This study evaluates the impact of the local perimeter confinements implemented during the second wave of COVID-19 in Madrid. This Autonomous Community adopted a special model of public health measures based on perimeter closures by BHZ depending of the epidemiological situation across time. Our mathematical models showed no statistical differences in cumulative incidence for BHZs with and without perimeter closures.
In Spain, the universal use of masks (indoors and outdoors) is mandatory throughout the whole territory since May 19th 2020 [4]. On October 25th, the second state of emergency started and additional public health measures were implemented, with a range of normatives that varied across the Autonomous Communities. These included curfews (set at 23:00 in Madrid for the time under study), limited seating capacities at restaurant premises and gathering limitations (maximum of 6 people in outdoors restaurant facilities and in-house meetings, and 4 people in indoors restaurant facilities in Madrid), and other general restrictions [15]. Prior to the state of emergency, the perimeter confinements by BHZ were also activated, in contrast to the municipality-level closures adopted at other Autonomous Communities (Galicia [16], Cantabria [17], among others).
We assessed whether the perimeter confinements of BHZs had a significant influence on the evolution of the epidemic curve by modeling them as explanatory covariables in several mathematical models. We found that the estimations provided by the models that included the perimeter confinements as an explanatory variable and those that did not were statistically very similar, indicating that the perimeter confinements did not have a significant impact on the 14 days accumulated CI.
Several factors limit the effectiveness of the BHZ closures system. For example, due to the high permeability between neighboring BHZs and associated difficulty in the evaluation of the citizens’ compliance to the measure, it has not been possible to determine if the policy was implemented effectively. In addition, a low risk perception towards the COVID-19 pandemic has been identified in the Spanish population during the time period under study [18], which could have been resulted in a decreased adherence to the policy [19, 20].
While local mobility restrictions are effective in a theoretical modelling framework [21, 22], evidence suggests that an informed and coordinated approach is required for the effective implementation of such a response measure [23]. Being a rare policy, few studies that focus on the effect of such selective confinements of such small units as BHZs are available. Fotán-Vela et al. [24] also analyze the case of the BHZ closures in Madrid. Their analysis shows that the decrease in the epidemic curve in Madrid started before the impact of the perimeter closures could be reflected. Other than Madrid, the only other context were a similar policy has been adopted is Chile, to our best knowledge. Cuadrado et al. [25] and Li et al. [26] study the local lockdowns active during the first wave of the COVID-19 pandemic in this country, obtaining, respectively, a reduction in effective reproductive number (with a wide confidence interval, nevertheless), and a highly variable effectiveness of the policy (depending on duration of intervention and spillover effect from neighboring areas).
Limitations of the analysis
The average BHZ is an epidemiologically small unit, both in terms of population (22.750 inhabitants) and area (28 km2). Because of this, the usual joint point methods for trend analysis will presumably not reveal meaningful conclusions at local BHZ level, lacking statistical significance. This is the case as well for trend analysis on models that incorporate information from all the BHZs, due to the asynchronicity in the implementation of the perimeter confinements among each of the BHZs. For the very same reason, precise estimations are not expected to be obtained from models fitted to this data. We thus chose to employ the present approach, sensible to general tendencies in models that have been adjusted differently, and focused on statistical assessments rather than in accurate predictions. GAM models are expected to capture a greater influence of the additional explanatory variables included than trend analysis models [12], and we incorporated higher significance by a leave-one-out cross-validation process over the 286 BHZs that involves the choice of the best of 15 models in each step.
An additional confounding effect is due to the fact that perimeter confinements (and COVID-19 related restrictions in general) have been introduced on an a posteriori basis. That is, restrictions are activated as a response to the increment of the 14 days CI, and therefore there is a natural correlation between BHZs with high CI and perimeter confined BHZs. Again, an approach that does not focus on assessing the explicit, precise impact of these restrictions and rather on its statistical effect is thus preferred, as misleading associations may be inferred otherwise.
Finally, the epidemiological threshold triggering the closures changed during the study period. On September 21st, weekly BHZ perimeter confinements were activated at those BHZs where the 14 days cumulative incidence surpassed the 1.000 cases per 100.000 inhabitants. This threshold was decreased to 750 cases on October 12th, 500 cases on October 26th, and 400 cases on November 23rd [11]. We have not included the possible effect of this variation in our analysis, as we focused on the effect of the actual perimeter confinements and not on their dependence to the current epidemiological status.
