-term (5-months) and shorter-term (4 -8-weeks) and under different normalizations has shown the robustness of the QTA process and this is currently being documented.

The longer-term (5-months) QTA for TAB and TNT sewersheds demonstrated that PMMoV-normalized (C vb ) and median flow-PMMoV normalization (L vb ) generally provide more conservative results, in terms 480 of increased WVS prevalence, over flow-normalized (L v ) SARS-CoV-2 RNA data. The shorter-term (5 -8 weeks) QTA on THC and TNT sewersheds provided similar findings to the longer-term analysis and provided 482 some additional insights into the potential impact of the data at or below the method LOD.

Overall the major WVS and CCC trends were captured by the three normalizations (C vb , L v or L vb ) 484 however, significant differences in trends were observed particularly when the data contained a large percentage (over 50%) of data below the reported LOD. To assess the more conservative and recommended 486 normalization approach, for any given sewershed, it is suggested that an initial comparative assessment of the QTA process using different normalizations be conducted and assessed based on a linear and non-linear 488 assessment based on the AICcWt criteria to determine the appropriate normalization mode. Further, the extent of clinical testing and the underlying assumption that it reflects the general population cases, should 490 be considered carefully as part of the integration of the WVS and CCC for interpretation. It is expected, that situations will arise (e.g., due to reduced severity of symptoms clinical testing becomes less prevalent) 492 that may make this integration less relevant. In such situations, the WVS may become a determining stand alone representative metric.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted March 13, 2022. ;

Since the ratio L vb /C vb is the median PMMoV divided by the sample PMMoV ( M b P M M oVi ) these normalization modes may yield similar QTA results particularly when this ratio is relatively constant during the 496 sampling campaign within a given sewershed.

Further testing on the QTA method is ongoing to further automate the process to generate complete 498 analysis and interpretations for different sewershed types and normalizations.

The work presented in this manuscript also served as a case study and pilot to assess the benefits of the 500 QTA process for production level reporting. The analysis, based on PMMoV-normalized wastewater signals, was well received with a positive feedback from the TPH and PHAC. It is noteworthy, that in the case of the 502 sewersheds within the TPH jurisdiction, the QTA analysis showed that the WVS was generally not predictive or did not lead the CCC. Despite this, the QTA served as a complimentary and confirmatory metric. To 504 this end it is recommended that QTA continue to be evaluated and used to generate timely interpretative quantitative reports to support public health decisions as a complementary tool to clinical data information.

This work was supported through the Ontario Wastewater Surveillance Initiative of the Ontario Ministry 508 of the Environment, Conservation and Parks (MECP).

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) 

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted March 13, 2022. ; https://doi.org/10.1101/2022.03.13.22272304 doi: medRxiv preprint

Wastewater Surveillance of the COVID-19 Genetic Signal in Sewersheds Recommendations from Global Experts. WRT Summit Handout V3b

Show us the Data: Global COVID-19 Wastewater Monitoring Efforts, Equity, and Gaps

Data filtering methods for SARS-CoV-2 wastewater surveillance

A wastewaterbased epidemic model for SARS-CoV-2 with application to three canadian cities

COVID-19 monitoring in rural communities: First comparison of lagoon and pumping station samples for wastewater-based epidemiology 2021

Catching a resurgence: Increase in SARS-CoV-2 viral RNA identified in wastewater 48 h before COVID-19 clinical tests and 96 h before hospitalizations

Wastewater-based epidemiology for tracking COVID-19 trend and variants of concern in ohio, united states

Metrics to relate COVID-19 wastewater data to clinical testing dynamics

Outline of Wastewater Surveillance Data Interpretation (Draft Guide). Canada: Public Health Agency of Canada (Lead)

Public Health Agency of Canada. Draft -SARS-CoV-2 wastewater-based surveillance: A guide for public health action

Interval estimation for the breakpoint in segmented regression: A smoothed score-based approach

Transfer Payment Agreements (Internal MECP Agreements)

MECP Wastewater Surveillances Best Practices Technical Guide (MECP Draft Internal Guide

The public health environmental surveillance open data model (PHES-ODM or ODM)

Open Data Model for collecting, quality-ensuring and sharing of SARS-CoV-2 data and metadata

Sewage Sentinel System for SARS CoV-2 -5th Town Hall Meeting, e-poster

MECP Wastewater Surveillance Initiative, Protocol for Evaluations of RT-qPCR Performance Characteristics (ISBN 978-1-4868-5481-3 Internal Laboratory Protocol)

Highland Creek Treatment Plant 2020 Annual Report

Elegant graphics for data analysis. ggplot2

Welcome to the tidyverse

Dates and times made easy with lubridate

Reshaping data with the reshape package

Scales: Scale functions for visualization. scales

Miscellaneous functions for "grid" graphics

Cowplot: Streamlined plot theme and plot annotations for 'ggplot2

Kowr: Utility functions

RStudio: Integrated development environment for r

Authoring books and technical documents with R markdown

LaTeX: A document preparation system

Reproducible research data and project management in r. Reproducible Research

Enumerating asymptomatic COVID-19 cases and estimating SARS-CoV-2 fecal shedding rates via wastewater-based epidemiology