The COVID-19 pandemic prompted worldwide lockdowns, but the need to restart socioeconomic activities necessitates careful strategies for relaxing containment measures. The challenge lies in developing effective models to guide decision-making by simulating various scenarios and their impact on epidemic trajectories. SARS-CoV-2 presents unique modeling challenges due to inapparent infections and the geographically uneven spread of the virus. In Italy, the outbreak's latitudinal progression revealed considerable delays in the onset of local epidemics. The model accounts for various epidemiological compartments (susceptible, latent, infectious, asymptomatic, symptomatic, recovered) and inter-community mobility.

The study builds upon a previously published spatially explicit model of COVID-19 spread in Italy. This model incorporated factors such as inter-community mobility, infection seeding timing, mobility restrictions, and social distancing. The authors note limitations in existing models, particularly the lack of spatial detail and the difficulties of accounting for inapparent infections in well-mixed models designed for single cities or entire countries. Empirical evidence shows mildly symptomatic individuals may be as contagious as symptomatic ones, and presymptomatic transmission is significant. The authors acknowledge that available epidemiological data are only approximations, suffering from underreporting of confirmed infections and deaths due to variations in testing capacity.

The researchers used a spatially explicit model, solving a system of coupled ordinary differential equations for each of Italy's 107 provinces. The model subdivides the population into epidemiological compartments (susceptible, latent, infectious, asymptomatic/mildly symptomatic, symptomatic, recovered). Inter-community mobility is incorporated, with infection force depending on both local and connected communities' epidemiological variables. Model parameters were estimated using a Bayesian framework, calibrated against daily hospitalization data from February 24 to May 1, accounting for progressively restrictive measures implemented. The model considers heterogeneity in transmission rates post-March 22. Hospitalization data, rather than infection counts, were used to minimize underreporting bias. The model parameters were estimated using the DREAMzs implementation of the Markov chain Monte Carlo algorithm. Scenarios were generated to simulate the effects of lifting lockdown restrictions, focusing on the impact on the epidemic curve and the isolation effort required to prevent resurgence. The model uses three different assumptions on the heavy symptomatic fraction to investigate the role of inapparent infections (σ = 10%, 25%, and 50%). The model also considers the effects of isolating exposed and highly infectious individuals to control transmission, exploring the daily percentage and number of individuals that would need to be isolated to maintain the epidemic curve on the decreasing baseline trajectory.

The model accurately reproduced cumulative hospitalizations in Italian provinces up to May 1. A large reduction in effective transmission rates (between 0.3 and 0.4) was estimated during the lockdown period. The study projects scenarios based on different levels of transmission increases after lockdown relaxation. A 20% increase leads to a slower decline in hospitalizations, whereas a 40% increase results in a significant resurgence in most regions. The results showed robustness across different assumptions of the proportion of infections with heavy symptoms. The fraction of susceptible individuals varies substantially across regions, with northern regions showing greater reduction. Estimated infection fatality rates varied based on the proportion of symptomatic infections. An isolation effort of ~5.5% daily isolation of exposed and highly infectious individuals (E and P compartments) was estimated to prevent a resurgence, and this target depended on the assumed fraction of infections developing heavy symptoms. Achieving the isolation target requires targeting not only primary infections but also secondary infections. Delaying lockdown release significantly reduces the isolation effort. An ex-post assessment, using data up to June 17, 2020, reveals that the actual post-lockdown transmission was largely consistent with the baseline scenario (no resurgence), except for Lombardy which showed an increase in transmission. The analysis also estimated changes in transmission rates between the last phase of lockdown and the period following its easing across different regions. Most regions showed a decrease in transmission, with some exceptions like Lombardy and Molise. At the national level, new daily hospitalizations remained slightly lower than the baseline prediction.

The findings highlight the importance of considering inapparent infections and spatial heterogeneity in modeling COVID-19. The model's ability to closely match hospitalization data enhances its reliability for scenario planning. The estimated reduction in transmission during lockdown (~65%) is largely attributed to implemented measures, rather than acquired immunity. The study emphasizes the need for ongoing monitoring of the epidemic curve and timely adjustments to control strategies. The analysis of isolation efforts emphasizes the importance of contact tracing, testing, and isolation of infectious individuals. The ex-post assessment suggests that the partial relaxation of restrictions and other factors such as increased isolation efforts, the potential impact of warmer weather, and control of infections in long-term care facilities, contributed to preventing a major resurgence. The limitations of neglecting demographic stochasticity when the case count is low should also be noted.

This study provides valuable insights into managing COVID-19 outbreaks by combining spatial modeling with an ex-post analysis. The findings highlight the importance of considering inapparent infections and spatial heterogeneity in forecasting epidemic trajectories and designing control strategies. The results suggest that a combination of continued monitoring, flexible control measures, enhanced contact tracing, and targeted isolation efforts are crucial for preventing future surges. Future work could incorporate more granular data and refine the model by including seasonal effects and stochasticity.

The study acknowledges limitations in the data availability and the inherent uncertainties in modeling infectious disease dynamics. Underreporting of cases and variations in testing capacity can impact the accuracy of estimations. The model simplifies some aspects of human behavior and disease transmission. The lack of granular data on contact tracing efforts limits a detailed analysis of its impact. The model does not explicitly consider seasonal variations in transmission, although this is a factor that could be explored in future studies.