Reduced-Dose Schedule of Prophylaxis Based on Local Data Provides Near-Optimal Protection Against Respiratory Syncytial Virus

Infants at high risk for severe respiratory syncytial virus (RSV) infection receive monthly prophylactic injections during the RSV season based on national guidelines. We considered whether a reduced-dose (4-dose) schedule tailored to the local RSV season would provide adequate protection.

Abstract

Background. Respiratory syncytial virus (RSV) is a major cause of respiratory infections among young children and can lead to severe disease among some infants. Infants at high risk for severe RSV infection receive monthly injections of a prophylactic monoclonal antibody during the RSV season based on national guidelines. We considered whether a reduced-dose schedule tailored to the local RSV season in the continental United States would provide adequate protection.

Methods. Hospitalization data for 1942 counties across 38 states from 1997 to 2009 were obtained from the State Inpatient Databases (Agency for Healthcare Research and Quality). We assessed the timing of RSV epidemics at the county and state levels using a 2-stage hierarchical Bayesian change point model. We used a simple summation approach to estimate the fraction of RSV cases that occur during the window of protection provided by initiating RSV prophylaxis during different weeks of the year.

Results. The timing of RSV epidemic onset varied significantly at the local level. Nevertheless, the national recommendations for initiation of prophylaxis provided near-optimal coverage of the RSV season in most of the continental United States. Reducing from 5 to 4 monthly doses (with a later initiation) provides near-optimal coverage (<5% decrease in coverage) in most settings. Earlier optimal dates for initiating 4 doses of prophylaxis were associated with being farther south and east, higher population density, and having a higher percentage of the population that was black or Hispanic.

Conclusions. A 4-dose schedule of prophylactic injections timed with local RSV epidemics could provide protection comparable to 5 doses and could be considered as a way to improve the cost-effectiveness of prophylaxis.

(See the Editorial Commentary by Panozzo and Hampp on pages 515–6.)

Respiratory syncytial virus (RSV) is a leading cause of hospitalizations among young children in the United States and globally [1]. The virus infects most children by the age of 2 years [2]. Premature infants and those with certain underlying cardiopulmonary conditions have a particularly high risk for severe lower respiratory disease [3–5]. No vaccine is available to protect children from RSV, but infants at risk for severe disease can receive monthly injections of a prophylactic monoclonal antibody (palivizumab) during the RSV season [6, 7]. The initiation of the monthly injections needs to be timed so that the child is protected for the duration of the RSV season [8]—starting too late or ending too early will result in the child being unprotected, while doses administered before or after the RSV season are not cost-effective. In the United States, up to 5 doses of palivizumab are recommended for high-risk infants, typically beginning in November; fewer doses may be administered depending on when during the RSV season a child is born [9].

Spatiotemporal variations in RSV epidemics make it difficult to determine when to initiate the monthly series of prophylaxis and to determine how many doses to use [8]. The earliest epidemics in the United States occur in summer in southern Florida and in autumn in northern Florida and other parts of the southeast. In other parts of the United States, epidemics occur later throughout the fall and winter [10–14]. In some regions, epidemics have a biennial pattern with large, early epidemics one year followed by mild, late epidemics the next year [15, 16], which might be linked to the strength of seasonal variations in environmental conditions [17]. Nearby locations can have different epidemic start times [12–14], and epidemic duration can differ with population characteristics [15, 16].

We used data from a comprehensive nationwide hospitalization database to estimate local variations in the timing of RSV epidemics and to determine whether a reduced-dose prophylaxis schedule would provide adequate protection if appropriately timed with local epidemic patterns. Our results, combined with cost-effectiveness analyses, can be used to inform recommendations for the use of RSV prophylaxis across the continental United States.

METHODS

Data Sources on RSV Hospitalizations and Demographics

Weekly hospitalization data were obtained from the State Inpatient Databases of the Healthcare Cost and Utilization Project, maintained by the Agency for Healthcare Research and Quality through an active collaboration. This database contains all hospital discharge records from community hospitals in participating states [18]; we used data from July 1997 to June 2009. Cases were identified by the presence of the RSV diagnostic discharge codes (International Classification of Diseases, Ninth Revision [ICD-9] codes 079.6, 466.11, 480.1) listed anywhere in the patient's record. Weekly time series were created for all RSV cases among children aged 0–23 months. All counties that contributed data for at least 3 RSV seasons (July–June) were included in the analyses. Information about population size and correlates of socioeconomic status (including proportion of the population that is black or Hispanic) for each county was obtained from Census Bureau statistics compiled by the Surveillance, Epidemiology, and End Results program (http://seer.cancer.gov/popdata/singleages.html). The urban–rural classifications of the counties (2006 version) were obtained from the National Center for Health Statistics (NCHS; http://www.cdc.gov/nchs/data_access/urban_rural.htm). All analyses were performed using SAS software, version 9.3 (SAS Institute, Cary, North Carolina) or WinBUGS [19].

Estimating Epidemic Onset and Duration at the County and State Levels

We estimated the start (onset) and end of the RSV epidemics as well as epidemic duration (end week ‒ beginning week) in each July–June season using a 2-stage Bayesian hierarchical model. The first stage consisted of a Bayesian change point model that estimated the start date, end date, and duration of the epidemic separately in each year and county. The second stage was a hierarchical Bayesian model that borrowed information across the estimates from individual locations and years to obtain a stabilized estimate of the epidemic onset and duration in each year and state or county. These models were fit to either state-level or county-level data. The first-stage model was fit using PROC MCMC in SAS version 9.3, and the second-stage models were fit using WinBUGS. Further details on the models are shown in the Supplementary Appendix.

Identifying Optimal Dates for the Initiation of Prophylaxis With 4 or 5 Doses

National recommendations stipulate that 5 monthly doses of RSV prophylaxis should be initiated in southeast Florida on 1 July; in north-central and southwest Florida on 15 September; and in most other areas of the United States on 1 November [7, 9]. Once prophylaxis begins, we assume the period of protection lasts for 24 weeks (5 doses, spaced 1 month apart) [9]. This is a conservative estimate, as the recommendations state that protection lasts for at least 24 weeks [9]. We calculated the percentage of cases occurring within the 24-week window beginning on the recommended start date and compared this to the percentage of cases occurring within an optimal window of protection. We also evaluated the potential coverage of a 4-dose series of monthly shots, which would provide protection for 20 weeks. Therefore, the goal was to identify the 24- or 20-week window in each county that covers the greatest fraction of all RSV cases across all years. We estimated the fraction of all RSV cases that would occur within different 24- or 20-week windows beginning between July and December (95% of epidemics at the state level began before the end of December). The optimal window was considered to be the one that included the greatest fraction of the total cases during the year:

$$\text{Max}\left(\sum _{(j=i)}^{(j=i+23)}(\text{RS}{\text{V}}_j)/\sum _{(j=1)}^{(j=52)}(\text{RS}{\text{V}}_j)\right),$$

where i represents the number of weeks since the beginning of July, ranging from 1 to 26, with the optimal week i being the one resulting in the maximum fraction. This statistic was estimated separately for each July–June period and for each state or county.

Optimal Initiation of Prophylaxis at the State Level

For each state, we calculated the average optimal start week across all available years. The optimal window of protection began in this week and lasted for 24 weeks. We then estimated the percentage of cases that occur during this optimal window in each year and took the mean of this value across all available years. Because some states exhibit biennial epidemic patterns (strong, early epidemic one year followed by a weak, late epidemic the next), we also tested whether the average optimal week differed significantly between odd and even years using linear regression.

Identifying Correlates of Optimal Dates for the Initiation of Prophylaxis at the County Level

We first evaluated the univariate associations between the optimal week for initiating RSV prophylaxis and a variety of demographic and geographic factors, including the proportion of the population that was black or Hispanic (logit transformed), population density (log-transformed), and latitude and longitude of the county (PROC CORR, SAS version 9.3). We also evaluated whether the optimal week differed by the NCHS urban–rural classification scheme (6 levels ranging from large central metropolitan areas to noncore areas) (linear regression, PROC GENMOD, SAS version 9.3). The observations were weighted by the number of cases of RSV occurring in each county and July–June period.

Next, we built a model to estimate the optimal week to initiate prophylaxis in each county based on demographic and geographic characteristics. We randomly selected 80% of the counties with available data to form a training dataset and reserved the remaining 20% as a validation dataset. Using the training dataset, we fit 22 different candidate models, each of which contained a different set of variables (Supplementary Data), including state dummy variable (ie, state average), latitude and longitude (cubic spline), county-level characteristics alone (as in univariate regression) or in combination, a dummy variable for being an odd or even year, and interactions among the variables. Observations were weighted by the number of RSV cases in each county and year. The Bayesian Information Criteria were compared to evaluate model fit. To evaluate predictive performance, we estimated the correlation between the observed values in the validation dataset and the predicted values (weighted by the observed number of RSV cases in each county and year), and we calculated the percentage of all cases that fell within the predicted optimal window.

RESULTS

RSV Hospitalization Patterns

There were 769 301 RSV hospitalizations among children aged 0–23 months that occurred between July 1997 and June 2009 and were captured in our dataset. These data were drawn from hospitals in 1942 counties across 38 states (Supplementary Figure 1). There was an average of 59 381 cases of RSV per year across the available states, drawn from an average population of 5.1 million children aged <2 years.

Variations in Epidemic Onset and Duration at the State and County Levels

Consistent with previous reports, there was considerable variability in the average timing of RSV epidemics between states. The earliest epidemic onsets occurred in Florida, and, in general, the epidemic onsets occurred later in the northern and western states (Figure 1A). The latest epidemic onsets, on average, occurred in Oregon and Maine, 7–10 weeks after the epidemics in the southeastern states (Figure 1A and 1B). Epidemic onset varied substantially between years (Figure 1B).

Figure 1.

A, Average weekly incidence of respiratory syncytial virus (RSV) hospitalizations in each state. The black bar represents the 24-week period when a child would be protected by prophylaxis, based on the national guidelines of 5 doses beginning 1 November. B, Year-to-year variations in the estimated week of epidemic onset at the state level. The states are ordered by latitude, with the southernmost states at the bottom. C, Variations in the estimated week of epidemic onset between counties and years in each state, with each bubble representing an estimate from each county and year. The estimates are obtained using a 2-stage model, where onset is estimated separately for each state/county and year, then entered into a hierarchical model that borrows information between years and geographic areas to obtain stabilized (shrinkage) estimates. The size of the bubbles in (A) and (B) is proportional to the inverse posterior variance of the estimate. Darker shading indicates that multiple bubbles are overlapping.

In addition to the variations in epidemic onset between states, there was variability in the average onset of epidemics at the county level within a state (Figure 1C). Epidemics lasted longer in large central metro counties (16.0 [95% confidence interval {CI}, 15.3–16.7] weeks) and large fringe metro counties (15.8 [95% CI, 15.3–16.4] weeks), compared with medium metro counties (15.5 [95% CI, 15.0–16.1] weeks), small metro counties (15.1 [95% CI, 14.5–15.6] weeks), micropolitan counties (14.1 [95% CI, 14.9–15.9] weeks), and nonurban counties (14.4 [95% CI, 14.0–14.9] weeks). The epidemics in the large central metro, large fringe metro, medium metro, small metro, and micropolitan counties were significantly longer than the epidemics in the nonurban counties (95% credible intervals for differences did not include zero; Supplementary Results).

Optimal Timing for the Initiation of RSV Prophylaxis With 5 Doses

At the state level, the national recommendations provided good coverage of the RSV season—94% of cases that occurred within the optimal 5-dose window of protection also occurred within the window of protection based on the national recommendations (Figure 2A, Table1). Notably, a range of start dates for each state and county would provide near-optimal coverage of the RSV season (Figure 2A and 2B).

Figure 2.

Percentage of cases that occur during the 24-week or 20-week window of protection provided by respiratory syncytial virus (RSV) prophylaxis if the dosing series is started in each week between the beginning of July (week 0) and the end of December. The maximum on each curve indicates the average optimum week for initiating prophylaxis in that state or county. The curves are calculated separately by year and state/county and then averaged across all available years. A and B, Protection provided by 5 doses at the state (A) and county (B) level. C and D, Protection provided by either a 5-dose series (blue) or a 4-dose series (red) at the state level (C) or county level (D). The vertical dotted lines are placed at 1 November, when the national recommendations suggest starting prophylaxis for most states. The horizontal dashed line represents 90% coverage. Only counties with an average of at least 25 cases are shown in (B) and (D). Counties in Florida are excluded from (B) and (D).

Table 1.

Average Optimal Start Date for 5 or 4 Doses of Prophylaxis and the Percentage of Respiratory Syncytial Virus Cases Occurring Within the Window of Protection for Each Schedule

State	Optimal Date of First Dose of Prophylaxisa		Cases Occurring Within Optimal Window of Protectionb, %			Proportion of Cases Covered by 4 vs 5 Doses
	5 Doses	4 Dosesc	National Recommendationd	5 Doses (Optimal)	4 Doses (Optimal)
Arizona	9-Nov	23-Nov	97.3	97.9	95.8	0.98
Arkansas	26-Oct	2-Nov	94.6	95	91.6	0.96
California	2-Nov	16-Nov	94.8	95	92	0.97
Colorado	16-Nov	30-Nov	94.3	96.1	93.2	0.97
Florida	31-Aug	14-Sep	77.7	77.7	70.4	0.9
Georgia	5-Oct	19-Oct	89.3	91.2	86	0.94
Illinois	9-Nov	16-Nov	94.4	94.6	90.9	0.96
Indiana	9-Nov	23-Nov	95.3	95.7	92.6	0.97
Iowa	23-Nov	30-Nov	91	93.4	89.9	0.96
Kansas	16-Nov	30-Nov	93.9	95	92.2	0.97
Kentucky	9-Nov	23-Nov	92.2	92.4	87.9	0.95
Maine	30-Nov	21-Dec	90.7	96.2	92.6	0.96
Maryland	2-Nov	9-Nov	93	93	88.7	0.95
Massachusetts	2-Nov	16-Nov	95.1	95.1	91.5	0.96
Michigan	16-Nov	30-Nov	93	94.7	91.6	0.97
Minnesota	9-Nov	23-Nov	94.3	94.8	91.2	0.96
Missouri	2-Nov	23-Nov	93.9	94.6	90.9	0.96
Nebraska	16-Nov	30-Nov	92.7	94.1	90.1	0.96
Nevada	16-Nov	23-Nov	88.8	89.9	86.2	0.96
New Hampshire	9-Nov	30-Nov	95.9	96	94	0.98
New Jersey	19-Oct	26-Oct	92.4	92.8	88.1	0.95
New York	19-Oct	2-Nov	91.2	91.5	85.7	0.94
North Carolina	26-Oct	2-Nov	93.8	93.7	89.4	0.95
Ohio	16-Nov	23-Nov	94.6	95.7	92.2	0.96
Oklahoma	2-Nov	16-Nov	96.5	96.5	94.2	0.98
Oregon	30-Nov	14-Dec	91.2	95.5	92.6	0.97
Pennsylvania	9-Nov	30-Nov	95.1	96.1	92.6	0.96
Rhode Island	2-Nov	9-Nov	96.4	96.4	92.3	0.96
South Carolina	19-Oct	26-Oct	90.1	90	84.9	0.94
South Dakota	23-Nov	30-Nov	88.9	92.1	86.8	0.94
Tennessee	26-Oct	16-Nov	93.5	93.5	89.1	0.95
Texas	5-Oct	19-Oct	92.4	93.1	87.2	0.94
Utah	23-Nov	7-Dec	93.9	96.4	93.7	0.97
Vermont	23-Nov	7-Dec	94.9	96.6	93.9	0.97
Virginia	26-Oct	9-Nov	92.4	92.4	87.9	0.95
Washington	16-Nov	30-Nov	91.5	93.7	89.5	0.96
West Virginia	16-Nov	30-Nov	92.8	93.7	90	0.96
Wisconsin	16-Nov	7-Dec	95.3	97	94.6	0.97

Analysis based on hospitalization data from the State Inpatient Databases of the Healthcare Cost and Utilization Project (Agency for Healthcare Research and Quality).

^a The week beginning on this date.

^b Average across all years of the percentage of cases that occur during the 24- or 20-week window starting on the optimal start date for 5 doses or 4 doses, respectively.

^c The standard error for the 4-dose optimal start dates ranged from 0.4 to 4.4 weeks, with an average standard error of 1.9 weeks.

^d Date of 1 November, except for Florida (15 September). All dates are based on the 2014 calendar.

Because RSV epidemics have a biennial pattern, we also considered whether the optimal date for initiating prophylaxis differed between odd and even years. Eleven states (Colorado, Iowa, Kansas, Kentucky, Maine, Maryland, Minnesota, Missouri, Nebraska, South Dakota, and Utah) showed some evidence of biennial variations (P < .1 comparing average onset in even and odd years), with 3–5.5 weeks between the optimal date of initiation in even and odd years (Supplementary Table 1).

Effect of Eliminating 1 Dose of Prophylaxis on Protection

We considered whether the use of 4 doses of palivizumab, rather than 5 doses, would provide adequate coverage of the typical RSV season. Across all states and counties, 90%–98% of the cases occurring within the optimal 24-week window of protection also occurred during the optimal 20-week window of protection (Table 2, Figure 2C and 2D). With the reduced-dose series, the optimal start date for initiating prophylaxis would be 1–3 weeks later in the season compared with the optimal start date for 5 doses (Table 2).

Table 2.

Univariate Associations Between Optimal Week for Initiating a 4-Dose Prophylaxis Series and Urban–Rural Characteristics (N = 1407 Counties)

Characteristic	Difference in Optimal Weeka	95% CI
Urban–rural classification
Large central metro	Ref	Ref
Large fringe metro	−0.03	−.21 to .14
Medium metro	0.88	.71 to 1.05
Small metro	2.03	1.84–2.23
Micropolitan	2.86	2.65–3.07
Noncore	4.76	4.44–5.08
Even vs odd years
Odd year	Ref	Ref
Even year	0.29	.18–.41
Correlationb (ρ)
Latitude (north)	0.41	.39–.42
Longitude (east)	−0.23	−.25 to −.21
Percentage black (logit)	−0.43	−.45 to −.42
Percentage Hispanic (logit)	−0.28	−.29 to −.26
Population density (log)	−0.38	−.40 to −.37
Percentage of population <5 y	−0.09	−.11 to −.07

Analysis based on hospitalization data from the State Inpatient Databases of the Healthcare Cost and Utilization Project (Agency for Healthcare Research and Quality).

Abbreviations: CI, confidence interval; Ref, reference.

^a Estimated with linear regression, with a dummy variable for each category.

^b Pearson correlation.

Predictors of Optimal Week of Initiation of RSV Prophylaxis at the County Level

Because not all counties had available hospitalization data (Figure 3A), we developed and validated a model of the optimal timing for the initiation of RSV prophylaxis with a 4-dose series for all counties in the continental United States to obtain an estimate of epidemic timing for counties with and without data (Figure 3B). In univariate analyses, an earlier optimal initiation date was associated with several socioeconomic indicators, including higher urbanization, having a higher percentage of black or Hispanic residents, higher population density, being further south or east, and being an odd vs even year (Table 1). The best multivariate model (Figure 3B) included latitude and longitude of the county (spline), correlates of socioeconomic status (population density, percentage of the population that was Hispanic, percentage of the population that was black), being an odd vs even year, and interaction terms (Supplementary Data). There was a moderate correlation between the optimal date predicted from the best model and the optimal date observed in the training and validation samples (ρ = 0.75 and ρ = 0.65, respectively). If a 4-dose prophylaxis series was initiated based on the predicted optimum start dates in each county, 90.1% of all cases would occur within the window of protection (90.0% in the training set, 90.7% in the validation set), compared with 93.2% if the nationally recommended start date for 5 doses was used (92.9% in the training set, 93.8% in the validation set).

Figure 3.

A, Optimal week for initiating respiratory syncytial virus (RSV) prophylaxis among counties for which data were available. B, Predicted optimal week for initiating RSV prophylaxis in each county in the continental United States, based on county-level characteristics. Darker colors indicate later weeks. The color scale indicates the number of weeks since the beginning of July.

DISCUSSION

We present evidence that despite large variations in RSV epidemic timing across the continental United States, national recommendations for the initiation of a 5-dose series of prophylaxis provide good coverage of the RSV epidemic period in most settings. Moreover, we find that eliminating 1 of the 5 monthly doses and initiating this reduced-dose series later in the season would not result in a substantial decline in protection if the initiation of prophylaxis accounts for variations in RSV epidemic timing between states and counties.

These results build on previous studies of RSV epidemic timing and prophylaxis and make several key advances. A previous study based on laboratory surveillance in 19 select US locations [8] compared the number of weeks a child would be unprotected during the RSV season if prophylaxis was initiated based on the national recommendations vs the average onset date. However, because RSV transmission and disease risk will be greatest at the peak of the epidemic and lower at the beginning and end, simply counting the number of weeks of the RSV season when a child is unprotected can be misleading. Our method allowed us to identify the optimal week for initiating prophylaxis and to demonstrate that a reduced-dose schedule, with a later start date, could provide adequate coverage of the RSV season in most locations. Additionally, our analyses used data from a comprehensive hospitalization database, which provides a more complete picture of the geographic patterns of RSV activity in the United States than analyses based on laboratory data.

Practically, the start dates provided in the tables provide an approximation of the optimal date to initiate prophylaxis at the state or county level. There is some flexibility in when to start prophylaxis, and in most cases, starting the 4-dose series a few weeks earlier or later than the optimal date would have a fairly small impact on protection (Figure 2). Due to the complexity of using county-level recommendations, the optimal start date at the state level (Table 2) would provide a good approximation of the optimal start date at the county level. For states without data, the optimal start dates predicted from the county-level model could provide a reasonable approximation of the best start date (Supplementary Table).

The cost-effectiveness of RSV prophylaxis has been debated [20–22]. To evaluate the impact of changing from a 5-dose schedule to a 4-dose schedule based on local timing, a formal cost-effectiveness analysis is needed that takes into account factors such as the incidence rate in the target population, the efficacy of prophylaxis, and the relationship between infant weight and dosage. However, we can approximate the impact using previous estimates of the incremental cost-effectiveness ratio (ICER). The optimal 4-dose schedule would capture 96.2% of hospitalized cases captured by the optimal 5-dose schedule (Table 2). Assuming that average dosage remains approximately the same, and that the administration and other costs are equally distributed across each dose, the total cost of prophylaxis under the 4-dose schedule should be 80% of what it would be under the 5-dose schedule. Therefore, the total effect of switching from a 5-dose to 4-dose schedule can be approximated by multiplying the ICER by 0.83 (=0.8/0.962). This is unlikely to alter the conclusions of Hampp et al, who estimated the ICER to be US$302 103 among premature infants <6 months of age in the Florida Medicaid population [20]. However, other studies have reported more favorable ICERs (eg, [21, 22]).

Our study is subject to several limitations. The county information assigned to each case is based on the county where the hospital is located rather than the county of residence. This could lead to some misclassification, particularly for rural areas with limited healthcare facilities. Furthermore, we lacked data for states in the Northern Plains and the South; our results may not be generalizable to these regions. Our analyses of epidemic onset assumed that there was a single RSV epidemic in each year. However, in some rural areas, the epidemics could instead be characterized by multiple, independent, and localized outbreaks. This would lead to less precise estimates of epidemic onset, but would not influence our analyses of optimal timing of initiation of prophylaxis. We assumed that a child has equal risk of developing disease outside the window of protection at the beginning or end of the RSV season. However, antibody protection wanes gradually, so an individual might have some residual protection at the end of the RSV season. There was a moderate correlation between the observed and expected optimal start dates estimated from the regression model, but other factors are not accounted for in the model, including waning of population-wide immunity, viral changes, and stochastic introductions. Finally, we relied on ICD-9 codes to define a case as being caused by RSV. This approach might be more sensitive but less specific for detecting RSV cases compared with a definition based on viral testing. However, the strong correlation [23] between hospitalizations coded as “RSV” and those coded as “bronchiolitis” (a syndromic definition) suggests that the epidemic patterns are not due to testing biases.

The key question is when to administer prophylaxis to high-risk infants. Our results suggest that although national recommendations provide good coverage of the RSV season for most US counties, a 4-dose series based on local epidemic timing would perform nearly as well in most settings. Such a change in the dosing schedule would represent a significant cost savings with little effect on the impact of the intervention.

Supplementary Data

Supplementary materials are available at Clinical Infectious Diseases online (http://cid.oxfordjournals.org). Supplementary materials consist of data provided by the author that are published to benefit the reader. The posted materials are not copyedited. The contents of all supplementary data are the sole responsibility of the authors. Questions or messages regarding errors should be addressed to the author.