The impact of COVID-19 vaccination campaigns accounting for antibody-dependent enhancement

Background COVID-19 vaccines are approved, vaccination campaigns are launched, and worldwide return to normality seems within close reach. Nevertheless, concerns about the safety of COVID-19 vaccines arose, due to their fast emergency approval. In fact, the problem of antibody-dependent enhancement was raised in the context of COVID-19 vaccines. Methods and findings We introduce a complex extension of the model underlying the pandemic preparedness tool CovidSim 1.1 (http://covidsim.eu/) to optimize vaccination strategies with regard to the onset of campaigns, vaccination coverage, vaccination schedules, vaccination rates, and efficiency of vaccines. Vaccines are not assumed to immunize perfectly. Some individuals fail to immunize, some reach only partial immunity, and—importantly—some develop antibody-dependent enhancement, which increases the likelihood of developing symptomatic and severe episodes (associated with higher case fatality) upon infection. Only a fraction of the population will be vaccinated, reflecting vaccination hesitancy or contraindications. The model is intended to facilitate decision making by exploring ranges of parameters rather than to be fitted by empirical data. We parameterized the model to reflect the situation in Germany and predict increasing incidence (and prevalence) in early 2021 followed by a decline by summer. Assuming contact reductions (curfews, social distancing, etc.) to be lifted in summer, disease incidence will peak again. Fast vaccine deployment contributes to reduce disease incidence in the first quarter of 2021, and delay the epidemic outbreak after the summer season. Higher vaccination coverage results in a delayed and reduced epidemic peak. A coverage of 75%–80% is necessary to prevent an epidemic peak without further drastic contact reductions. Conclusions With the vaccine becoming available, compliance with contact reductions is likely to fade. To prevent further economic damage from COVID-19, high levels of immunization need to be reached before next year’s flu season, and vaccination strategies and disease management need to be flexibly adjusted. The predictive model can serve as a refined decision support tool for COVID-19 management.

Introduction controversial due to its fast emergency use authorization (EUA) without a phase III 23 study [8]. Nevertheless, mass vaccination started in Russia on December 5, 2020. The 24 modified chimpanzee adenovirus vector-based candidate from AstraZeneca, AZD1222, is 25 currently under phase III study, will cost only 4 USD per dose, and has a capacity of 26 400 million doses for Europe and 300 million doses for the USA. 27 China gave EUA to two vaccines that trigger an immune response by inactivated 28 SARS-CoV-2 variants. BBIBP-CorV has a capacity of 1 billion doses for China in 2021 29 at a cost of less than 75 USD per dose and was fully approved, while CoronaVac costs  NVX-CoV2373 by Novavax, seeking approval in Mexico, is a vaccine that uses 34 SARS-CoV-2 recombinant spike protein nanoparticles with adjuvants to trigger an 35 immune response [9]. 36 Two modRNA-based candidates are currently in phase III studies, which either seek 37 approval or were granted EUA. Tozinameran (BNT162b2) by BioNTech (20 USD per 38 dose), was approved in Canada and Europe, and received EUA in the UK and the USA. 39 A second modRNA-based candidate, mRNA-1273 by Moderna, is currently in phase III 40 trials, and received EUA in Canada and the USA. 41 Vaccination campaigns aim for herd immunity. There is an ongoing debate on the 42 optimal deployment of the vaccine. Some countries have ambitious deployment 43 strategies, e.g., Morocco plans to immunize up to 80% of the population. Globally the 44 trend is to deploy vaccines voluntarily and free of charge, with a general agreement to 45 prioritize vulnerable risk groups (e.g., senior citizens, people with co-morbidities, etc.) 46 and people of systemic importance (e.g., healthcare workers, police, public services) 47 before making the vaccine available to the general public [10]. Incentives for getting 48 voluntary vaccines have been proposed, e.g., recently Qantas airlines announced to make 49 the vaccine mandatory for their passengers [11,12]. 50 Nevertheless, skepticism about vaccines and their potential side effects are 51 widespread, resulting in vaccine hesitancy [13]. One of the potentially negative effects of 52 March 5, 2021 2/16 a vaccine is the occurrence of antibody-dependent enhancement (ADE) or, more general, 53 enhanced respiratory disease (ERD) [14,15]. ADE is best understood in Dengue fever 54 and was observed also in SARS-CoV and MERS-CoV both in vitro and in vivo [16]. In 55 SARS-CoV-2, ADE occurs most likely via enhanced immune activation [17]. Here, 56 sub-optimal antibodies form immune complexes with the virus that deposit into airway 57 tissues and activate cytokine and complement pathways. This triggers inflammation, 58 airway obstruction, and even acute respiratory distress syndrome [17]. By this 59 mechanism, vaccines could potentially result in more severe symptoms upon infection 60 with SARS-CoV-2.

61
Here, we employ predictive modelling to explore the outcome of vaccination 62 strategies on the COVID-19 pandemic. We explore the influence of (i) the vaccination 63 coverage reflecting vaccine hesitancy, the commonness of contraindications, and access 64 to the vaccine; (ii) the vaccination rate, reflecting supply and infrastructure; and (iii) 65 the immunizing effect on disease incidence, prevalence, and mortality. We further 66 investigate the impact of the launch of the vaccination campaign relative to the 67 epidemic peak. Intentionally, our model accounts for the occurrence of ADE, ERD, and 68 other deleterious side effects of the vaccine -subsumed here as ADE. By this approach, 69 we seek to address the following questions: What is the benefit of launching the Vaccination either results in (i) full immunity, (ii) partial immunity, (iii) no immunity 122 (the vaccine had no effect), or (iv) ADE (the vaccine had a deleterious effect). Partial

127
Individuals are vaccinated only once (one vaccination cycle, subsuming all necessary 128 doses). The waiting time during which immunization is pending reflects the waiting 129 time for the vaccination cycle to be completed (it can be one or several doses). In the 130 model we need to distinguish between infections of individuals that are (i) partially latent, and prodromal phases, and during the fully infectious and late infectious phases 135 if the infections remain asymptomatic and undetected; see Figure 3). After some 136 waiting period (during which the effect of the vaccination is pending) these individuals 137 are either successfully, partially, or not immunized, or they might be affected by ADE. 138 The likelihood of these outcomes depends on the phase of the infection. Particular 139 consideration is given to individuals that are vaccinated during the fully infectious or 140 late infectious phase. Namely, when the disease has already progressed, the effect of the 141 vaccine potentially changes. dependent. For the simulations here we assume an initial "hard lockdown" followed by a 152 phase of relief, a "soft lockdown", a second "hard lockdown", and a final relief phase 153 before contact reductions are lifted.

154
Contact rate 155 Susceptible individuals get infected by random contacts with infected individuals that 156 can transmit the disease and are not isolated. The contact rates are mediated by general 157 contact reduction. The basic reproductive number R 0 is assumed to fluctuate seasonally. 158 Model implementation 159 The model as described in detail in S1 Mathematical Appendix was implemented in 160 Python 3.8. We used a 4th order Runge-Kutta method using the function solve ivp as 161 part of the library Scipy. Graphical output was created in R [19]. The Python code 162 with the model implementation is available at GitHub 163 (https://github.com/Maths-against-Malaria/COVID19_ADE_Model.git).

165
Here, we study the effectiveness of different hypothetical vaccination 166 programs/strategies to immunize the population in the ongoing COVID-19 pandemic in 167 terms of disease incidence (or prevalence) and overall mortality. Vaccination campaigns 168 differ in (i) their onset and vaccination rate reflecting the available infrastructure reflecting the willingness of the population to get vaccinated and access to medical care; 171 and (iii) vaccination schedules, immunogenicity, and efficacy/effectiveness summarizing 172 vaccine-specific properties. The ultimate goal of any vaccination campaign is to reach 173 herd immunity, mitigate SARS-CoV-2, and "return to normality" as soon as possible. 174 We explore how fast herd immunity is reached assuming that contact reductions cannot 175 be sustained for too long. The approach is conservative, as we focus on the potential 176 negative effects of the vaccine. 177 We report disease incidence and prevalence. Disease prevalence is defined as the sum 178 of all infected individuals (in the latent, prodromal, fully infectious, and late infectious 179 phases). By incidence we refer to 7-day incidence, which is the number of new cases 180 within the last 7 days. This is defined as the integral of the force of infection over the 181 time interval from t − 7 to t. Approximately, incidence and prevalence differ only by a 182 multiplicative factor here and can be used synonymously. Both are reported for the 183 readers convenience to facilitate comparison with publicly accessible data.

184
Regarding the results, we use model parameters reflecting the situation in the Federal 185 Republic of Germany that so far intervened successfully in the COVID-19 pandemic. To 186 illustrate the model's applicability to other countries, we also parameterized it to reflect 187 the situation in the USA. The results for the USA are presented in S2 Results for the 188 USA. The parameters used for Germany are listed in S1 Table -S5 Table. The 189 population size of Germany was set to N = 83 million. We assumed the first COVID-19 190 cases were introduced in late February 2020 (corresponding to t = 0). A basic yearly 191 average reproductive number ofR 0 = 3.4 was assumed, which fluctuates seasonally by 192 43% with a peak in late December (t R0 max = 300). The average durations for the latent, 193 prodromal, fully infectious and late infectious phases were set to D E = 3.7, D P = 1, increased for those that developed ADE (see S4 Table). These parameters were justified 201 by CovidSim 1.1 [18] and are a combination of COVID-19 and influenza estimates. Regarding general contact reductions, we assumed a "hard lockdown" from early 206 April (t Dist1 = 40) to mid-May 2020 (t Dist2 = 82) that reduced p Dist1 = 70% of all 207 contacts, followed by a period of relaxation until the end of October (t Dist3 = 246) 208 during which p Dist2 = 40% of the contacts were avoided. The first "soft lockdown" from 209 late October was sustained until the beginning of December (t Dist4 = 280) with a 210 contact reduction of p Dist3 = 50%. This was followed by a second "hard lockdown" from 211 early December until late March (t Dist5 = 397) and a phase of relief resulting in a 212 p Dist4 = 68% contact reduction sustained until May 2021 (t Dist6 = 450), after which all 213 general contact reductions are lifted reflecting the worst-case scenario, in which 214 compliance with social distancing can no longer be sustained in the face of the vaccine 215 becoming available. Contact reductions were deduced by assuming roughly the imposed 216 contact restriction in Germany in schools, at work, at home, and other locations. These 217 restrictions reduced the age-dependent contact-rate estimates available from [20], which 218 where then averaged over all age strata, weighted by their relative sizes.

219
The simulation results until early March 2021 (t = 375) match the disease incidence 220 in Germany. Notably, only the number of confirmed, not of actual cases is known. The 221 true incidence (which is modelled) is obviously unobservable and might be substantially 222 March 5, 2021 6/16 higher than the confirmed incidence (true incidence multiplied with the probability of 223 detection). Consequently, the incidences in Figs 4-9 exceed the number of reported cases 224 in Germany -however, the numbers are plausible and match in the order of magnitude. 225 Regarding the vaccination campaigns, we use the following reference scenario: 60% of 226 the population will get vaccinated, a vaccination rate of 1/180 (i.e., an average time of 227 180 days to get vaccinated), a launch of the campaign in late December 2020 (t = 310), 228 and a vaccination schedule of 28 days (the vaccine becomes effective after 28 days).

229
No vaccination 230 We consider the situation in which no vaccination campaign (VC) is launched as a 231 reference case for comparisons. Figs 4A-D show disease incidence, which matches the 232 reported numbers in Germany. The base reproductive number is at the seasonal 233 maximum at times t = 300 and t = 665. Although R 0 declines after t = 300, the end of 234 the second "hard lockdown" in late March (t = 397) leads to a moderate increase in 235 incidence. In fact, it will start to decline later in spring until summer due to seasonal 236 reductions in R 0 . After contact reductions are lifted at time t = 450, disease incidence 237 would increase drastically and the epidemic peak would be reached around t = 590 (late 238 October 2021) with almost 40% of the population being infected at this time point.

239
This is a hypothetical worst-case scenario.

240
Onset of vaccination campaigns 241 The onset of the vaccination campaign (VC) has a profound effect on disease incidence 242 (see Fig 4). The earlier the onset of the VC, the earlier and the stronger the reduction in 243 incidence after the second "hard lockdown". Even if the VC starts at t = 300 incidence 244 will increase after the second "hard lockdown". However, the increase is less than half of 245 that observed without a VC. Even launching the campaign in Dead only affects mortality, not incidence.) As a baseline comparison, the black lines show incidence and mortality in the absence of the vaccine. The vertical dashed line indicates time t = 450 at which contact reductions are lifted. Seasonal fluctuations in R 0 are shown by the grey dashed lines corresponding to the y-axis on the right-hand side. Plot parameters are given in S1 Table -S3 Table   Fraction

265
It takes a vaccination coverage of about 70% to avoid a high epidemic peak after the 266 contact reductions are lifted. The lower the proportion of the population being 267 vaccinated, the earlier and higher will the epidemic peak be (Fig 5B). If 75% of the 268 population gets vaccinated, an epidemic peak emerges in spring 2022 (t = 750) -notice 269 this prediction assumes that no contact reducing interventions are in place after summer 270 2021 (t = 450). This peak will exceed the one in early 2021. If 80% of the population is 271 vaccinated, a pandemic peak that is in between the first and second wave of 2020 272 emerges. This peak will be avoided if 85% of the population gets vaccinated.  Note that vaccination coverage is not the same as the herd immunity threshold. The 277 latter is the percentage of the population that needs to be immune for the disease to

Vaccination schedule and immunogenicity 291
Another important factor in VCs is the time until immunization is reached, as 292 determined by the vaccination schedule and immunogenicity. There is a visible effect if 293 the time to immunization is increased from 28 days to 42 days. However, the effect is 294 not as strong as that of the rate of infection or the proportion being vaccinated. In 295 relative terms, the short-term effect is more pronounced (cf. Figs 7A with B and C with 296 D). The reason is that the number of infections is rising at the onset of the vaccination 297 campaign. During this period early immunization reduces the spread of COVID-19.

298
Once incidence is low, the time to immunization is not as important until the final 299 epidemic peak emerges. This will emerge earlier and will be higher if the time to

Effectiveness of the vaccine 305
The better the immunizing effect of the vaccine, the more effective is the VC. We 306 compared four scenarios summarized in Table 1, one being a worst-case scenario which 307 is unrealistic and just used as a comparison. When effectiveness increases from 78%  Table 1. No ADE-induced increased mortality is assumed.   The lower the vaccination coverage, the higher the increase in mortality. This is not  the model itself can be adapted to other countries to obtain quantitative results, their 347 appropriateness needs to be taken with caution. Namely, the model neglects an explicit 348 age-structure. Therefore, it is applicable to industrial nations with demographics similar 349 to Germany. Adaptations will be necessary for low and middle-income countries with a 350 large young population. These adaptations can be done similarly as in CovidSim 2.0 351 http://covidsim.eu/. To illustrate that the model is applicable to other countries (with 352 similar demographic structure), we parameterized the models also to reflect the 353 situation in the USA (see S2 Results for the USA).

354
The impact of the vaccination in terms of incidence and mortality depends on the 355 contact-reducing interventions in place. Here, it was assumed that a "hard lockdown" 356 will be sustained until the end of March 2020 followed by a "relief period" still with 357 relatively strong contact reduction, until summer 2021. Afterwards no contact reduction 358 was assumed. These assumptions are obviously questionable. The rationale behind 359 them was that case numbers in 2021 will first require action to reduce disease incidence 360 by sustaining contact reduction. Once people get vaccinated and incidence is decreasing, 361 compliance with distancing measures will fade after the summer season. As soon as 362 attendance of cultural events (e.g., concerts, museums, sports events, etc.) and 363 unrestricted air travel will be possible and mandatory wearing of facial masks will be 364 lifted for vaccinated individuals, it will be difficult to control distancing interventions in 365 the population. Notably, we assume that case isolation (quarantine and home isolation 366 of confirmed cases) is further sustained. time-dependent fashion. This is a straightforward generalization. Here, we decided to 377 simulate a plausible range of fixed parameters to quantify their effects, rather than 378 making assumptions on time-dependence that would be purely speculative. According 379 to our predictions, vaccination campaigns will have a strong impact on the reduction in 380 disease incidence. In the short term, a swift onset of the vaccination campaign 381 contributes to a substantial reduction in incidence and mortality in the first quarter of 382 2021. The later mass vaccination starts, the smaller the reduction in incidence or 383 mortality. The onset of the campaign mainly depends on the logistics to initially 384 distribute the vaccine efficiently.

385
Importantly we assumed the optimistic case that the vaccine protects from infection 386 and transmission. This however, is not clear yet. Vaccines might just protect from 387 severe disease. In our model this situation can be accommodated, by assuming that the 388 vaccine leads with a high probability only to partial immunity, which results in 389 symptomatic infections with significantly reduced probability. However, data from Israel 390 suggests that the BioNTech vaccine protects from transmission [21]. Moreover, we did 391 not assume the British or South African mutation. Not all vaccines might protect from 392 these variants [22][23][24]. Also these situations can be accommodated by the model. These 393 mutations are characterized by a higher base reproductive number. Thus they will 394 spread, which can be captured in our model, by increasing the yearly average base 395 reproductive number in a sigmoidal fashion and adjust the parameters reflecting the 396 effects of the vaccine in a similar fashion.

397
The vaccination schedule also has a substantial impact on incidence and mortality. 398 Our simulations showed a substantial improvement if the time to immunization is 399 shortened from 42 to 28 days. This depends on the vaccination schedule, which requires 400 typically two doses with a waiting time of two to three weeks between them. After that, 401 the immunizing effect is reached within about 14 days. Efficient logistic planning and 402 properly scheduling appointments to receive the two required doses can help to 403 minimize the time to immunization.

404
Not surprisingly, the higher the vaccination rate (i.e., the faster the population is season in order to prevent another epidemic outbreak. Such an outbreak might be more 423 difficult to control as interventions will be a balancing act between restrictions tolerable 424 by the vaccinated part and necessary to protect the unvaccinated part of the population. 425 Notably, even if the herd immunity threshold to prevent a COVID-19 outbreak is 426 substantially lower than 75%, this threshold must be reached on time. Importantly, the 427 March 5, 2021 11/16 vaccination coverage in our predictions also reflects the propensity to get vaccinated 428 early. Furthermore, vaccination does not lead to immunization in all cases, a fact that 429 was addressed in our model. In particular, we studied the consequences of varying 430 vaccine efficiencies, which have a substantial effect. According to clinical trials, efficacy 431 varies between 78% − 95% among the most promising vaccines [26][27][28][29]. With lower 432 efficiency vaccination coverage must increase to reach herd immunity, i.e, to reach 433 immunization in 60% if the population 63% needs to be vaccinated if efficiency is 95%, 434 while 77% need to be vaccinated if efficiency is just 78%.

435
Our predictions are conservative in as far as negative side effects of the vaccine were 436 considered. In particular, we incorporated ADE or more generally ERD. These are well 437 known in corona viruses and it has been explicitly warned about ADE in the context of 438 vaccination campaigns [17]. The effects of ADE are notoriously difficult to predict [30]. 439 Here, we assumed that a fraction of vaccinated individuals develops ADE, which results 440 in a higher likelihood to develop symptomatic infections and higher mortality. Although 441 we assumed mortality to be substantially increased (20% rather than 7% mortality in 442 symptomatic infections), the overall effect was minor. More precisely, the reduction in 443 mortality due to immunization achieved through vaccination always outweighs increased 444 mortality due to ADE. This is an encouraging result that justifies neglecting ADE in 445 future models. In fact, our model can be substantially simplified if ADE is neglected.

446
While our predictions adequately reflect disease incidence, the predictions for 447 mortality have to be interpreted with caution. Namely, vaccination campaigns will 448 target risk groups suffering from elevated mortality first. In fact, 50% of 449 COVID-19-related deaths occur in long-term care facilities (LTCFs), although less than 450 1% of the German populations live inside such a facility. Under thorough contact 451 reducing measures, the spread of COVID-19 inside LTCFs can be efficiently maintained 452 by regularly testing employees [31]. Concerning the interpretations of our results, 453 mortality has to be understood qualitatively rather than quantitatively. However, 454 adequate quantitative predictions can be easily deduced from our simulations by 455 multiplying mortality with an adjustment factor. The relative effect of model 456 parameters remains unaffected by an increase or decrease in mortality.

457
Our results for the USA (see S2 Results for the USA) are similar than for Germany, 458 although we predict that a lower vaccination coverage is sufficient to avoid a further 459 epidemic peak. Notably, these results also do not assume the more infectious British 460 mutation. However, they serve as a benchmark for comparison.

461
In conclusion, vaccination campaigns should be launched as early as possible.

462
Logistics should be well planned to utilize the maximum capacity of the vaccination 463 infrastructure. Failure to immunize a sufficient part of the population by the beginning 464 of the flu season in 2021 will result in a high endemic peak, by far exceeding current 465 levels of incidence. Adverse effects of the vaccine such as ADE are by far outweighed by 466 the benefits of the vaccine. In fact, the higher vaccination coverage, the lower the risks 467 associated with ADE. We predict that vaccination coverage of 80% would result in 468 sufficiently high levels of herd immunity to allow a return to normality by summer 2021. 469 Nevertheless, it is important to sustain the vaccination campaign until the herd 470 immunity threshold is actually reached. This will require sustained incentives to get 471 vaccinated after disease incidence drops, e.g., through general contact reductions 472 measures being tight to vaccination coverage.