Benefits and Challenges in Using Seroprevalence Data to Inform Models for Measles and Rubella Elimination

We use simulations to highlight how (accounting for the dynamical context) high-quality measles and rubella serological surveys can be used to inform key control and elimination questions if the challenges of conducting, analyzing, and interpreting them are overcome.

Infectious diseases can persist in populations if there are enough individuals susceptible to infection to acquire and transmit infection. Infection from fully immunizing pathogens, such as measles and rubella viruses, leads to lifelong immunity, thus depleting the number of susceptible individuals in a population. Therefore, pathogen persistence is only possible if susceptible individuals are replenished via births and immigration of susceptible persons. Measles and rubella vaccines are highly effective, reducing the rate of accumulation of susceptible individuals. These vaccines provide indirect protection to susceptible individuals by reducing the probability of effective contact between infectious and susceptible individuals, in addition to direct protection of immunized individuals. As a result, successful immunization programs can curtail or eliminate pathogen transmission [1]. More importantly, a small proportion of individuals fail to develop a long-lasting immune response after vaccination [2,3]: thus, we refer to "vaccination" as the administration of vaccine and "immunization" as the induction of a protective immune response via vaccination.
High measles vaccination coverage worldwide has reduced measles incidence and mortality to low levels in most countries, although progress in achieving high coverage with the first dose of measles-containing vaccines (MCVs) has slowed recently [4]. Although MCVs have been in widespread global use for over 40 years, rubella-containing vaccines (RCVs) have only recently been introduced into low-income countries [5]. The World Health Organization (WHO) Region of the Americas was certified as achieving elimination of endemic rubella and measles in 2015 and 2016, respectively [6]. The remaining 5 WHO regions have measles elimination targets, and 3 have set rubella control or elimination targets for 2020 [5]. Elimination efforts include attaining and sustaining high coverage with the first dose of MCV, scaling up of routine vaccination with a second dose of MCV, introduction of RCVs, and supplemental immunization activities ([SIAs] or campaigns) of both MCV and RCV [5].
Mathematical models indicate that elimination requires the achievement of a threshold level of population immunity (ie, the percentage of the population immune). This threshold differs between measles and rubella and between settings, but it is determined by the pathogens' transmission potential [1] and the birth rate [7]. Elimination must be maintained by sustaining this level of population immunity through immunization or preventing reintroduction of the virus. The WHO recommends that programs use good quality data to monitor population immunity by identifying and responding to large numbers of susceptible individuals [8].
Although analytic techniques can be used to infer population immunity profiles from vaccination coverage [9] and incidence data [10], these methods necessarily rely on indirect inference of immunity rather than direct measurement. Vaccination coverage data are often of poor quality [11] and can misrepresent population immunity because most countries lack data on vaccine effectiveness under field conditions and, assuming an average vaccine effectiveness, may overestimate population immunity in countries with weak cold chain systems. Furthermore, where multiple doses are offered (eg, a routine second dose or SIAs), doses may be disproportionately delivered to individuals who received a first dose [12]. Disease surveillance data may be biased due to nonspecific diagnosis, preferential reporting of disease in young children, and gross underreporting (eg, in 2016, WHO estimated approximately 7 million (95% confidence interval, 4.2-28.7 million) global cases [4], but only 132 137 were reported to the WHO [13]). For rubella, reporting is even less sensitive because surveillance was introduced recently, and 20%-50% of rubella cases are subclinical or asymptomatic [14].
Serology can provide a direct measure of population immunity. After infection or immunization, pathogen-specific immunoglobulin (Ig)M and IgG antibodies are produced. Immunoglobulin M antibodies persist for a few weeks, whereas IgG antibodies persist for years to decades, although levels may decline over time. Measles virus-specific and rubella virus-specific IgG antibodies are recognized correlates of immunity, and antibody concentrations in the blood exceeding a threshold are deemed protective against infection or disease [15,16]. Thus, in principle, high-quality serological surveys, ie, cross-sectional household surveys in which these antibodies are measured, allow direct measurement of a population's immunological profile. Depending on their design, serological surveys can reveal "immunity gaps" (ie, age groups or spatial locations where immunity is lower than expected, or below some operational threshold), thus identifying areas for additional vaccination efforts [17].
Because natural and vaccine-derived immunity cannot be distinguished, interpreting serological data requires accounting for historical changes in disease incidence and vaccination coverage (achieved both through routine services and SIAs), as we illustrate in this paper. We first delineate expectations for age-specific immunity profiles across a spectrum from endemicity to elimination. We describe how inferring immunity profiles from vaccination coverage data and/or reported case data (the current method used by most countries) can result in biases, showing the potential added value of serology. We then describe how serology can be used to address 2 key questions for the control and elimination of measles and rubella: (1) what are the most effective vaccination strategies to control and eliminate infection, and (2) how effective are current vaccination programs? We conclude by discussing challenges of serological surveys and provide suggestions for improving their feasibility and sustainability.

Age-Specific Serological Profiles: Expectations Across a Transmission Spectrum From Endemic to Elimination
Seropositivity in young infants results from the transplacental transfer of maternal IgG antibodies to the fetus. Antibody levels then decay exponentially as the infant ages. The proportion of seropositive children then increases at a rate determined by the rate of immunization through vaccination or infection. Age-specific seroprevalence profiles vary over time because vaccine coverage and infection transmission vary, as described below and illustrated in Figures 1 and 2 (model assumptions are described in Tables 1 and 2, methods are decribed in Supplementary Materials S1).

Age-Specific Serological Profiles in Endemic Settings
In settings where vaccination coverage is low or absent, the age-specific serological profile is largely dominated by natural immunity ( Figure 1A). The rate of acquisition of immunity with age (after loss of maternal immunity) is determined by the rate of transmission in the population. If transmission is low, acquisition of infection and thus immunity is slow. Susceptible individuals may not encounter an infected individual until adolescence or older. R 0 , the basic reproductive number representing the number of new infections per infectious individual in a completely susceptible population, is a commonly used measure of transmission potential that informs expected age-specific patterns of seropositivity [1]. Data detailing changes in the number of cases over time [10] or their distribution across ages [18] can be combined with mathematical models to estimate R 0 and infer expected age-specific immunity patterns. In addition, the profile of cases accumulated across age strata may be used to reflect the cumulative proportion of immune individuals by age, although biases may emerge due to age-specific sensitivity of reporting.
Insensitivity of disease surveillance data can affect estimation of immunity gaps. Assuming optimistically that 5%-15% of cases are reported across a spectrum from endemicity to elimination (Table 2), Figure 2A illustrates how using case data to estimate the fraction of the population naturally immune over age, r(a) ( Figure 2A2), results in biased estimates of the proportion seropositive by age, p(a) ( Figure 2A3), in an endemic setting. Poor quality of reported vaccination coverage further complicates interpretation of case data for understanding age profiles of immunity. Countries without national rubella vaccination programs often have low levels of private healthcare sector vaccination [19] (Table 1). If these estimates are unreported, they could cause bias estimates related to measures of transmission (eg, R 0 ) and age-specific seroprevalence profiles ( Figures 2A1 and 2A3); see Supplementary Materials S2 for an empirical example.

Coverage
In areas where vaccination programs are well established, vaccine-derived immunity dominates the age-specific serological profile at younger ages. Older cohorts may have acquired natural immunity before high vaccination coverage was achieved, and maternal and natural immunity will continue to play some role.
Different strategies for vaccine administration (ie, routine vaccination or SIAs) mean that the history of vaccine delivery can result in distinct age-specific seroprevalence profiles ( Figure 1B-D). In low-income countries, routine MCV was designed to reach children during the first year of life. More recently, children have access to a second dose of MCV, commonly in the second year of life [8]. Supplemental immunization activities are conducted periodically (usually every 2-5 years) over brief timeframes (usually days/weeks, although potentially months/years in large countries) at the national or subnational level and target specific age groups (eg, 9 months-5 years of age). Outbreak response campaigns also occur. Information on the history of routine and SIAs allow inference into the age profile of vaccine-derived immunity, but data on vaccination coverage are often inaccurate [7,11,12]. Without individual vaccine histories, combining SIA coverage [20], outbreak response campaigns [21], and (where they exist) routine second dose programs may overestimate susceptibility reduction, because vaccination may be disproportionately delivered to immune individuals. For example, one study estimated that 31% of eligible populations were never accessible by routine or campaign vaccination in Sierra Leone [12]. Even if vaccine coverage areas were known precisely, everyone does not develop protective immunity after vaccination [2], and vaccine-induced antibody levels may wane below the threshold for seropositivity [3], resulting in discrepancies between vaccination history and age-specific seroprevalence.
Uncertainties in vaccination coverage data combined with underreported incidence (Table 2) result in errors in estimates of the proportion seropositive by age ( Figure 2B Table 1). Bottom panels A-D display age profiles of immunity after 6, 11, 16, 26 years, respectively; dashed red vertical lines indicate age ranges affected by the supplemental immunization activities (SIAs) in preceding years. Bottom panel shows (A) proportion immune under low vaccination coverage results in a gradual increase over age after the decay of maternal immunity. (B) Increased routine vaccination to 50% slows the rate of acquisition of immunity through natural infection, but vaccine-acquired immunity increases at rates reflecting routine vaccination delivery and SIAs, resulting in further age-specific increases in targeted age groups (affecting 2-to 7-year-olds here). (C) During periods of control, immunity in relevant age classes reflects routine vaccination coverage in ages 1 to 6 years, corresponding to low incidence between year 10 and year 13 (see incidence time series). In year 15, a resurgence affects individuals just outside the target age group of the first SIA, ie, individuals older than age 6 in year 10, and older than age 12 in year 16. (D) Erratic age-specific immunity profiles emerge under increased routine coverage: although immunity in ages 1-6 closely reflects the 90% vaccination coverage, the dip in 8-year-olds occurs because the second SIA reduced incidence, and therefore these children had a low risk of natural infection but were too young to be immunized during the second SIA, or the increase in routine coverage that occurred in year 20.
estimates of the proportion immune from natural infection, r(a) ( Figure 2B2, 2C2, and 2D2), can further bias inferred seropositivity, p(a) ( Figure 2B3, 2C3, and 2D3). By improving case surveillance, managers can reduce the degree of bias ( Figure 2). Immunity derived from natural infection remains a source of seropositivity in expanded control settings, but it is an erratic, potentially misleading one. For example, extended periods of low incidence after vaccine introduction may allow accumulation of susceptible individuals [1]. Eventually, their proportion may grow sufficiently large to sustain an outbreak. Thus, incidence provides a poor indicator of population immune status: low case numbers can reflect either sustained, high levels of population immunity or an increasing risk of an outbreak. Imperfect surveillance combined with uncertainty in  Figure 2. Comparison of inferred age-specific immunity from inaccurate coverage and case surveillance data to "true" age-specific immunity using a simulated population. Right panel displays time series of incidence in a changing context of increased vaccination coverage (see top panel, Figure 1). Left panels A1-D3 display age-specific proportion vaccinated (v(a); A1, B1, C1, and D1), proportion recovered from natural infection (r(a); A2, B2, C2, and D2), and proportion immune (p(a); A3, B3, C3, and D3) at years 6, 11, 16, 26, respectively. Blue lines represent the "truth", and black dashed lines represent the inferred age trajectories, assuming some error in vaccination and incidence data (see Table 2 for data error assumptions). False positives from estimated v(a) and false negatives from estimated r(a) both contribute to the error in estimated p(a). The added value of a high-quality representative immunoglobulin G serological data (assuming minimal diagnostic testing error) is clear because uncertainties associated with vaccination coverage and incident case data greatly bias estimates of age-specific immunity.
vaccination coverage data further complicate estimates of age-specific population immunity.

Age-Specific Serological Profiles in Near-Elimination Settings
In near-elimination settings, the prevalence of immunity derived from natural infection decreases and vaccine-derived immunity increases. Reconstructing immunity profiles without serology is complicated by vaccination coverage uncertainty, absent or rare case data from surveillance, and the resultant decline in positive predictive value of cases reported [22], unless all cases are laboratory-confirmed. Estimating the age-specific immunological profile is an important factor in identifying vaccination age targets to achieve and sustain elimination. One particular near-elimination issue is maternally derived immunity. Given that maternal antibodies neutralize vaccine virus, vaccine efficacy increases with age as maternal antibodies wane ( Figure 3A). Accordingly, administration of the first dose of MCV is usually delayed until 9 or 12-15 months of age to ensure that infants are free of maternal antibodies. However, vaccinated mothers transfer a lower level of measles virus-specific antibodies to their children than naturally infected mothers [23], potentially leaving these children susceptible to measles at an earlier age and raising questions about shifting the age of vaccination younger (see below). The ability to robustly characterize such nuanced patterns without serology is challenging.

Strategizing Effective Targeting of Vaccination to Control and Eliminate Infection
In mature vaccination programs, additional efforts would ideally target immunity gaps that could allow outbreaks [8].
Populations are at risk of an outbreak when the number (density) of susceptible individuals is sufficiently large. Susceptible individuals are distributed over age (and space; see Supplementary Materials S3). Age-specific seroprevalence estimates, combined with data on mixing-patterns by age and vaccination coverage, can help estimate susceptibility and characterize outbreak risk [7,24], thus helping program managers prioritize strategies to close immunity gaps.
One way to tackle immunity gaps is by applying modifications to the timing and targeting of vaccination campaigns. For example, low seropositivity in age classes thought to be important for transmission (eg, school children) suggests a need to increase vaccination coverage in this age group (assuming that a sensitive laboratory assay and the appropriate seropositivity cutoff have been used [25], because antibody levels wane after vaccination, especially in the absence of boosting from exposure to wild-type virus [15,16]). For example, observing low immunity in children aged 7-10 years ( Figure 1B) before  Abbreviations: mo, months old; RCV, rubella-containing vaccine; SIA, supplemental immunization activity; yo, years old. a The first 10 years of low vaccination coverage represent an example of RCV administered in the private-sector only [19]. Increased routine vaccination at 10 and 20 years represents introduction of RCV into national vaccine schedules and then expansion of the programs to capture more infants, respectively. SIAs at 10 and 18 years represent typical SIA efforts to vaccinate many young age groups.
a planned vaccination campaign strongly supports extending the age range of the campaign beyond age 5 years to close this immunity gap. Serological data could also be used to help determine the need for a vaccination campaign. Serosurveillance is conducted in high-income countries such as Japan [26], Australia [27], the United Kingdom, and the Netherlands [28], and it has contributed to vaccine policy recommendations such as initiating catch-up campaigns [29,30]. A serosurveillance system that triggers vaccination campaigns may be particularly valuable in near-elimination settings because sparse case data are a poor indicator of increases in population susceptibility, and immunity gaps may go undetected until an outbreak occurs [31]. Historically, these outbreaks have had age distributions that deviate from expected patterns of age susceptibility based on historical vaccination coverage data (eg, Malawi [21]). Serosurveys can reveal changes in the age distribution of immunity, providing an opportunity to conduct campaigns to fill immunity gaps before outbreaks occur [32]. More importantly, survey results must be available promptly, so that vaccination campaigns can be planned and implemented rapidly in response to serological data.
One additional benefit of age-specific seroprevalence is the potential to fine-tune age of routine vaccination in high-vaccine settings. The ideal age of the first dose of MCV will optimize vaccine effectiveness without increasing the number or severity of measles cases [33]. Although the WHO Americas Region successfully sustained measles elimination after an increase in the age for the first dose of measles vaccine [6], in other regions cases before the age of 9 months remain a concern. Because mothers with vaccine-induced immunity have lower antibody levels than those with infection-induced immunity, babies born to the former have lower levels of maternal antibody and become susceptible at an earlier age. Therefore, the transition to elimination could result in more unprotected infants below the age of routine vaccination [23]. In the WHO African region, high birth rates mean that this could eventually translate to  [40]. Associated estimates of the magnitude of transmission can be combined with age fertility to infer the burden of congenital rubella syndrome (CRS) prevaccination. For this example, assuming that age-specific contact patterns and private-sector vaccination coverage were comparable in both settings, these patterns suggest higher transmission in urban than rural settings, because the increase in age-specific seroprevalence is faster in urban settings. Accounting for differences in fertility, higher transmission of rubella in urban Vellore contributed to lower estimated risk of CRS compared with rural Vellore [40].
large numbers of susceptible individuals capable of maintaining measles virus transmission [34], an effect potentially amplified by high rates of human immunodeficiency virus (HIV) infection, which reduces the efficiency of transplacental transfer of maternal measles-specific antibodies [35]. Because more countries have extended periods of low measles incidence and more women of child-bearing age have vaccine-induced immunity, the optimal age for routine administration of first and second doses of MCV becomes an important policy question that serology could potentially inform. However, given the narrow age window affected, the programmatic scale of the interventions implied, and the scope of impact of any change, detailed data beyond cross-sectional, age serology (eg, clinical trials) may be required to inform policy decisions.

Evaluating the Effectiveness of Vaccination Programs Against Infections
Serology can be used to assess the effectiveness of vaccination activities, the impact on disease burden, and the progress towards elimination.
The role of SIAs in achieving measles elimination is receiving increasing scrutiny because of the high costs and human resource needs required to successfully conduct campaigns. Serosurveys have been suggested as one option to evaluate their success [17]. However, cross-sectional serosurveys cannot currently distinguish immunity from natural infection from vaccine-induced immunity (Figure 2), which complicates the use of a single postcampaign serosurvey to evaluate SIA impact (although detection of IgM antibodies within weeks of an SIA may indicate immunization of a susceptible individual, because individuals with prior immunity should not mount an IgM response). Conducting preand postcampaign serological surveys does allow managers to measure the increase in immunity due to SIAs, as demonstrated in England and Wales [30], Australia [36], and research settings in Ethiopia [37] and Kenya [38].
If a precampaign serosurvey is not feasible, serosurveys can determine whether target immunity prevalence has been reached, without drawing specific conclusions about the vaccination campaign. Otherwise, nuanced inferential tools combining data from serosurveys with age-specific disease incidence and the history of vaccination can estimate vaccination campaign effectiveness, although uncertainties with these data sources (see above) remain a limitation.
Age-specific serological data are also a valuable resource to indirectly infer disease burden. The burden of congenital rubella syndrome (CRS) is difficult to measure directly, given the complexity of diagnosis and difficulty of reporting in settings with limited medical resources [39]. Estimates obtained by combining the (1) age-specific force of infection derived from age-specific serological data with the (2) age profile of fertility and risk during pregnancy provide the only estimate of CRS in many countries [40] (see Figure 3B). Finally, serological surveys provide an additional source of population immunity estimates, a necessary line of evidence for measuring progress towards elimination and verifying elimination [41,42]. The WHO Americas Region success in eliminating measles and rubella largely without reliance on serological data to inform vaccination policy is likely to have been a result of the robustness and consistency of their vaccination programs. Other WHO regions may require broader data streams to describe their unique transmission dynamics. In African countries, vaccination coverage varies substantially within and between countries [43], and data on SIA coverage are suboptimal, especially regarding their effectiveness in reaching previously unimmunized persons [12]. Vaccine refusal in European countries, and increasingly elsewhere, can result in patches of susceptible individuals associated with considerable outbreak risk, given high importation rates of infected individuals due to inter-and intramigrations [44]. Seroprevalence profiles are correspondingly hard to anticipate [7,21].

Serological Survey Challenges
High-quality, cross-sectional household serological surveys with blood specimen collection require the following: (1) substantial financial resources and time commitment, (2) logistical capacity and skilled personnel to design and conduct a serosurvey that is generalizable to the target population, (3) laboratories and laboratory expertise to perform serological assays with quality control and assurance, and (4) expertise in statistical analysis to interpret serological data [17]. Thus, serosurveys are typically not prioritized in low-or lower-middle income countries, but there is potential to make them more feasible and sustainable by the following: (1) expanding capacities in conducting high-quality household surveys such as vaccination coverage surveys [11]; (2) including serology in vaccination coverage and/or multipurpose household surveys (as already is done for HIV [45]); (3) expanding existing sentinel site surveillance systems to include measles/rubella serology; (4) expanding the scope of serological surveillance to other vaccine-preventable and emerging infectious diseases (eg, multiplex assays); and (5) standardizing specimen collection, testing, and interpretation of serological results (see Supplementary Materials S4 for further discussion).
The WHO is rolling out updated guidance for the conduct of high-quality vaccination coverage surveys [46]. Established household survey programs such as the Demographic and Health Surveys offer opportunities for inclusion of serology in many countries. However, major barriers remain, including the following: difficulty in ensuring high participation rates, especially for invasive samples (eg, blood, which often has the benefit of reduced laboratory uncertainty) [47]; difficulty in standardizing different laboratory assays [47,48]; and challenges in defining appropriate cutoffs for seropositivity [48] (although expanded deployment of statistical methods such as mixture models could formally address individual variability in cutoffs [49]).
For measles, a cutoff of 120 IU/L is proposed as indicating protection from infection, from plaque reduction neutralization assay results on blood samples obtained before and after a measles outbreak in American college students [50]. This cutoff may not be appropriate (1) if less sensitive, nonfunctional assays such as enzyme immunoassays (EIAs) are used [47] and (2) when seeking evidence of past immunization (in which case a lower threshold may be appropriate), irrespective of whether antibody levels have persisted above the putative fully protective threshold. Likewise, for rubella, antibody levels wane after vaccination and EIAs lead to a high number of false-negative results [48]. Careful examination of the data can lead to a more appropriate cutoff choice, aligned to the survey objectives, rather than using universal cutoffs (eg, 120 IU/L for measles; 10-15 IU/mL for rubella). Much remains to be done to identify, or develop, field-friendly assays that consistently provide readily interpretable data from surveys in low-and middle-income countries [17,25].

CONCLUSIONS
As countries approach measles and rubella elimination goals, the age profile of immunity and the relative contribution of natural and vaccine-derived immunity change. High-quality serosurveys allow explicit characterization of the distribution of immunity at a particular time point, but they also provide mangers with an opportunity to evaluate assumptions made about natural transmission dynamics and vaccine program performance. However, given the costs and challenges inherent in deploying serological surveys, the potential gain in inference should be considered carefully before such surveys are planned. Dynamic models can link serology, clinical, and programmatic surveillance to generate robust estimates of the profile of immunity. Immunity gaps identified through multiple data sources, including serology, can be used to target specific interventions and improve routine programs to prevent future gaps. Finally, investment in serosurveillance for the goal of measles and rubella surveillance could form the foundation of broader serosurveillance efforts as multiplex assays become increasingly available.