Keeping Vulnerable Children Safe from Pertussis: Preventing Nosocomial Pertussis Transmission in the Neonatal Intensive Care Unit
Objective. To examine the impact of different acellular pertussis booster vaccination strategies on the probability of a nosocomial pertussis outbreak occurring and the distribution of outbreak sizes observed for each intervention strategy.
Setting. Neonatal intensive care unit.
Methods. We developed a stochastic, agent‐based simulation model to examine the impact of booster vaccination strategies for pertussis on health care–related transmission.
Results. Our results demonstrate that healthcare worker booster vaccination decreases the probability of secondary transmission from 49% (base case, no boosting) to 2% (if 95% of healthcare workers are boosted) and decreases final outbreak size. Boosting family caregivers did not have a clinically significant impact on nosocomial disease transmission.
Conclusion. The provision of booster vaccine to healthcare workers in the neonatal intensive care unit substantially reduces the risk of hospital‐centered pertussis outbreaks in a manner that enhances the health of hospitalized children. A formal health economic analysis of this finding is currently under way. Policies to protect patient safety in pediatric facilities should include compliance with the United States Advisory Committee on Immunization Practices, which recommends provision of pertussis booster vaccination to healthcare workers.
Received December 17, 2008; accepted June 2, 2009; electronically published September 28, 2009.
Pertussis (whooping cough) is caused by the bacterium Bordetella pertussis.1 The pathogen is highly contagious and easily transmitted between individuals by respiratory droplets.2 The disease is characterized by prolonged cough illness that may be accompanied by paroxysms, vomiting, or inspiratory whoop.1,3 Pertussis infection in infants (ie, children aged less than 1 year) can result in significant morbidity, including pneumonia, convulsions, apnea, encephalopathy, acute respiratory distress, and death.4 Premature infants are especially vulnerable to infection as a result of immunologic immaturity.5,6 Routine pediatric pertussis immunization has helped to decrease the disease burden in young children. However, in recent years, pertussis incidence has increased in adolescents and adults as a result of waning immunity. These older cohorts have the potential to transmit pertussis to unimmunized children or neonates.7,8
Many infected adults do not have classic pertussis symptoms, and for this reason, mild or asymptomatic infections in healthcare workers (HCWs) pose the greatest risk to pediatric healthcare facilities. In several reported outbreaks, a mildly symptomatic HCW acted as the index case.2,3,9‐11 The mobility of HCWs in a hospital is expected to make them highly influential for the transmission of respiratory pathogens such as B. pertussis. It is believed that boosting of HCWs would prevent pertussis outbreaks, although empirical data to support this contention are lacking. Although such outbreaks can be successfully contained once they occur, containment requires hours of contact tracing by infection control practitioners, laboratory tests, antimicrobial prophylaxis, droplet precautions, booster vaccination, isolation of patients, and furlough of HCWs.8,12 Explicit recommendations for boosting HCWs in the United States have been made, but such recommendations are lacking in Canadian guidelines.
In this article, we present a theoretical framework to describe pathogen transmission in a neonatal intensive care unit (NICU). There were 3 steps to this research. The first step was to develop a simplified, stochastic simulation model of a NICU. The second step was to use the base model to examine the impact of different acellular pertussis booster vaccination strategies for adults on the probability of a nosocomial outbreak occurring and the distribution of outbreak sizes observed for each intervention strategy. Our hypothesis was that the provision of booster pertussis vaccine to HCWs would substantially reduce the risk of hospital‐centered pertussis outbreaks in a manner that enhances the health of hospitalized children. The third step was to use the data obtained from the model to suggest that policies to protect patient safety in pediatric facilities should include the provision of adult booster vaccination to HCWs.
Methods
Model Description
To examine the impact of implementing a pertussis booster vaccination program in a pediatric healthcare facility, we developed a stochastic, agent‐based simulation model to describe the dynamics of a directly transmitted respiratory pathogen in a NICU (Appendix A). In the simulation, one infected HCW who reported to work with a mild cough illness (pertussis) acted as the index case, and the extent to which secondary transmission occurred was the outcome of interest. Transmission could occur between HCWs during social interactions, between HCWs and patients and/or family caregivers during patient care activities, and between family caregivers and patients during close contact activities. Individual health states included susceptible to infection (S), infected and contagious to others (I), and recovered from infection and immune to reinfection (R). On each day of the 3‐month simulation, patients were admitted and discharged according to hospital administrative data from September 1 through November 30, 2007. HCWs and family caregivers contacted patients a mean of once per hour. Additional parameters used in the model can be found in the Table. The model was developed using AnyLogic (XJ Technologies), a JAVA‐based modeling tool. Model calibration and validation are described in detail in Appendix A.
Intervention Strategies Examined
To evaluate the impact of introducing adult booster vaccination on the spread of pertussis in the NICU, we varied the proportion of HCWs who were susceptible to the infection by adding an additional health state that represented vaccinated individuals. We tested 4 different vaccination strategies for HCWs (25%, 50%, 75%, and 95%) and 2 different vaccination strategies for family caregivers (0% and 100%). Inherent to this strategy is the assumption that family caregivers received booster dosing at least 2 weeks before the admission of their infant to the NICU in order for the caregivers to exhibit full immunity. The success of each intervention was assessed by calculating the probability that the introduction of one index case (HCW) resulted in secondary transmission to patients and by examining the final outbreak sizes.
Study Population
Our model was developed using data from the 38‐bed NICU at the Hospital for Sick Children, a major tertiary care center in Toronto, Ontario, Canada. Our hypothetical study population comprised patients admitted to the NICU from September 1 through November 30, 2007. The mean length of stay for these patients was 12.9 days. Infants are housed in isolettes in 9 rooms with 4 infants per room. Two additional rooms house only one infant per room, for a total of 11 rooms. HCWs in the model did not provide one‐on‐one nursing care (HCWs were randomly assigned to infants at each time step).
Results
Base Model Dynamics
The results of 1,000 simulation replicates of the base model (in which all individuals are susceptible to pertussis infection and no interventions are in place) demonstrated that 38% of the simulations resulted in successful secondary transmission. We found that the probability of an infected HCW causing secondary infections in family caregivers was relatively low (0.15), compared with the probability of secondary transmission occurring in colleagues (0.48) or patients (0.49). In the base model, the largest outbreak observed out of 1,000 simulations was an outbreak that infected 22 HCWs (in addition to the index case) and 15 infants. The probability that secondary transmission occurred in the base model was relatively insensitive to reasonable variation in all parameter values for disease transmission (
, where 
is sensitivity).
Intervention Dynamics: HCW Vaccination
Adding booster vaccination (25%–95%) for HCWs in the model decreased the probability of an outbreak in the patient population from 49% with no vaccine intervention to 32% when 25% of HCWs were vaccinated. Further nonlinear reductions were observed after vaccinating greater proportions of HCWs in the model, until 95% of all HCWs were boosted, resulting in an outbreak probability of 2%. Increasing vaccination also resulted in smaller outbreak sizes when outbreaks did occur (Figure 1). After the vaccination of 95% of HCWs in the model, no outbreak clusters larger than 3 cases of secondary transmission were observed, compared with cluster sizes of 1–13 cases of secondary transmission with only 25% of HCWs boosted (Figure 1).
Figure 1. Bar graphs of proportion of simulation runs of various outbreak sizes predicted by models with an increasing proportion of healthcare workers (HCWs) boosted for each group of individuals in the model. A, 25% of all HCWs boosted (
); B, 50% of all HCWs boosted (
); C, 75% of all HCWs boosted (
); and D, 95% of all HCWs boosted (
). PAR, parent; PT, patient.
Introducing 100% booster vaccination to family caregivers did not result in a clinically significant reduction in the probability of an outbreak occurring in either the HCW (43%) or patient (44%) groups, compared with the probability in the base model. A comparison of median outbreak size and 95% credible intervals for the 4 different levels of HCW vaccination with and without family caregiver boosting directly overlapped one another, indicating that the introduction of family caregiver boosting had little effect on secondary transmission in HCWs or patients.
Sensitivity Analysis
Model outputs were most sensitive to variation in parameters related to transmission from HCWs to patients (Figure 2) and transmission between HCWs (Figure 3), but these effects were attenuated by high levels of vaccination (75% and 95%). The model appeared insensitive to the parameter value for transmission from family caregivers to patients, despite the fact that caregivers had as much or more direct contact with patients than did HCWs. As we increased the probability of transmission from patients to HCWs, we observed very little change in the number of patients who became infected. In all sensitivity analyses, increasing the uptake of booster vaccination in the HCW population decreased the maximum number of patients who became infected.
Figure 2. Graph of best fit lines for the maximum number of patients (PTs) infected for each iteration of the model for each level of the healthcare worker (HCW)– boosting intervention (25%, 50%, 75%, and 95%). Increasing the probability of pertussis transmission from HCW to PT results in increasing numbers of patients becoming infected.
Figure 3. Graph of best fit lines for the maximum number of patients (PTs) infected for each iteration of the model for each level of the healthcare worker (HCW)–boosting intervention (25%, 50%, 75%, and 95%). Increasing the probability of pertussis transmission from HCW to HCW results in increasing numbers of PTs becoming infected.
Discussion
Tremendous gains have been achieved in the control of pertussis in the community since the introduction of vaccination in the late 1940s. However, nosocomial pertussis in pediatric settings, including the NICU, continues to occur, with large associated morbidity and costs. Recommendations for HCW vaccination are not universal, and randomized trial data to support the effectiveness of this approach for outbreak prevention are lacking. We used a modeling approach to project, on the basis of the best available data, the likely impact of routine, proactive HCW booster vaccination for outbreak mitigation and prevention in the NICU setting.
Our model demonstrates a good qualitative and quantitative fit to hospital‐based outbreaks of pertussis that are described in the literature. From a patient safety standpoint, we are interested in both the probability of an outbreak occurring and the size of outbreaks. By means of the best available data, we predicted that moving increasing proportions of HCWs to the model vaccinated class reduces both the likelihood of outbreaks and the final outbreak size in a nonlinear manner. Our results demonstrate a clinically important impact of pertussis boosting in HCWs, even when only small numbers of HCWs are boosted.
To prevent outbreaks entirely, one needs to boost larger proportions of HCWs. Whether such a goal is attainable is likely to be a function of several economic variables, including the cost of ensuring high levels of vaccine coverage and HCW beliefs and attitudes with regard to the importance of vaccination for their own health and that of their patients. However, such considerable effort and resource use may be more than counterbalanced by resource expenditures related to contact tracing, laboratory tests, antimicrobial prophylaxis, healthcare resources for droplet precautions, booster vaccination, isolation of patients, and furlough of staff when nosocomial pertussis outbreaks do occur.3,7,14‐16
Although our results support the position of the US Advisory Committee on Immunization Practices on the vaccination of HCWs, some of our findings were surprising and (to us) counterintuitive. Introducing additional acellular pertussis booster vaccine to all family caregivers of NICU patients produced results that were similar to those observed without boosting this group. These results suggest that although boosting family members in the NICU may provide “cocoon” protection of the infant once the infant is discharged from the NICU and returns home, boosting caregivers in our model did not change the probability of an outbreak occurring or the final outbreak sizes observed, compared with the probability and the sizes in the model that included no caregiver boosting. The high mobility of HCWs and the frequent contacts between HCWs in the context of care provision made this group a more attractive target for booster vaccination than parents, at least in the NICU context and with constrained resources.
Young infants who are susceptible to pertussis can suffer serious morbidity and mortality as a result of the infection. Infants in a NICU are at increased risk of infection because of immature immune systems, long hospital stays, and frequent close contacts with HCWs that are sufficient for the transmission of a respiratory pathogen. Pertussis outbreaks are disruptive, dangerous, and costly in terms of both morbidity and healthcare resources. It would be preferable to avoid rather than to contain these outbreaks. Our findings suggest that the provision of booster vaccine to HCWs substantially reduces the risk of hospital‐centered pertussis outbreaks. A formal health economic analysis of these findings is currently under way. Policies to protect patient safety in pediatric facilities should include compliance with the US Advisory Committee on Immunization Practices, which recommends provision of pertussis booster vaccination to HCWs.
Acknowledgments
Infection control practitioners from Mount Sinai Hospital, The Hospital for Sick Children, and Sunnybrook Health Sciences Centre provided excellent feedback on earlier versions of the model.
Financial support. The Hospital for Sick Children (REB 1000011797); Government of Ontario (Early Researcher Award to D.N.F.).
Potential conflicts of interest. Unrestricted matching funds for the Early Researcher Award to D.N.F. were received from Sanofi Pasteur, which manufactures acellular pertussis vaccine. Personnel from Sanofi Pasteur did not have a role in the formulation, design, or execution of this study or in the drafting of this article. A.L.G. reports no conflicts of interest relevant to this article.
Appendix A
Base model assumptions. Each simulation run of the base model begins with the introduction of one infected HCW into the neonatal intensive care unit and then tracks the spread of the disease during a 3‐month period. A number of simplifying assumptions were made during model development:
| 1. | Each patient is accompanied by a family caregiver who has contact with the patient and healthcare workers (HCWs) for the duration of the hospital stay. | ||||
| 2. | All patients admitted are susceptible to pertussis (S) because of the rapid decay of maternal antibody.5,6 | ||||
| 3. | Family caregivers have direct contact with only their own child, and children do not interact with one another. | ||||
| 4. | When HCWs are not providing patient care, they interact with one another at a central location. | ||||
| 5. | No HCWs or family caregivers have received an adult booster vaccination for pertussis. | ||||
| 6. | Recovery from pertussis results in full immunity (for the duration of the simulation). | ||||
Base model validation. We identified nosocomial disease transmission events documented in the published biomedical literature in which an infected HCW was identified as the probable index case in the pertussis outbreak described (
).2,3,9‐11,15,17‐19 We assumed that the literature values were skewed toward larger outbreaks, because small outbreaks and instances in which an index case resulted in no transmission to the patient population are less likely to be written up for publication (ie, there is publication bias).
To calibrate the base model to the empirical data, we varied the probability of transmission between HCWs and ran 500 replicates for each value, to identify the value that generated a base model that provided the best fit to the empirical data. To compare the distributions of outbreak sizes obtained from the base model simulations to the distribution of outbreak sizes in the literature, we used an Anderson Darling goodness‐of‐fit test.20 This test is used to identify whether a sample is likely to be derived from a specific distribution. If the Anderson Darling statistic is greater than the critical value, then we reject the null hypothesis that the literature data come from the distribution generated by the base model. We used an Anderson Darling statistic because the test emphasizes agreement at the extremes of the distribution (tails) and is suitable for small sample sizes.
We found that increasing the probability of pertussis transmission between HCWs resulted in more outbreaks of pertussis and a gradual increase in the size of the observed outbreaks. Histograms of model‐predicted outbreak sizes for different values of the transmission parameter were described by γ distributions with similar shape parameter (α) values (range, 0.27–0.33). The scale parameter of the distribution (β) increased as the parameter value increased, demonstrating that as the probability of transmission between HCWs increased, outbreak size distributions became more right‐skewed.
We tested the goodness of fit between the γ distributions and determined that simulations that used 0.0006 as the probability of transmission per contact between HCWs provided the best fit to the empirical data (the Anderson Darling Statistic was −6.0203, which is less than the critical value of 2.5018; therefore, we cannot reject the null hypothesis that the data from the model and the data from the literature come from the same γ distribution). For all subsequent analyses, we used the base model parameter values that best replicated the outbreak sizes described in the literature (Table).
Sensitivity analysis. The mean sensitivity of the model was evaluated by using the probability that an outbreak would occur in the patient population as our output of interest, because we are ultimately interested in preventing infections in children. Model sensitivity (
) is defined as the proportional change in the probability of an outbreak (V) for a given parameter value (P) based on pairs of parameter values, with one parameter value being the default value for the base model.21,22 
Sensitivity (
) allows us to identify the relative importance of the input parameters to the output parameter of interest, which in our case is outbreak size in children. Large values of sensitivity indicate that changes in the probability of an outbreak occurring exist in a manner that is greater than linear, and this would be an undesirable characteristic.22 To examine model sensitivity in a more qualitative fashion, we also varied the probability of pertussis transmission from 0.2 per contact (
) to 0.4 per contact (
) to 0.6 per contact (
) for each transmission probability and plotted the maximum number of patients infected per model run. We repeated this method for each of the vaccine intervention strategies (25%, 50%, 75%, and 95%).
References
- 1. Crowcroft NS, Pebody RG. Recent developments in pertussis. Lancet 2006; 367:1926–1936.
- 2. Friedman DS, Curtis R, Schauer SL, et al. Surveillance for transmission and antibiotic adverse events among neonates and adults exposed to a healthcare worker with pertussis. Infect Control Hosp Epidemiol 2004; 25:967–973.
- 3. Baggett HC, Duchin JS, Shelton W, et al. Two nosocomial pertussis outbreaks and their associated costs—King County, Washington, 2004. Infect Control Hosp Epidemiol 2007; 28:537–543.
- 4. O’Brien JA, Caro JJ. Hospitalization for pertussis: profiles and case costs by age. BMC Infect Dis 2005; 5:57.
- 5. Healy CM, Munoz FM, Rench MA, Halasa NB, Edwards KM, Baker CJ. Prevalence of pertussis antibodies in maternal delivery, cord, and infant serum. J Infect Dis 2004; 190:335–340.
- 6. Heininger U, Riffelmann M, Leineweber B, von Koenig CHW. Maternally derived antibodies against Bordetella pertussis antigens pertussis toxin and filamentous hemagglutinin in preterm and full term newborns. Pediatr Infect Dis J 2009; 28:443–445.
- 7. Daskalaki I, Hennessey P, Hubler R, Long SS. Resource consumption in the infection control management of pertussis exposure among healthcare workers in pediatrics. Infect Control Hosp Epidemiol 2007; 28:412–417.
- 8. Weber DJ, Rutala WA. Pertussis: a continuing hazard for healthcare facilities. Infect Control Hosp Epidemiol 2001; 22:736–740.
- 9. Bassinet L, Matrat M, Njamkepo E, Aberrane S, Housset B, Guiso N. Nosocomial pertussis outbreak among adult patients and healthcare workers. Infect Control Hosp Epidemiol 2004; 25:995–997.
- 10. Boulay BR, Murray CJ, Ptak J, Kirkland KB, Montero J, Talbot EA. An outbreak of pertussis in a hematology‐oncology care unit: implications for adult vaccination policy. Infect Control Hosp Epidemiol 2006; 27:92–95.
- 11. Pascual FB, McCall CL, McMurtray A, Payton T, Smith F, Bisgard KM. Outbreak of pertussis among healthcare workers in a hospital surgical unit. Infect Control Hosp Epidemiol 2006; 27:546–552.
- 12. Edwards KM, Talbot TR. The challenges of pertussis outbreaks in healthcare facilities: is there a light at the end of the tunnel? Infect Control Hosp Epidemiol 2006; 27:537–540.
- 13. Arino J, McCluskey CC, Van den Driessche P. Global results for an epidemic model with vaccination that exhibits backward bifurcation. Siam J Appl Math 2003; 64:260–276.
- 14. Calugar A, Ortega‐Sanchez IR, Tiwari T, Oakes L, Jahre JA, Murphy TV. Nosocomial pertussis: costs of an outbreak and benefits of vaccinating health care workers. Clin Infect Dis 2006; 42:981–988.
- 15. Ward A, Caro J, Bassinet L, Housset B, O’Brien JA, Guiso N. Health and economic consequences of an outbreak of pertussis among healthcare workers in a hospital in France. Infect Control Hosp Epidemiol 2005; 26:288–292.
- 16. Davis JP. Clinical and economic effects of pertussis outbreaks. Pediatr Infect Dis J 2005; 24:S109–S116.
- 17. Bryant KA, Humbaugh K, Brothers K, et al. Measures to control an outbreak of pertussis in a neonatal intermediate care nursery after exposure to a healthcare worker. Infect Control Hosp Epidemiol 2006; 27:541–545.
- 18. Vranken P, Pogue M, Romalewski C, Ratard R. Outbreak of pertussis in a neonatal intensive care unit—Louisiana, 2004. Am J Infect Control 2006; 34:550–554.
- 19. Gehanno JF, Pestel‐Caron M, Nouvellon M, Caillard JF. Nosocomial pertussis in healthcare workers from a pediatric emergency unit in France. Infect Control Hosp Epidemiol 1999; 20:549–552.
- 20. Stephens MA. EDF statistics for goodness of fit and some comparisons. J Am Stat Assoc 1974; 69:730–737.
- 21. Webb CT, Brooks CP, Gage KL, Antolin MF. Classic fleaborne transmission does not drive plague epizootics in prairie dogs. Proc Natl Acad Sci USA 2006; 103:6236–6241.
- 22. Keeling MJ, Gilligan CA. Bubonic plague: a metapopulation model of a zoonosis. Proc R Soc Lond B Biol Sci 2000; 267:2219.
-
Presented in part: Understanding and Controlling Infectious Diseases: An Agenda for the 21st Century; Paris, France; 2008 (Abstract P‐32); and Epidemics; Asilomar, California; 2008 (Abstract P1.72).