BACKGROUND: The \'third wave\' of COVID-19 in Hong Kong, China was suppressed by non-pharmaceutical interventions (NPIs). Although social distancing regulations were quickly strengthened, the outbreak continued to grow, causing increasing delays in tracing and testing. Further regulations were introduced, plus \'targeted testing\' services for at-risk groups. Estimating the impact of individual NPIs could provide lessons about how outbreaks can be controlled without radical lockdown. However, the changing delays in confirmation time challenge current modelling methods. We used a novel approach aimed at disentangling and quantifying the effects of individual interventions.
METHODS: We incorporated the causes of delays in tracing and testing (i.e. load-efficiency relationship) and the consequences from such delays (i.e. the proportion of un-traced cases and the proportion of traced-cases with confirmation delay) into a deterministic transmission model, which was fitted to the daily number of cases with and without an epi‑link (an indication of being contact-traced). The effect of each NPI was then calculated.
RESULTS: The model estimated that after earlier relaxation of regulations, Re rose from 0.7 to 3.2. Restoration of social distancing to the previous state only reduced Re to 1.3, because of increased delay in confirmation caused by load on the contact-tracing system. However, Re decreased by 20.3% after the introduction of targeted testing and by 17.5% after extension of face-mask rules, reducing Re to 0.9 and suppressing the outbreak. The output of the model without incorporation of delay failed to capture important features of transmission and Re.
CONCLUSIONS: Changing delay in confirmation has a significant impact on disease transmission and estimation of transmissibility. This leads to a clear recommendation that delay should be monitored and mitigated during outbreaks, and that delay dynamics should be incorporated into models to assess the effects of NPIs.
BACKGROUND: City University of Hong Kong and Health and Medical Research Fund.

BackgroundGiven its high economic and societal cost, policymakers might be reluctant to implement a large-scale lockdown in case of coronavirus disease (COVID-19) epidemic rebound. They may consider it as a last resort option if alternative control measures fail to reduce transmission.AimWe developed a modelling framework to ascertain the use of lockdown to ensure intensive care unit (ICU) capacity does not exceed a peak target defined by policymakers.MethodsWe used a deterministic compartmental model describing transmission of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) and the trajectories of COVID-19 patients in healthcare settings, accounting for age-specific mixing patterns and an increasing probability of severe outcomes with age. The framework is illustrated in the context of metropolitan France.ResultsThe daily incidence of ICU admissions and the number of occupied ICU beds are the most robust indicators to decide when a lockdown should be triggered. When the doubling time of hospitalisations estimated before lockdown is between 8 and 20 days, lockdown should be enforced when ICU admissions reach 3.0-3.7 and 7.8-9.5 per million for peak targets of 62 and 154 ICU beds per million (4,000 and 10,000 beds for metropolitan France), respectively. When implemented earlier, the lockdown duration required to get back below a desired level is also shorter.ConclusionsWe provide simple indicators and triggers to decide if and when a last-resort lockdown should be implemented to avoid saturation of ICU. These metrics can support the planning and real-time management of successive COVID-19 pandemic waves.