Understanding Epidemic Drivers Using GEMS

Modeling

1.   ANDRZEJ JARYNOWSKI (1), ALEJANDRA RINCÓN HIDALGO (2), ASHISH THAMPI (2), STEVEN SCHULZ (2), MARLLI ZAMBRANO (1,2), ALISA SERGEEVA (1), PHILIP EL-DUAH (3), REXFORD MAWUNYO DUMEVI (3), HANNA-TINA FISCHER (3), JOHANNA WILDEMANN (3), JULIANE BÖNECKE (4), JONATHAN STRÖBELE (4), VERENA STRUCKMANN (5), VINCENT FINDEISS (5), DANIEL OPOKU (5,6), JOHN AMUASI (4,5), VITALY BELIK (1), ON BEHALF OF THE BUA FLATTENING THE CURVE CONSORTIUM. (1) SYSTEM MODELLING GROUP, INSTITUTE OF VETERINARY EPIDEMIOLOGY AND BIOSTATISTICS, FREIE UNIVERSITÄT BERLIN, GERMANY (2) MACHINE LEARNING UNIT, DEPARTMENT OF ENGINEERING, NET CHECK GMBH, BERLIN, GERMANY (3) CHARITÉ CENTER FOR GLOBAL HEALTH, CHARITÉ – UNIVERSITÄTSMEDIZIN BERLIN, GERMANY (4) BERNHARD NOCHT INSTITUTE FOR TROPICAL MEDICINE (BNITM), HAMBURG, GERMANY (5) KUMASI CENTRE FOR COLLABORATIVE RESEARCH (KCCR), KWAME NKRUMAH UNIVERSITY OF SCIENCE AND TECHNOLOGY (KNUST), KUMASI, GHANA (6) DEPARTMENT OF HEALTH CARE MANAGEMENT, TECHNISCHE UNIVERSITÄT BERLIN, GERMANY Understanding Epidemic Drivers Using GEMS Microsimulation and Bayesian Transmissibility Framework Applicable to LMICs

2. ACCOUNTING FOR SPREADERS We define a contact as two devices that are co-located within an area of 8m x 8m within 2 minutes.

3. Transmissibility estimation (Germany) CONTACT INDEX CONTACT DRIVEN mobile phones!!! TRANSMISSIBILITY TRANSMISSION EFFICIENCY PER CONTACT (VARIANT- AND IMMUNIZATION-DRIVEN)

4. Transmissibility estimation from contacts vacc 1 vacc 2 booster % difference (to baseline) - 59.4 % - 33 % +8% VACCINATION II VACCINATION I ALPHA DELTA OMICRON BOOSTING BASELINE

5. Flattening the Curve: epidemiology time Number of cases/hospitalizations/deaths Without intervention With intervention: reduction of physical contacts, reduction of infection probability, shortening of infectious period Health care system capacity

6. SP4 Population Model SP2 Behavioral Adaptation SP5 Simulation (GEMS) SP6 Policy Making SP3 Epidemiologic Parametrization SP1 Contact Parametrization

7. What Else Can You Do With GEMS Building custom transmission mechanics Model vaccination strategies Building custom contact mechanics Model immunity and waning Add custom post- processing steps Run GEMS on HPC clusters Building custom populations Add custom loggers for simulation runs

8. GEMS Model Architecture Simulations Scenarios Batches Populations Interventions

9. Lockdown testing - Ghana The full lockdown in the Ashanti/Greater Accra Region, which began on March 30, 2020, was lifted on April 20, 2020. @ Desmond Sarfo Boampong, Gloria E. Swilla, Amquador Acquah Keelson, Gideon Kwarteng Acheampong No NPI: 26200 infections (8%) Peak ~5,600. 21 days Lockdown: 17,750 infections (5.4%) Peak ~4,000 (↓32%)

10. Large Household Size Urban Residence Larger Population Density Lower Access to healthcare (HCA) COVID-19 registered cases Small Share of elderly people

11. Key findings Average delay (Ct → Rt): WT 17d, Alpha 15.5d, Delta 16.5d, Omicron 14dDelta: +47% transmissibility vs Alpha, Omicron: +32% transmissibility vs Delta Primary vaccination transmissibility ↓ 71% reduction vs natural pos infection immunity Rt shaped by social (contacts) + biological (variants/vaccines) factors

12. CONCLUSIONS We recover transmissibility and separate the effects of contact patterns , variant fitness , and immunization ML-based surveillance + GEMS scenario modeling = real-time epi analysis tool Mobile data–based transmissibility estimation is globally applicable, especially in LMICs (Rt estimates feasible from sentinel or wastewater surveillance ) Provides actionable insights for public health planning and intervention, without sophisticated labs!

13. CONCLUSIONS We recover transmissibility and separate the effects of contact patterns , variant fitness , and immunization ML-based surveillance + GEMS scenario modeling = real-time epi analysis tool Mobile data–based transmissibility estimation is globally applicable, especially in LMICs (Rt estimates feasible from sentinel or wastewater surveillance) Provides actionable insights for public health planning and intervention, without sophisticated labs! Andrzej Jarynowski