This analysis is part of a larger project describing the impact of PCV dosing schedules on IPD, immunogenicity, nasopharyngeal carriage, pneumonia and indirect effects.^5–8 Details on the literature search terms and methods used in this systematic review are described in the Methods Appendix.⁹ In brief, a systematic literature review was performed to collect all available English language data published from January 1994 to September 2010 (supplemented post hoc with studies from 2011) on the effect of various PCV vaccination schedules among immunized children on immunogenicity, NP colonization, IPD, pneumonia and on indirect effects among unvaccinated populations. Articles published in 14 databases, from ad hoc unpublished sources and abstracts from meetings of the International Symposium on Pneumococci and Pneumococcal Disease (1998–2010) and the Interscience Conference on Antimicrobial Agents and Chemotherapeutics (1994–2010), were searched. We included all randomized-controlled clinical trials, nonrandomized trials, surveillance database analyses and observational studies of any PCV schedule on 1 or more outcomes of interest. Studies were included for abstraction if PPV was used as a booster dose, but not as a primary dose. Titles and abstracts were reviewed twice and those with relevant content on 1 of the 5 outcomes (immunogenicity, carriage, invasive disease, pneumonia and indirect effects) underwent full review using a standardized data collection instrument. Details on the search methods are provided in the Methods Appendix.⁹

Citations recovered through the literature search went through several stages of independent review to determine their eligibility, as described earlier. Citations meeting inclusion criteria were categorized on an outcome-specific basis into “study families,” where each family included abstracts or publications generated from a single protocol, population, surveillance system or other data collection system relevant to that outcome. Investigators identified primary data from the individual studies making up each study family for inclusion in the analysis. The primary data were selected as the most current and complete data available for that study family. In some cases, these data were drawn from more than one publication within a family. We also defined “study arms” as a group of children distinguished by immunization schedule or PCV product.

We abstracted core information on the following: number of children in a “study arm”; PCV manufacturer, valency and conjugate protein; coadministered vaccines; country; age at each dose and date of study and publication. Additional data abstracted for the immunogenicity outcome included antibody levels; age at each dose and blood draw and antibody assay methods. Geometric mean antibody concentration (GMC) with confidence interval was abstracted for immunoglobulin G (IgG) antibody determined by enzyme-linked immunoabsorbent assay (ELISA). We also abstracted the percentage of children with serotype-specific IgG concentration >0.35 µg/mL [or >0.2 µg/mL if the GlaxoSmithKline ELISA method was used (GSK, Middlesex, United Kingdom)], defined previously as the correlate of observed efficacy for VT-IPD across the randomized trials.^10–13 Results of other assays were abstracted if performed, such as opsonophagocytic assay (OPA) and avidity. If age at vaccination for PCV product or coadministered vaccines were not described, recommended ages for receiving doses from that country’s national immunization plan for the year when the study was conducted were used.

Study arms meeting the following criteria were included in analyses: subjects immunized with at least 2 doses of PCV, with the first dose ≤4 months and last primary dose ≤6 months of age; licensed PCV with 7 or more antigens, or unlicensed PCV but sufficiently similar to a licensed PCV, and ELISA IgG GMC or percentage >0.35 µg/mL (or >0.2 µg/mL if GSK ELISA method used) provided for any of the 6 serotypes of interest (1, 5, 6B, 14, 19F and 23F). Data following immunization with PPV23 were excluded. Results from populations at high risk for PD were excluded (ie, HIV-infected, sickle cell disease, those with chronic illness and indigenous populations). Antibody responses to serotypes 1 and 5 were excluded from analyses for PCVs not containing these serotypes (eg, PCV7, PCV8). Antibody responses to the Aventis PCV11 were substantially higher than those of other PCV products and studies of this product were limited to a 3-dose primary series schedule; therefore, we excluded studies of Aventis PCV11 because they obscured our ability to assess the effect of dosing schedule on the antibody response.

Any study arm with immunogenicity data after a second primary dose, including study arms eventually receiving a third primary dose, was defined as “2 primary doses”; “3 primary doses” was defined as any study arm with immunogenicity data after a third primary dose, whether a booster dose was given. Schedules “2+0” and “2+1” refer to 2 primary doses without and with 1 PCV booster dose, respectively; “3+0” and “3+1” schedules refer to 3 primary doses without and with 1 PCV booster dose, respectively. A booster dose was defined as immunization with PCV between 9 and 18 months of age in infants who had completed a 2-dose or 3-dose primary series. Mean age at immunization, if available, defined age at each dose; otherwise, scheduled age was used. To collapse into schedules, age was rounded to the nearest 2 weeks. Interval between doses was determined by number of months between first and second primary dose; all but 1 study retained the same interval between first and second as between second and third primary doses.

Because most studies did not directly compare 1 PCV dosing schedule to another, we followed an ecological regression approach to compare dosing schedules across studies. We fitted random effects meta-regression models of log-transformed GMC levels by serotype, weighting by the inverse of their variances, to account for intrastudy variation for studies assessing more than one study arm. We calculated robust standard errors to account for multiple arms within studies.^14,15 We adjusted for study-specific covariates: age at first dose, geographic region, PCV product, type of coadministered diphtheria-tetanus-pertussis (DTP) vaccine [ie, whole-cell (DTwP) vs. acellular (DTaP) pertussis], and ELISA method. For comparisons of immunogenicity at different time points, such as postprimary versus postboost and preboost versus postboost, only studies with data for both time points were included in the analyses. For studies that did not report the variance of the GMC (between 10% and 18% among analyses), we assigned the median of the variances reported by studies with the same region, coadministered DTP vaccine and age at first dose. To investigate the effect of this simple imputation approach, we conducted analyses with and without the studies missing variance; there was no association between absence of variance estimates and GMC.

To calculate 95% confidence intervals for the proportion of children with antibody concentrations above the cut-off, we used the normal approximation interval with proportion (p)=(X+2)/(n+4), a near approximation to the Wilson interval method, which is valid for extreme values (ie, p = 100%).^16,17

Statistical significance was defined as P < 0.05; there was no adjustment for multiple testing. SAS version 9.2 (SAS Institute Inc., Cary, NC) was used for all analyses.

Of 61 studies included in the 2-dose versus 3-dose primary series analysis, 6 directly compared these different schedules and 9 additional studies provided after dose 2 results for 13 study arms eventually receiving a third primary dose. GMC results were available for 25 two-dose and 86 three-dose study arms (Table 1). The 3-dose primary schedule produced significantly higher postprimary GMC antibody response than the 2-dose schedule for all except serotype 1 (Fig. 1A,B). However, the number of primary doses did not meaningfully affect GMCs measured in the second year of life [ie, prebooster (Fig. 1B) or postbooster GMCs (Fig. 1C)], except the postboost response to serotype 6B, which was significantly higher in children who received 3 primary doses.

We evaluated the age at first dose, interval between doses and age at last dose, while adjusting for whether 2 or 3 primary doses were administered; albeit, each is a factor of the other 3 so they cannot be disentangled entirely. Administering the first dose at 1 month of age (eg, at 3 months vs. 2 months of age) did not meaningfully affect the postprimary GMCs (Fig. 1D). However, increasing the interval between doses from 1 to 2 months significantly increased the postprimary GMCs for serotypes 6B, 14 and 23F (Fig. 1E). Increasing the age at last dose had a similar effect to increasing months between doses although the effect size was smaller and did not reach statistical significance (Fig. 1F). Among the studies in the meta-analysis, only one directly compared interval of dosing while holding age at first dose constant, and it did so in a randomized trial setting.³³ That study’s findings support the meta-analysis results; it showed that children randomly assigned to a 2-month interval between 2 primary doses had significantly higher postprimary GMCs for serotypes 6B, 14 and 23F and preboost GMCs at 1 year of age for serotypes 4, 6B and 23F compared with children who received their 2 primary doses with 1-month interval.

Of 53 study arms evaluating a PCV booster dose (generally in the second year of life), 11 (21%) had received a 2-dose primary schedule and 42 (79%) a 3-dose primary schedule. The booster dose increased GMCs from pre- to postbooster for all serotypes regardless of the priming schedule (Fig. 3A). Postbooster GMCs (mean age 13.5 months) were also significantly higher than postprimary GMC (mean age 7.2 months) for all serotypes regardless of the priming schedule (Fig. 3B).

There were 58 studies with immunogenicity data on 2+1 or 3+0 schedules; GMC results were available for 86 “3+0” and 11 “2+1” study arms. The unadjusted median postbooster GMC response (in µg/mL) of 2+1 schedules (median age at blood draw 12.5 months) compared with the postprimary response of 3+0 schedules (median age at blood draw 7.0 months) was 5.1 versus 2.5, 3.4 versus 3.0, 7.4 versus 1.4, 12.0 versus 4.7, 7.7 versus 3.6, 4.5 versus 1.7 for serotypes 1, 5, 6B, 14, 19F and 23F, respectively. After adjusting for geographic region, age at first dose, coadministration of DTaP versus DTwP, PCV product and ELISA laboratory method, the mean 2+1 postbooster antibody response (median age at blood draw 12.5 months) was significantly higher than the 3+0 postprimary response (median age at blood draw 7.0 months) for all serotypes (Figs. 4 and 5), although the covariates are tightly correlated and so confounding might not be entirely controlled for.

The percentage of children with IgG antibody concentrations >0.35 µg/mL (or >0.20 µg/mL for GSK ELISA) was high and comparable for both 2-dose and 3-dose primary schedules for all serotypes except for 6B and 23F (Fig. 6). For these serotypes, the proportion of children above the threshold was higher for those receiving a 3-dose than a 2-dose primary schedule; 17/24 (70.8%) 3-dose study arms reported over 80% of subjects above the cut-off for serotype 6B compared with 3/13 (23.1%) 2-dose study arms. For serotype 23F, results were 21/24 (87.5%) compared with 4/11 (36.4%), respectively.

Only one 2+1 study presented percentage response data following the booster dose, which showed >96% response (IgG antibody concentrations ≥0.2 µg/mL) using the GSK ELISA for all serotypes except for 6B which had 88% response; response was also high (>87% for all serotypes) using the 0.35 µg/mL cut point.⁷²