Predictive modelling approaches

Analysis of relevant predictive modelling approaches suitable to the database from the TKI report mentioned a preference for Partners for Urban Water method. [1] However, in order to apply predictive modelling, a good correlation is needed between parameters. Therefore a KPI has been developed to show the value of infections, hospitalisations and WWTP measurements relative to a selected date. The TKI project shows that a good correlation was present during the second and third waves in the Netherlands, but recently correlation has been lost due to vaccinations and the onset of the Omicron variant. Currently, the testing strategy has also changed, also due to the introduction of self-tests at home. Therefore, we believe that the current number of infections is underestimated.

Figure 8. Correlation between infections, hospitalisations and nationwide WWTP measurements for the Netherlands relative to 29 October 2020 (=100%)

Since the latest trends show that with the reduction of severity of the latest SARS-COV-2 variant (omicron) there is no longer a correlation between infections, hospitalisations and WWTP data. It is no longer feasible with the current variant to use clinical infections or WWTP measurements to forecast hospitalisations. Therefore, no predictive modelling was undertaken.

Moreover, data science requires a large number of data points. Sampling at every 2-14 days does not result in enough WWTP data points to implement data science.

Instead of using data science based on long-term trends in time, KPIs (key performance indicators) were introduced into the model to show the growth rate of infections and hospitalisations, which could function as early warning signs.

For consistency, it would be best if the same sampling and measurement strategy is applied to i) WWTP measurements and ii) number of infections, to obtain a consistent dataset over time. We recommend to continue sampling the same WWTPs as long as COVID-19 poses a problem. Ideally, infections too should be measured in the same way in the future (i.e. via PCR testing), but with the arrival of self-test kits, this is no longer feasible. Hospitalisations will be reported as before, but the data will become increasingly contaminated with people with COVID-19, who are hospitalised for other/multiple issues. This highlights that only WWTP data allows for a consistent comparison of SARS-CoV-2 detection since the outbreak of COVID-19 in February 2020, across all COVID variants. This is one of the most important points to support the use of wastewater-based surveillance and a point that should be emphasized more often when arguing for the utilisation of wastewater-based epidemiology (WBE) to monitor outbreaks.

  • 1 TKI Rijnmond project - 2021