Review of Mathematical Models of Vaccination for Preventing Congenital Cytomegalovirus Infection

Abstract

Background

Several cytomegalovirus (CMV) vaccine candidates are under development. To reduce the burden of congenital CMV infection, potential strategies under consideration include vaccination of adult women, adolescent girls, and/or young children (both sexes).

Methods

We reviewed 5 studies that used infectious disease modeling to assess the potential impact of vaccination for preventing congenital CMV infection. All models assumed CMV vaccination would prevent primary infection and 2 models also assumed prevention of reinfections and reactivations.

Results

Despite differences in structure, assumptions, and population data, infant vaccination (both sexes) was the optimal strategy in all models, but in 1 model vaccinating seronegative women at 19–21 years of age was also optimal (for duration of vaccine protection ≥8 years). In 3 models, infant vaccination increased average age at primary infection as a result of decreased secondary transmission (herd immunity) combined with waning vaccine-induced immunity. This effect could increase the risk of congenital CMV infections in populations where primary CMV infection occurs early in childhood but could be minimized by administering a second dose of vaccine during adolescence.

Conclusions

Understanding vaccine efficacy and duration of immunity, and how these might vary depending on CMV serostatus and age at vaccination, will be key to defining CMV vaccination strategies.

Congenital cytomegalovirus (CMV) infection is the leading infectious cause of developmental disabilities and sensorineural hearing loss in developed countries, occurring in 2 to 7 per 1000 infants [1, 2]. Congenital CMV infection can occur as a consequence of maternal primary infection, reinfection, or reactivation. In 1999, a CMV vaccine was rated as a highest priority for vaccine development by the Institute of Medicine in the United States [3]. Several CMV vaccine candidates are currently in preclinical or clinical development [4]. Various target groups are under consideration for effective use of CMV vaccines [5, 6], including women of childbearing age, adolescents, and infants. The safety and efficacy of a CMV glycoprotein B (gB) vaccine candidate was evaluated in phase 2 studies among CMV-seronegative women and adolescents [7, 8]; phase 1 studies included infants/young children [9] and CMV-seropositive women [10]. An ongoing phase 2 study among CMV-seronegative women 16–35 years of age will evaluate the safety and efficacy of a replication-defective CMV vaccine candidate, which expresses the gH/gL/pUL128-131 pentameric complex [11].

Published studies using infectious disease modeling to assess the potential impact of different vaccination strategies and target groups for preventing congenital CMV infection and/or associated disability have predicted variable effects in different populations [12–16]. These effects ranged from potential elimination of CMV in the long term to a paradoxical increase in congenital CMV infections in settings where primary CMV infections occur early in childhood [12–16]. In this review, we summarize these CMV vaccination models and describe common findings, potential sources of variation in their predictions, and knowledge gaps that will be important to address to inform the design of future CMV vaccine clinical trials and recommendations on CMV vaccination.

SUMMARY OF CMV VACCINATION MODELS

Five studies published from 2001 to 2015 assessed the potential impact of CMV vaccination for prevention of congenital CMV infection in the United Kingdom [12], Brazil [13, 14], and the United States [14–16]. The key features of each model (Table 1) are useful for identifying potential limitations that should be considered in interpreting their main outcomes (Table 2). The model strata describe the natural history of infection assumed in each model and the mechanism of vaccine protection. Four models included a latent infection strata [13–16] but only 2 included reinfections [14, 15]; 3 models assumed the vaccine would prevent primary CMV infection in susceptible individuals only [12, 13, 16], and 2 models assumed the vaccine would also prevent reinfections and reactivations among latently infected individuals [14, 15].

Table 1.

Description of Key Features of the 5 Models

Features	Griffiths et al 2001 [12]	Azevedo et al 2011 [13]	Lanzieri et al 2014 [14]	Hogea et al 2015 [15]	Alfaro-Murillo et al 2016 [16]
Country	United Kingdom	Brazil	Brazil and United States	United States	United States
Model type	Catalytic	Dynamic transmission	Dynamic transmission	Dynamic transmission	Dynamic transmission
Model strata describing the natural history of infection	Susceptible Infected Immune	Susceptible Incubating Infectious Latent	Susceptible Primarily infected Latently infected Reactivated Reinfected	Susceptible Primarily infected Latently infected ReactivatedReinfected	Susceptible Primarily infected Latently infected
Age structure (age transition of the model population)	No	Not specified	0–12 months, 13–18 months, 19 months to 5 years, 6–9, 10–11, 12–14, 15–19 years, and 10-year age groups from 20 to 49 years	1-year age groups from age >1 to <86 years; 6-month age groups from 0 to 1 year	1-year age groups (individuals remain in the same age stratum until the end of the year and then move into the next age stratum, resembling real-world conditions)
Contact rates	None	Three possible patterns of contact rates were derived from seroprevalence data. Pattern I had peaks in transmission from children to children and from adults to adults; pattern II had peaks in transmission from children to adults and vice versa, with CMV thus not transmitted among children; and pattern III combining the peaks observed in both pattern I and II	Adapted from Azevedo et al [13]; contact rate with best fit assumed higher transmission probabilities from mother-to-infant through breastfeeding and between young children, but lower transmission from children to adults	Based on European survey of contact between persons of different ages [17]	Assumed highest contact rates among newborns and other age groups, decreasing by age until 7 years old, and remaining stable after that
Population parameters	All persons die at 70 years of age	Not specified	Brazil and US birth and death rates to reproduce country-specific age distribution	US census data and age-specific fertility rates	US census data
Infectiousness duration	Constant across all ages	Age dependent	Age dependent ≤5 years of age: 2 years 6–19 years of age: 1 year ≥20 years: 6 months Reduced in reactivated and reinfected individuals	Age-dependent average duration of infectiousness was estimated by the model	3–6 months following primary infection Reduced in individuals with nonprimary infection
Mother-to-fetus transmission rates	Not included	Not included	Not included	Primary infection: 33.4% Nonprimary infection: 8.5% Both estimated by the model	Primary infection: 32.3% [18] Nonprimary infection: 1.4% [18]
Mother-to-infant transmission rates via breastfeeding	Not included	Not included	Specific rate not included, but accounted for in the contact rates	Not included	38.7% [19]
Vaccination strategies	Susceptible individuals At birth	Susceptible individuals 2–6 months 10–11 years 2–6 months + 10–11 years	Susceptible and latently infected individuals 0–12 months 12–18 months 10–11 years 15–19 years 20–29 years (12–18 months + 15–19 years)	Susceptible only or susceptible and infected individuals 6 months 10 years 10–17 years (females only vs males and females; seronegative only vs all; with/without booster dose)	Susceptible individuals identified by serological screening 0 to 35 years (seronegative females in older ages; both sexes for infant vaccination)
Vaccine efficacy	80%–100%	10%, 30%, 50%, 70%	0%–100%	70%, 90%	50%, 75%, 95%
Vaccine coverage	59%–62% (estimated by the model)	90%	0%–100%	30%, 70%	20%, 60%, 90%
Duration of protection	Lifelong	2, 10, 20 years, and lifelong	2–50 years, describes 5 and 20 years specifically	10 and 20 years	4 patterns (half-life): short (4 years), intermediate with gradual decline (8 years), intermediate with steep decline (10 years), long (25 years)
Seroprevalence data used for model fitting	Women in antenatal clinics, United Kingdom, 1975–1985 (n >14 000)	Persons 0–40 years of age, Sao Paulo, Brazil, 1990 (n = 443) [20]	Persons 0–40 years of age, Sao Paulo, Brazil, 1990 (n = 443) [20] Persons 6–49 years of age, United States, 1988–2004 [21]	Persons 6–49 years of age, United States, 1988–2004 [21]	Persons 1–49 years of age, United States, 1988–2012 (unweighted) [22]

Abbreviation: CMV, cytomegalovirus.

Table 2.

Description of Main Outcomes of the 5 Models

Output Parameters	Griffiths et al 2001 [12]	Azevedo et al 2011 [13]	Lanzieri et al 2014 [14]	Hogea et al 2015 [15]	Alfaro-Murillo et al 2016 [16]
Force of infection (seroconversion rate per year)	Constant (3.1%–3.5%)	Age dependent (up to 60% in infants)	Age dependent (not described)	Age dependent (up to 8% in infants; 1.2% among pregnant women)	Age dependent (not described)
R0	2.4–2.7, using constant force of infection	Not described	Brazil = 5.1, United States = 1.9, using next generation matrix	Not described	Not described
Estimated proportions of congenital CMV infections by type of maternal infection	Not assessed	Not assessed	Primary, reinfection, reactivation Brazil: 15%, 38%, 47% United States: 16%, 12%, 72%	Primary: 29% Nonprimary: 71%	Not assessed
Optimal ages for vaccination	Assumed vaccination at birth, did not assess other ages	2–6 months if no immunity waning 10–11 years if shorter duration of protection	0–12 months for both countries 12–18 months + 15–19 years	Infancy	19–21 years Infancy for shorter duration of vaccine protection
Outcome measures	Number of CMV infections in 16–40 year olds, before and after vaccination	Number of congenital CMV infections with and without vaccination	Number and reductions in congenital CMV infections in 10, 20, and 50 years post vaccination CMV birth prevalence	Reductions in congenital CMV infections, 10, 20, 50, and 100 years post vaccination	Number of congenital CMV-associated disabilities averted over a period of 20 years of vaccination
Main effect	Potential CMV elimination	Reduced CMV transmission and increase in the average age at primary CMV infection, depending on duration of vaccine protection	Potential CMV elimination in both countries, with vaccination at 0–12 months even with short duration of vaccine protection	Notable herd protection with infant vaccination, even at low effective coverage (30%)	Faster decline in congenital CMV disability with infant vaccination compared to vaccination at childbearing ages
Potential increase in congenital CMV infections	No	Yes, if vaccine protection wanes before 20 years	Yes, for congenital CMV infections due to primary maternal infections in Brazil No, United States	No	No
Comments		May overestimate potential increase in congenital CMV infections, because the model does not account for type of maternal infection Time to achieve population effect not described	May overestimate prevalence of congenital CMV infection by type of maternal infections because the model does not account for mother-to-fetus transmission rates by type of maternal infection	Model was calibrated to 1.1% prevalence of congenital CMV infection in the United States, which is higher than the 0.5% observed in a recent multicenter screening study	Unweighted CMV prevalence is about 10% higher than weighted estimates, thus diminishing the pool of susceptible for vaccination, further reducing potential impact of vaccination

Abbreviation: CMV, cytomegalovirus.

With the exception of the model by Griffiths et al [12], which was a catalytic model, all were dynamic transmission models [13–16]. Catalytic models assume infection occurs at a constant age- or time-dependent rate, while dynamic transmission models incorporate contact rates and transmission between different age groups [23, 24]. The input parameters, used in the mathematical equations to specify the rate of transition between strata, varied with the natural history of infection assumed in the models. The infectiousness duration was age dependent in 3 models [13–15], as suggested by natural history studies, but constant across all ages in 2 models [12, 16]. Two models included mother-to-fetus transmission rates [15, 16] and 1 model also included mother-to-infant transmission rate via breastfeeding [16]. Vaccination parameters, that is strategy, efficacy, coverage, and duration of protection, are described in Table 1.

Some models provided information on dynamics of CMV infection without vaccination, such as estimates of force of infection (seroconversion rate per year) [12, 13, 15], basic reproduction number (R₀, the number of new infections generated per infectious individual in a totally susceptible population) [12, 14], and proportions of congenital CMV infections by type of maternal infection [14, 15]. The potential impact of vaccination was assessed using the number of CMV infections [12], congenital CMV infections [13–15], or congenital CMV-associated disabilities [16]. One model assumed vaccination of both sexes at birth and did not assess other ages [12]. Vaccination of both sexes during infancy was found as the optimal strategy in all other models [12–16], though in 1 model it was optimal if the vaccine had short duration of protection (4 years) [16]. For vaccine protection ≥8 years, vaccination of seronegative women at 19–21 years of age was the optimal strategy for preventing cases of congenital CMV-associated disabilities [16]. A potential increase in congenital CMV infections was predicted by models that assessed the impact of vaccination in populations with high baseline seroprevalence [13, 14]. Here, we describe the key features and main outcomes of each model.

OVERVIEW OF CMV VACCINATION MODELS

Griffiths et al, United Kingdom

The model by Griffiths et al [12] assumed susceptible individuals acquire a primary CMV infection and become immune for life, without reinfections or reactivations. The model was fitted to CMV seroprevalence data from 2 large antenatal populations (>14 000 women) in the United Kingdom during 1975–1985. The force of infection estimated using the constant force of infection method was 3.1%–3.5% per seronegative persons per year, with an R₀ of 2.4–2.7. Assuming vaccination given at birth would confer lifelong protection against primary CMV infection, the critical vaccination proportion required for eradication of CMV was estimated at 59%–62% for a vaccine with 100% efficacy and 66%–75% for a vaccine with 80%–90% efficacy. Vaccine impact on congenital CMV infections was assessed based on the number of primary infections among persons 16–40 years of age, rather than in pregnant women. Because in this population the median age of CMV infection (29–32 years) without vaccination was greater than the median age at pregnancy (26–28 years), vaccination would be unlikely to result in a paradoxical increase in congenital CMV infections. Although this first model provided encouraging prospects for development of a CMV vaccine, the main limitation was the simplified natural history, assuming lifelong immunity with no reinfection or reactivation following primary CMV infection or vaccination.

Azevedo et al, Brazil

The model by Azevedo et al [13] represented the natural history of CMV infection by using susceptible, incubating, infectious, and latent strata, and assumed vaccination would prevent CMV infections among susceptible individuals, with no efficacy in preventing reinfection or reactivation. The force of infection was estimated to be highest among infants, with a second peak during adolescence. The impact of vaccination was assessed for 3 age groups (2–6 months, 10–11 years, and a combination of both), assuming various levels of efficacy, and a fixed 90% coverage level. Assuming lifelong protection, the optimal age for vaccination was 2–6 months. However, if the vaccine protection lasted <20 years, congenital CMV infections could increase because women could become susceptible to natural CMV infection during their childbearing years. Because this model did not include reinfections, all congenital CMV infections would likely result from primary maternal infections. Therefore, the overall increase in congenital CMV infections might be overestimated because data from other studies in Brazil suggest that about 10% of congenital CMV infections are due to primary maternal infections [25].

Lanzieri et al, United States and Brazil

In the model by Lanzieri et al [14], the natural history of CMV infection was represented by susceptible, primarily infected, latently infected, reactivated, and reinfected strata; vaccination was assumed to prevent primary infections among susceptible individuals, and reactivations and reinfections among latently infected individuals. The population was stratified into broad age groups (Table 1). To understand the impact of CMV vaccination in populations with moderate and high baseline maternal CMV seroprevalence, separate models were built for the United States and Brazil, where maternal seroprevalence is 56% and >95% [21, 25], respectively. The US model was fitted to data on CMV seroprevalence from the National Health and Nutrition Examination Survey (NHANES), 1988–2004 [21], and the Brazilian model was fitted to the data previously used by Azevedo et al [13, 20]. The best fit for both US and Brazilian seroprevalence data was achieved by using contact rates that accounted for higher transmission probabilities from mother-to-infant through breastfeeding and between young children (due to their long duration of viral excretion, high viral titers in body fluids, and high contact rate), but lower transmission from children to adults. The model assumed an age-dependent force of infection, which was reduced for reactivations and reinfections, and estimated the values of R₀ using the next generation transmission matrix method [26]: R_{0 =} 1.9 for the US and R_{0 =} 5.1 for Brazil. The model-estimated proportions of congenital CMV infections by maternal primary infection, reinfection, and reactivation were 16%, 12%, and 72% in the US, and 15%, 38%, and 47% in Brazil, respectively.

The impact of vaccination was assessed for vaccination at 0–12 months of age, 12–18 months, 10–11 years, 15–19 years, 20–29 years, and a combination of 12–18 months and 15–19 years; and for varying vaccine efficacy (0% to 100%), coverage (0% to 100%), and duration of protection (2 to 50 years), in a timeline of 10, 20, and 50 years post vaccination. These models predicted potential elimination of CMV in both countries several decades after vaccination introduction at 0–12 months of age even with short duration of vaccine protection (2.5 years). Considering other childhood vaccination strategies, vaccination at ages 12–18 months with booster vaccination at 15–19 years in both settings would result in 40%–80% reduction in the overall number of congenital CMV infections, if ≥50% of individuals were vaccinated. With vaccination at 12–18 months of age, a potential increase in congenital CMV infections due to primary maternal infection was predicted for Brazil, resulting from an increase in the average age at primary CMV infection. This effect was not predicted for the United States. The main limitations of this model were that it did not account for mother-to-fetus transmission rates by type of maternal infection and the prevalence of congenital CMV infection obtained as a model output was 1%. Data from a recent multicenter US study found the prevalence of congenital CMV infection was 3–5 per 1000 live births [27].

Hogea et al, United States

In the model by Hogea et al [15], the natural history of CMV infection was represented using similar strata to Lanzieri et al [14]. Individuals were stratified into 1-year age groups, and 6-month age groups for <1 year of age. The model was fitted to data on CMV seroprevalence among persons 6–49 years of age from the NHANES, 1988–2004 [21], and used data on contact rates for air-borne infections from a European survey [17]. The model included mother-to-fetus transmission rates of 33.4% for primary infection and 8.5% for nonprimary infections, and the prevalence of congenital CMV infection was fitted to 1.1%. The model estimated age-dependent duration of infectiousness, transmission, and reactivation rates, with the highest force of infection among children 0–2 years of age and a secondary peak among persons 20–40 years of age. The annual seroconversion rates were estimated at up to 8% among infants and 1.2% among pregnant women. An estimated 29% of congenital CMV infections were due to primary maternal infections and 71% to maternal reinfections and reactivations together.

The impact of vaccination was assessed for various scenarios, with vaccination at 6 months, 10 years, and 10–17 years, for 70% and 90% efficacy, 30% and 70% coverage, and 10 and 20 years duration of protection, in a timeline of 10, 20, 50, and 100 years post vaccination. The vaccination strategies included females only, both sexes, CMV-susceptible individuals only, and all individuals regardless of CMV serostatus. The optimal strategy was vaccination of all infants at 6 months (without booster or catch-up), which resulted in substantial indirect (herd) protection, even at low coverage (30%). None of the scenarios predicted a potential increase of congenital CMV infections in the United States.

Alfaro-Murillo et al, United States

Different from other models described above, the model by Alfaro-Murillo et al [16] evaluated the optimal age for CMV serostatus screening followed by vaccination of CMV-seronegative females to prevent congenital CMV-associated disability. Individuals were stratified into susceptible, primarily infected, and latently infected strata, and 1-year age groups. The model was fitted to unweighted data on CMV seroprevalence among persons 1–49 years of age, from NHANES, 1988–2012 [22]. The contact rate was assumed to be highest between newborns and all other age groups. The contact rate decreased from immediately after the newborn period through 7 years of age, after which it remained stable. The model assumed a 3 to 6-month duration of infectiousness following primary CMV infection. Mother-to-fetus transmission rates were assumed to be 32.3% for primary infection and 1.4% for nonprimary infections, regardless of reinfection or reactivation [18], and 38.7% transmission rate via breastfeeding [19]. The model assumed 18% of newborns with congenital CMV infection would develop disabilities [28]. It is likely that most cases of disabilities would result from primary maternal infections based on the assumptions of mother-to-fetus transmission rates by type of maternal infection [18].

The impact of CMV serostatus screening followed by vaccination of CMV-seronegative females was assessed for vaccination between 0 and 35 years of age, and for various levels of vaccine efficacy (50%, 75%, and 95%), screening coverage (20%, 60%, and 90%), and duration of protection (short, 4 years; intermediate with gradual decline, 8 years; intermediate with steep decline, 10 years; and long, 25 years). The optimal age for screening and vaccination was 19 to 21 years, which predicted the largest reductions in cases of congenital CMV-associated disabilities over 20 years of vaccination (0.3% to 4%); likely because women 20–34 years of age have the highest birth rates [29]. For short duration of protection, infant vaccination would be the optimal strategy and would result in a faster decline in congenital CMV-associated disability than vaccination at childbearing ages. While the duration of vaccine protection should be relatively long to confer direct protection to women in their childbearing years, infant vaccination would provide substantial indirect protection even with relatively short duration of protection. None of the scenarios predicted a potential increase of congenital CMV-associated disabilities. The potential impact of vaccination could have been underestimated—by using unweighted CMV prevalence, which produces slightly higher estimates, the pool of susceptibles for vaccination was likely smaller.

DISCUSSION

Mathematical models of CMV infection in the United Kingdom, Brazil, and the United States help elucidate the dynamics of CMV transmission in populations with moderate and high baseline seroprevalence, and facilitate assessment of the potential impact of vaccination for prevention of congenital CMV infection or associated disability. In this review, we described the key features and main outcomes of 5 CMV models, which represented the natural history of infection in variable levels of complexity to determine optimal age and groups to target for CMV vaccination, vaccine efficacy and duration of protection, and critical level for vaccination coverage. Despite differences in structure, assumptions, and population data, in 4 models infant vaccination (both sexes) was the optimal strategy for reducing congenital CMV infections. In the fifth model, the optimal strategies for preventing cases of congenital CMV-associated disabilities were either infant vaccination if duration of protection was 4 years, or vaccination of seronegative women at 19–21 years of age if duration of protection was ≥8 years [16].

Assuming a CMV vaccine would prevent infection, all models found that vaccination at young age would have one of the highest predicted impacts as a result of direct vaccine protection and indirect vaccine protection from reduced transmission (herd immunity). However, reduced transmission combined with waning vaccine-induced immunity could increase the average age when persons acquire a primary natural CMV infection. In populations where primary CMV infection occurs early in childhood, such as in Brazil, this age shift in CMV susceptibility to childbearing age increased the risk of congenital CMV infections. This phenomenon has been well documented for congenital rubella syndrome in countries with low rubella vaccine coverage [30, 31]. In 1 model, the potential increase in congenital CMV infections in Brazil was only in those resulting from primary maternal infection, which accounted for about 10% of all congenital CMV infections in a large Brazilian study [25], and 15% based on the model estimates [14]. A second dose of CMV vaccine given during adolescence could minimize this effect [13, 14]. In the future, observational studies to assess the impact of vaccination on herd immunity and any changes in transmission patterns would be warranted.

Comparing predictions from models structured in different ways and that used data from different countries is challenging. Variability in the findings between countries may result from a combination of factors, including differences in underlying population dynamics of infection and model structure and assumptions, all of which are difficult to assess. By using the same model structure and assumptions, and country-specific population data, the Lanzieri models provide useful insight into the potential impact of vaccination in populations with high (Brazil) and moderate (United States) CMV seroprevalence [14]. It is reasonable to assume that the variability in the predictions for the 2 countries results from underlying population differences in the force of infection. The R₀ for Brazil (5.1) was much higher than for the United States (1.9) [14]. Nonetheless, it is important to note that these models were fitted to the overall seroprevalence data, which may obscure significant differences across population subgroups.

The prevalence of congenital CMV infection in the United States differs across racial groups [27]. Congenital CMV infection disproportionately affects black and multiracial infants [27], but it is not known if long-term sequelae and CMV-associated mortality differ. One model found that the optimal age to screen and vaccinate CMV-seronegative women in the United States was 19–21 years. However, age-specific CMV seroprevalence among US women varies substantially [21]. While 34% of non-Hispanic white females are CMV seropositive by 12–19 years of age, 57% of non-Hispanic blacks and 70% of Hispanics are CMV seropositive in the same age group [21]. Incorporating race/Hispanic origin in dynamic transmission models would be challenging but seems to be an important step for assessing the optimal age for screening and vaccination.

In populations with high maternal CMV seroprevalence, repeated reinfections with CMV among mothers [32] may pose a risk to the fetus. However, primary CMV infection during pregnancy is reported to pose a greater risk to the fetus [33]. The models described in this review were based on the assumption that a CMV vaccine would prevent infection, thereby preventing further transmission. No model assessed the impact of a vaccine that would not prevent CMV infection but rather prevent mother-to-fetus transmission of CMV.

Infectious disease modelling can inform the design of future clinical trials and recommendations on CMV vaccination. The existing models indicate that understanding vaccine efficacy and duration of immunity, and how these might vary depending on CMV serostatus and age at vaccination, will be key to defining CMV vaccination strategies. Clinical trials designed to assess vaccine efficacy among women of childbearing age, adolescents, or young children have unique challenges [5]. The finding of a potential increase in congenital CMV infections that are due to primary maternal infections in settings with high CMV seroprevalence highlights some of the main knowledge gaps in the natural history of CMV infection that need to be addressed in future studies. The frequency of reinfections and reactivations in specific populations, the mother-to-fetus transmission rates that follow these type of infections, and their relative contribution to the burden of congenital CMV disease and associated sequelae are unknown [34] and need to be elucidated.

Notes

Acknowledgment. We thank Mary Ann Hall for her editorial assistance and thoughtful review of the paper.

Disclaimer. The findings and conclusions in this article are those of the authors and do not necessarily represent the official position of the Centers for Disease Control and Prevention.

Supplement sponsorship. This supplement was sponsored by NIAID and NICHD.

Potential conflicts of interest. All authors: No reported conflicts of interest. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.