Distributed lag model for estimating incidence using SARS-CoV-2 RNA concentration from wastewater solids

In a recent study posted on medRxiv* A preprint server, an interdisciplinary team of researchers from the United States (US), has confirmed in the laboratory coronavirus disease 2019 (SARS-CoV-2) via the coronavirus 2 (SARS-CoV-2) ribonuclear syndrome. Prediction of incidence (IR) of COVID-19) was evaluated. Acid (RNA) concentration in wastewater solids using a robust dispersion lag model (DLM).

The team reduced sampling in different sewer sheds of different sizes and features to analyze the effect of sampling frequency on IR prediction errors.

Background

During the COVID-19 pandemic, researchers around the world used wastewater-based epidemiology to monitor SARS-CoV-2 RNA levels in wastewater and predict COVID-19 incidence in the community. ..

DLM uses the effect of explanatory variables that occur over time rather than once on the dependent variable. DLM has been applied to previous studies of COVID-19 wastewater surveillance, but of new SARS-CoV-2 variants for optimal sampling frequency and the association between COVID-19 cases and SARS-CoV-2 RNA levels. Not used to identify the impact. Wastewater.

This study was conducted to estimate whether wastewater monitoring helps predict COVID-19 incidence in real time, using a regression model scaled to a fairly cost-effective sampling frequency.

Research design

This study analyzed four public treatment facilities (POTW) in California. Two of these POTWs serve the population of Santa Clara County (SJ and PA), and each of the remaining POTWs is in Sacramento County (Sac), San Mateo County (PA), and Yolo County (Dav). It worked as part. From November 2020 to September 2021, approximately 50 ml of solid wastewater samples were collected daily from each POTW.

They used a high-throughput procedure consisting of an automated device and a liquid processing robot to measure the concentrations of the SARS-CoV-2 nucleocapsid (N) gene and pepper mild mottle virus (PMMoV) RNA, and externally used bovine coronavirus. (BCoV) was used. Control.

The team downsampled the datasets collected daily for N / PMMoV at each POTW to four low sampling frequencies (every 2 days, 3 days, 4 days, and weekly for model fitting). Case studies of COVID-19 incidents from each sewer hut were collected from georeferenced home addresses and portrayed using POTW-specific Geographic Information Systems (GIS) shapefiles. The incidence of COVID-19 was used as a dependent variable and was calculated by the estimated population in each sewer hut.

Various regression models were fitted to the dataset sampled daily for each POTW, such as the linear model (Eq 1), DLM (Eq 2), and the variation of Eq 2 using the percent delta as the explanatory variable. To reduce sampling frequency, the team fitted both linear and DLM models to each POTW and created a DLM with two predictors that fit each sampling frequency.

Researchers calculate the root-mean-squared error (RMSE) in-sample (May 11, 2020 to July 19, 2021) and out-of-sample period (July 20, 2021 to September 15, 2021). By doing so, I predicted the performance of the model. Reported as IR (Case / 100000).

Survey results

The SARS-CoV-2 RNAN gene and PMMoV range were not detected at 3.71×10 in the four POTWs studied.⁶ cp / g and 7.12×10⁷ Up to 3.74×10^Ten cp / g. Maximum and minimum concentrations of the N / PMMoV and SARS-CoV-2RNA N genes were observed prior to the delta surge.

The researchers used the N / PMMoV data during the sample period of Equation 1-2 and the entire sample period of Equation 3-4 to select the most effective model from the candidate models. Based on the Bayesian Information Criterion (BIC), a model that fits SJEq2 with U = 3 was selected rather than a linear model. In particular, for all POTWs, the coefficient estimates for the explanatory variables are very different from 0 and are positive, reflecting the high logs._Ten N / PMMoV was associated with increased logs_Ten IR.

The linear and DLM models were fitted using data on the reduction in wastewater sampling frequency across four POTWs. At each sampling frequency based on the increase in R in the sample, DLM was prioritized over the linear model.² value. The difference in BIC between the DLM and the linear model is less than 1.0, indicating no priority. With the exception of the Sac weekly model, the regressor’s coefficient estimates were positive, significantly different from 0.

The team observed a maximum increase in immediate impact effect only at SJ (t = 0) for DLM compliance at lower sampling frequencies compared to Dav, PA, and Sac. The linear regression coefficients across the four POTWs of the daily sampling of N / PMMoV ranged from 0.51 to 0.84. For daily sampling DLM, the coefficient estimates were the same as the standard error and the sampling frequency was reduced.

Researchers noted that the off-sample model performed well. In the case of SJ and PA, the RMSE outside the sample decreased compared to the case inside the sample, but in the case of DA, there was an increase of 3 cases / 100000. Also, the Sac RMSE outside the sample was 10 cases / 100,000 higher than the RMSE inside the sample.

For model fits with reduced sampling frequency across POTW, the median RMSE outside the sample was predominantly 3 cases / 100,000 higher than the median RMSE inside the sample. However, the one with the highest Sac difference of about 7 is excluded. Case / 100,000.

Throughout POTW, surges were observed in DLMs captured in and out of the sample, as predicted by IR traces. When predicting IR using a reduced sampling frequency, the predictions varied widely depending on the particular sample date.

In particular, for all POTWS, the RMSE of the model based on daily sampling was smaller than the RMSE of the low frequency model, both inside and outside the sample. During the sample period, researchers found that when daily sampling was reduced to weekly sampling, the maximum RMSE increased by 20 cases / 100,000 in SJ, followed by 16, 8, and 4 cases / 100,000 in Dav, Sac, and PA. Observed to increase. , Each. The maximum increase in RMSE from daily sampling was approximately 7 cases / 100,000 while sampling once every 2, 3, and 4 days.

Conclusion

The results of this study revealed that DLM has strong predictive power to estimate COVID-19 infection rate in California through wastewater monitoring of the SARS-CoV-2N gene. DLM used Model Fit to monitor the spikes and declines in COVID-19 cases between delta variants and other new variants.

This study enables real-time prediction of COVID-19IR with fewer errors, even when SARS-CoV-2 variants are circulating, by reducing the sampling frequency once every 4 days, weekly. I showed that.

*Important Notices

medRxiv publishes unpeer-reviewed preliminary scientific reports and should not be considered definitive, guide clinical / health-related behaviors, or be treated as established information.