Categorizing the Status of COVID-19 Outbreaks Around the World

Category A – Susceptibility-Driven Dynamics with Limited Mitigation

The evolution of reported COVID-19 cases shows characteristics of susceptibility-driven dynamics with only limited mitigation from control measures in 5 countries with about 168 million people (2.2% of the global population). For these countries in Category A, reported cases typically have narrow peaks with doubling times of less than 10 days, corresponding to values of R_{0_e} values of 1.4 or above (from the formula for the doubling time of t_g * ln(2) / ln(R_{0_e}), where the mean generation time t_g is assumed to be 5 days [22]). Countries in this category include Central African Republic (for which data is presented in panel Aa of figure 5) and Madagascar (panel Ab).

Categories B and C – Susceptibility-Driven Dynamics Mitigated by Measures

In a further 32 countries with about 2.24 billion people (28.8% of the global population), the COVID-19 epidemic also shows characteristics of susceptibility-driven dynamics, but the case curve was affected substantially due to social distancing and disease control measures.

Countries in Category B experienced substantial flattening of their case curves due to disease control measures, with doubling times of 10 days or more and R_{0_e} values which were typically below 1.4 but still above 1. There are 23 countries (with 2.05 billion people or 26.3% of the global population) in Category B, as of 31 January 2021, located mainly in Africa and Asia, and including Angola, Iraq (panel Ba), India (panel Bb), Morocco, and Venezuela, among others.

Category C includes countries which experienced susceptibility-driven peaks having previously controlled the virus through social distancing and other measures – such as Bosnia & Herzegovina (panel C). In all, this category included, as of 31 January 2021, 9 countries with 190 million people (2.5% of global population).

Categories D and E – Susceptibility-plus-Measures-Driven Dynamics

In 31 countries, with 1.45 billion people (18.6% of global population), reported cases have increased slowly and have not yet either reached a peak or fully declined from their first peak, and estimated infection levels are moderate or high, indicating that control measures have had significant effects but not suppressed the virus.

Of these, the 19 countries in Category D have passed an apparent peak, mostly fairly recently, including, for example, Ukraine (panel D1) and Colombia (panel D2). A further 12 countries in Category E are before or just at a peak, including Indonesia (panel E1) and Lebanon (panel E2). For these countries, control measures significantly reduced R_{0_e}, potentially to values falling in the range between about 1.3 and 1.05 (corresponding to case doubling times of between 13 days and 60 days if the mean generation time t_g = 5 days). As a result, the threshold levels beyond which cases have declined, or will decline in the future, was significantly reduced due to the control measures, and the numbers of people infected should be much lower than in an unmitigated epidemic.

Relaxation of control measures – or emergence of higher-transmissibility variants – in such circumstances, especially if it happens before cases have declined fully from the peak, can lead to substantial resurgences in cases. This pattern appears to have occurred already in several countries, including Brazil, Colombia and Paraguay (all in Category D) and Lebanon, Mexico, Mozambique, Togo and the United Arab Emirates (all in Category E).

Countries in Categories D and E will eventually transition to Category B or C once cases pass their peak and decline fully (unless changes are made which push R_{0_e} below 1 and “crush the curve”). Many of the countries in Categories B and C were previously in Categories D and E. India and Iraq, for example, would have been classified as Category E up to September and then as Category D until December, when they moved to Category B. Panama would have been classified as Category E until it recently moved to Category C. Ukraine, in Category D on 31 January 2021, will likely soon move into Category B.

Categories F and G – Measures-Driven Dynamics

Disease control measures, including combinations of social distancing, hygiene, testing & tracing and quarantines in different combinations, have constrained the SARS-CoV-2 virus in 74 countries, with 2.49 billion people (31.9% of global population).

Category F includes 54 countries in which the disease has been suppressed through control measures to date and is currently under control. In 5 countries, outbreaks were suppressed in the first half of 2020 and the disease has since been kept largely out, including in China (panel F1) and New Zealand. In 11 countries, the disease was kept largely completely out at first, and outbreaks which happened in the second half of 2020 were suppressed – including, for example, Bhutan Georgia, Trinidad & Tobago and Uruguay. In 38 countries, control measures “crushed the curve” during two or more waves, including Ireland (panel F2) and many other European countries. This category covers a wide range of experiences, including countries with infection levels to date that are likely well below 10% as well as countries for which infection levels might be as much as 20% or more (for which the relatively high infection level can reduce the strength of control measures required to suppress further transmission of the virus).

Category G included – as of 31 January 2021 – 20 countries in which the disease was previously excluded – such as Mongolia (panel Ga) – or controlled – such as Portugal (panel Gb) – but in which reported cases are currently increasing. Many countries in Category G may bring their current outbreaks under control (and thus move to Category F, as has happened to several countries that were in Category G in December and to Portugal in the period since 31 January 2021), some may not (and thus move to Category C); the future evolution of the disease depends on whether R_{0_e} is or will be pushed below 1 by current or future control measures (which will require stronger measures in places with new variants of the virus that have higher “natural” basic reproduction numbers).

Category H – Index-Case-Control Dynamics

COVID-19 appears to have been largely kept out, through strict quarantines and testing & tracing (together with lockdowns in a few cases), of all 39 countries that report fewer than 500 cases to date. These are almost all island states and territories (including Fiji shown in panel Ha of figure 5), except for Brunei, Cambodia (panel Hb of figure 5), Laos and Timor-Leste. The total population of countries and territories in Category H is 29 million or about 0.4% of the global population.

Some countries which largely kept SARS-CoV-2 virus out in the early months of the pandemic have subsequently experienced outbreaks, and some of these are shown in figure 6. In all, 82 countries and territories had fewer than 500 cases up to 30 June, of which 70 had fewer than 250 reported cases. Most of the countries which subsequently experienced substantial outbreaks managed to suppress the disease through control measures; 17 of them are in Category F (including Trinidad & Tobago shown in panel F of figure 6) and 9 in Category G (including Mongolia, previously shown in panel Ga of figure 5) as of 31 January 2020. However, 8 countries experienced outbreaks which have not been suppressed by control measures (and the situation in the other 9 countries is uncertain): as of 31 January 2021, 3 countries which kept the disease largely out up to 30 June 2020 were in Category B (Angola, shown in panel B of figure 6, Belize and Myanmar), 4 were in Categories D and E (Botswana, shown in panel E of figure 6, Guyana, Sint Maarten and Syria), and 1 was in Category J (Suriname, shown in panel J of figure 6).

Category I – Inconsistencies between Evolution of Reported Cases and Estimates for Total Infection Levels

In 5 countries, the evolution of reported cases relative to timing of changes in the stringency index suggests one of Categories A-H but the total numbers of reported cases and/or deaths are inconsistent with the infection levels expected for that category. In all these countries, the evolution of reported cases suggests that the disease dynamics have been driven by susceptibility levels, but the total reported cases and/or deaths are lower than would be expected if most people had been infected, after accounting for plausible case detection and death detection rates.

For three of the countries – Oman, Qatar and Saudi Arabia – it is likely that large outbreaks have occurred within segments of the population – most likely segments of the large migrant worker populations in these countries – while lower infection levels prevailed in the rest of the population. Two other Gulf countries, Bahrain and Kuwait, are included in Category B2, but have reported cases that are close to the bottom of the plausible ranges if true infection levels are high; cases have started to increase again, likely due to the fact that large sections of the population remain uninfected to date.

Singapore, which is included in Category F, is another country in which the data suggests disease spread within a population segment – also migrant workers living in dormitories – but not much in the rest of the population. For Singapore, the case fatality ratio (CFR) is only 0.08 times the value of the infection fatality ratio (IFR) expected given the age-structure of Singapore’s population. The CFR may be reduced if many cases are undetected, but the very low value for Singapore suggests that deaths are also much lower than expected – a sign that the disease has likely spread mainly in a population segment with younger people than the general population.

Two other countries fall into Category I. For Yemen, it seems likely that the case detection rate is even lower than the rate of 1 in 2,000 assumed to be the lowest plausible level for LICs. For Chile, it is possible that fewer cases are being detected than expected for HICs, or that the virus is spreading mainly in a segment of the population, or that the evolution of reported cases is not due purely to susceptibility-driven dynamics.

Category J – Resurgences after Susceptibility-Driven Peaks

In 17 countries, there were peaks in reported cases in mid-2020 characteristic of susceptibility-driven dynamics, but large resurgences of cases occurred after the end of the first peak. For some of these countries, such as Afghanistan and Pakistan, the first peaks appeared to be due to largely unmitigated outbreaks (similar to Category A), but resurgences have reached 25% or more of the maximum number of reported daily new cases during the first peak. For other countries, such as Egypt and Kenya, the first peaks were flattened to some degree (similar to Category B), but the resurgences have led to numbers of reported cases which exceeded 50% of the maximum case numbers during the first peak.

There are some clusters among the countries in Category J, both in geography and in the timing of the resurgences in cases. In South America, Bolivia, Peru and Suriname show steep surges in December-January. In South and Central Asia, Pakistan, Afghanistan, Kyrgyzstan and Kazakhstan show relatively slow-growing resurgences starting in September or October. In West Africa, Côte d’Ivoire, Ghana and Nigeria show steep surges starting in early or late December. In Southern Africa, Malawi and South Africa have resurgences starting in late December and late November, respectively. Egypt and Sudan experienced resurgences starting in October. Kenya, Democratic Republic of Congo and Dominican Republic experienced resurgences whose timing or shape don’t appear to match the evolution of reported cases in any of their neighbours.

For all these countries, the scale of the resurgences appears to be greater than would be expected given the share of population expected to have been infected during the first peak. Consequently, the resurgences are an indication of more complex dynamics – potentially including effects from new virus strains with higher transmissibility or ability to evade immune responses, from population segments having low infection levels after the first wave, or from waning of immunity acquired from infection in the first wave. These possibilities will be considered – to understand better the dynamics in Category J countries – at the end of section 5.

Category K – Uncertain Dynamics Due to Ambiguities in Evolution of Reported Cases

There are 17 countries for which we classify the disease dynamics as uncertain because we cannot definitively allocate them to one of the earlier categories, since the evolution of reported cases could arguably fit with two or more of those categories and the total numbers of reported cases and deaths are consistent with each of the possible categories (considering the range of plausible values of the case detection rate and death detection rate for the country).

For 6 countries in West Africa (Benin, Gambia, Mali, Mauritania, Senegal and Sierra Leone) the first waves of cases continued growing for a month or more after the introduction of control measures before reaching a peak, and cases declined at the same time as the stringency index was declining or constant. On balance of probabilities, it seems likely that the peaks in cases were due to susceptibility-driven dynamics – with curve flattening due to measures, especially in Mali, Senegal and Sierra Leone. However, since the peaks in cases were reached within about 1-2 months after measures were introduced or the first cases appeared, there is a possibility that the peaks might have been due to control measures. In all 6 countries, cases have grown rapidly in the past 2-3 months, similar to the West African countries in Category J.

For 5 countries in Southern Africa (Eswatini, Lesotho, Namibia, Zambia and Zimbabwe), the first significant waves of cases reached peaks in July or August and cases declined to low levels by October. In most of the countries, the stringency index increased in July, but only by small amounts, and cases declined even as the stringency index was reduced in September. In all of these countries, there were surges in cases in December, and both the surges and the prior peaks around August are similar to those observed in South Africa and Malawi, which are in Category J.

For the countries in West Africa and Southern Africa in Category K, it is uncertain whether the second waves are due to spreading of the virus among uninfected people in the general population (similar to countries in Categories C and G) or due to resurgences generated by new variants or population segments with previously low infection levels (similar to countries in these regions that are in Category J). These possibilities will be considered at the end of section 5.

Two countries in the Eastern Mediterranean – Palestine and Turkey – appear to fall close to the boundary between Category C and Category F, and they have similar curve shapes to other countries in the Balkans and the Levant, some of which are included in Category C and some in Category F.

The remaining countries in Category K are Djibouti, Jamaica, Rwanda and South Sudan.

Category L – Uncertain Due to Insufficient Data

Three countries – North Korea, Tanzania and Turkmenistan – do not track COVID-19 cases and deaths and do not provide reports to the World Health Organization.

Three countries – Belarus, Burundi and Nicaragua – had values of the stringency index which never exceeded 30 points, making it difficult to use the index to determine the strength of control measures.

In nine other countries and territories – the largest in population of which are Guinea-Bissau, Kosovo and Equatorial Guinea – there is no stringency index metric for the country, and, without this data, we could not categorize the disease dynamics reliably. (The Oxford COVID-19 Government Response Tracker does not have stringency index metrics for 55 countries, but 30 of these fall into Category H because they have fewer than 500 reported cases, and 13 more can be categorized based on the shape of the reported case curves and estimates of actual infection levels from total reported cases and deaths.)

Effects of variability of reproduction number and case detection on reported cases

Several factors can affect the evolution of reported cases, with the potential in theory to create patterns that might be misinterpreted as indicating one category of disease dynamics when in fact the disease is behaving very differently. First, the effective reproduction number, R_{0_e}, can vary due to alteration of control measures, changes in compliance levels, or seasonal patterns. Second, there may be geographical differences in the onset date of the epidemic and wide variations in local values of R_{0_e}. Third, case detection rates – the ratio of reported cases to real cases – can change as the availability of and regulations for testing change.

These complications do in fact affect the evolution of reported cases to some degree in all countries. They explain why no country has a smooth and simple reported cases curve like the simulation results shown in figure 1.

Figure 7 shows four countries where the evolution of reported cases likely implies deviations from simple assumptions of consistency in R_{0_e} and case detection rates across geography and time. Iran’s reported cases (shown in panel A) have increased, decreased and stayed roughly constant at various times, likely reflecting periods when the average R_{0_e} was above 1, below 1 and around 1.

• Download figure
• Open in new tab

Figure 7: Reported case data and diagnostic findings for selected countries, which show features indicative of variations in reproduction number and/or case detection rates.

• Chile’s cases (panel B) show a single prominent peak in June, and the stringency index remained roughly constant before and after the peak – suggestive of susceptibility-driven dynamics. However, reported cases have experienced a ‘long tail’ and a resurgence since December. The Santiago capital region, which has 40% of the population, experienced a peak in cases around June. Cases peaked in northern regions mostly in July and August but experienced resurgences in December (perhaps related to the surge in cases across the border in Bolivia and Peru). Meanwhile, in most of southern Chile, which includes some of the least densely populated regions, cases continued to increase gradually throughout the year – suggesting that the disease arrived later in more remote regions and spread more slowly in less densely populated regions.

• Ethiopia’s cases (panel C) show a simple curve with a peak in late August, except for several undulations. These features are suggestive of changes in the case detection rate, and indeed some of the undulations correspond to notable dips in the number of tests reported in Ethiopia.

• Other countries, mainly large countries such as the United States of America (panel D), likely show the effects of variations, across geography and time, in both R_{0_e} and case detection rates.

In countries where R_{0_e} is usually close to 1, or with geographical regions or population segments that are strongly isolated from each other, it is likely to be important for policymakers to understand disease dynamics in detail to determine how R_{0_e} is changing over time and for different regions or population segments (in response to policy choices). However, our findings suggest that these complexities do not obscure the basic drivers of disease dynamics – in other words, which of our disease dynamics categories applies – for most countries.

The most significant assumption underlying the diagnostic tool is that the evolution of reported cases reflects roughly the evolution of actual cases, despite reported cases accounting for only a small proportion of all actual cases. Three considerations support this assumption. First, in most countries, the number of reported new cases has varied to a greater extent than the number of tests and the number of tests per confirmed case has varied substantially – even though one might expect the demand for tests to vary somewhat based on actual cases. Second, the numbers of reported deaths and reported cases follow a very similar pattern, with changes in reported deaths lagging changes in reported cases, even for countries with very low reported numbers of both – which suggests that changes in reported cases and deaths are genuinely reflecting the underlying disease dynamic. Third, for countries found to have high or medium infection levels (Categories A-E), the reported cases and deaths are a small fraction of the presumed actual numbers – but there is good evidence that case and death detection rates for such countries, mostly low- and middle-income countries, are in fact quite low [3-13], and, indeed, our approach specifically checks to make sure that estimated infection levels (from reported cases and reported deaths) are not inconsistent with the disease dynamics suggested by the evolution of reported cases.

This is not to say that the case detection rate, the ratio of reported to actual cases, has not changed. It has. The number of reported deaths as a proportion of the number of reported cases – the case fatality ratio or CFR – declined in about three-quarters of countries in the second half of 2020 compared to the first half of the year, and by factors of 5-10 or more in many HICs. The reductions in CFR values are likely due to a combination of lower mortality rates, due to improved treatment of COVID-19 patients, and to an increase in the proportion of cases detected. These increases in case detection rate, however, were likely gradual and thus would not change any of the categorizations, which are principally based on changes in cases within a month or two of changes in the stringency index.

Another important assumption is that the evolution of reported cases is indicative of the evolution of actual cases in a country’s entire population. We have already found some countries where that assumption was not reasonable. For five Arabian Peninsula countries and Singapore, we found that reported cases might principally reflect the disease outbreak in one segment of the population. For some of the countries in Category J, it seems likely that resurgences in cases were due to disease spread, after relaxation of social distancing practices, in wealthier population segments which experienced lower infection levels than the rest of the population during the initial spread of the disease. For some countries, there are differences in disease dynamics in different regions (e.g., Chile and USA), and a few countries have unusually low case fatality rates which suggest that more younger people are being infected than older people (e.g., Singapore, Thailand, Martinique, Iceland and Réunion). However, aside from the countries just mentioned, there is little evidence to suggest that the disease evolution is markedly different for specific population segments (by geography or age or any other characteristic) to such a degree that the reported cases might not reflect fairly well the disease dynamics of the national population. Even in low-income countries, in some of which testing is disproportionately conducted in the capital cities or other large urban areas, it is likely that the reported cases do reflect the national picture: serology testing in India, Kenya and Nigeria [18,23,24], for example, indicate that the disease spread quickly to rural areas after outbreaks started in urban centres.

Effects of waning immunity

Evidence suggests that antibodies last for at least 3-4 months [25] but start to wane in some people for longer periods of time [26,27], with the largest study, by the REACT Study Team, finding a decline in detectable IgG antibodies of 26.3% between 12 and 24 weeks after the peak in infections in England in April [27]. However, antibodies are only one component of the immune response to SARS-CoV-2, which also includes T cells (CD8⁺ and CD4⁺) and B cells. There is evidence that B cells [28] and T cells [29,30] can create strong immune responses to SARS-CoV-2 (with samples from about a third of people in studies in Germany [29] and Sweden [30] showing cross-reactive T cell responses to SARS-CoV-2 from prior exposure to other coronaviruses but not to SARS-CoV-2 itself). A recent study by J.M. Dan et al. [31] looked at five components of immune response (RBD IgG antibodies, RBD IgA antibodies, RBD memory B cells, total SARS-CoV-2-specific CD8⁺ T cells, and total SARS-CoV-2-specific CD4⁺ T cells) and found that “[b]y 5+ months after COVID-19, the proportion of individuals positive for all five of these immune memory compartments had dropped to 40%; nevertheless, 96% of individuals were still positive for at least three out of five SARS-CoV-2 immune memory responses”. They found wide variations in the magnitude of adaptive immune response across individuals, suggesting that a fraction of people may become susceptible to reinfection quickly after infection with SARS-CoV-2 (or vaccination).

These findings are consistent with observations of COVID-19 reinfections. In the most rigorous study to date, by the SIREN Study Group [32], infection rates were tracked in a cohort of 6,614 UK public hospital staff with SARS-CoV-2 antibodies due to prior infection and compared to those in an antibody-negative cohort. The study found that “[a] prior history of SARS-CoV-2 infection was associated with an 83% lower risk of infection” – meaning that 17% of previously infected people were susceptible to reinfection – after a median observation time of 5 months following primary infection. The reduction in susceptibility to reinfection found by the SIREN Study is lower than the estimated decline in numbers of people with detectable IgG antibodies of 26.3% after 24 weeks. This finding is tentatively consistent with the idea that other components of the immune response to SARS-CoV-2, besides antibodies, can play a role in protecting people from reinfection.

The impact of immunity on disease dynamics, however, depends not on whether people become sick after reinfection with SARS-CoV-2, but on whether they contract the virus at all and whether, and to what extent, they transmit the virus to others. Only neutralizing antibodies provide sterilizing immunity, which stops a virus from replicating within the body and hence completely stops reinfection or risk of transmission to others. There is some evidence that T cell responses decrease viral shedding in other coronaviruses [33], and hence it is plausible that such responses could decrease infectiousness of individuals. If future studies can provide estimates of the percentages of infected people who become susceptible to reinfection over time, and estimates of their likelihood to transmit the disease to others (compared to first infection), then models can simulate the likely impact on the dynamics of the disease in a population.

If immunity is lost over time, and people who were previously infected (or vaccinated) become able to contract and transmit the virus again, this could lead to future outbreaks. However, such outbreaks are likely to be significantly more modest in scale than the initial outbreaks when SARS-CoV-2 first spread during 2020, for several reasons. First, as discussed, people may retain partial immunity – with reduced susceptibility to reinfection compared to their first infection, and potentially a reduced likelihood of transmitting the disease to others. Second, of course, vaccines are likely to be available, which can continually replenish the overall level of immunity in the population.

Effects of new virus strains with higher transmissibility or resistance to immunity

Many new strains of SARS-CoV-2 have emerged since the first variant of the virus started to circulate in human populations. Most of these have had similar characteristics to the original virus – causing similar disease symptoms and mortality, reproducing at similar rates, and generating and being controlled by similar immune system responses. However, late in 2020, at least three more concerning mutations have occurred. The B.1.1.7 variant, first identified in the United Kingdom and since found in more than 75 countries (as of 9 February 2021 [34]), has higher transmissibility, and is estimated to have increased the reproduction number in the UK by between 0.4 and 0.7 [35]. The B.1.351 variant, first identified in South Africa and since found in 32 countries (as of 9 February 2021 [34]), appears to have mutations which cause both higher transmissibility and decreased effectiveness of the immune system response in people who developed immunity to earlier variants (through infection or vaccination) [36, 37]. The P.1 variant, a descendant of the B.1.1.28 variant, first identified in Brazil and since found in 14 countries (as of 9 February 2021 [34]), has similar significant mutations to the B.1.351 variant, and thus appears likely to have higher transmissibility and resistance to existing immunity. It seems likely that more variants with immunity resistance and/or higher transmissibility will emerge in the coming months.

The emergence of new variants of SARS-CoV-2 with different characteristics would be expected to lead to deviations in the evolution of reported cases from the simple patterns expected in Categories A-H of figure 1. Consequently, keeping track of the trends in reported cases, compared to changes in the stringency of control measures, can help to identify such variants, especially in countries that do not regularly test for specific strains of SARS-CoV-2 in COVID-19 cases. The signatures of immunity-resistant and higher-transmissibility variants in reported case data should be different depending on the current status of the COVID-19 epidemic in a country.

Determinants of resurgences in cases observed in Category J and K countries (and indicators of new strains in some other countries)

Categories J and K include 28 countries in which reported cases have surged after first waves which were likely or possibly susceptibility-driven, with curves flattened to various extents as a result of control measures which mitigated the epidemics. In addition, there have been notable surges or accelerations in several other countries in South & Central America, in Northeast Africa, in West Africa and in Southern Africa besides the ones in Categories J and K. The complete list of countries to consider includes the following:

1. Kenya

2. Democratic Republic of Congo

3. Pakistan, Afghanistan, Kyrgyzstan and Kazakhstan

4. Egypt and Sudan

5. Southern Africa – Malawi and South Africa (in Category J); Eswatini, Lesotho, Namibia, Zambia and Zimbabwe (in Category K); Botswana and Mozambique (both in Category E but showing accelerations in reported new cases in January)

6. West Africa – Cote d’Ivoire, Ghana and Nigeria (in Category J); Benin, Gambia, Mali, Mauritania, Senegal and Sierra Leone (in Category K); Togo and Burkina Faso (in Categories E and C, respectively, but showing rapid growth in cases in December-January)

7. South and Central America – Bolivia, Peru and Suriname (in Category J); Brazil, Mexico, Argentina, Chile, Ecuador, El Salvador, Guatemala, Honduras (in other categories but showing substantial increases in cases in the November to January timeframe) Charts showing the reported cases and stringency index for selections from these countries are shown in figure 8.

• Download figure
• Open in new tab

• Download figure
• Open in new tab

• Download figure
• Open in new tab

Figure 8: Reported cases and stringency index for selected countries which show features indicative of possible susceptibility-driven dynamics but with significant “second waves” after the first peak, or which show surges of accelerations in new reported cases.

See text for discussion.

The resurgences and accelerations in new cases are potentially an indication of some or all of the more complex types of disease dynamics described in this section. To try to use the evolution of reported cases to diagnose the types of dynamics that are taking place, we consider five possible causes for resurgences after a susceptibility-driven peak that seem plausible:

1. Relaxation of control measures, leading to additional cases if the share of population infected at the end of the first peak is lower than the full herd immunity threshold when no measures are in place. (This scenario does not really constitute “complex dynamics”, but we consider it in order to understand what kinds of resurgences can be created purely by relaxation of measures after a susceptibility-plus-measures-driven peak.)

2. Waning of neutralizing immunity to the virus, which is expected to happen in some people after about 4-6 months.

3. Low infection levels in large population segments during the first peak, causing rapid spread of the disease after relaxation of measures within those segments. The effect on reported cases would be enhanced if the segments had higher-than-average case detection rates (as might be expected, for example, for upper- and middle-class communities in LMICs).

4. Emergence of new strains of the virus with higher transmissibility, if the effective herd immunity threshold with the new strain is increased to more than the share of population infected at the end of the first peak.

5. Emergence of new strains of the virus which are able to fully or partly evade immunity acquired from infection by earlier virus strains.

Figure 9 shows the evolution of reported cases that might be expected for each of these five possibilities, for scenarios in which R_{0_e} = 1.25 and 1.5 during the first peak (corresponding to doubling times of about 15.5 days and 8.5 days if t_g = 5 days), which would result in infection levels of about 38% and 59%, respectively, of the population after the first peak.

• Download figure
• Open in new tab

• Download figure
• Open in new tab

Figure 9: Evolution of numbers of disease cases in different scenarios following a susceptibility-driven outbreak with control measures.

Calculated values for new daily cases per million people, from a simple SIR (susceptible-infected-recovered) model, under different assumptions. Each scenario starts with a complete outbreak during which control measures reduce the effective basic reproduction number to R_{0_e} = 1.25 (in the first column) or R_{0_e} = 1.5 (in the second column), resulting in infection levels of about 38% or 59%, respectively, by the end of the initial outbreak. (In all scenarios, the mean generation time is taken to be t_g = 5 days).

Subsequent features of the scenarios are as follow:

(A) Control measures are fully relaxed and R_{0_e} increases to R₀ = 2.75 – after 250 days in panel A1 and after 175 days in panel A2.

(B) Immunity wanes in the population, assuming a uniform distribution of immunity loss between 4 and 8 months after initial infection. The solid and dashed lines in these panels show the expected evolution when immunity wanes fully and wanes by 2/3, respectively.

(C) Segment of the population, for which interactions within the segment are much more common than with the rest of the population, had a lower R_{0_e} = 0.80 during the initial outbreak but increases to R₀ = 2.75 together with the rest of the population – after 325 days in panel C1 and after 250 days in panel C2.

(D) Emergence of a new virus strain with higher transmissibility, increasing R_{0_e} to 2.5 after 225 days (both panels).

(E) Emergence of a new virus strain which can evade immune responses, starting from 250 days. The solid and dashed lines in these panels show the expected evolution when the new strain is assumed to make all and 2/3, respectively, of the previously infected people susceptible to second infections.

Note that the maximum value of the scale for new daily infections is greater for panels C1 and C2 than for the other panels.

The simulations indicate the circumstances which can generate larger, smaller or no resurgences ‐ specifically:

• Second peaks in cases can reach higher maximum values and be narrower than the first peaks when R_{0_e} increases due either to a relaxation of measures or emergence of a higher-transmissibility strain, and either (a) the first peak was substantially flattened or (b) a segment of the population, representing a substantial portion of the total population, experienced low infection levels during the first peak.

• Second peaks in cases which have lower maximum values and are broader than the first peaks are generated when there is waning immunity or immunity-resistant strains – unless the loss of immunity (to contracting and re-transmitting the virus) is complete (which causes second peaks that are similar to the first peaks), or there is simultaneously an increase in R_{0_e} due to relaxation of measures or virus strains which also have higher transmissibility.

• Second peaks, however, are not possible, or will only be very low and broad, if (1) the first peak was not significantly flattened, (2) there are no large population segments with low infection levels after the first peak, and (3) there is no significant reduction in population-level immunity due to waning or emergence of immunity-resistant strains.

The timing of second peaks depends, naturally, on the timing of the triggering events. In the case of waning immunity, the time difference between the first and second peak will always exceed the average time for people to lose immunity, and will increase beyond this lower bound for lower values of R_{0_e} and if population-level immunity loss is less than complete – for COVID-19, this means that waning immunity (alone) cannot generate a second peak in less than 6-8 months.

The results of the simulations suggest that some explanations are more or less likely to explain the resurgences in cases observed in each of the regions in which they have been observed.

In Kenya and Democratic Republic of the Congo, we note three features: (i) second peaks appear to have started following declines in stringency index in each country; (ii) neighbouring countries do not show rapid increases in cases around the same times; and (iii) the rates of increase of cases before the second peaks were somewhat faster, but not much faster, than before the first peaks. These features all suggest the explanation that the second outbreaks were driven by relaxation of control measures (cf. figure 9, panel A1), in which the virus spread among people who remained uninfected after the first, flattened peaks. It is possible that the second waves may have disproportionately affected upper- and middle-class population segments with lower-than-average infection levels at the end of the first peaks (cf. figure 9, panels C1 and C2).

In Pakistan, Afghanistan, Kyrgyzstan and Kazakhstan, the resurgences: (i) started around the time or soon after control measures were relaxed; and (ii) led to second peaks that were wider and lower than the first peaks. It seems like that declines in disease control measures led to the resurgences (cf. figure 9, panels A1 and A2). The second peaks are somewhat larger than might be expected from relaxation of control measures alone; however, they could be readily explained if case detection rates were higher for the second wave (which might have happened if, for examples, the second waves had a disproportionate effect on upper- and middle-class population segments).

Egypt and Sudan experienced second waves in October to January, following first peaks which appeared to be susceptibility-plus-measures-driven (i.e., Category B). The second wave in Sudan can be attributed to a large decline in stringency index, and the second peak is narrower that the first, as expected with a large increase in R_{0_e} (cf. figure 9, panel A1). The cause of Egypt’s second wave is less clear: Egypt’s stringency index declined only by a small amount before its second wave (although it is possible that social distancing may have relaxed to a greater extent in practice after the first outbreak abated) – and the second wave in Egypt had a slower doubling time than the first wave.

In Southern Africa, the main features to note are: (i) second waves happened in all countries in November or December; (ii) cases increased in countries irrespective of whether, and how much, the stringency index changed (including in Botswana and Mozambique which previously had experienced slow but steady trends in case numbers over time); and (iii) increases in cases were steep in all of the countries. These features point to the emergence of a new virus strain which has higher transmissibility – leading to rapid rates of growth in case numbers (cf. figure 9, panel D1) – and likely also resistance to immunity from prior infections – leading to sharp increases in countries with high prior infection levels as well as countries with low to moderate prior infection levels (cf. figure 9, panel E2, compared to figure 9, panel D2 which shows that higher transmissibility alone will not generate a large second peak in a country if large majority of people have previously been infected). Of course, it is known that the B.1.351 variant, first identified in South Africa, has both higher transmissibility and immunity resistance [36,37,6], and that it has been responsible for most new COVID-19 cases in South Africa in the past 3 months. It is possible that relaxation of control measures and increased movements of people, especially around the holiday season in December, may have added to the pace with which the virus spread in some or all of these countries.

In West Africa, the main features have some similarities and some differences with those observed in Southern Africa: (i) second waves happened in most countries across the region, starting at different times between early November and January, but have not been observed in Guinea (and possibly not in Guinea-Bissau); (ii) cases increased in some countries after declines in the stringency index but without changes in other countries; and (iii) increases in cases after resurgences were rapid (with doubling times of less than 15 days) but similar to or slower than the increases during the first waves in some countries. The features suggest the presence of a new virus strain with higher transmissibility and possible resistance to immunity, but the evidence is somewhat less definitive than for Southern Africa. If there is a new virus variant circulating in West Africa, it is most likely to be the B.1.1.7 variant, which has been found in Ghana and Nigeria, but it is possible that the B.1.351 or P.1 or additional unidentified variants may also be present.

In South and Central America, Bolivia and Suriname experienced second waves which started in early December, and followed an upturn in cases in Brazil in early November. The second peak in Bolivia followed a significant decline in control measures, and could be explained by that alone (cf. figure 9, panel A1), because the curve was flattened substantially during the first wave. Similarly, the upturn in cases in Brazil followed a period during which the stringency index declined gradually. The second wave in Suriname started without any prior change in stringency index – and might be due to a higher-transmissibility variant (perhaps the P.1 variant that was first identified as coming from the Brazilian Amazon). Several other countries have experienced increases in cases from around the start of January – including Argentina, Chile, Peru and Colombia in South America as well as Mexico, Guatemala, Honduras and El Salvador in Central America. The simplest explanation for these increases is relaxation of social distancing behaviours over the Christmas and New Year holiday season (even if relaxations in practice are not reflected in the stringency index), although a role for virus variants cannot be discounted entirely.

COVID-19 epidemic status is correlated with income, but most categories include countries of all income levels

Countries of all income levels appear in nearly all the main disease dynamics categories. Nevertheless, there are clear correlations between income and COVID-19 status categories.

Most high-income countries have controlled the spread of SARS-CoV-2 through measures – accounting for 79 of the 113 countries in Categories F, G and H. Of the 15 HICs in other categories, the most populous is the United States of America, which had, as of 31 January 2021, “flattened the curve” but had not suppressed the virus spread, and five Arabian Peninsula countries in which the epidemic appears to have disproportionately spread in only parts of their populations.

Middle-income countries are spread across all categories. They account for 45 of the 63 countries which have slowed the disease significantly but not fully suppressed it (Categories B, C, D and E) – including populous countries such as India, Indonesia, Brazil, Bangladesh, Russia, Mexico and the Philippines. Furthermore, 25 out of 34 countries in Categories J and K are MICs – including Pakistan, Nigeria and Egypt – which have seen large resurgences in cases after susceptibility-driven peaks (or other uncertainties in categorizing the dynamics of their COVID-19 epidemics). Middle income countries also account for about half of the countries which have kept the disease out (Category H), and some others which have suppressed the disease to very low levels (including in Categories F and G) including several countries in South-East Asia, China, and many island states.

Some low-income countries have experienced largely unmitigated susceptibility-driven dynamics (7 of 29 countries, including countries in Categories A and some in Category J), while others have “flattened the curve” to varying degrees (11 countries in Categories B, C, D, E and J).

Regional trends have emerged in COVID-19 epidemic status – but there are exceptions in each region

There are clear regional trends in the state of COVID-19 epidemics as of 31 January 2021. This is not surprising for three reasons. First, countries within a given region often have similar income levels. Second, the disease dynamics in one country is likely to influence in the dynamics in neighbouring countries, due to transmission of the disease across land borders. Third, governments and citizens look to neighbouring countries in determining disease control measures. There was more diversity in the state of the epidemic within regions earlier in the pandemic, but regional patterns have become clearer over time, especially as some countries which kept the disease out or under control during the early months of the pandemic have subsequently experienced more significant outbreaks.

In the Americas, the disease has spread slowly but has not been suppressed (Categories D and E), up to 31 January 2021, in most countries – including those with the largest populations, including the United States, Mexico, Colombia, Brazil and Argentina. Many of the Caribbean islands have kept SARS-CoV-2 out or under control (Categories F, G and H), but some have not, in particular Dominican Republic, Haïti, Jamaica and Puerto Rico. Several countries in South and Central America have showed prominent resurgences in cases, perhaps indicating the spread of a new variant with higher transmissibility and potentially also immunity resistance – including Bolivia, Peru and Suriname, in which cases had declined to low levels after susceptibility-driven peaks before increasing again (Category J).

Western and Northern European countries have, for the most part, controlled the disease through social distancing and hygiene measures, through two or three waves, and fall in Categories F and G. Some countries appear – based on estimates of excess deaths [8] as well as reported cases – to have reached infection levels well above 10% and perhaps beyond 20%, including Belgium, Spain and the United Kingdom, while other countries appear to have kept infection levels to low single-digit percentages, likely including Germany, Denmark, Finland and Norway.

Across Eastern Europe, the Levant, the Caucuses and Iran, all countries have constrained growth of the disease significantly, but infection levels in most have grown to moderate levels. The countries fall into three broad groups based on the COVID-19 status categories, although some countries likely lie at the boundary between categories. In many of countries, the virus was controlled or crushed during the early months of the pandemic. In the second half of 2020, case numbers grew substantially. In several countries in the Balkans, as well as Armenia and Iran, cases appear to have gone through peaks and declined due to rising immunity levels in the population (albeit likely at fairly low infection levels due to reductions in R_{0_e} caused by control measures) – and these countries fall into Category C. In other countries in Eastern Europe, as well as Georgia and Azerbaijan, increases in cases in the second half of the year were brought under control through measures – putting them in Categories F or G – although likely only after significant numbers of people were infected. For Belarus, Kosovo, Palestine and Turkey, it isn’t possible to determine whether Category C or Category F applies, and they are classified in Categories K or L. In some other countries in the region – including Moldova, Ukraine, Russia, Syria, Lebanon and Iraq – cases have shown, on average, a relatively steady increase over time, putting those countries into Categories D and E (or Category B in the case of Iraq).

In South and Central Asia, the virus has spread widely in most countries and cases have declined. In India, Bangladesh, Nepal and Uzbekistan, the case curve was flattened considerably, and current infection levels are likely moderate (Category B). In Pakistan, Afghanistan, Kyrgyzstan, and Kazakhstan (all in Category J), there have been two peaks in cases; the second peaks may have been generated purely due to relaxation of control measures, leading to spreading of the disease to previously uninfected people, but the possibility cannot be discounted that a new variant with some immunity resistance and/or higher transmissibility may have contributed to the second waves. Bhutan has contained the outbreaks of the virus to date (Category F).

Most countries in East and South-East Asia have largely kept the disease under control. Viet Nam, Laos, Cambodia, Thailand and Taiwan (all in Categories F or H) have largely kept the disease out – as has China in the time since it suppressed the initial outbreak of the disease in February and March of 2020. Malaysia, Mongolia and Myanmar also kept the virus until the second half of 2020, but have experienced widespread outbreaks since then. The disease has spread very slowly in Indonesia (in Category E) – which has one of the longest doubling times in the world – but has not been suppressed to date. The Philippines appears to be past the peak of its epidemic to date, and the curve was flattened significantly – meaning, like for other countries in Category B, that significant numbers of people have been infected but equally that sizable numbers remain susceptible to the disease (even before new variants are taken in account).

In Australia, New Zealand and most Pacific Island States, SARS-CoV-2 has been excluded through quarantines, together with testing and tracing and lockdowns when the virus has spread beyond quarantined individuals (Categories F and H).

There are significant differences in the status of COVID-19 epidemics across the African continent. Many countries appear to have experienced widespread epidemics followed by declines in case numbers. Disease spread in those countries was largely unmitigated in a few countries, but the curve was flattened in most. In some countries – Tunisia, Libya, Togo, Botswana and Mozambique – the effective reproduction number was reduced to very close to 1 and cases spread very slowly during 2020 (falling into Category D or E). A few countries – Eritrea as well as Chad (and possibly Niger and other Sahelian countries) – appear to have kept the disease out and suppressed small outbreaks, and a few, such as Algeria and Burkina Faso, may have experienced widespread outbreaks after keeping the virus under control during earlier outbreaks. As described earlier, several countries in Southern Africa, in West Africa, and in Northeast Africa experienced rapid growth in case numbers in December and January. Spikes in cases have occurred in countries which appeared to have previously experienced susceptibility-driven peaks (previously Category B, now Category J), in countries where the nature of earlier peaks was uncertain (Category K) and in countries with slowly rising numbers of cases (Category E).

Actual infections and deaths are likely much higher than previously estimated

For each country, we estimate the actual number of infections and deaths to date, based on the overlap between two plausible ranges. The first set of ranges comes from the total numbers of reported cases and deaths, and plausible ranges of case and death detection rates for the country’s income level. The second set of ranges comes from the expected range of infection levels for the country’s disease dynamics category, and the estimated infection fatality rate (IFR) for the country based on its population by age. Figure 12 shows these estimates for different groups of countries. Globally, the estimates suggest that roughly 1.3–3.0 billion people have been infected by SARS-CoV-2 to date, or about 17–39% of the population, which is between 13 and 30 times the number of confirmed cases, and perhaps twice to four times as much as previous estimates of total infection numbers [38]. The estimate for actual deaths to date is 4.6–10.0 million, between 2.1 and 4.5 times the number of confirmed deaths attributed to COVID-19.

• Download figure
• Open in new tab

Figure 12: Estimates of actual and potential additional cases and deaths if 90% of the population were infected, for different groups of countries and territories.

Estimates are shown for different disease dynamics categories, income levels and regions.

High-income countries, China and India account for about two-thirds of potential extra deaths (before vaccinations or reinfections)

We next estimate the number of people who remain vulnerable to infection, assuming that up to a total of 90% of the population (somewhat higher than the proportion expected if R₀ = 2.75 and the outbreak were completely uncontrolled) could be infected in the absence of control measures or vaccination. We also estimate the number of additional deaths, based on published estimates of age-specific IFRs [20] and the population by age in each country. There is some evidence that IFRs may have declined in recent months due to improvements in treatment of COVID-19 patients; it is also likely that health systems in LMICs and LICs have less capacity to handle COVID-19 cases and will be unable to provide the same standard of care. Consequently, we make a second estimate of potential additional deaths, by scaling the numbers for HICs and UMICs by 1/3. The estimates for different groups of countries also appear in figure 12. Not surprisingly, Category A/B/C countries have the lowest estimates for additional cases and deaths (as a share of population), while Category F/G/H countries have the highest estimates. The gaps between categories are even greater for estimated deaths, because Category A/B/C countries have younger populations than the global average and Category F/G/H countries have the oldest populations. Allowing for better COVID-19 treatment in richer countries reduces the effect because Category A/B/C countries are mainly LICs and LMICs while Category F/G/H countries are mainly HICs and UMICs.

Across the world, we estimate that if 90% of the population were infected, COVID-19 would cause an additional 8.9–12.5 million deaths allowing for reduced mortality in HICs and UMICs due to improvements in COVID-19 care (or 20.0–25.7 million deaths without such reductions in age-specific infection fatality ratios). HICs account for about 2.4–2.9 million of the potential additional deaths (assuming reduced mortality), China for about 2.1 million, and India for about 1.7–2.9 million (large numbers of people remain uninfected because India’s case curve was flattened significantly and reductions in infection fatality rates from improved COVID-19 care are not likely to apply). In all, HICs, China and India account for about two-thirds of the potential extra deaths that could occur if 90% of all people were infected (69% of the lower estimates and 63% of the upper estimates). Vaccinations, of course, will reduce these numbers substantially: as of 16 February 2020, HICs had administered about 40 million vaccine doses (which corresponds to 17% of the over-65 population of HICs, although some people have already received two doses, and not all vaccines have been administered to people over-65), China had administered about 41 million doses (corresponding to 3% of the over-65 population), and India had administered about 9 million doses (corresponding to 0.7% of the over-65 population). New disease variants that can evade immune responses (whether from prior infection or vaccination) could allow for infection rates of more than 90% of the population – but it remains to be seen whether the infection fatality ratios from reinfections will be comparable to those from first infections or much lower as might be expected.

Testing for total infection and immunity levels

It is important to get accurate information on the current state of the COVID-19 epidemic in various countries. Ideally, this would be achieved through immunity testing of representative population samples. It may be impractical for every country to implement serological testing programmes – but even studying a selection of countries across the different categories would help to test our conclusions that countries in Categories A-E and I-K (i.e., other than Categories F-H) have high to moderate infection and immunity levels, and the findings of this report could be applied to other countries which cannot deploy immunity testing quickly.

Of course, many serological studies have already been undertaken – and their estimates of infection and immunity levels are consistent with the findings of this study. Seroprevalence levels in several European countries, the United States, Canada and China all indicate low immunity levels [18,20], while studies in several cities and regions in Brazil, India, Kenya, Nigeria, Pakistan, Qatar, and South Africa [3-7,18,20] reported finding antibodies for SARS-CoV-2 in large percentages of the studied populations.

Note that serological testing will underestimate the number of people who have been infected by SARS-CoV-2 once more than 4–6 months have passed, due to waning of SARS-CoV-2 antibodies. As discussed earlier, it is possible, therefore, that serological testing may understate the actual degree of immunity in a population, because some such people may have antibodies at levels below the detection threshold of the serology tests or may have memory B cell or T cell responses, either or both of which may reduce their vulnerability to infection and/or reduce their likelihood to pass on the virus to other people.

In some places, reliable estimates of actual deaths due to COVID-19 may be a substitute for immunity testing to determine the share of population infected to date, at least approximately. Such estimates have been made for many HICs and a few MICs – including Bolivia, Ecuador, Peru, Mexico and South Africa – by comparing total deaths in since the start of the COVID-19 pandemic with expected baselines based on deaths in recent years [9]. Estimates have been made using novel methods for several LICs, including for Sudan using a social media survey, a joint serological and molecular survey and modelling [10], for Syria using community-uploaded obituary certificates and modelling [11], for Yemen using geospatial analysis of burial surface areas in cemeteries [12] and for Zambia from a systematic post-mortem surveillance study [13]. By comparing the estimates of actual deaths with estimates of expected mortality rates, one can infer estimates for the share of population infected to date. These estimates infer low infection levels for 34 HICs, a wide range of infection levels for UMICs, and moderate to high infection levels for the LMICs and LICs studied [8].

The implications of immunity levels on the disease will become complex to understand during the year ahead, for several reasons:

• Immunity from first infections will wane. More research is needed to determine to what extent people are at risk of reinfection after immunity begins to wane, how severe the symptoms from second infections will be, and whether reinfected people will transmit the virus at the same rates as people infected for the first time.

• People will gain immunity through vaccinations. More research will be needed to determine whether, and to what extent, vaccinated people are protected from contracting and re-transmitting the virus, and to determine for how long immunity from vaccines remains before waning. For countries in Categories A-E, which likely have high to moderate infection levels, the total numbers of immune people as vaccination programmes ramp up will depend on whether vaccinations can be targeted to people who have not already contracted SARS-CoV-2 or are distributed generally irrespective of prior infection.

• New virus strains will render previously immune people susceptible again. Several new strains, such as the B.1.351 variant first identified in South Africa and the P.1 and P.2 variants (descended from the B.1.1.28 variant) first identified in Brazil, appear likely to reduce immunity levels. However, it is likely that previously infected or vaccinated people will have partial immunity to these variants, reducing the severity of their symptoms and possibly reducing the degree to which they transmit the new strain to others. Further research on these variants, and any others that emerge, will be necessary.

Disease control strategies

Countries in different categories ought to pursue different approaches to disease control. Figure 13 helps to illustrate the choices facing countries in different disease dynamics categories, by showing calculations of the expected total infections in the event of different decisions on relaxation of control measures. In both panels, the solid black squares and line show the total infections, for different values of the effective reproduction number R_{0_e} under control measures, if measures are not relaxed. In both panels, the maximum infection level is 92.4%, the share of the population infected if an epidemic with R₀ = 2.75 is unmitigated; note that this is higher than the herd immunity threshold, which is 63.6%, because infections continue even after new case numbers start to decline. Panel A shows the total infection levels expected if control measures are fully relaxed (causing R_{0_e} to revert to R₀), for different timings of the relaxation of control measures after the peaks in cases. Panel B shows the total infection levels expected if control measures are relaxed well after the peak in cases (200 days) but are only partially relaxed; black squares show infection levels in the scenario in which measures are not relaxed (prior to vaccination), while black ‘X’ marks show infection levels in the scenario in which measures are fully relaxed. In both panels, arrows indicate the outbreak dynamics categories (A, B/C, D/E, F/G) that roughly correspond to the initial values of R_{0_e} in the simulations – although this correspondence is not exact because R_{0_e} varies over time for most countries in most categories.

• Download figure
• Open in new tab

Figure 13: Expected total share of population infected during an epidemic for different levels of control measures and different timings and degrees of relaxation of control measures.

Calculated values from a simple SIR model, with R₀ = 2.75 and t_g = 5 days.

(A) Control measures reduce R_{0_e} to between 0.5 and 2.75 at the start of the outbreak, and then are relaxed to allow R_{0_e} to return to R₀, at different times after the peak in cases.

(B) Control measures reduce R_{0_e} to between 0.8 and 2.0 at the start of the outbreak, and then are relaxed at 200 days after the peak in cases, to different values of R_{0_e} post-relaxation.

Arrows indicate the outbreak dynamics categories (A, B/C, D/E, F/G) that roughly correspond to the initial values of R_{0_e} in the simulations – noting that this correspondence is only approximate as R_{0_e} varies over time for most countries.

As shown in both panels, for these simulations, if there is a complete first wave of the epidemic mitigated by measures, it is possible to relax measures to some degree without causing a second wave and additional infections – if the infection level at the time that measures are relaxed is less than the effective herd immunity threshold under the relaxed measures. If the first wave is complete, and if R_{0_e} = R₁ during the first wave, the total infection level Z₁ at the end of this wave is determined by the equation [39]. If R_{0_e} = R₂ after measures are relaxed, the effective herd immunity threshold is 1 − 1/R₂. Consequently, there will no second wave if Figure 14 shows the maximum values of R₂ for which there is no second wave after relaxation of measures, from this inequality. However, if measures are relaxed before the first wave is complete, the infection level will be lower than Z₁, and the maximum value of R₂ for which there is no second wave will be lower than that determined by the above inequality. If R₂ = R₀ satisfies the inequality, then it is possible to relax measures fully.

Virus variants with higher transmissibility increase R_{0_e} even if control measures stay constant, and thus can generate second waves of sizes determined by figure 13. Virus variants that evade immune responses, or waning of immunity from infection, would allow for re-infections, and thus for greater numbers of additional infections after first waves than those suggested by figure 13.

• Download figure
• Open in new tab

Figure 14: Maximum values of R_{0_e} after relaxation of measures, for different values of R_{0_e} during the first wave under control measures, for which there is no second wave after relaxation of measures.

Values determined from the inequality presented in the text, which assumes relaxation of measures only after the first wave if fully complete, that there are no significant homogeneities in the population and that there is no waning of immunity or emergence of immunity-resistant variants. Arrows indicate the outbreak dynamics categories (A, B/C, D/E, F/G) that roughly correspond to the values of R_{0_e} during the first wave – noting that this correspondence is only approximate as R_{0_e} varies over time for most countries.

Vaccination strategies

Multiple vaccines have been developed, and are being deployed, which have high levels of effectiveness in preventing COVID-19 [40]. The effectiveness of these vaccines may be reduced for some new variants of the SARS-CoV-2 virus. It is not yet clear to what extent the vaccines reduce transmission of the virus, either of the early variants or the more recent variants with higher transmissibility and/or ability to evade immune responses.

Announcements made by various companies suggest that perhaps 12.5 billion doses of the major vaccines could be produced in 2021, enough for 6.25 billion people (as the vaccines require two doses to be effective), or most of the world’s population. However, supplies are limited at present and will continue to be limited globally at least until the middle of 2021. And supplies will be limited in most of the world well beyond that, with most countries only expected to achieve widespread vaccination coverage by 2022 or 2023, according to the Economist Intelligence Unit [41]. The limitations on supply of vaccines raises several questions. First, to which segments of the population should vaccines be deployed first? Second, how should vaccines be allocated across countries? Answers to these questions should be driven principally by ethics and equity, among countries and individuals. Making these choices requires information on the potential benefits, in terms of deaths averted or life-years saved, of different choices. Key analytical questions include: With a limited supply of vaccine, is it better to vaccinate directly those who are most at risk of death or severe illness, or is it better to vaccinate the general population, or key transmitters within the population, to constrain disease spread and hence protect (indirectly) greater numbers of vulnerable people? What allocation of vaccine doses across countries will protect the greatest numbers of people – taking into account both different numbers of vulnerable people and the potential benefits of possible vaccination strategies in different countries?

A.B. Hogan et al. have used mathematical models to study some of these questions and reach several important findings [42]. The number of cases and deaths averted through vaccination is, not surprisingly, greater if the proportion of people already infected is smaller at the time vaccination starts and/or if the immunity conferred by past infection has a shorter duration. More deaths per thousand people are averted in HICs and UMICs than in LMICs and LICs, because HICs and UMICs have older populations, if the ability of health systems to save lives of COVID-19 patients is ignored, but the numbers are roughly similar across country income groups if the constraints on health systems are factored into the estimates – and the numbers of life-years gained is higher in LICs and LMICs. They find that, “within a country, for a limited supply (doses for <20% of the population) the optimal strategy is to target the elderly and other high-risk groups” but that, if a larger supply is available within a short time, “the optimal strategy switches to targeting key transmitters (i.e., the working age population and potentially children) to indirectly protect the elderly and vulnerable” – for simulations which assume that small numbers of people have previously been infected.

Current policies in several countries call for deployment of vaccines first to healthcare workers and then by age cohort, starting with the oldest [43].

For countries in Category A, our findings imply that early vaccination should concentrate on elderly and vulnerable people. There is no alternative strategy to consider in the short term, because vaccination is not required to create herd immunity in the population to the initial variants of the virus – although vaccination to create herd immunity could become relevant for future variants if they evade immune responses and if vaccines are available to protect against such new variants. The greatest danger, both now and into the future, lies in the threat to elderly and vulnerable people from endemic SARS-CoV-2 over time, as the disease is unlikely to be eliminated completely.

For countries in Categories F and G and in Category H, which have low current infection levels, the simulations of Hogan et al. are most applicable, and vaccinations should be given first to frontline healthcare and other essential workers and to elderly and vulnerable groups (starting with the oldest and most vulnerable).

For countries in Categories D and E, there may be situations in which the optimal strategy for early vaccination switches to targeting key transmitters (mainly working-age people) and indirectly protecting the elderly and vulnerable, even if the supply of vaccine doses is enough for only a small share of the population. In some of these countries, if key transmitters are vaccinated (with vaccines that are capable of limiting transmission and not just reducing severity of disease and mortality), and if current disease control measures are maintained and higher-transmissibility virus variants don’t become common, then it may be possible to halt transmission of the disease completely, while waiting for further vaccine supplies to arrive (after which disease control measures could be released). The strategy to start with key transmitters to suppress the disease has already been adopted by Indonesia [44], and a recent paper proposed that mass vaccination of adults aged 20-49 should be considered in the United States to bring the disease under control because the reproduction number for this group has remained above 1 and sustained resurgences in cases [45]. For each country in Categories D and E, careful modelling would be necessary to determine if the key transmitter strategy is optimal, taking into account the likely supplies of vaccines and when they will be available. Vaccination of elderly and vulnerable groups would likely be the most prudent choice in the absence of good country-specific modelling and an effective plan for identifying and vaccinating key transmitters. In countries, such as Indonesia, that opt for a strategy to vaccinate working-age people first, it will be important to keep control measures in place and to keep high-transmissibility variants of the virus out; otherwise, the benefits of the key transmitter vaccination strategy will be lost, and the COVID-19 cases will continue to grow in number.

For countries in Category B, vaccinations should probably be given first to frontline healthcare and other essential workers and to elderly and vulnerable groups (starting with the oldest and most vulnerable), similar to countries in Categories F, G and H. However, if there are resurgences in cases across the population (and not just in limited segments of the population), due to higher-transmissibility or immunity-resistant variants or otherwise, then the optimal strategy may switch to targeting key transmitters, similar to some countries in Categories D and E.

In allocating vaccine doses across countries, Hogan et al. found that, “the strategy that maximises the total deaths averted is one in which the available vaccine doses are allocated preferentially to higher-income countries who have the highest at-risk elderly populations,” but noted that “the optimality of this allocation is sensitive to many assumptions and will vary depending both on the vaccine characteristics and the stage of the epidemic in each country at vaccine introduction,” and concluded that, “[g]iven this uncertainty, allocating vaccine doses according to population size appears to be the next most efficient approach.”

Our findings reinforce the uncertainty strongly: it is very likely that that stage of the epidemic varies greatly across countries. For most countries, the optimal allocation of vaccines doses is likely still to be according to population size (or according to population over 60, scaled up by factors of perhaps 3-4 for LICs and LMICs with less effective healthcare) – and then for those countries to give doses first to elderly and vulnerable people. However, the global optimal allocation strategy might include providing somewhat greater supplies, during the next few months, to Category D and E countries where using the vaccine to halt spread of the disease might be possible (provided that disease control measures are maintained in those countries). It is clear, in any case, that further modelling of vaccine allocation strategies is essential, taking into account the actual vaccine efficacies (in limiting both illness and virus transmission) and projected available doses by month, as well as allowing for disease stage categories in different countries.

Higher-transmissibility variants of the SARS-CoV-2 virus will not affect early vaccination strategies for most countries – other than to increase the urgency of distributing vaccines, especially in some Category F and G countries which may not be able to keep the disease suppressed once the higher-transmissibility variant begins to dominate new infections. The exceptions are any Category D or E countries that consider starting with vaccinations of key transmitters to suppress the disease spread; such countries will need to be sure that they can keep the disease suppressed even if higher-transmissibility variants become common.

Immunity-resistant variants of the virus may reduce the effectiveness of current vaccines. However, the immune responses generated by current vaccines are likely to be at least partially effective against new variants, even if the variant can avoid being fully neutralized by antibody responses.

In the longer term, vaccination strategies may need to plan for multiple cycles of vaccinations. Immunity-resistant virus variants will necessitate development and deployment of new vaccines that are fully effective against the new variants. Immunity acquired through vaccination is likely to wane over time – like immunity acquired through infection. In making long-term plans, countries will face a wide range of options for who to vaccinate (elderly and vulnerable populations, key transmitters or entire populations) and for frequency of vaccination (every 6 months, annual, or once if residual benefits are sufficient). Optimal strategies for each country will be complicated to determine, as the choice will depend on many factors, including vaccine effectiveness in reducing mortality and in reducing transmission, how effectiveness wanes over time, mortality rates and transmissibility of new variants (in general and in previously infected or vaccinated people), and, once the risks to life and health from endemic COVID decrease to the point where COVID-19 is not an overriding issue, comparison with other health and budgetary priorities.