Hepatitis B vaccination as an elimination tool assessed in a paediatric cohort and simulated in a model

58 Sustainable Development Goals set a challenge for the elimination of hepatitis B virus (HBV) 59 as a public health concern by 2030. We evaluate the current and future role of HBV vaccination 60 and prevention of mother to child transmission (PMTCT) as tools for elimination, through the 61 combined scrutiny of a paediatric cohort in South Africa and a model to simulate transmission 62 and prevention. Existing efforts have been successful in reducing prevalence of infection 63 (HBsAg) in children to <1%. Our model anticipates that current combination efforts of 64 vaccination and PMTCT can significantly reduce population prevalence (HBsAg) by 2030, but 65 will reduce the prevalence of HBV e-antigen positive carriers more slowly, with potential 66 implications for public health control. With strategies and resources already available, 67 significant, positive public health impact is possible, although time to HBV elimination as a 68 public health concern is likely to be longer than that proposed by current goals. 69 70


Serological evidence of exposure to HBV infection 146
From our cohort of 402 children in Kimberley, South Africa, three were HBsAg-positive (0.7%; 147 Table 1). This HBsAg prevalence is significantly lower than in adults in a comparable study 148 population (e.g. 11.1% in a previous study (9); p<0.0001). Exposure to HBV infection was 149 measured using anti-HBc antibody; this was detected in three children (0.7%), one of whom 150 was also HBsAg-positive. The other two were HBsAg negative, indicating previous HBV 151 exposure and clearance. 152 153  HIV-positive children with anti-HBs titres ≥100mIU/ml were significantly younger than those 184 with lower antibody titres (median age 17 months vs. 31 months, p=0.0008), while no such 185 difference was observed within the HIV-negative group (Fig 2A). Using the lower threshold of 186 ≥10mIU/ml, we found no significant difference by age in either the HIV-positive or the HIV-187 negative groups (p=0.17 and 4.48 respectively, data not shown). To expand our view of the 188 HIV-positive group, we also added analysis of an older cohort (92 children aged >60 months), 189 and demonstrated that anti-HBs titres were significantly lower in this older group (p<0.0001), 190 with only 2/92 subjects (2.2%) achieving an anti-HBs titre of ≥10mIU/ml (Fig 2B). Anti-HBs 191 titres waned significantly with age up to age 60 months in HIV-positive children (Fig 2C;  192 p=0.004). We observed a similar trend in the HIV-negative cohort, but this did not reach 193 statistical significance (Fig 2C; p=0.07). The proportion of HIV-positive subjects with a 194 detectable anti-HBs titre declined steadily with age in the cohort, contrasting to the trend in 195 HIV-negative subjects, where individuals maintained protective anti-HBs titres despite a trend 196 towards decreasing mean titres (Fig 2C). Although the numbers of children in this cohort are 197 small, and we did not collect longitudinal data, these results support previous literature reports 198 that HBV vaccine-mediated immunity wanes over time independently of HIV serostatus, but 199 faster for HIV positive individuals (22,23). For HIV-positive children aged ≤60 months, ART treatment data were available for 79% of 204 subjects. Within this group, 71% were receiving ART at the time we tested for anti-HBs, and 205 had received a median of 20 months of treatment (IQR 6-33 months). Comparing anti-HBs 206 titres between ART-treated vs. untreated children, we found no significant difference (p=0.72; 207 76 ART-treated, median anti-HBs 13.3 mIU/ml and 31 untreated children, median anti-HBs 208 14.1 mIU/ml, data not shown). There was also no difference between anti-HBs titres of children 209 treated for ≤12 months vs. >12 months (p=0.50, data not shown). We did not examine the 210 effect of ART on anti-HBs titres in children >60 months due to the low numbers of subjects 211 with a detectable anti-HBs titre (n=2). 212 213

Fitting of a dynamic model to local HBV epidemiology 223
We set out to use our clinical data to inform the development of a dynamic model to provide 224 insights into the long-term outcomes of sustained immunization, and to suggest how 225 prevention strategies can be optimized, for example by enhancement of PMTCT or extended 226 vaccination campaigns targeting older age groups. 227

228
In summary, the model takes into consideration the susceptible proportion of the population 229 (S), the chronic (C) and acute (I) carriers, the immune (R) and the vaccinated (V) (Fig 3A). To 230 be able to parameterize HBV or vaccine-related epidemiological traits in age, such as age-231 specific probability of chronicity or decay of vaccine-induced protection, susceptible (S) and 232 vaccinated (V) individuals are divided into three subgroups representing infants (i, <1 years of 233 age), children (c, 1-6 years of age) and older individuals (comprising older children, 234 adolescents and adults, a, >6 years of age). Chronic carriers, C, are divided into HBeAg-positive (C+) and HBeAg-negative (C-) to further allow for different parameterization between 236 these two biologically distinct states. 237 238 Informed by the cohort data described above, natural decay and the effects of HIV sero-status 239 on vaccine-induced protection are also taken into account. We used a Bayesian Markov-chain 240 Monte Carlo (bMCMC) approach to fit the dynamic model to the local demographic and 241 epidemiological setting of Kimberley before projecting the impact of interventions. We used 242 informative priors for model parameters for which robust literature support exists, and 243 uninformative (uniform) priors otherwise. For full details on the model and fitting approach, see 244 the Methods section. 245 246 The dynamic model was able to closely reproduce the target (fitted) variables -HBV 247 prevalence (HBsAg), prevalence of HBV exposure (anti-HBc) (Fig 3 B1), and relative 248 proportion of HBeAg-negative and HBeAg-positive among chronic carriers ( Fig 3B2). For 249 parameters for which little or no support was found in the literature (Fig 3C), the resulting 250 posteriors were well behaved. For parameters using informative priors taken from the literature 251 (Fig 3 D, E), the resulting posteriors matched well. Overall, the obtained bell-shaped posteriors 252 highlighted no identifiability issues with the fitting approach (Fig 3 C, D, E). 253

254
The posterior for the rate of seroconversion from HBeAg-positive to HBeAg-negative (θ) 255 suggested slow progression, with a median period of ~18.5 years (95% CI [14.3, 21.9]). We 256 note here that although we used an uninformative (uniform) prior for θ, its posterior with 257 median ~5.3% a year, here not accounting directly to age-specificity, is compatible with 258 empirical estimations (24) of yearly rates of less than 2% for <3 years of age and 4-5% for As expected, both HBsAg incidence (Fig 4 A1) and HBeAg-positive prevalence (Fig 4 B1) 278 reduce faster with increasing neonatal immunization coverage, resulting in shorter times to 279 reach the elimination targets (Fig 4 A2, B2). Importantly, even immunization of 100% of 280 neonates is predicted to take ~99 years (95% CI 61 -186) for the HBsAg incidence target to 281 be achieved (Fig 4 A2), and ~175 years (95% CI 103 -278) for the HBeAg-positive prevalence 282 target (Fig 4 B2). Such long timeframes are supported by a previous modelling study (21). The model suggests that reaching either of the elimination targets will require different 297 intervention coverage and different time scales. In particular, the target for reducing HBsAg 298 incidence is easier to achieve than reducing HBeAg-prevalence. This implies that for a certain 299 vaccination coverage or PMTCT effort, reductions in HBsAg incidence must be interpreted 300 with caution, as such positive trends will potentially mask the fact that HBeAg-positive 301 prevalence, critical for public health, will not be responding at the same rate. Based on the premise that interventions in the South African population have been most 306 consistently deployed since roll-out of the HBV vaccine in infancy since 1995 (6), we used our 307 model to determine the impact of combined interventions by the year 2030 (Fig 5 A1, B1), and 308 to predict the year at which the 90% reduction in HBsAg incidence and 0.1% HBeAg-positive prevalence targets would be reached (Fig 5 A2, B2). Strikingly, HBsAg incidence could already 310 have been reduced by >90% (Fig 5 A1) if both neonatal vaccination and PMTCT had been 311 deployed at 100% coverage since they became widely available in 1995 (mean predicted year 312 of elimination 2017; Fig 5 A2). In reality, complete coverage of such interventions is not 313 possible, and we therefore projected outcomes based on <100% intervention coverage. For 314 example, combining neonatal vaccination and PMTCT with 90% coverage of each since 1995 315 is projected to achieve the HBsAg incidence target by 2028; if this is reduced to 80% coverage 316 then goals will be attained by 2044. To achieve the target reduction in HBeAg Adding catch-up vaccination campaigns makes no impact on the probability of reaching either 340 of the elimination targets (Fig 6 A1, A2, blue and cyan lines). Routine vaccination at 6 years 341 of age as an alternative for PMTCT, even when delivered at 100% coverage, is markedly less 342 effective than any other projected intervention (Fig 6 A1, A2, magenta line). 343

344
Projecting the impact of HIV on the probability of achieving elimination targets As our clinical cohort is centred in South Africa, at the epicentre of the HIV pandemic, we also 346 used our model to investigate the impact of co-endemic HIV on the success of interventions 347 for HBV. We considered a baseline scenario defined by the epidemiological setting fitted by 348 our model in the context of Kimberley, in which local HIV prevalence was taken into 349 consideration for each of the modelled age groups (Fig 6 B1, B2, solid line). We then 350 performed a sensitivity exercise, considering alternative scenarios in which HIV prevalence 351 was altered to zero or higher prevalence, projecting HBV interventions into the future. to the time taken to achieve a 50% chance of reaching the goals (Fig 6 B1). We also simulated 357 the effect of higher population HIV prevalence (x2, x3 and x4 baseline data for Kimberley) to 358 investigate the potential impact of coinfection in high-risk populations. Increasing HIV 359 prevalence, as expected, has a negative impact on the success of combined interventions for 360 HBV, but the effects are relatively modest. In particular, doubling HIV prevalence would shift 361 the 50% probability endpoint into the future by ~4 years for the HBsAg incidence target, and 362 ~7 years for the HBeAg prevalence target. With increasing HIV prevalence, the negative 363 impact on HBV interventions increases, particularly with respect to reduction in HBeAg 364 prevalence. Encouragingly, as ART is now offered at the point of HIV diagnosis and uptake 365 is consistently increasing, any detrimental impact of HIV coinfection is likely to diminish over 366 time, with more of the HIV-infected population retaining near intact immunity. Our evaluation provides the advantages of both clinical data and a mathematical 406 model, with close links between our cohort and simulations, and strengths in 407 interpretation of data derived through different approaches. In so doing, we have also 408 been able to specifically address the impact of co-endemic HIV that has not been 409 factored into previous evaluations, using unique cohort data to implement a data-410 driven approach into the dynamic model. 411 ii.
In contrast to approximating model behaviour to a wide range of epidemiological 412 settings across many geographical regions, we focus on a particular population for 413 which we derive unknown epidemiological parameters and apply a robust data-driven 414 approach to others. Our Bayesian framework therefore stands alone (as a tool) that 415 can be applied to any population for which empirical support of key HBV 416 epidemiological parameters is missing. By supplying the model's code, we can 417 facilitate the use of the tool by other academics. 418 iii.
As outputs, we have used targets for reductions in both HBsAg incidence and HBeAg- Our results also underscore that a major public health impact is possible even without 485 achieving elimination. Careful adjusting of expectations and aims, according to the scale on 486 which particular changes occur, may inform the setting of realistic targets (e.g. reduction in 487 the prevalence of HBeAg-positive carriers could be the most informative outcome measure). 488 The wrong choice of either target or timescale could result in unnecessary abandonment of a 489 strategy that could have a major impact in a few decades. In addition to informing rational use 490 of interventions that have a positive population impact, our study is also important in cautioning 491 against the use of strategies that may have little or no lasting population impact. This is 492 illustrated by our results for catch-up HBV vaccination, which adds little in situations where 493 high coverage of both neonatal immunization and PMTCT can be attained. Considerable 494 political drive, investing in increased surveillance and reducing barriers to treatment access 495 will also be required in order to accurately monitor progress towards the elimination targets 496 (49). 497 498

Impact of HIV and ART on achieving the 2030 sustainable development goals for HBV 499
Our clinical cohort highlights the day-to-day challenges of drug provision and monitoring within 500 this setting: we did not have access to detailed prospective ART treatment data, guidelines 501 have changed numerous times since 2002, and 3TC was intermittently used as a substitute 502 for nevirapine (NVP) due to supply issues. During the period covered by our study, ART was 503 only introduced in children achieving certain immunological criteria (as per old treatment 504 guidelines), while in future, infected children will be started on treatment as soon as diagnosed, 505 which could restore vaccine responses to similar levels as seen in the HIV-negative 506 population; further studies will be required to assess this over time. ART treatment is relevant We used cohort data to parameterize vaccine-induced protection depending on HIV 516 serostatus and time since vaccination. As far as we know, this is the first data-driven approach 517 to project the effects of HIV prevalence on HBV interventions using a dynamic model. Our 518 projections propose that HIV does have a negative effect on HBV interventions, although HIV 519 prevalence only marginally increases time to reach elimination targets, which may not be 520 significant in light of the long overall time-frames that we project even in the absence of HIV. 521 The high HIV prevalences modelled can occur in specific high-risk groups including sex 522 workers and men who have sex with men (50) and it is likely that increased intervention will 523 be required in these groups to minimise HBV transmission. 524 525

Caveats and limitations 526
Different approaches to recruitment of our HIV-positive and HIV-negative cohorts may have 527 introduced unintentional bias. By using respiratory admissions to hospital for the KReC cohort, 528 we were able to identify and recruit a sufficient number of HIV-negative children, but the KReC 529 children may be less healthy than a comparable group of HIV-negative children in the 530 community, and this approach predominantly selected younger children (on average 9.4 531 months younger than the HIV-positive cohort). 532

533
We set out to focus on children aged <60 months in order to collect data from the RTHB. In 534 practice, we did not capture good RTHB data and data collection from the RTHB is itself 535 subject to bias, as families who attend with such records may be those who are most likely to 536 have immunised their children. Numerous complex social factors are also relevant in 537 determining whether children are immunised; babies born to mothers who have HIV and/or 538 HBV are more likely to be in disadvantaged by poverty, and by illness and death in the family, 539 such that they might be less likely to present for (or respond to) vaccination. However, in this 540 setting (and others where antenatal HBV screening is not routinely deployed (12,51,52)), we 541 deem it unlikely that there is a significant difference in vaccination rates between infants born 542

Study cohorts 599
Recruitment was undertaken in Kimberley, South Africa. A previous study of HBV serology in 600 adults in the same setting found HBsAg prevalence of 9.5% (55/579) (7). Children were 601 recruited as part of the Co-infection in South-African Children (COSAC) study as previously described (30,56). The lower age limit of recruitment was 6 months in order to limit the 603 detection of maternal anti-HBs. For the purpose of analysis, we divided these into two groups according by age: 617 i.
Age 6-60 months; n=136. This group was selected to match the age range of 618 the HIV-negative group, and also included five children who were initially 619 screened for the KReC cohort but tested HIV-positive. 620 ii.

Laboratory assessment of HBV status 633
Testing for Hepatitis B serum markers and DNA was performed as previously described; for 634

HIV-positive children this is in keeping with recent implementation of HBV screening in 635
Kimberley (30). Briefly, HBsAg testing was carried out in Kimberley Hospital, South Africa 636 using the Magnetic parcel chemiluminometric immunoassay (MPCI; Advia Centaur platform). 637 Confirmatory HBsAg testing was carried out by the UKAS accredited clinical microbiology 638 laboratory at Oxford University Hospitals (OUH) NHS Foundation Trust, Oxford, UK (Architect 639 i2000). For all samples, anti-HBs and anti-HBc testing were carried out by the OUH laboratory 640 (Architect i2000). Limit of detection of the anti-HBs assay was 10 mIU/ml. 641 642

Threshold for vaccine mediated immunity 643
An absolute threshold for vaccine-mediated immunity is difficult to define, and studies variably 644 quote anti-HBs titres of ≥10 mIU/ml or ≥100 mIU/ml as a correlate of protection. UK 645 recommendations for testing HBV immunity often rely on the more stringent criterion of an 646 anti-HBs titre of ≥100 mIU/ml (14). However, early vaccine studies have highlighted that a titre 647 of ≥10 mIU/ml is likely to be a clinically relevant threshold for protection; a study of children in 648 The Gambia showed that children who attained an anti-HBs titre of ≥10 mIU/ml were most 649 likely to be immune (15), and another study demonstrated increased risk of infection when 650 antibody titres fell <10 mIU/ml (57). Due to the varying use of different thresholds, we have 651 presented our results pertaining to both thresholds of ≥10 mIU/ml and ≥100 mIU/ml. 652 653

Statistical analysis 654
Data from the cohort was analysed using GraphPad Prism v.7.0. We determined significant 655 differences between sub-sets within the cohort using Mann-Whitney U tests for non-656 parametric data, Fisher's exact test for categorical variables and correlation between data 657 points was assessed using Spearman's correlation coefficient. 658 659

Mathematical model of HBV transmission and prevention 660
A mathematical model was developed using ordinary differential equations (ODE) and is 661  We assume that the probability of developing chronic infections decreases with age, with 685 ψ=0.15, ε=0.4, and γ=0.95 (58-60). When developing chronic infection, we assume that all 686 individuals become HBeAg-positive but may lose this status and become HBeAg-negative at 687 a rate θ (61). HBeAg-negative carriers may clear infection spontaneously at a rate ρ, entering 688 the recovered class (R). Acute infections (I) are assumed to last 6 months (62) and are cleared 689 at a rate σ, entering the recovered class (R). 690

Births and Mortality 697
The population is assumed to be of constant size with equal births b (eq. 12) and deaths (µ, 698 µ'). Due to HBV-associated mortality, the lifespan of chronic HBeAg-positive (C + ) individuals 699 is taken to be lower (50 years) than the general lifespan (59 years (11)). In the absence of 700 control, the total number of births (b) is divided into Z (eq. 13), W (eq. 14) and Z' (eq. 17) 701 depending on the probability of vertical transmission (A1, A2) and proportion vaccinated at birth 702

Routine vaccination 714
Routine vaccination is implemented under three general strategies: coverage of neonates (Z', 715 eq. 8, 17), coverage of 1-6 years old by vaccinating individuals leaving the susceptible <1 716 years old class (term cωcSi in eq. 9), and coverage of 6+ years old by vaccinating individuals 717 leaving the susceptible 1-6 years old class (term aωaSc in eq. 10). In essence, we model 718 vaccination occurring either at birth, or at particular ages (1 year, 6 years). 719 720

Catch-up vaccination 721
For simplicity, catch-up is modelled in a single event (time step tcu), by moving a proportion of 722 susceptible individuals into the age-corresponding vaccinated classes. In practice, this is an 723 impulse event in the ODE system. Catch-up proportions are age-specific with parameters Ki 724 for <1 years old, Kc for 1-6 years old, and Ka for 6+ years old. 725 726

Markov-chain Monte-Carlo fitting approach 727
In two independent steps, we fit certain ODE model outputs to empirically observed variables 728 in the South African population, to set demographic and transmission backgrounds before

Fitting transmission background 762
After fitting demographic parameters and before considering interventions, we fitted the model 763 to a transmission background. This is done using the above described fitting approach, with 764 fixed aging rates a and c. The target variables are set to the percentage of the population that 765 is HBsAg-positive (total carriers), percentage that are anti-HBc positive (R), and relative 766 prevalences of chronic carriers HBeAg-positive (C + ) and HBeAg-negative (C -) for the 767 population of study. We used target Gaussian distributions (standard deviation 1) with mean 768 30% for anti-HBc, mean 8.3% for total carriers, mean proportion of 73% for HBeAg-negative 769 and 23% for HBeAg-positive (9,66). In this step, the posteriors of the parameters β, ρ, α1, α2, 770 θ and βm are obtained. 771 772

Fitted parameters and priors for transmission setting 773
We fitted six parameters for the local transmission setting (β, ρ, α1, α2, θ and βm). Gaussian are used with ranges of 0 to 30 for β and 0 to 1 for θ and ρ. In the main results we demonstrate 780 that the posteriors for ρ and θ follow the scarce knowledge of these parameters. 781 Vaccine-induced protection is modelled in the dynamic system using the term (1-∆x) in 855 equations 4 and 7-10, where x relates to a specific age class. The term (1-∆x) therefore models 856 a reduction in risk of infection, with ∆x being the protection offered by the vaccine. Given that 857 vaccine-induced protection is dependent on HIV status, ∆x takes the following forms: 858 859 860 Where Px + is the HIV prevalence at a certain age x, vx + the vaccine-induced protection at a 861 certain age x for HIV-positive individuals, and vxthe vaccine-induced protection at a certain 862 age x for HIV-negative individuals (as determined in the approach detailed above

29.
Chu CM, Liaw YF. HBsAg seroclearance in asymptomatic carriers of high endemic 949 areas: Appreciably high rates during a long-term follow-up. Hepatology.