SUMMARY Background Surveillance of SARS-CoV-2 in wastewater offers an unbiased and near real-time tool to track circulation of SARS-CoV-2 at a local scale, next to other epidemic indicators such as hospital admissions and test data. However, individual measurements of SARS-CoV-2 in sewage are noisy, inherently variable, and can be left-censored. Aim We aimed to infer latent virus loads in a comprehensive sewage surveillance program that includes all sewage treatment plants (STPs) in the Netherlands and covers 99.6% of the Dutch population. Methods A multilevel Bayesian penalized spline model was developed and applied to estimate time- and STP-specific virus loads based on water flow adjusted SARS-CoV-2 qRT-PCR data from 1-4 sewage samples per week for each of the >300 STPs. Results The model provided an adequate fit to the data and captured the epidemic upsurges and downturns in the Netherlands, despite substantial day-to-day measurement variation. Estimated STP virus loads varied by more than two orders of magnitude, from approximately 1012 (virus particles per 100,000 persons per day) in the epidemic trough in August 2020 to almost 1015 in many STPs in January 2022. Epidemics at the local levels were slightly shifted between STPs and municipalities, which resulted in less pronounced peaks and troughs at the national level. Conclusion Although substantial day-to-day variation is observed in virus load measurements, wastewater-based surveillance of SARS-CoV-2 can track long-term epidemic progression at a local scale in near real-time, especially at high sampling frequency.

Background Surveillance of SARS-CoV-2 in wastewater offers a near real-time tool to track circulation of SARS-CoV-2 at a local scale. However, individual measurements of SARS-CoV-2 in sewage are noisy, inherently variable and can be left-censored. Aim We aimed to infer latent virus loads in a comprehensive sewage surveillance programme that includes all sewage treatment plants (STPs) in the Netherlands and covers 99.6% of the Dutch population. Methods We applied a multilevel Bayesian penalised spline model to estimate time- and STP-specific virus loads based on water flow-adjusted SARS-CoV-2 qRT-PCR data for one to four sewage samples per week for each of the more than 300 STPs. Results The model captured the epidemic upsurges and downturns in the Netherlands, despite substantial day-to-day variation in the measurements. Estimated STP virus loads varied by more than two orders of magnitude, from ca 1012 virus particles per 100,000 persons per day in the epidemic trough in August 2020 to almost 1015 per 100,000 in many STPs in January 2022. The timing of epidemics at the local level was slightly shifted between STPs and municipalities, which resulted in less pronounced peaks and troughs at the national level. Conclusion Although substantial day-to-day variation is observed in virus load measurements, wastewater-based surveillance of SARS-CoV-2 that is performed at high sampling frequency can track long-term progression of an epidemic at a local scale in near real time.

Capture mark recapture (CMR) models allow the estimation of various components of animal populations, such as survival and recapture probabilities. In recent years, incorporating the spatial distribution of the devices used to detect an animals presence has become possible. By incorporating spatial information, we explicitly acknowledge the fact that there will be spatial structuring in the ecological processes which give rise to the capture data. Individual detection probability is not heterogeneous for a range of different reasons, for example the location of traps within an individuals home range, the environmental context around the trap or the individual characteristics of an animal such as its age. Spatial capture recapture models incorporate this heterogeneity by including the spatial coordinates of traps, data which is often already collected in standard CMR approaches. Here, we compared how the inclusion of spatial data changed estimations of survival, detection probability, and to some extent the probability of seroconversion to a common arenavirus, using the multimammate mouse as our model system. We used a Bayesian framework to develop non spatial, partially spatial and fully spatial models alongside multievent CMR models. First, we used simulations to test whether certain parameters were sensitive to starting parameters, and whether models were able to return the expected values. Then we applied the non-spatial, partially spatial and fully spatial models to a real dataset. We found that bias and precision were similar for the three different model types, with simulations always returning estimates within the 95% credible intervals. When applying our models to the real data set, we found that the non-spatial model predicted a lower survival of individuals exposed to Morogoro virus (MORV) compared to unexposed individuals, yet in the spatial model survival between exposed and non-exposed individuals was the same. This suggests that the non-spatial model underestimated the survival of seropositive individuals, most likely due to an age effect. We suggest that spatial coordinates of traps should always be recorded when carrying out CMR and spatially explicit methods of analysis should be used whenever possible, particularly as incorporating spatial variation may more easily capture ecological processes without the need for additional data collection that can be challenging to acquire with wild animals.