Decline, impact of reopening: where does the Covid-19 epidemic stand and what can we expect for the fall?
Reopening of restaurants and cultural venues, lifting of curfews, end of mandatory mask wearing outdoors... The Covid-19 epidemic is slowing down in France, and is even declining significantly faster than models predicted last April. Mircea Sofonea, senior lecturer in epidemiology and infectious disease evolution at the University of Montpellier, explains why, and gives an update on the assumptions for the coming months.
Mircea T. Sofonea, University of Montpellier
The Conversation France: The decline in the epidemic has been faster than models predicted in April. What happened?
Mircea Sofonea: It should be noted that mechanistic epidemiological models (i.e., those based explicitly on explicit transmission dynamics) do not produce predictions, but projections. Formally, they are akin to a logical implication: if condition A is met, then we can expect situation B to occur.
If the working assumptions are not actually satisfied, the scenarios produced become invalid and the simulations must be updated. This problem arises mainly when the signal of the effect of a new measure is still incomplete in the hospital data on which our models have been based since the start of the pandemic.
Our initial projections for the third lockdown were developed at the end of April. At that time, the impact of the third lockdown was limited compared to the first two, as on average 10 people infected 9 others, compared to 8 (or even 7) during the previous two lockdowns. Our most optimistic scenario was therefore based on the assumption that this trend would not reverse.
The pleasant surprise came a week later, in early May, with the lockdown measures becoming more effective. From then on, we were able to update the simulations.
For the past month, the epidemic has followed the most optimistic trend in the model, without requiring any adjustments.
TC: Why didn't you imagine an "optimistic" scenario, with such a sharp drop in the reproduction number?
MS: The scientific approach is based on an important principle, the principle of parsimony or "Ockham's razor" (named after the Franciscan philosopher William of Ockham who formulated it): "the simplest sufficient hypotheses should be preferred."
In the absence of solid evidence allowing for quantitative predictions of reproduction number dynamics that are not already included in the model (typically, the effect of vaccination and immunization through infection), the minimal and methodologically neutral assumption is to extrapolate the dynamics of the epidemic based on the latest data. In April, this extrapolation was itself supported by the analogy with the first two lockdowns. On both previous occasions, the estimated reproduction number had reached its minimum level about ten days after the measures were introduced.
However, the kinetics of the third lockdown were different: during the first 10 days, the reproduction number stagnated between 0.9 and 1, then fell sharply to below 0.8, before rising slightly in early May.
(Editor's note: The reproduction number is an estimate, over the last 14 days, of the average number of individuals infected by one infected person. We refer to the basic reproduction number (or R0) at the start of an epidemic, in the absence of transmission control measures and when the population is entirely susceptible to the virus. During the epidemic, this number changes: it is referred to as the effective or temporal reproduction number (Rt). If it is less than 1, the epidemic is declining; if it is greater than 1, it is progressing.)
While it now seems clear in retrospect that the third lockdown led to a slower decline in the reproduction number than the first two lockdowns, it was not possible to anticipate this at the end of April, just as there was no evidence to suggest a reversal of the trend during the lockdown, a scenario that we did not consider for the same reasons.
TC: Is the reliability of the models being called into question?
MS: No, but it is important to understand that several methodological and biological factors complicate the estimation of reproduction numbers shortly after the introduction or lifting of a restriction measure.
Firstly, there is a difference between reality and what can be understood through measurements. When restrictions are implemented (lockdown, curfew, etc.), the reproduction number drops overnight in everyday life. However, this discontinuity is not reflected in the figures collected, simply because various biological parameters vary from one person to another (incubation time, onset of contagiousness, appearance of symptoms, etc.).
The discontinuity caused by restrictions could be visible if all these events occurred uniformly across the board, assuming that testing was carried out at exactly the same time, with results obtained within identical timeframes. But this is not the case. In reality, there is a smoothing effect, and the effects are only seen after two weeks, indirectly, in hospitalizations and daily admissions to critical care units (data that is more reliable than testing, especially when public holidays are involved).
Secondly, the methods used to calculate the reproduction number also use smoothing techniques, in particular to overcome the "weekend effect": the 7-day moving average allows the decline in case detection activity at weekends (when laboratories are closed) to be spread over the whole week, thereby eliminating the impact of irrelevant fluctuations. The disadvantage is that this approach buffers variations that reflect a recent change in trend, for example when restrictive measures are put in place.
It is therefore necessary to continue research efforts in modeling. In particular, transmission patterns will need to be refined and, above all, inference from weak signals will need to be improved. Similarly, the acquisition and cross-referencing of complementary datasets is a real challenge. In the United Kingdom, for example, epidemiological surveillance in schools, contact tracing, random population testing, and sequencing provide valuable sources for improving the reliability of models.
Despite everything, parsimonious models still have a role to play, even a year and a half after the start of the pandemic. Although they may not be the most accurate in the short term, they make it easy to explore all the possibilities in the medium term, a timeframe that is of particular interest to decision-makers.
In this respect, they are well suited to informing forward-looking strategies, particularly in the context of an epidemic outbreak where any delay can have an exponential impact on health. This may even involve reassessing the schedule every two weeks as estimates become more reliable.
TC: Looking back, do we have any idea why this decline was faster than during previous lockdowns?
MS: Today, we still lack the perspective (and time) to explain it causally. However, as we move further away from the events, we will be able to use other statistical methods dedicated to studying the distant past to examine the course of the epidemic more precisely and assess the contribution of various factors. This is work that, for the moment, cannot be prioritized given the urgency of other demands.
One hypothesis is that school closures made a major contribution to the lockdown effect (in the first case, closures were widespread, and in the second, school vacations had already begun). However, as children are less symptomatic, it takes longer to see the effects of such a measure.
The stagnation in the reproduction number observed during the first week of the third lockdown could be explained by the fact that transmission chains remained intact among adults (particularly in the workplace). However, when the school holidays began, the chains of transmission that had started in schools no longer existed at that point, and the reproduction number fell sharply, as many adults were also on vacation.
TC: This result once again raises the question of the role of schools...
MS: Yes, especially since the "vaccine" effect has been favorably integrated.
However, it cannot really be said that there was a greater "school effect" during this lockdown than during the first one, as the conditions were different: the presence of the Alpha variant (formerly known as the "British" variant, which is more contagious than the original strain, particularly among younger people), and the vaccination of those most at risk of complications.
In addition, other factors may have had an impact: parks were open, the weather was better than in the fall (but milder weather alone cannot explain the decline). However, in prospective modeling, it remains difficult to incorporate a parameter such as weather into projections, given that forecasts beyond a week are uncertain.
TC: Where are we today? Are we seeing any effect from the reopenings?
MS: When schools reopened on April 26 (and secondary schools at half capacity the following week), then outdoor cafes on May 19, we saw a slight increase in the reproduction number, which rose to between 0.8 and 0.9, before stabilizing at around 0.8 at present.
We can therefore clearly see the effect of schools reopening and, to a lesser extent, outdoor cafés reopening, but nothing that is likely to cause the epidemic to resurge, which continues to decline, albeit at a slightly slower rate than in early May. This suggests that outdoor cafés could have been reopened earlier, particularly in areas where the incidence was already lower. In general, a territorial approach to lifting restrictions (and not just to implementing them) makes it possible to generate data that can inform decision-making for areas that have been less spared.
Furthermore, vaccination plays a key role, as shown by the various scenarios: even if the reproduction number rose again to slightly above 1, the progress of vaccination could cause it to stabilize or decline rapidly.
For the time being, it is important to remain vigilant with regard to indoor gatherings with limited air circulation. The reopening of theaters and indoor dining areas is still too recent to assess their effect on the epidemic.
TC: Wearing a mask outdoors is no longer mandatory. What do you think about that?
It should be noted that the reopening of outdoor dining areas took place in a context of recognition of the contribution of airborne transmission (via aerosols) to the spread of SARS-CoV-2 and the dynamics of the epidemic.
However, recent literature suggests that the risk of outdoor transmission is very low. Nevertheless, it may persist in situations of prolonged close contact without airflow, if a person is exposed for several minutes to the aerosol cloud produced by an infectious person without the latter having had time to dissipate.
TC: What can we expect in the fall? What variants can we expect to see circulating, and in which segments of the population?
MS: Again, the goal of mechanistic models is not to predict how many hospitalizations there will be within a given number of days, but rather to determine, for example, how much relaxation we can afford without fear of hospital saturation, or what the death toll of a fourth and final wave might be.
Currently, the alpha variant is predominant (although the beta variant, which originated in South Africa, now appears to be spreading more rapidly than the alpha variant in the Île-de-France and Hauts-de-France regions, perhaps because it is more likely to evade natural immunity; see our latest work published in the European Center for Disease Prevention and Control journal).
Given that the alpha variant spreads more easily among young people and that older people have been vaccinated as a priority, it is natural to expect that younger people will become the reservoir for the virus in the coming months.
The real question is what our ambition is for the start of the new school year: do we want, for example, to completely do away with masks, not only outdoors but also indoors? To allow all cultural events to resume, regardless of audience size? In that case, continuing the vaccination campaign this summer will be crucial.
TC: What are the estimates in terms of potential deaths?
MS: Currently, even if 90% of the adult population is vaccinated with two doses by the start of the school year, we would still be facing the threat of 15,000 hospital deaths nationwide—the equivalent of a major seasonal flu epidemic. This is an order of magnitude, under current conditions (excluding immune escape from a variant, which does not seem to be the case at the moment).
Are we prepared to accept 15,000 more deaths? Given that we have already collectively accepted more than 110,000 deaths from this disease, there is little reason to imagine otherwise...
From a purely pragmatic point of view, the risk is that if these 15,000 deaths occur in quick succession, they could once again put pressure on certain local hospitals. For this reason, it is crucial to start preparing for the new school year now. In particular, hospitals will need to determine how best to organize themselves to manage the remaining Covid-19 activity: should they maintain a dedicated unit or distribute hospitalizations among different departments?
TC: To prevent this type of problem, is it necessary for vaccination to be consistent across the entire country?
MS: Not necessarily. In a model developed by Olivier Thomine based on OpenStreetMap data, which has not yet been peer-reviewed, the spatial heterogeneity of the epidemic suggests that it is important to achieve high levels of vaccination coverage, especially in large cities.
This calls for differentiated regional measures, including for vaccination coverage: having lower vaccination coverage in the Gers than in Paris, Lyon, or Seine-Saint-Denis is not necessarily a problem. However, if the virus were to reach areas with very low vaccination coverage, there would be a risk of local epidemics. This is what has been observed, for example, in the Netherlands with measles.
In addition to the issue of the Delta variant (originally from India), whose spread in France is being closely monitored, the end of the epidemic in France will be determined by vaccination coverage. Adherence may wane as the hospital situation improves and the perceived risk of infection decreases.
Epidemiological models do not yet incorporate human behavior, although this is an active area of research. In the meantime, models must produce scenarios based on a range of realistic and documented assumptions, so that they cover a range of possibilities capable of anticipating the risk of an epidemic.
Mircea T. Sofonea, Associate Professor of Epidemiology and Evolution of Infectious Diseases, MIVEGEC Laboratory, University of Montpellier
This article is republished from The Conversation under a Creative Commons license. Readthe original article.