An economic analysis of poliovirus risk management policy options for 2013–2052

Analytical framework

We develop a model to characterize prospectively the economic outcomes associated with long-term poliovirus risk management policy options. The 40-year analytical time horizon runs from the beginning of 2013 (T₀) through the end of 2052 (T_end). The model considers 200 countries included in both the United Nations World Population Prospects (2012 revision) [33] and the World Bank list of economies (as of 2013) [34]. We categorize countries as low-income (LOW), lower middle-income (LMI), upper middle-income (UMI), and high-income (HIGH) according to the 2013 World Bank levels [34] to approximate the variability in conditions throughout the world. The analysis takes a global, societal perspective and thus includes all costs and benefits regardless of who pays or receives them. We use a 3 % discount rate [35, 36] for future costs and polio cases to report 2013 net present values and we use the United States Consumer Price Index [37] to convert all financial estimates to 2013 United States dollars ($), unless indicated otherwise.

Table 1 lists the policy options we consider, including two reference cases (RCs) that continue the status quo indefinitely with or without continued supplemental immunization activities (SIAs) (i.e., RC with SIAs, RC without SIAs). In all of our analyses, we assume that countries using an IPV-only or IPV/OPV sequential RI schedule at T₀ will continue to use IPV for the entire analytical time horizon. However, for countries that use OPV at T₀ we consider the current strategic plan through 2018 [19] followed by global minimum policies of IPV use for 5 or 10 years following OPV13 cessation (i.e., IPV5, IPV10), and IPV use through the end of the analytical time horizon (IPV through T_end). We also consider a policy that follows the current plan for OPV2 cessation in 2016 and OPV13 cessation in 2019 except that countries using OPV-only at T₀ do not introduce IPV (i.e., No IPV). We assume that these global IPV options reflect minimum requirements, with the expectation that countries can always opt to do more than the minimum recommended policy [31]. The model assumes that LOW and LMI countries that currently use OPV-only would opt for the minimum policies, but UMI countries that use OPV-only or IPV/OPV at T₀ will use IPV through T_end regardless of the global minimum policies (Table 1).

We compute the incremental cost-effectiveness ratio (ICER) and incremental net benefits (INB) of each alternative option compared to each RC, which continues the status quo indefinitely and encompasses the spectrum of SIA frequencies that might occur with continued OPV use. We express the ICER in $ per prevented (paralytic) polio case and $ per disability-adjusted life-year (DALY) [38] averted and the INB in $ (see equations in Additional file 1). Negative values for ICERs distort the interpretation of these ratios [27, 39]. For example, an intervention with $100 in incremental costs but one more case compared to the status quo represents an undesirable option but receives the same ICER of −100 $/case as a desirable intervention that saves $100 and prevents one case. Therefore, we refer to ICERs with negative incremental costs and negative prevented cases as cost-saving but life-costing (CSLC), those with negative incremental costs but positive prevented cases as cost- and life-saving (CLS), and those with positive incremental costs but negative prevented cases as dominated [26]. Given complications associated with aggregating ICERs across different income levels, we report the ICER separately for each income level, while we report the INB both by income level and as a global aggregate [26, 40].

We implemented the model in JAVA^TM using Eclipse^TM and perform model runs on the Amazon Elastic Compute Cloud. We perform 100 stochastic iterations of the model for all considered policy options, and we use the results to characterize the economic metrics based on the average annual costs and cases by income level. For the RCs, one iteration suffices, because they do not include the stochastic poliovirus reintroduction events after OPV cessation, and we confirmed limited impact of random poliovirus exportations in the context of continued OPV use.

Economic inputs

Table 2 shows the economic model inputs. The GPEI Financial Resource Requirements (FRRs) [41] of the current six-year plan for 2013–2018 [19] include a total of $4.1 billion in resources budgeted to individual countries, and another $1.4 billion of global programmatic costs for technical assistance, surveillance and the Global Polio Laboratory Network, and other global-level and regional-level costs not budgeted to any individual countries. We assume that all policy options would incur approximately the same global programmatic costs, and therefore we do not include these in the incremental results. We calculate immunization costs associated with delivering vaccine doses based on unit cost estimates from prior work [18, 40, 42]. We attribute any difference between our estimates of aggregate costs based on this approach and those budgeted in the FRRs to costs not covered by GPEI resources, such as national resources for RI (including in-kind contributions), volunteer time [43], and any bilateral funds not included in the FRRs.

We base our SIA cost estimates directly on the unit cost inputs in Table 2 and the targeted numbers of children vaccinated during SIAs estimated by the model and adjusted for wastage [44]. Table 2 assumes that the administration costs per OPV dose during outbreak response SIAs (oSIAs) equal 1.5 times those during regular, planned preventive SIAs (pSIAs) [42]. In addition to treatment costs, the INB accounts for the societal costs associated with lost wages and suffering due to polio. In the absence of any direct estimates, we approximate these by equating each DALY associated with polio with the appropriate average annual per capita gross national income [26, 27, 36, 40, 45, 46].

Global poliovirus transmission model

We expanded and revised the poliovirus transmission model [28] used in the 2008 economic analysis [26, 27], to address the more complex policy space [47]. Specifically, the differential equation-based expanded poliovirus transmission and OPV evolution model (i.e., the DEB model): (1) characterizes each serotype separately (to analyze serotype-specific vaccination policies and risks), (2) considers explicitly both fecal-oral and oropharyngeal transmission (to account for the differential impact of IPV on fecal and oropharyngeal excretion), (3) uses 8 recent immunity states to reflect immunity derived from maternal antibodies, only IPV vaccination, only LPV infection, or both IPV vaccination and LPV infection (to more realistically capture the differences in immunity derived from IPV and LPV), (4) includes multi-stage waning and infection processes (for more realistic characterization of these processes), (5) characterizes OPV evolution as a 20-stage process from OPV as administered to fully-reverted polioviruses with assumed identical properties to typical homotypic WPVs (to allow cVDPV emergence to occur within the model), and (6) accounts for heterogeneous preferential mixing between mixing age groups and subpopulations [47] (Additional file 1).

For this analysis, we adopt all generic model inputs from the DEB model [47, 48]. We further scale the model up to a global level by characterizing global variability and mixing between populations. In the context of limited information to characterize all countries and heterogeneity within them and finite computing resources, we developed a simplified global model that does not explicitly identify individual countries or populations. Instead, it stratifies the world into 71 epidemiological blocks that each consist of 10 subpopulations with approximately 10 million people at T₀ (i.e., a global population of 7.1 billion people). A subpopulation corresponds to a population with spatially homogeneous but age-heterogeneous mixing such as a country, state, or large city within a large country, or a group of well-connected smaller countries of the same income level. A block corresponds to a larger epidemiological area such as parts of very large countries (e.g., Western Uttar Pradesh in India, the northern states of Nigeria), large countries (e.g., Egypt, Ethiopia, Philippines), or groups of connected countries (e.g., Central Africa, West Africa minus Nigeria). Table 3 provides a breakdown of the global population as of 2013 [33] by World Bank income level [34] and polio vaccine use as of October 2012 [16] for all 200 countries with available data. Table 3 also allocates the 71 blocks to the different combinations of income level and polio vaccine use at T₀. For this allocation, in some cases we assigned countries smaller than a block to a block with a higher or lower income level because of geographic proximity to countries that used the same polio vaccine in 2013 that probably incurred similar vaccine expenditures and transmission conditions.

We characterize random periodic infective interactions with people from other subpopulations and blocks. Specifically, we track the cumulative number of effective infections (CEI, defined as the cumulative prevalence of infectiousness-weighted infectious people) in each subpopulation, by virus reversion stage. Whenever the CEI of a reversion stage reaches a certain exportation threshold (E*) we trigger a potentially effective introduction of virus from the same reversion stage into another subpopulation and reset the CEI to zero. The exportation of poliovirus may or may not lead to an effective introduction that establishes transmission depending on micro-level dynamics not explicitly captured in the DEB model, and chance (e.g., the precise location of the virus introduction that impacts whether the infection spreads beyond the first household(s)) [49]. Therefore, we randomly determine if the exportation leads to an effective introduction, defined as an introduction that establishes subpopulation-wide transmission (i.e., transmission beyond the individual(s) importing the virus and its immediate surrounding household or community), using a function for the probability of an effective introduction (PEF), which logically depends on the immunity level of the receiving subpopulation. We model the PEF as a function of the mixing-adjusted net reproduction number (R_n), which represents the average number of secondary infections generated by a single infection accounting for population immunity calculated as the basic reproduction number (R₀) multiplied by the effective susceptible proportion [50]. R_n in a subpopulation depends on the baseline R₀ of the subpopulation and the virus strain (i.e., different R₀ values for different serotypes and reversion stages) [47, 48] and changes with time depending on vaccination policies, any immunity derived from LPV exposure, and seasonality. Thus, PEF depends on all of these factors through R_n (Additional file 1).

Effective virus introductions may or may not lead to an outbreak (i.e., at least one polio case) depending on the population immunity level in the receiving subpopulation and the kinetics of the initial infections relative to the seasonally changing R₀. However, if they do, an outbreak can unfold very quickly in the model due to the assumption of homogeneous mixing within relatively large subpopulations of approximately 10 million people. The homogenous mixing assumption implies faster propagation of the virus than would occur if in reality the subpopulation remains more heterogeneous [47, 51]. Given the inability to observe ineffective introductions that die out locally due to chance or locally effective introductions that do not continue to circulate due to high surrounding population immunity, and the reality of spatial heterogeneity in mixing within subpopulations, estimating the exportation threshold E* from data on long-range exportations remains challenging.

Based on the relatively localized transmission of cVDPVs to date despite presumably large numbers of infections (e.g., Nigeria) [13], we determine E* such that a cVDPV2 outbreak in a subpopulation within a year following OPV2 cessation yields approximately one expected effective exportation to another subpopulation, assuming an aggressive and effective response in the subpopulation of the initial outbreak. This criterion leads to an estimate of E* of 200,000 CEIs and remains consistent with our current experience associated with cVDPV and WPV importation outbreaks in the context of populations with recent widespread LPV exposure. We remain uncertain about the kinetics of poliovirus transmission between populations in the unprecedented context of no recent global LPV exposure as the time since OPV cessation increases, but our model assumes that the inherent frequency of potentially effective exportations per CEI (i.e., E*) does not change over time. However, the probability that an exportation becomes effective in the receiving subpopulation increases in the absence of recent LPV exposure through the dependence of PEF on R_n, which increases rapidly after OPV cessation in high-R₀ populations, regardless of IPV use [24, 52]. Similarly, the outbreak kinetics following an effective introduction speed up as the time since OPV cessation increases.

Assuming preferential mixing between the subpopulations in an epidemiological block, we assume that 24 out of 25 (96 %) exportations go to random subpopulations within the same block, while the remaining 1 out of 25 (4 %) go to random subpopulations of other blocks (i.e., inter-block exportations). Thus, inter-block exportations occur once every 5 million CEIs on average (i.e., 1/200,000 × 1/25). To determine the importing block for an inter-block exportation, we group all 71 blocks into 9 regions with variable number of blocks, corresponding to large geographical regions (i.e., Africa, Australasia, China and neighbors, East and Central Asia, Europe, India, Latin America and the Caribbean, North America, and South Asia) (Additional file 1). We assume that 3.5 % of all exportations (i.e., 87.5 % of all inter-block exportations) go to a random block in the same region and that the remaining 0.5 % (i.e., 12.5 % of all inter-block exportations) go to a random block in a different region. To characterize the global variability in conditions relevant to poliovirus transmission, we vary a number of model inputs by subpopulation related to viral transmission (i.e., R₀ and its seasonal fluctuations, the relative importance of oropharyngeal and fecal-oral poliovirus transmission) and immunization program performance (i.e., OPV take rates, RI and SIA intensity, and surveillance quality)(Additional file 1). To approximate the WPV prevalence and global immunity levels at T₀ we run the model for a “burn in” period to begin the policy comparisons starting with initial conditions that approximate actual demographic profiles and exposure histories [47] (Additional file 1).

Simulation of post-OPV cessation risks

We assume tOPV intensification leading up to OPV2 cessation maximizes population immunity at OPV2 cessation and avoids cVDPV2 emergences after OPV2 cessation [24, 52] and sufficient bOPV use before OPV13 cessation to avoid subsequent cVDPVs. Thus, we focus on other risks, including the small, but non-zero probabilities of unintentional or intentional release of LPV and introduction of iVDPVs from prolonged or chronic excretors [8, 53, 54]. Table 4 provides estimates for the non-cVDPV risks based on the currently available evidence [54] and updated from prior work [8].

For the iVDPV risks, we constructed a discrete-event simulation (DES) model of long-term iVDPV excretor prevalence to estimate iVDPV prevalence until and after OPV cessation of each serotype [54]. For each stochastic iteration of the global model, we use one stochastic realization of the DES model to generate random introductions of iVDPV into the general population after OPV cessation. We randomly generate contacts with the general population for each individual with active long-term iVDPV excretion after OPV cessation. To estimate the rate of general population contacts, we assume that R₀ provides a measure of the average number of contacts per approximately 30 days for immunocompetent individuals, assuming approximately 30 days of excretion for fully susceptible individuals [47]. While we model R₀ as ranging from 4–13 globally [47], we assume that any primary immunodeficiency disease (PID) patients surviving long enough to become long-term excretors in any setting mix much less intensely with others than immunocompetent individuals in the general population (i.e., their continued survival depends on relatively good hygiene and limited mixing), with their R₀ values effectively ranging from 1–4. We further assume that the majority of contacts (i.e., 95 %) involve close contacts (e.g., in the same household) with individuals who possess sufficient immunity to prevent further spread due to their ongoing exposure to the long-term excretor. This leaves between 0.05 and 0.2 contacts (i.e., R₀ of 1–4 times 5 % of contacts that are not close) per 30 days with the general population for a long-term excretor, or an average time of approximately 150–600 days between potential contacts that may lead to an iVDPV infection in the general population (Table 4). We draw a random contact rate for each individual long-term excretor from this range with a uniform distribution. Based on the contact rate for the individual long-term excretor, we randomly determine the time between general population contacts and include as potentially effective iVDPV introductions all contacts until (1) the excretor dies, (2) the excretor recovers and stops excreting, or (3) the time of the next contact exceeds the analytical time horizon (i.e., it would occur after 2052). In addition to using the DES model [54] to track the prevalence of and generate potentially effective introductions from long-term excretors infected prior to OPV cessation, we also use it to account for the possibility of creating new iVDPV excretors exposed to any mOPV used to respond to outbreaks after OPV cessation (Additional file 1). As with LPV importations, iVDPV excretor contacts with the general population or other releases of poliovirus may or may not lead to effective introductions depending on micro-level dynamics and chance, and therefore we apply the PEF to determine whether the introduction establishes transmission.

For IPV production sites releases, we assume that ongoing production of IPV from WPV seed strains will continue indefinitely in 5 fixed different subpopulations in HIGH blocks, which may generate potential WPV introductions at any time. We assume that a further 7 (for IPV5 or IPV10) or 10 (for IPV through T_end) facilities in non-HIGH subpopulations from a selected list of blocks that use OPV-only at T₀ will produce IPV from Sabin seed strains (Additional file 1). These sites may generate potential OPV introductions as long as IPV remains in use in the corresponding block, which depends on the policy option. The literature documents 4 reported containment failures during the past 25 years of IPV production [8, 55, 56], and one additional breach of containment from an OPV production site [8]. While improved containment guidelines may reduce this risk, some releases may have gone unnoticed due to currently very high population immunity to transmission, and global IPV production will increase. Therefore, we assume a continued rate of IPV production site releases of 1 per 5 years, independent of the number and locations of IPV production sites. Upon triggering a vaccine production site release, the model randomly selects one of the assumed production sites. The location determines the type of virus released (i.e., WPV from a current production site in a HIGH block or OPV from a non-HIGH Sabin-IPV production site), while the model randomly selects the serotype (each with equal probability). The selection of the virus determines its transmission properties, and we randomly determine the probability that the release comprises an effective introduction based the R_n-dependent PEF, as for other releases or virus importations. Although potential future research may develop non-replicating IPV seed strains, we do not consider that possibility in this analysis.

We assume much lower rates for other releases, translating into an approximately 10 % chance of a release of unreturned OPV during the first year after OPV cessation, and 10 % chance of any other intentional or unintentional release at any point during the analytical time horizon (Table 4). If any of these releases occurs, we randomly select the OPV (i.e., in the event of an unreturned OPV release) or WPV serotype released with equal probability, and the receiving subpopulation according to the assumed distribution of the risk by income level in Table 4. To ensure comparability across policies, we use the same list of potential post-OPV-cessation introduction events for all policy options, and in some cases the policy choice affects whether the potential introduction takes place.

Characterization of outbreak response after OPV cessation

Table 4 includes model inputs related to oSIAs. Once a block eliminates WPV, but before OPV cessation, we start accumulating the incidence of polio cases in each subpopulation resulting from effective importations or indigenous cVDPV emergences. If the cumulative incidence of WPV or fully-reverted VDPV cases per 10 million people reaches more than the subpopulation-specific detection threshold (i.e., 1, 2, or 3 polio cases), then this triggers outbreak response SIAs (oSIAs) in the subpopulation that override any scheduled pSIAs and that start at 45 days after detection. After the oSIAs, the subpopulation returns to its post-WPV-elimination schedule (Additional file 1) and again begins accumulating polio cases from WPV or fully-reverted VDPV until any new detection occurs. After global OPV cessation of a serotype, the nature of outbreak response changes. We accumulate polio cases from any LPV (i.e., all OPV-related viruses) to trigger oSIAs and we modify the response strategy to reflect sufficiently aggressive response to minimize the chances of failing to fully control outbreaks after OPV cessation, as shown in Table 4. In the event of a subpopulation-specific response after OPV cessation, we assume that all other subpopulations in the same block remain on “high alert” between detection and the completion of the last oSIA in the outbreak subpopulation, characterized as high acute flaccid paralysis (AFP) surveillance quality (i.e., detection after occurrence of 1 cumulative paralytic case per 10 million people) and a short response delay (i.e., 30 days between detection and the first oSIA).

We assume that subpopulations that use IPV-only at T₀ would use only IPV for oSIAs any time after they switch to IPV-only and particularly after global OPV cessation, based on the unavailability of OPV for outbreak response in the United States (i.e., the largest IPV-only country to date) [57, 58], although some IPV-only countries responded to outbreaks with both IPV and OPV [22, 59]. For all other subpopulations, the vaccine choice depends on time and the detected serotype. Specifically, before OPV cessation of any serotype oSIAs use tOPV (if serotype 2 poliovirus detected) or bOPV (if no serotype 2 poliovirus detected), during the first 5 years after OPV cessation of any serotype they use mOPV of the detected serotype, and any time after that they use IPV, because we assume that the risk related to reintroducing large amounts of LPV becomes too large to use OPV that long after OPV cessation. The optimal duration of mOPV use for oSIAs after OPV cessation remains uncertain, but 5 years resulted in a very low probability (i.e., <1 %) of exported OPV-related viruses establishing transmission in other subpopulations or blocks, given all other model assumptions. We do not constrain the amount of mOPV and IPV available for oSIAs after OPV cessation, which allows us to estimate potential vaccine needs from the stockpile, based on the total targeted population in all oSIAs after OPV cessation, adjusted for the estimated wastage rates during SIAs (Table 2). We report the fraction of stochastic iterations in which for at least one serotype the number of mOPV doses needed for oSIAs exceeds the 500 million total and 100 million filled mOPV doses of each serotype currently planned for the stockpile.

Variations of the IPV5 policy

In the context of the IPV5 policy, we consider the potential impact of the adoption of polio antiviral drugs (PAVDs) to treat iVDPV excretors from 2017 forward, which would potentially clear their infections [54]. We consider IPV5 with PAVD passive use as one option, which assumes 40 % effectiveness in clearing the infection with treatment of 50 % of excretors with paralysis on January 1, 2017 and of those who subsequently develop paralysis (i.e., IPV5, PAVD40%). We also consider IPV5 with PAVD active use, which assumes 90 % effectiveness in clearing the infection and treatment of 90 % of all excretors with an ongoing infection after January 1, 2017 (i.e., IPV5, PAVD90%). We randomly pre-determine which excretors would recover from their iVDPV infections as a result of PAVD treatment based on the probabilities for the two scenarios, which provide some bounds on the combined effectiveness of the PAVD compound(s) and the degree of passivity of efforts to identify and treat iVDPV excretors. We also consider the impact of a failure to intensify tOPV use leading up to OPV2 cessation, which assumes continuation of the SIA schedule from before the year 2015 up until the time of OPV2 cessation (Additional file 1). Finally, we consider the impact of a higher E* and lower cumulative paralytic case thresholds used to trigger an OPV restart (compared to the base case threshold of 50,000 cumulative polio cases after 2016 above which we assume countries that used OPV as of 2013 would restart using OPV).

Expected future burden of polio cases

Table 5 reports the average total number of cases (including VAPP) over the analytical time horizon, broken down by iterations with or without OPV restart, the number of iterations with uncontrolled outbreaks leading to OPV restart in all countries that use OPV at T₀ (i.e., the number of runs reaching 50,000 cumulative cases), and the number of iterations with any detected outbreaks that trigger a response. For all policies involving OPV cessation, ≥96 % of iterations involve one or more outbreaks after OPV cessation, which implies expected use of the vaccine stockpile and outbreak response plans. The majority of the outbreaks trace back to long-term iVDPV excretors, who can re-introduce polioviruses years after OPV cessation when population immunity to transmission becomes low enough to allow these viruses to establish transmission and cause outbreaks. However, the longest expected survival of iVDPV excretors occurs in lower-R₀ settings with less fecal-oral transmission [54] in which IPV provides more impact on poliovirus transmission. In higher-R₀ settings we expect few long-term iVDPV excretors to survive beyond the 5-year window during which our model allows mOPV use for oSIAs.

Aggressive outbreak response rapidly controls the majority of the expected outbreaks, avoiding exportations to other subpopulations and blocks that would lead to a high number of cases after OPV cessation and eventual OPV restart. However, for IPV5 and IPV10, 2 of 100 iterations led to poliovirus reintroductions that occurred at a time and place with very low IPV-only-induced population immunity (i.e., due to a combination of high enough R₀ and contribution of fecal-oral transmission and introduction long enough after OPV cessation) that triggered an OPV restart. One of these traced back to an iVDPV1 introduction in a very high-R₀ block relatively soon after OPV cessation, which triggered mOPV SIAs that infected a PID patient who became a new long-term excretor and reintroduced an iVDPV1 at a time when the model no longer allows mOPV use for outbreak response. The other traced back to an unintentional or intentional “other” release (i.e., from an accidental breach in laboratory containment or a bioterrorism event) of WPV3 in a LMI block with an R₀ of 8 in the second half of the year 2049. As in most relatively higher R₀ populations, we observed for the first OPV restart iteration that even a very large number of oSIAs with IPV with a block-wide geographical scope and increasingly wide target age range could not control the outbreak. While the IPV oSIAs kept the incidence relatively low and delayed spread to other blocks for many years, eventually enough effective exportations occurred to trigger new large outbreaks and accumulate over 50,000 cases. These two iterations provide two examples from a larger number of possible scenarios that could potentially lead to an OPV restart and they average between approximately 300,000-700,000 expected polio cases, depending on whether the OPV restart involves resumed SIAs. However, based on 100 stochastic iterations, any such scenario represents a relatively rare event in the context of our assumed aggressive outbreak response and frequency of spread between populations. The 98 iterations that control all outbreaks with IPV5 average an expected 340 post-OPV-cessation polio cases.

The policy of IPV through T_end led to a total of 10 OPV restarts, including the 2 that occurred with IPV5 or IPV10 and 8 additional iterations associated with release of Sabin seeds strains from Sabin IPV (sIPV) production sites after the year 2035 and located in blocks with a R₀ between 9 and 11. This led us to the general observation that in some relatively high-R₀ blocks (e.g., R₀ ≥ 9), any releases of Sabin seed strains can eventually establish uncontrollable transmission. Thus, based on the historical rate of releases from poliovirus vaccine production sites during the last 25 years, the use of any LPV strains in high-R₀ populations presents an important risk that requires management. Releases of WPV and Sabin seed strains also occurred in lower-R₀ blocks in some iterations, but these either did not establish any transmission (e.g., Sabin seed strain releases) or led only to smaller, controlled outbreak in high-income blocks (WPV seed strain releases).

The policy of No IPV use assumes that all blocks that use OPV-only at T₀ do not add IPV at any time before or after OPV cessation. The absence of any IPV use in these blocks allows population immunity to drop more rapidly, particularly in medium-R₀ populations (i.e., between 6–8) in which we expect better survival of long-term iVDPV excretors. Consequently, for this policy option we observed 5 iterations in which iVDPV introductions triggered an eventual OPV restart (in addition to one OPV restart associated with an “other” release that caused an OPV restart for all policies), including one new iVDPV excretor associated with mOPV use for outbreak response after OPV cessation (Additional file 1).

For the global IPV policies (i.e., IPV5, IPV10, or IPV through T_end), the estimated number of mOPV doses needed from a stockpile for use in oSIAs after OPV cessation exceeded the currently planned 100 million filled mOPV doses for at least one serotype in 32 stochastic iterations (i.e., including 23 iterations for mOPV1, 12 for mOPV2, 8 for mOPV3). Given that all mOPV use in the model occurs within 5 years of OPV cessation of each serotype, this suggests a high probability of needing to fill some of the bulk mOPV stock soon after OPV cessation. In 2 stochastic iterations, the estimated number of mOPV doses exceeded the currently planned total stockpile of 500 million mOPV doses for at least one serotype (i.e., 1 for mOPV1 and 1 for mOPV3). One of those 2 iterations led to an OPV restart even in the event of an unlimited stockpile. In the other iteration, exhaustion of the entire mOPV stockpile would result in an eventual OPV restart due to the lack of a viable alternative oSIA vaccine to prevent ultimate exportations of the outbreak virus to other populations with low immunity levels. Thus, an insufficiently large stockpile carries some risk of ultimately leading to OPV restart (i.e., an insufficient stockpile may lead to much higher OPV demands associated with OPV restart).

The average numbers of cases for each policy show a clear dichotomy between iterations that typically control outbreaks rapidly and iterations that led to OPV restart (Table 5). Iterations with controllable outbreaks represent the most common outcome and yielded fewer than 1,000 expected cases on average between OPV cessation of each type and the end of the analytical time horizon. IPV10 instead of IPV5 reduced the expected average number of post-OPV-cessation cases in those iterations by more than half from 340 to 120, while No IPV more than doubled the expected average number of cases to 840. IPV through T_end increased the expected average number of cases with no OPV restart compared to IPV5 or IPV10, because in 4 iterations transmission resulting from a late release from an sIPV site continued until T_end (without resulting in an OPV restart), which drove up the average. Iterations that restart OPV typically involve one year with a very high burden of paralytic cases followed by all remaining years with either high burden (i.e., OPV restart without SIAs) or low burden (i.e., OPV restart with SIAs). With both options, for runs with OPV restart, the year prior to the OPV restart drives the average number of cases to over 100,000 cases. Despite the relative rarity of OPV restarts, they significantly increase the expected average cases. Nevertheless, the expected average cases for all OPV cessation policies remain far below the expected cases of RC no SIAs, which results in an ongoing VAPP burden in all OPV-using populations and re-established cVDPV transmission in many of them.

Figure 1 shows the average annual incidence (i.e., undiscounted cases) for the full analytical time horizon based on 100 stochastic iterations of the model for the different policies. Clearly, RC no SIAs represents the worst option in terms of annual cases (Fig. 1a and b). On this scale, the burdens for RC with SIAs and all OPV cessation options remain negligible until the time when larger, uncontrolled outbreaks occur in a small number of iterations, resulting in bumps in the average numbers of cases. Subsequent years either exhibit noticeable average numbers of cases (i.e., Fig. 1a assuming OPV restart without SIAs) or very small numbers of cases (i.e., Fig. 1b assuming OPV restart with SIAs) compared to the reference case without SIAs. To better show the behavior for typical runs that do not involve uncontrolled outbreaks, Fig. 1c omits RC no SIAs and all iterations that lead to an OPV restart. For IPV5, the dynamics in Fig. 1c reflect the average incidence from relatively common but small outbreaks between OPV2 cessation in 2016 and the end of both IPV use for RI and mOPV use for oSIAs in 2024. The average numbers of cases in subsequent years reflect lower probability events with higher consequences due to iVDPV introductions in relatively medium- or low-R₀ populations that get controlled by IPV oSIAs and/or local burn-through of susceptible individuals before the virus can spread more widely, resulting in more peaky behavior associated with larger outbreaks in a few stochastic iterations. IPV10 substantially reduces the expected probability and consequences of these events compared to IPV5. However, for IPV through T_end, Fig. 1c shows the occurrence of some late releases of Sabin seed strains from IPV production sites, including 4 iterations in which LPV transmission continued until T_end without triggering an OPV restart. Despite the possibility of outbreaks after OPV cessation, the expected number of annual cases for the subset of typical runs that do not result in an OPV restart remains below the expected annual number of VAPP cases of RC with SIAs in each year, and very far below the expected annual burden of RC without SIAs (off-scale in Fig. 1c).

Expected future vaccination costs

Figure 2 shows the expected vaccinations costs over time, which do not include the global programmatic costs associated with preparing for OPV cessation or any specific risk management activities after OPV cessation other than RI and oSIAs. The panels break down the results by OPV restart outcome similar to the panels of Fig. 1. The first years show a marked increase in vaccination costs associated with the introduction of IPV in RI. For all OPV cessation options the costs drop dramatically after OPV13 cessation in 2019, which coincides with the cessation of SIAs for RC without SIAs. A further drop in costs occurs at the time of cessation of universal IPV use (if applicable). Nevertheless, significant costs continue to occur throughout the analytical time horizon due to the assumed continued use of IPV in RI in UMI and HIGH blocks that already use IPV at T₀. For IPV through T_end, the continued IPV use everywhere combined with the need to respond to releases of Sabin seed strains from sIPV production sites in some iterations imply much higher costs than the other OPV cessation policies and RC no SIAs, but these costs remain below the expected costs of RC with SIAs. The average costs of No IPV highlight the important costs associated with oSIAs after OPV cessation, for which we assume relatively higher costs than pSIAs, particularly for repeated and widespread but relatively ineffective IPV oSIAs during years when we assume mOPV no longer represents an option for oSIAs. Given that this occurs in some iterations, the reactive behavior associated with No IPV results in higher expected vaccination costs than preventive investments in IPV RI with IPV5 or IPV10. Notably, the No IPV option, thus, includes outbreak response with IPV when it represents the only polio vaccine option, including use in countries that do not use IPV in RI.

Economic analyses

Table 6 summarizes the results of the economic analysis for IPV5 compared to the two RCs, broken down by income level and aggregated over the 40-year time horizon. In HIGH blocks, IPV5 implies a different path than the RCs in only 2 out of 10 blocks, which would switch to IPV-only after OPV13 cessation as opposed to continuing sequential IPV/OPV use in the RCs (Table 1). This change results in a relatively small increase in costs, but if divided by a very small number of VAPP cases prevented yields very high ICER of approximately $3-5 million per DALY averted, consistent with the known high societal willingness-to-pay to prevent VAPP in developed countries that already made the switch from OPV to IPV [60]. In UMI blocks, the expected cases similarly reflect VAPP cases, which decrease with the move to an eventual IPV-only schedule, but at a high incremental cost of $12 billion (and thus high ICERs), particularly compared to the costs of RC no SIAs. The impact of switching to IPV in UMI and HIGH blocks appears much less dramatic if expressed in terms of INBs, with relatively small negative INBs, except for the comparison of IPV5 to RC no SIAs in UMI blocks, which yields INBs of $-3.5 billion and significantly decreases the positive global INBs. LOW and LMI countries appropriately remain the focus of attention, because the impact of the policy choices on costs and cases remains much more significant.

Table 6 suggests that RC no SIAs results in many more cases than expected with IPV5. Despite the inclusion of 1 IPV dose in RI during 2015–2024 with IPV5, RC no SIAs remains more costly in LOW and LMI blocks. Consequently, IPV5 represents a cost-and life-saving option compared to RC no SIA, yielding almost $20 billion in INBs in the combined LOW and LMI blocks. If instead we compared IPV5 in LOW and LMI blocks with RC with SIAs, then Table 6 suggests significant savings with a small incremental number of expected cases (LOW blocks) or a small number of expected prevented cases (LMI blocks, not visible with the two significant digits shown in Table 6). Consequently, we find that compared to RC with SIAs, IPV5 represents a cost- and life-saving policy in LOW blocks and a cost-saving but life-costing policy in LMI blocks. Unlike the ICERs that change dramatically with small denominators, the INBs remain more robust to the choice of RC, with the difference in vaccination costs for OPV with SIAs roughly equivalent to the societal costs of paralytic polio cases associated with RC no SIAs. The global INBs amounts to approximately $16 billion or $17 billion for comparison to RC no SIA or RC with SIAs, respectively. The negative INBs in UMI and HIGH blocks reduce the global INBs, which equal $18-20 billion if we exclude the two highest income levels.

Table 7 shows the expected global costs, cases, and INBs for the other main policy options and the modeled IPV5 variations. IPV10 increases costs compared to IPV5 while preventing a relatively small number of additional cases, which decreases the global INBs by approximately $0.8 billion. As shown above, IPV through T_end increases the expected number of cases while increasing costs, leading to an overall reduction of the global INBs of approximately $6 billion. No IPV does not save costs due to relatively high numbers of oSIAs required, and it also yields more expected cases compared to IPV5. Consequently, for No IPV the global INBs decrease by approximately $0.5 billion.

The IPV5, PAVD40% variation only prevented a small number of outbreaks and did not notably affect the global net benefits. In contrast, active PAVD use with IPV5, PAVD90% prevented many iVDPV-associated outbreaks, including one of the two events that triggered an eventual OPV restart with IPV5 (i.e., the OPV restart caused by a long-term iVDPV excretor). Consequently, the expected number of cases for IPV5, PAVD90% decreased by almost 4,000, resulting in an expected increase in the global INBs of approximately $0.5 billion, which provides some economic justification for significant investment (e.g., $500 million) in the development of PAVDs and strategies for actively identifying and managing iVDPV excretors. The PAVD90% scenario also decreased the number of stochastic iterations for which the estimated number of mOPV oSIA doses required from the stockpile exceeded 100 million for at least one serotype from 32 to 6, and the number of stochastic iterations for which the estimated number of mOPV oSIA doses needed from the stockpile exceeded 500 million for at least one serotype from 2 to 1.

No tOPV intensification prior to OPV2 cessation led to a cVDPV2 outbreak in all 100 stochastic iterations as a result of insufficient population immunity to serotype 2 poliovirus transmission in one subpopulation at the time of OPV2 cessation. Aggressive outbreak response with 4 block-wide oSIAs controlled the cVDPV2 outbreak in all 100 stochastic iterations, leading to a relatively modest increase in average expected cases from the cVDPV2 outbreak. However, No tOPV intensification also affects population immunity to serotype 2 transmission in other subpopulations which would otherwise intensify tOPV use. Consequently, in one of the 100 stochastic iterations, we observed a different realization of exportations related to an iVDPV2-associated outbreak that ultimately led to an OPV restart. Thus, failure to intensify tOPV use prior to OPV2 cessation resulted in a notable increase in the average expected number of cases by almost 3,000, and the need for mOPV2 use to respond to the cVDPV2 outbreak that increased the expected average costs by approximately $0.2 billion and decreased the global INBs by $0.3 billion. Failure to intensify tOPV use prior to OPV2 cessation did not affect the number of stochastic iterations for which the estimated number of mOPV oSIA doses exceeded 500 million for at least one serotype. However, the cVPDV2 outbreaks significantly increase expected demand from the stockpile by almost 120 million mOPV2 doses, which exceeds the size of the currently planned filled mOPV2 stockpile.

The assumed speed of viral spread between subpopulations affects the ability of aggressive oSIAs to control outbreaks after OPV cessation, prevent further spread, and avoid eventual OPV restarts. For IPV5, we explored the impact of changing the exportation threshold E* from 200,000 to 100,000 CEIs, which effectively doubles the frequency of exportations. For IPV5, this change resulted in 3 additional iterations with an eventual OPV restart and typically larger outbreaks in iterations without an eventual OPV restart. As shown in Table 7, the increased frequency of exportations leads to higher expected costs, more expected cases, and a decrease in the global INBs by approximately $1.2 billion compared to IPV5. This analysis demonstrates the importance of the uncertainty associated with the potential for international spread in an unprecedented world with no recent LPV exposure. Table 7 further shows the absence of any impact of varying the model choice for the threshold for restarting OPV between 50,000 and 15,000 cumulative cases, with only a small reduction in INBs for a reduction in the threshold to as low as 1,000 cumulative cases. The effect remains small because uncontrolled outbreaks typically include many thousands of cases in the year before OPV restart, so that crossing the threshold typically occurs in the same year regardless of the choice of threshold. Thus, our economic results remain robust to realistic choices for this threshold.