[Paper Review] An Epidemiological Model for contact tracing with the Dutch CoronaMelder App

Peter Boncz|arXiv (Cornell University)|May 18, 2021

COVID-19 Digital Contact Tracing4 references5 citations

TL;DR

This study evaluates the effectiveness of the Dutch CoronaMelder digital contact tracing app by modeling its impact on reducing COVID-19 infections, hospitalizations, and deaths, comparing it to manual contact tracing by GGD services. Using real-world data on infections and hospitalizations, the model estimates the app averted approximately 45,088 cases, while manual tracing averted 773,599 cases—indicating a 17:1 effectiveness ratio, though potential improvements in adoption and risk scoring could enhance future impact.

ABSTRACT

We present an epidemiological model for the effectiveness of CoronaMelder, the Dutch digital contact tracing app developed on top of the Google/Apple Exposure Notification framework. We compare the effectiveness of CoronaMelder with manual contract tracing on a number of metrics. CoronaMelder turns out to have a small but noticeable positive influence in slowing down the COVID-19 pandemic, an effect that will become more pronounced in an opened-up society where adoption of CoronaMelder is increased.

Motivation & Objective

• To quantitatively assess the effectiveness of the Dutch CoronaMelder digital contact tracing app in reducing COVID-19 transmission.
• To compare the app’s performance against manual contact tracing by GGD services in the Netherlands during the second wave (Oct 2020–May 2021).
• To estimate the number of averted infections, hospitalizations, ICU cases, and deaths attributable to both digital and manual contact tracing.
• To analyze the impact of the app and manual tracing on the effective reproduction number Rt, using a 4-day serial interval model.
• To explore how future improvements in app adoption and risk scoring could enhance digital contact tracing effectiveness in post-lockdown scenarios.

Proposed method

• The study uses a model-based approach to estimate the number of averted infections, hospitalizations, and deaths by comparing observed infection and hospitalization data with counterfactual scenarios without contact tracing.
• It computes the effective reproduction number Rt using a 4-day serial interval, modeling the growth rate of daily infections relative to infections 4 days prior.
• The model incorporates real data on positive PCR tests and hospitalizations to calibrate estimates, using serology studies to inform overall pandemic size.
• It estimates the app’s effectiveness by modeling the number of infections averted through digital warnings, assuming a 16% app adoption rate and a 40% compliance rate with warnings.
• The model accounts for the infectious window (2 days before symptom onset to day of GGD call) and uses TEK uploads from confirmed cases to trigger warnings in the app.
• It compares two scenarios: manual contact tracing alone and manual tracing enhanced by the CoronaMelder app, estimating incremental reductions in transmission.

Experimental results

Research questions

• RQ1 How many COVID-19 infections, hospitalizations, and deaths were averted by the Dutch CoronaMelder app during the second wave?
• RQ2 How does the effectiveness of the CoronaMelder app compare to that of manual contact tracing by GGD services?
• RQ3 To what extent did the app reduce the effective reproduction number Rt, and how does this compare to manual tracing?
• RQ4 What factors could improve the app’s future effectiveness in a post-vaccination, open-society setting?
• RQ5 How sensitive are the estimates to assumptions about app adoption, compliance, and risk scoring accuracy?

Key findings

• The CoronaMelder app averted an estimated 45,088 infections during the study period (October 12, 2020 – May 16, 2021), based on a 16% adoption rate and 40% compliance with warnings.
• Manual contact tracing by GGD services averted an estimated 773,599 infections, representing 4.5% of the Dutch population (17.4 million).
• The app averted 595 hospitalizations, 113 ICU cases, and 271 deaths, based on infection-to-hospitalization and infection-to-death ratios.
• The app reduced the effective reproduction number Rt by an average of 0.0005 per week, compared to 0.0091 for manual contact tracing alone.
• The app’s effectiveness is estimated to be 17 times lower than manual tracing, but could improve significantly with higher adoption or better risk scoring.
• In a post-lockdown, post-vaccination environment, the app could become relatively more effective than manual tracing due to its ability to detect anonymous contacts.