Of 4232 cases enrolled in the PERCH study, 4136 had available S. pneumoniae colonization and density data. Of those, data on MCPP status were available on 4091 cases (56 MCPP and 4035 non-MCPP cases); 45 cases were excluded owing to missing data required to define MCPP status. Of 5325 controls, the analysis included 1226 controls with and 3962 without RTI in whom S. pneumoniae colonization and density were measured by PCR analysis of the NP/OP specimen. An additional 3 MCPP cases, 82 non-MCPP cases, 11 cases with unknown MCPP status, and 137 controls did not have analyzable NP/OP PCR results because of missing or insufficient samples (2.3%).

Among the 56 MCPP cases, 21% were aged 1–5 months, 23% were 6–11 months, 30% were 12–23 months, and 25% were 24–59 months; 52% were male. Age and sex distribution were similar across MCPP, non-MCPP, and control groups (mean age, 14 months), except that a higher proportion of non-MCPP cases (41%) were aged <6 months compared with MCPP cases (21%). Cases with MCPP were identified at all 5 African sites (15 in The Gambia, 5 in Kenya, 24 in Mali, 5 in South Africa, and 7 in and Zambia) but at neither of the 2 Asian sites (Bangladesh and Thailand) (Table 1).

MCPP cases were more likely to be colonized with S. pneumoniae (by culture or PCR, 100% [56 of 56]) compared with non-MCPP cases (75.7% [3055 of 4035]), all controls (81.4% [4224 of 5188]), controls with RTI (85.5% [1048 of 1226]), and controls without RTI (80.2% [3176 of 3962]), and were more likely to be HIV infected (23.2%) than non-MCPP cases (5.6%) (P ≤ .01 for each). Non-MCPP cases were more likely than those with MCPP to have received antibiotics before NP/OP specimen collection (46% vs 29%; P < .01). Antibiotic use before NP/OP specimen collection occurred in 3 of 14 MCPP cases in The Gambia (data missing for 1), 2 of 5 in Kenya, 3 of 24 in Mali, 5 of 5 in South Africa and 3 of 6 in Zambia (data missing for 1).

Among children who had a positive density value, median S. pneumoniae colonization density was highest in MCPP cases (17.28 × 10⁶ copies/mL) relative to non-MCPP cases (0.75 × 10⁶) and controls (0.60 × 10⁶) (P < .001 for each) (Table 1). However, in South Africa, the only site where all MCPP cases had received prior antibiotics, MCPP cases had lower median density (0.25 × 10⁶) than both non-MCPP cases (0.70 × 10⁶) and controls (0.77 × 10⁶), although differences were not statistically significant. For each case and control group, the median colonization density was lower in children with prior antibiotic use than in those without, and lower in those with NP culture negative versus positive for S. pneumoniae (Table 1).

Density among MCPP cases varied by site (Table 1 and Figure 1; P < .001); median density differed by >100-fold between the site with the highest density, Mali (35.81 × 10⁶ copies/mL), which that also had the highest proportion of MCPP cases (3.6%), and the sites with the lowest density, Kenya and South Africa (0.35 and 0.25 × 10⁶ copies/mL), both with 5 MCPP cases (<0.2%). Among non-MCPP cases and controls, density distributions were similar across sites (Figure 1). Median densities were lowest in Thailand in all groups. The all-site density distribution curves were shifted toward higher densities in MCPP cases versus controls, but the distributions of these groups overlapped substantially (Figure 2). The colonization density distribution among MCPP cases pretreated with antibiotics was shifted toward lower densities compared with MCPP cases without antibiotics before NP specimen collection.

The optimal colonization density threshold for discriminating MCPP cases from controls was >6.9 log₁₀ copies/mL (sensitivity, 64.3%; specificity, 92.2%; age-, sex-, and site-adjusted odds ratio, 17.9 [95% confidence interval 9.9–32.4]). The threshold was unchanged when restricted to controls without RTI and when limiting the comparison to HIV-negative children. When restricted to those MCPP cases (n = 40) and controls (n = 5074) without prior use of antibiotics, the optimal threshold was 6.6 log₁₀ copies/mL (sensitivity, 77.5%; specificity, 85.3%), and it was 4.4 log₁₀ copies/mL when restricted to MCPP cases (n = 16) and controls (n = 114) exposed to antibiotics (sensitivity, 100%; specificity, 52.6%).

The proportion of cases and controls with densities >6.9 log₁₀ copies/mL among those positive varied by site (Figure 3), sex, HIV status, antibiotic pre-exposure, and pneumococcal culture positivity (Table 2). The proportion of MCPP cases with density >6.9 log₁₀ copies/mL ranged from 0 of 5 in Kenya to 21 of 24 (87.5%) in Mali. Across sites, this proportion was lower among MCPP cases who received antibiotic pretreatment than in those who did not (P = .04). The proportion of controls with density >6.9 log₁₀ copies/mL ranged from 1.2% in Thailand to 15.6% in Mali.

Among all PERCH cases, high colonization density was associated with clinical and severity measures considered suggestive of bacterial pneumonia (Table 3). Increasing density was associated in a dose-dependent manner with very severe pneumonia, white blood cell count >15/μL, C-reactive protein (CRP) ≥ 40 mg/L. and coinfection with any virus for which testing was performed. CXR-confirmed pneumonia, consolidation on CXR, HIV infection, oxygen saturation <92% with room air, and respiratory syncytial virus coinfection were all associated with density >6.9 log₁₀ copies/mL, but without clear evidence of increasing strength of association with increasing densities.

Compared with MCPP cases with density ≤6.9 log₁₀ copies/mL, those with density >6.9 log₁₀ copies/mL had higher frequencies of very severe pneumonia and fatal outcome, and lower frequencies of prior antibiotic use, CXR-confirmed pneumonia, and consolidation on CXR (Table 4), but these differences were not statistically significant. Among non-MCPP cases, those with density >6.9 log₁₀ copies/mL (n = 500; 12.4%) were more likely than those below the threshold to have very severe pneumonia, CXR-confirmed pneumonia, consolidation on CXR, oxygen saturation <92%, HIV infection, CRP ≥40 mg/L, or any virus coinfection, and they were less likely to have been previously treated with antibiotics. MCPP cases, regardless of colonization density, were similar to non-MCPP cases with density >6.9 log₁₀ copies/mL, for frequency of elevated white blood cell count, oxygen saturation <92%, prior antibiotic use, or any virus coinfection, but they were more likely to be HIV positive or to have very severe pneumonia, CXR-confirmed pneumonia, alveolar consolidation on CXR, CRP ≥40 mg/L, or fatal outcomes, after adjustment for age, sex, and site.

The serotype of the invasive pneumococcal isolate was available for 46 (98%) of 47 culture-positive MCPP cases, and that of the NP isolate was available for all 44 NP culture-positive MCPP cases. One MCPP case infected with serotype 18C, although NP culture-positive, was PCR negative for pneumococcus, so density could not be determined. Of 43 with serotype data for both the NP and invasive isolate and PCR data, 32 (72.7%) had matching invasive and NP serotypes, with 18 serotypes represented, including both vaccine and nonvaccine serotypes (Figure 4). Although the number of MCPP cases with each identified serotype was small (1–4 per serotype), the distribution of colonization densities seemed similar by serotype. However, the 2 MCPP cases infected with serotype 13 and serotype 14 had colonization densities ≤6.9 log₁₀ copies/mL; neither had received prior antibiotics. For serotypes identified in ≥10 controls, the percentages of controls with density >6.9 log₁₀ copies/mL were similar across serotypes and ranged from 2.3% to 15.6% (Figure 4), equivalent to 84.4% to 97.7% specificity.

In the PERCH study, pneumococcal colonization density was significantly higher among children with MCPP than among other pneumonia cases or community controls. The strength of the association increased with increasing colonization density, with an optimal density threshold of >6.9 log₁₀ copies/mL (64% sensitivity, 92% specificity) to distinguish MCPP cases from controls, but performance varied by site. The optimal threshold was lower (≥4.4 log₁₀ copies/mL; 100% sensitivity, 52.6% specificity) for children treated with antibiotics before specimen collection. Pneumococcal colonization density was associated in a dose-dependent manner with characteristics regarded as suggestive of bacterial pneumonia (alveolar consolidation on CXR, very severe pneumonia, and elevated CRP levels).

Pneumococcal colonization density was also was found to divide PERCH cases along a spectrum of disease severity from MCPP cases; MCPP cases with density >6.9 log₁₀ copies/mL had the greatest proportion with very severe pneumonia and fatal outcomes, followed by MCPP cases with density ≤6.9 log₁₀, non-MCPP cases with density >6.9 log₁₀, and non-MCPP cases ≤6.9 log₁₀, who had the lowest proportion with these characteristics. The association of colonization density with disease severity was observed in a previous study among HIV-infected adults with pneumonia in South Africa [24] but has not been reported among children.

Viral infections, especially influenza, have previously been associated with pneumococcal pneumonia and invasive pneumococcal disease in human studies [25, 26] and animal models [27, 28]. We found that high pneumococcal colonization density was associated with virus detection in the upper respiratory tract, and this finding was explained in part by respiratory syncytial virus coinfection. This finding may indicate that upper respiratory infection with viral pathogens enhanced the density of pneumococcal colonization, but it does not directly address whether these copathogen infections are themselves related to the lower respiratory tract disease. Our finding is consistent with a recent study in South Africa among hospitalized adults and children with acute lower respiratory tract infection (LRTI), which showed that pneumococcal colonization density was associated with the presence of respiratory viruses [10]. In a case-control study such as the PERCH study, we cannot assess the potential causal role of viral infection increasing pneumococcal density or even whether viral infection preceded pneumococcal colonization. Longitudinal cohort studies, such as the Drakenstein study [29], are more suited to address this question.

Our findings in children are similar to the reported association between pneumococcal colonization density and confirmed pneumococcal pneumonia in adults [8, 9, 11, 30]. Studies among children have found that higher colonization density was associated with alveolar consolidation on CXR [12–14], a proxy for pneumococcal pneumonia. In a study among 550 children hospitalized with LRTI in Vietnam [12], cases with consolidation on CXR had higher median NP pneumococcal density at PCR (6.9 log₁₀ copies/mL) than others with LRTI (6.1 log₁₀ copies/mL) and community controls (5.9 log₁₀ copies/mL). These studies did not identify a colonization density threshold that reliably predicted radiographically confirmed pneumonia.

A study among children and adults hospitalized with acute LRTI in South Africa found that invasive pneumococcal pneumonia was associated with increased colonization density; cases with density >1000 copies/mL had 18 times greater odds of invasive pneumococcal pneumonia than colonized cases with density <1000 copies/mL [10]. The South African study defined invasive pneumococcal pneumonia by detection of S. pneumoniae by PCR in the blood, a diagnostic not used in our study owing to poor specificity [31, 32]. We found that the best-performing threshold (6.9 log₁₀ [10^6.9] copies/mL) was much higher than that suggested by the South African study, but comparison of density thresholds between studies is limited by methodologic differences.

The association of pneumococcal colonization density with MCPP does not indicate its utility for patient care. Even in a population with a relatively high prevalence of pneumococcal disease (eg, children hospitalized with pneumonia), the positive predictive value would probably be too low to influence clinical decision making. In settings with lower pneumococcal disease prevalence (eg, countries using PCV), the positive predictive value would be even lower. Although the negative predictive value may be relatively high, it would not be high enough to justify withholding antibiotics in hospitalized children with clinical or radiographic evidence suggestive of bacterial pneumonia. Furthermore, to be useful in a clinical setting, local data on the pneumococcal colonization density distribution would be needed, and patient assessment would have to account for antibiotic pretreatment.

Although our findings are strengthened by the large study size, 7 country sites, and systematic enrollment of well-characterized cases and controls using standardized clinical criteria and laboratory procedures, there were limitations. The number of MCPP cases limited stratified analyses by study site and pneumococcal serotype and prevented calculation of site-specific density thresholds. The findings were largely driven by cases from the 3 sites with the most MCPP cases (The Gambia, Mali, and Zambia). Despite previous evidence of substantial pneumococcal disease burden in children in Bangladesh [33, 34] and Thailand [35, 36], no MCPP cases were identified among enrolled PERCH cases in either of those sites, limiting the evaluation of this threshold at those sites. However, Bangladesh and Thailand did have cases with colonization density above the threshold, the proportion of which in Bangladesh exceeded that in Kenya and Zambia.

The association between pneumococcal pneumonia and colonization density was derived using MCPP cases, but the potential application as a diagnostic assay would be most important to identify cases without pneumococcal detection from blood or other sterile body fluid, which represent the majority of cases with pneumococcal pneumonia [6]. Therefore, the sensitivity of the 6.9 log₁₀ copies/mL threshold for detecting pneumococcal pneumonia may be lower than we estimated based on the MCPP cases. Finally, our study design did not allow assessment of the temporal relationship of colonization density with MCPP. Our analysis aimed not to assess causality but rather to identify a diagnostic adjunct to improve pneumococcal case detection over detection from invasive specimens alone. In addition to study limitations, there are limitations inherent to the measurement of pneumococcal colonization density. Although the PERCH study made great efforts to standardize specimen collection [37] there was no way to standardize the specimen volume taken from the NP/OP space. Higher specimen volume resulting from, for example, coryza could increase measured colonization density.

Our findings provide strong evidence for the relationship between pneumococcal colonization density and pneumococcal pneumonia in children. Pneumococcal colonization density seems to improve detection of pneumococcal pneumonia beyond blood culture, which though highly specific, is insensitive and available only in settings with good microbiology capacity. However, the sensitivity of colonization density remains suboptimal, limiting its utility in clinical settings at the individual case level.