key: cord-0773538-87fcle23
authors: Løchen, Alessandra; Anderson, Roy M.
title: Dynamic transmission models and economic evaluations of pneumococcal conjugate vaccines: a quality appraisal and limitations
date: 2021-07-09
journal: Clin Microbiol Infect
DOI: 10.1016/j.cmi.2021.07.002
sha: 3b91932f98ad888ac9402eaaca3f80fe53b7987f
doc_id: 773538
cord_uid: 87fcle23

BACKGROUND: Of over 90 serotypes of Streptococcus pneumoniae, only 7 were included in the first pneumococcal conjugate vaccine (PCV). While PCV reduced the disease incidence, in part because of a herd immunity effect, a replacement effect was observed whereby disease was increasingly caused by serotypes not included in the vaccine. Dynamic transmission models can account for these effects to describe post-vaccination scenarios, whereas economic evaluations can enable decision-makers to compare vaccines of increasing valency for implementation. OBJECTIVE: The aim of this review was to examine epidemiological and economic models and their assumptions for their potential contributions to future research and immunisation policy. SOURCES: Pubmed, Scopus, Ovid, ISI Web of Knowledge, Centre of Reviews and Dissemination (CRD) databases were searched. CONTENT: Twenty-three dynamic transmission models and twenty-one economic models were retrieved and reviewed. Published models employed various templates, revealing several key uncertainties regarding the biology and epidemiology of pneumococcal infection. While models suggested that PCVs will reduce the burden of disease, the extent to which they are predicted to do so depended on various assumptions regarding features of pneumococcal infection and epidemiology that governed PCV cost-effectiveness as well. Such features include the duration of protection and competitive interactions between serotypes, which are unclear at present, but which directly relate to herd immunity and serotype replacement. IMPLICATIONS: Economic evaluations are not typically based on transmission dynamic models and hence omit indirect herd immunity effects. The two tools could be used in conjunction to inform decision-makers on vaccine implementation, but so far there have been few attempts to build economic evaluations on transmission dynamic models, and none in this field. Future directions for research could include studies to evaluate key parameters for the models involving herd immunity, serotype competition and the natural history of infection.

PCVs will reduce the burden of disease, the extent to which they are predicted to do so depended on 23 various assumptions regarding features of pneumococcal infection and epidemiology that governed 24 PCV cost-effectiveness as well. Such features include the duration of protection and competitive 25 interactions between serotypes, which are unclear at present, but which directly relate to herd 26 immunity and serotype replacement 27 Implications Economic evaluations are not typically based on transmission dynamic models and 28 hence omit indirect herd immunity effects. The two tools could be used in conjunction to inform 29 decision-makers on vaccine implementation, but so far there have been few attempts to build 30 economic evaluations on transmission dynamic models, and none in this field. Future directions for 31 Introduction one that has been used frequently is cost-effectiveness analyses (CEAs) which express outcomes in 48 natural units, such as the cost per infection averted. A type of CEA known as cost-utility analyses 49 (CUAs) quantifies the effectiveness of different interventions using utility metrics, which relate to a 50 person's level of wellbeing/health, and include quality adjusted life years (QALYs), disability adjusted 51 life years (DALYs) or life years gained (LYG). CUAs produce an incremental cost-effectiveness ratio 52 (ICER) for two comparable interventions, which is the difference between their costs divided by the 53 difference between their utility values. Health economic models are generally split into cohort or 54 individual-level models, with the former evaluating the costs and utilities of proportions of a 55 population following a health event, and the latter considering individuals with certain characteristics 56 and being more computationally intensive. The main types of cohort models are decision tree models 57 and Markov models. Decision tree models occur instantaneously with different branches for each 58 disease outcome, whereas Markov models have cycles over which the probability of a patient 59 J o u r n a l P r e -p r o o f models that captured herd immunity, and as such all models were static. Because of their relatively 167 recent licensure, much of the data on PCV10 and PCV13 used by the models was extrapolated from 168 PCV7 data. PCV10 and PCV13 clinical trial data have demonstrated immunogenicity, safety and 169 tolerability rather than clinical effectiveness. Currently, the latter is inferred from earlier vaccine 170

formulations. This demonstrates a limitation of the studies. Despite this, most papers compared 171 PCV10 and PCV13. 172

Country guidelines for health economic perspectives [50] were not necessarily followed [51, 52] , 173 particularly by studies comparing the same perspective for different countries [53] [54] [55] . However, the 174 studies that included multiple health economic perspectives [54, 56-58] generally had consistent 175 results on which vaccine was more cost-effective. 176

Interestingly, the included EEs are nearly all industry studies, which at times fundamentally biased the 177 parameters used (Table 4 , Suppl. Table 2) , and thus the interpretation of the results. A majority 178 reported conflict of interest with either GSK or Pfizer, the two competing PCV manufacturers who 179 funded or provided grants for these studies or employed one or more of the authors. The vaccine that 180 was manufactured by the funding/supporting company was always preferred or considered more cost-181 effective. Of the four papers that did not report a conflict of interest, two reported that PCV13 was 182 more cost-effective [56, 59] , and one reported that PCV10 was considered more cost-effective if the 183 impact on non-typeable Haemophilus influenzae (NTHi) was included [60] . The remaining reported 184 that neither were cost-effective, although it is worth noting that this study did not include indirect 185 effects in their analysis [57] . Additional published work from national public health agencies, or 186 detailed methodology including the completion of a reporting guidance checklist, such as the 187 Consolidated Health Economic Evaluation Reporting Standards (CHEERS) checklist, could help 188 eliminate the bias and make the results more transparent and robust. 189

One of the challenges in comparing these studies is that they all described different settings, with 190 different currencies and price years (Supplementary Table 4 

This review gathered PCV DTMs and EEs and highlighted some of the gaps in our knowledge in the 233 epidemiological aspects of vaccination. DTMs generally found PCVs to be effective in reducing the 234 burden of IPD and EEs found PCVs to be cost-effective, although both sets of papers made different 235 assumptions which suggests biological uncertainties. In both cases, the results relied primarily on herd 236 immunity and serotype replacement which are difficult to parameterize given their dynamic nature 237 with time and heterogeneous populations. 238

Indirect effects are crucial to vaccine evaluation and cannot be captured as accurately by static 239 models. DTMs show that PCV7 is an effective way to reduce the disease burden although they still 240 cannot quantify serotype replacement yet, particularly as a majority still group serotypes into VT and 241 NVT supergroups. One way to circumvent this and be computationally feasible while still capturing 242 the indirect effects might be to group serotypes according to serotype characteristics, such as 243 acquisition and clearance rates, as one of the models did [25] . How and whether one chooses to group 244 serotypes will depend on the goal of the study. A model investigating the mechanism of competition 245 might only need two serotypes, whereas a model investigating the effects of vaccination or multi-246 serotype coexistence and the mechanism of serotype replacement may require more. Additional 247 biological and epidemiological data, such as on immunity (natural, vaccine-induced, and cross-248 serotype) and duration of protection would allow for parameter estimation on herd immunity and 249 serotype replacement, especially for PCV10 and PCV13. Additionally, interaction between 250 individuals (age specific rates of infection and mixing patterns between age groups), something that is 251 essential when modelling infectious disease transmission and vaccine impact, is excluded, as is an 252 interaction between serotypes (competition). Several mechanisms of competition have been described 253

[90-98], but the precise mechanisms and intensity of inhibition by direct competition remain to be 254 explored. One postulated mechanism, for example, states that serotypes requiring less in a nutrient-255 deficient environment (i.e. are less metabolically demanding) are those that are more prevalent and 256 therefore those considered more "fit" [40, 99, 100]. Competition parameters cannot be directly 257 measured but are currently extrapolated from surveillance data. Large sample sizes are required in 258 surveillance data that models rely on to implement an age-dependent multiple carriage prevalence, 259

which is rarely the case in practice [42] . This would allow DTMs to better estimate the disease burden 260 after vaccination and to predict which serotypes are more crucial to include in the vaccine. This data 261 would also add robustness in the EEs such that bias from conflicts of interest would not prevail. Static 262 EE models simplify heterogeneity and do not allow an estimation of the changes in herd immunity or 263 serotype replacement over time, particularly as vaccine coverage increases. Static EEs may be 264 warranted if the time horizon is shorter, however serotype replacement has a long-term impact. They DTMs, even if much of the data required to precisely parameterize these effects is scarce. 272

At the core of the pneumococcal transmission dynamics is multiple serotype carriage. Currently only 273 bi-carriage has been modelled. The consequence of a higher multiple serotype carriage on 274 transmission dynamics is not yet understood and adds a degree of complexity to modelling that is 275 potentially unnecessary depending on the model's objective. Regardless, used in conjunction with an 276 invasiveness estimate (case-to-carrier ratio) to extract the number of disease cases. Should the data 277 become more abundant, one could also extract the number of specific disease outcomes (meningitis, 278 sepsis, etc) per serotype, which could link the DTMs to the EEs. 279

The reviewed transmission models never accounted for NIPD, whereas it was often included in EEs 280 and assumed to have a similar serotype profile to IPD. This may be because the PCV impact on NIPD 281 is not measured or required for the licensure [36], and because serotype-specific data on AOM and 282 pneumonia is sparse and therefore it is difficult to predict long-term effects of vaccination or indirect 283 effects. Nevertheless, AOM accounts for most pneumococcal disease costs because of its frequency, 284

and as a key driver in the published sensitivity analyses, its parameters merit exploration and 285

The population-level differences between countries signify that basing data from one country makes 287 predictions specific to this region. It is unknown at present whether these predictions apply to other 288 populations [49] . Higher carriage and IPD prevalence in, for example, developing countries and 289 native population settings may be due to less effective immune responses arising from malnutrition, 290 genetic differences, or other epidemiological risk factors such as HIV infection [4, 5] as well as from 291

increased number of contacts [30] . These factors affect the case-to-carrier ratios, which are age-and 292 serotype-dependent, and could be used to directly measure IPD prevalence. Furthermore, increasing 293 the transmission intensity might increase the occurrence of multiple carriage in a host population, 294 which would increase the indirect effects of competition [13] . In the EEs, most models assumed price 295 parity between vaccines even though this is highly unlikely in practice. Cost (vaccine dose price, 296 medical costs), a main driver in EEs, and different tendering outcomes, would change the conclusions 297 derived from economic models specific to one country. The extent of the benefits of PCVs on 298 populations of different sizes, age structures, or where differing pricing policies prevail will depend 299 on their respective epidemiological and economic characteristics. As such, predictions may not 300 currently be easily comparable. 301

While much past research has focused on DTMs and EEs separately, there has been no focus on 302 developing EEs based on DTMs and no DTMs including EEs. DTMs provide a template for economic 303 analyses of the relative benefit of adding more serotypes to the conjugate vaccine. This cannot happen 304 if they only model two supergroups. Assessing vaccine impact in terms of both carriage and disease 305 provides a greater understanding of the transmission dynamics that can ultimately improve our ability 306 to make informed policy decisions [36] . However, doing this effectively is challenging at present 307 given our poor understanding of the relationship between pneumococcal carriage and disease, 308

including multi-serotype coexistence, and PCV's role in it. Until we have a better understanding of 309 these factors, EEs will not be able to reflect the true value of implementing one PCV over another. 310

Recommendations for performing future models would however depend on the data available and the 311 research question of the study. for the determination of which serotypes to include in future vaccines, as well as which PCV to 319 implement and quantifying its public health impact in both economic and epidemiological terms. 320

Dynamic transmission models and economic evaluations rely on serotype replacement and herd 322 immunity effects to determine the epidemiological and economic impact of PCV. These indirect 323 effects depend on several factors: the interaction between VT and NVT and how it translates to a 324 competition parameter, antibiotic use, vaccine coverage, and the vaccine efficacy against each VT, 325 among many others. Considering all these factors and the variety of assumptions on which the models 326 rely, standardization of their frameworks must wait until more biological and epidemiological data 327 becomes available. At present, much is unclear about the duration of protection to any given serotype 328 and the competitive interactions between serotypes, which are key to estimating the overall indirect 329 effects of vaccination. More focus must be placed on acquiring a better understanding of the natural 330 history on pneumococcal infection before predictions based on mathematical models can be regarded 331 as reliable. Until this information becomes available, static economic evaluations will be unreliable as 332 they depend on the extent of these dynamic indirect effects. Clearly, indirect effects are central to 333

determining the impact of PCV, but a limited understanding of the mechanisms behind these prevents 334 us from unlocking the value of these tools. 335 

A mathematical model of serotype replacement in pneumococcal carriage following vaccination

7-Valent pneumococcal conjugate vaccination in England and Wales: is it still beneficial despite high levels of serotype replacement? PLoS One

Optimising Assessments of the Epidemiological Impact in the Netherlands of Paediatric Immunisation with 13-Valent Pneumococcal Conjugate Vaccine Using Dynamic Transmission Modelling

Assessing the efficiency of catch-up campaigns for the introduction of pneumococcal conjugate vaccine: a modelling study based on data from PCV10 introduction in Kilifi

Geographic variation in pneumococcal vaccine efficacy estimated from dynamic modeling of epidemiological data post-PCV7. Sci Rep

Expanding vaccine efficacy estimation with dynamic models fitted to cross-sectional prevalence data post-licensure

How direct competition shapes coexistence and vaccine effects in multi-strain pathogen systems

Vaccination against colonizing bacteria with multiple serotypes

Bacterial vaccines and serotype replacement: lessons from Haemophilus influenzae and prospects for Streptococcus pneumoniae. Emerg Infect Dis

Dynamic models of pneumococcal carriage and the impact of the Heptavalent Pneumococcal Conjugate Vaccine on invasive pneumococcal disease

Sustained reduction in vaccine-type invasive pneumococcal disease despite waning effects of a catch-up campaign in Kilifi, Kenya: A mathematical model based on prevaccination data

How Is Vaccine Effectiveness Scaled by the Transmission Dynamics of Interacting Pathogen Strains with Cross-Protective Immunity? Plos One

Timing of bacterial carriage sampling in vaccine trials: A modelling study

Transmission-dynamic model to capture the indirect effects of infant vaccination with Prevnar (7-valent pneumococcal conjugate vaccine (PCV7)) in older populations

Estimation of invasive pneumococcal disease dynamics parameters and the impact of conjugate vaccination in Australia

Short-and long-term effects of pneumococcal conjugate vaccination of children on penicillin resistance

A dynamic model of pneumococcal infection in the United States: Implications for prevention through vaccination. Vaccine

The influence of competition and vaccination on the coexistence of two pneumococcal serotypes

Carriage burden, multiple colonization and antibiotic pressure promote emergence of resistant vaccine escape pneumococci

The impact of specific and non-specific immunity on the ecology of Streptococcus pneumoniae and the implications for vaccination

Pneumococcal transmission and disease in silico: a microsimulation model of the indirect effects of vaccination

Niche and neutral effects of acquired immunity permit coexistence of pneumococcal serotypes

Mathematical Modelling Long-Term Effects of Replacing Prevnar7 with Prevnar13 on Invasive Pneumococcal Diseases in England and Wales

Cost-effectiveness of pneumococcal conjugate vaccination in Croatia. Vaccine

A few years later: update of the cost-effectiveness of infant pneumococcal vaccination in Dutch children

Cost-effectiveness of a 10-versus 13-valent pneumococcal conjugate vaccine in Denmark and Sweden

Cost-effectiveness of 13-valent pneumococcal conjugate vaccine

Cost-effectiveness of 2+1 dosing of 13-valent and 10-valent pneumococcal conjugate vaccines in Canada

Cost-effectiveness of 13-valent pneumococcal conjugate vaccine in Switzerland

Cost-effectiveness of new pneumococcal conjugate vaccines in Turkey: a decision analytical model. BMC health services research

Cost effectiveness of child pneumococcal conjugate vaccination in middle-income countries. International Health (RSTMH)

Outcomes and costs associated with PHiD-CV, a new protein D conjugate pneumococcal vaccine

Pharmacoeconomic evaluation of 10-and 13-valent pneumococcal conjugate vaccines. Vaccine

Simulation model for comparing the costs and effectiveness of different pneumococcal conjugate vaccines

Cost effectiveness of pneumococcal vaccination among Dutch infants: An economic analysis of the seven valent pneumococcal conjugated vaccine and forecast for the 10 valent and 13 valent vaccines

Overall effectiveness of pneumococcal conjugate vaccines: an economic analysis of PHiD-CV and PCV-13 in the immunization of infants in Italy

Cost-effectiveness analysis of routine pneumococcal vaccination in the UK: a comparison of the PHiD-CV vaccine and the PCV-13 vaccine using a Markov model

Health and economic impact of PHiD-CV in Canada and the UK: A Markov modelling exercise

Comparing health outcomes and costs of general vaccination with pneumococcal conjugate vaccines in Sweden: a Markov model

Economic evaluation of second generation pneumococcal conjugate vaccines in Norway. Vaccine

Cost-effectiveness of 10-and 13-valent pneumococcal conjugate vaccines in childhood KCE Reports

Modeling the cost-effectiveness of infant vaccination with pneumococcal conjugate vaccines in Germany

Public health and economic impact of the 13-valent pneumococcal conjugate vaccine (PCV13) in the United States. Vaccine

Value Added by the Prevnar 13 Childhood Immunization Program in Alberta

Pneumococcal capsular polysaccharides conjugated to protein D for prevention of acute otitis media caused by both Streptococcus pneumoniae and non-typable Haemophilus influenzae: a randomised double-blind efficacy study

Efficacy of Pneumococcal Nontypable Haemophilus influenzae Protein D Conjugate Vaccine (PHiD-CV) in Young Latin American Children: A Double-Blind Randomized Controlled Trial

J o u r n a l P r e -p r o o f Public sector only 8 [3, 5, 7, 10, [12] [13] [14] 18 ] Private only 5 [4, 9, 21] Societal (public and private) 9 [1, 2, 6, 8, 11, 15-17, 19, 20] 

Cost-effectiveness 20 [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] Cost-benefit 1 [21] 

Herd immunity 19 [1] [2] [3] [4] [5] [6] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] ] Serotype replacement 16 [1-4, 8, 9, 11-20] PCV10 cross-protection of NVT 11 [2, 7-9, 11, 13-18] Protection against NTHi 14 [2-5, 7, 9-11, 13-18] Includes transmission dynamics 0 Fitted to data 0 J o u r n a l P r e -p r o o f