Data Sources

RCAP cases were diagnosed at the Soroka University Medical Center, the only medical center in southern Israel, as described (13,14). The total population of children <3 years of age in the region in 2009 was ≈45,000, of whom 23,439 were Jewish children and 21,596 were Bedouin children. A case was defined as chest radiographic evidence of alveolar pneumonia in a patient <3 years of age who was seen in the pediatric emergency department or admitted to the hospital and in whom RCAP had not been diagnosed within the previous 28 days. Chest radiographs were analyzed according to the World Health Organization Standardization of Interpretation of Chest Radiographs Working Group (15). All readings were collected daily and read by 2 pediatric infectious diseases specialists (D.G. and R.D.) and independently confirmed by a pediatric radiologist, as described (13,14). The cases were aggregated into weekly time series for July 2004–June 2012 and stratified by ethnicity (Jewish vs. Bedouin) and by age group (<6 months, 6–17 months, and 18–35 months). Population size for each stratum was estimated on the basis of the number of births among the Jewish and non-Jewish populations in each calendar year in Beer-Sheva (16).

Uptake of PCV7 and 13-valent PCV (PCV13) was determined from an ongoing hospital survey starting in July 2009, when Israel’s national immunization program began covering PCV7. We enrolled the first 4 Jewish children and the first 4 Bedouin children <5 years of age seen on each working day at the pediatric emergency department of the Soroka Medical Center whose parents consented. The children’s vaccinating centers provided data on birth date, vaccination date, and whether the vaccine was PCV7 or PCV13. More than 95% of vaccinating centers responded to the questionnaires. During July 2009–April 2012, a total of 2,555 Jewish children and 2,666 Bedouin children were enrolled. Before the survey, no data were available on vaccine uptake. Health maintenance organizations (HMOs) began covering PCV7 in summer 2008; uptake among Jewish children was moderate (up to 25%) but was lower among Bedouin children. For the period before mid-2008, we assumed no vaccine uptake in either ethnic group. To fill in the values from the third quarter of 2008 through the second quarter of 2009, we assumed that coverage increased linearly from 0 in 2008 to observed vaccine uptake in 2009. We calculated the uptake in each quarter of the year and assigned this coverage to all weeks in that period.

Starting in July 2004, nasopharyngeal wash specimens were obtained from children during working days for virus testing. The decision to obtain nasopharyngeal wash specimens was made by the treating physicians and driven by clinical judgment. Specimens were obtained mostly from children admitted to the hospital and were processed in the virology laboratory within 6 hours. All samples were tested for RSV; influenza viruses A and B; and parainfluenza viruses 1, 2, and 3 by using either PCR or direct immunofluorescence antibody testing and commercial monoclonal antibodies (Chemicon, Temecula, CA, USA) supplemented by tissue culture (17,18). Weekly RSV and influenza counts were divided by the total number of swabs tested in that July–June year to adjust for biases in testing. These calculations were performed separately for Jewish and Bedouin children.

As an alternate measure of RSV activity, we used the weekly incidence of bronchiolitis among Jewish or Bedouin children. For this purpose, we retrieved from the hospital computer database all hospital visits (inpatients and outpatients) of children <3 years old for whom the final diagnoses included “bronchiolitis.” Monthly climate data on temperature, humidity, precipitation, and wind speed were obtained from the Central Bureau of Statistics (16) for the Beer-Sheva West meteorologic station.

Estimating Vaccine Impacts

Figure 1

Figure 1. . PCV uptake and decline in RCAP incidence 2004–05 through 2011–12, southern Israel. A) Uptake of >2 PCV doses among Bedouin and Jewish children 6–17 months of age. Black line...

Unadjusted incidence rate ratios were calculated by determining the incidence in a given year (July–June) or given season (November–April or May–October) and dividing it by the mean incidence of the corresponding prevaccine period. To estimate the effects of RSV, influenza, and vaccine uptake on RCAP incidence, we fit a Poisson regression (log-link) model using PROC GENMOD in SAS version 9.2 (SAS Institute, Cary, NC, USA). The outcome variable was the weekly incidence of RCAP. The predictor variables were weekly activity of influenza and RSV among Jewish and Bedouin children (as described above), an indicator variable for ethnicity; sine and cosine terms that had frequencies of 52.25 and 26.125 weeks and varied between ethnic groups; and a binary variable that indicated whether the PCV vaccination program in the Jewish or Bedouin population had reached maturation. The vaccination program was considered to be mature when at least 85% of the Jewish or Bedouin populations 6–17 months of age had received at least 2 doses of any PCV (85% was chosen to indicate an immunization program with stable rates [Figure 1]). We also considered using vaccine uptake as a continuous variable in the model, but the relationship between vaccine uptake and incidence appeared to have a threshold effect, as might be expected given the role of herd immunity (Technical Appendix Figure 1).

This model assumes that the effects of the viruses and vaccination are the same for both ethnic groups but enables the seasonal baseline to vary in timing and intensity between ethnic groups. Periodic variables were selected on the basis of Akaike Information Criteria in preliminary fixed-effects analyses, and other variables were included if they were significant and improved the Akaike Information Criteria score. Interaction terms between ethnicity and the virus variables or vaccine uptake variable were tested but not included in the final model. The distribution of the deviance residuals was approximately normal for all strata except for the <6-month-old Jewish children (where the data were sparse).

We determined 95% CIs for the parameters using an 8-week moving-block bootstrap (n = 1,000) to account for autocorrelation in the time series (19,20). The incidence rate ratio associated with high vaccine uptake in each age group was estimated by taking the exponent of the regression coefficient for the vaccine uptake variable. The percentage change was estimated as 100 × (incidence rate ratio – 1); a negative percentage indicates a decline in incidence. The effects of influenza and RSV were calculated by fitting the model and then obtaining predicted values by substituting either the observed value for RSV or influenza or 0 into the equation. The difference between the estimates constituted the virus-attributed incidence, and the attributable percentage was the virus-attributed incidence divided by total estimated incidence. To validate the approach, we fit an alternative model in which the 2011–12 season was excluded or a model that excluded RSV and made predictions for the 2011–12 season. We also tested for variations in climate across the study period, such as monthly temperature and precipitation, that might be associated with the observed declines in RCAP incidence.

Changes in Alveolar Pneumonia Incidence, 2004–2012

We first evaluated whether the incidence of RCAP changed after PCV was introduced into the national immunization program. RCAP incidence decreased significantly in 2011–12 compared with the prevaccine era, by ≈49%, ≈46%, and ≈37% among Jewish and Bedouin children <6 months, 6–17 months, and 18–35 months of age, respectively (Figure 1, panel B). For the preceding year, 2010–11, RCAP incidence declined modestly among 18–35-month-old Jewish children (28%, 95% CI 11.0%–41%), Bedouin children overall (17%, 95% CI –1% to 32%), and 6–17-month-old Bedouin children (17%, 95% CI 2%–30%). However, RCAP incidence did not decline among Jewish children <6 months or 6–17 months of age, despite high uptake of the vaccine.

We also compared changes in incidence during winter months with changes in incidence during summer months, when there is little respiratory virus activity. RCAP incidence declined significantly in the summer of 2010 compared with the prevaccine era in the Jewish population but was similar to baseline levels during the following winter (Technical Appendix Figure 2, panel B). Among Bedouin children, where vaccine uptake was slower (Figure 1, panel A), RCAP incidence did not decline significantly until summer 2011.

Association between RCAP Incidence and Activity of Respiratory Viruses

We next considered whether year-to-year variations in virus activity (Technical Appendix Figure 3) might influence estimates of vaccine impact and perhaps mask the effect of the vaccine during the first few years or lead to overestimation of vaccine effect during the weak RSV and influenza season of 2011–12. To evaluate the relationship between respiratory virus activity and RCAP incidence, we fit a regression model that accounts for virus activity, vaccine uptake, and baseline seasonal variations. RSV and influenza were associated with significant increases in RCAP incidence; RSV was associated with an especially strong increase. A total of 44% (95% CI 36%–52%), 21% (95% CI 15%–25%), and 16% (95% CI 11%–20%) of RCAP cases were attributed to RSV among children <6 months, 6–18 months, and 18–36 months of age, respectively (Technical Appendix Figure 4). Influenza was more modestly associated with RCAP incidence, with 1% (95% CI −6% to 6%), 4% (95% CI 1%–6%), and 2% (95% CI 1%–4%) of disease attributed to influenza among children <6 months, 6–18 months, and 18–36 months of age, respectively (Technical Appendix Figure 4).

When we controlled for virus activity, vaccination (>85% uptake) was associated with a significant decline in RCAP incidence of −31% (95% CI −51% to −15%), −41% (95% CI −52 to –32%), and −34% (95% CI −42% to −25%) among <6-month-old, 6–18-month-old, and 18–35-month-old children, respectively. Results from a model in which bronchiolitis hospitalizations, rather than RSV positivity, was used as a predictor gave similar results through the end of 2011 (data not shown). We also considered the possibility that the low RCAP seasons might be associated with anomalous climatic conditions but did not detect any significant differences in monthly temperature, humidity, precipitation, or wind across the study period that might explain the observed patterns; these variables were not included in the final model.

To evaluate the sensitivity of our results to the RCAP incidence in 2011–12, we refit the model to data from 2003–04 through 2010–11 and extrapolated the change in incidence for 2011–12. The model accurately predicted the observed change in RCAP incidence in 2011–12 for all age groups (Figure 2).

Importance of Including RSV and Seasonal Variables in the Analyses

As a sensitivity analysis, we tested whether inclusion of RSV in the model influenced the estimates of vaccine impact. When fitting the model to the entire dataset (2004–05 through 2011–12), removing RSV from the model had little or no effect on estimates of vaccine impact in any of the age groups (Table 2). However, when we fit the model to data only from 2004–05 through 2010–11 (including the first 2 years of PCV7 use) and then evaluate the impact of excluding RSV from the model, we found that RSV substantially influenced the estimates in children <6 months old (Table 2). Removing RSV from the model modestly affected the estimate of vaccine impact among the 6–17-month-old children and had no detectable on the estimates among the 18–35-month-old children (Table 2). This finding suggests that RSV masked the impact of vaccination on RCAP incidence in young infants in the first full year after vaccine introduction.

We also fit a simple model that did not include variables for virus activity or seasonal (harmonic) terms. Again, when the model includes data through the 2011–12 season, the omission of these variables had little impact on the estimates of vaccine impact (Table 2). However, when the model was fit just to data through the 2010–11 season, the estimate of vaccine impact was closer to 0, especially among younger children.