Procalcitonin and C-reactive protein as predictors of blood culture positivity among hospitalised children with severe pneumonia in Mozambique

Authors

  • N. Díez-Padrisa,

    1.  Barcelona Centre for International Health Research, Universitat de Barcelona, Spain
    2.  Centro de Investigaçaõ em Saúde da Manhiça, Manhiça, Mozambique
    Search for more papers by this author
  • Q. Bassat,

    1.  Barcelona Centre for International Health Research, Universitat de Barcelona, Spain
    2.  Centro de Investigaçaõ em Saúde da Manhiça, Manhiça, Mozambique
    3.  CIBER de Epidemiología y Salud Pública, Barcelona, Spain
    Search for more papers by this author
  • L. Morais,

    1.  Centro de Investigaçaõ em Saúde da Manhiça, Manhiça, Mozambique
    Search for more papers by this author
  • C. O’Callaghan-Gordo,

    1.  Barcelona Centre for International Health Research, Universitat de Barcelona, Spain
    2.  Centro de Investigaçaõ em Saúde da Manhiça, Manhiça, Mozambique
    Search for more papers by this author
  • S. Machevo,

    1.  Centro de Investigaçaõ em Saúde da Manhiça, Manhiça, Mozambique
    2.  Faculdade de Medicina, Universidade Eduardo Mondlane, Maputo, Mozambique
    Search for more papers by this author
  • T. Nhampossa,

    1.  Centro de Investigaçaõ em Saúde da Manhiça, Manhiça, Mozambique
    2.  Instituto Nacional de Saúde, Ministerio de Saúde, Maputo, Mozambique
    Search for more papers by this author
  • A. B. Ibarz-Pavón,

    1.  Barcelona Centre for International Health Research, Universitat de Barcelona, Spain
    Search for more papers by this author
  • L. Quintó,

    1.  Barcelona Centre for International Health Research, Universitat de Barcelona, Spain
    Search for more papers by this author
  • P. L. Alonso,

    1.  Barcelona Centre for International Health Research, Universitat de Barcelona, Spain
    2.  Centro de Investigaçaõ em Saúde da Manhiça, Manhiça, Mozambique
    Search for more papers by this author
  • A. Roca

    1.  Barcelona Centre for International Health Research, Universitat de Barcelona, Spain
    2.  Centro de Investigaçaõ em Saúde da Manhiça, Manhiça, Mozambique
    Search for more papers by this author

Corresponding Author Anna Roca, Centre de Recerca en Salut Internacional de Barcelona, Hospital Clínic/Institut d’Investigacions Biomèdiques August Pi i Sunyer, Universitat de Barcelona, C/Villarroel, 170, Barcelona 08036, Spain. Tel.: +34 932275706; Fax: +34 932279853; E-mail: aroca@mrc.gm

Abstract

Objectives  To evaluate the benefits of using procalcitonin (PCT) and C-reactive protein (CRP) as pre-screening tools to predict blood culture positivity among Mozambican children with clinical severe pneumonia (CSP).

Methods  586 children <5 years with CSP and no concurrent malaria fulfilled criteria to be included in the study groups. We determined PCT and CRP for all children with positive bacterial culture (BC+ group, = 84) and of a random selection of children with negative bacterial culture (BC− group, = 246).

Results  PCT and CRP levels were higher in the BC+ group than the BC− one (PCT: median 7.73 versus 0.48 ng/ml, < 0.001; CRP: 177.65 mg/l vs. 26.5 mg/l, < 0.001). In multivariate analysis, PCT was the only independent predictor of the group. To be used as pre-screening tool, PCT presented higher specificities for predetermined sensitivities (≥85%) than CRP. Pursuing a sensitivity of 95%, PCT could reduce the need for bacterial culture by 49% and overall diagnosis costs by 7–35% [assuming variable costs for PCT measurement (ranging from 10 to 30 USD) and a fixed cost of 72.5 USD per blood culture].

Conclusions  Among hospitalised children with CSP and absence of concurrent malaria, PCT pre-screening could help reduce the number of blood cultures and diagnosis costs by specifically targeting patients more likely to yield positive results.

Abstract

Objectifs:  Evaluer les avantages de l’utilisation de la procalcitonine (PCT) et de la protéine C-réactive (CRP) comme outils de présélection pour prédire la positivité de l’hémoculture chez les enfants mozambicains atteints de pneumonie clinique sévère (PCS).

Méthodes:  586 enfants de moins de 5 ans avec une PCS et sans paludisme simultané répondaient aux critères d’inclusion dans les groupes d’étude. Nous avons mesuré la PCT et la CRP de tous les enfants ayant une culture bactérienne positive (groupe CB+, n = 84) et d’une sélection aléatoire d’enfants avec une culture bactérienne négative (groupe CB-, n = 246).

Résultats:  Les taux de PCT et de CRP étaient plus élevés dans le groupe BC+ que dans le groupe BC- (PCT médiane: 7,73 contre 0,48 ng/ml, p < 0,001; CRP: 177,65 mg/l contre 26,5 mg/l, p < 0,001). En analyse multivariée, la PCT a été le seul facteur prédictif indépendant du groupe. Pour être utilisé comme outil de présélection, la PCT présentait des spécificités plus élevées pour des sensibilités prédéterminées (≥85%) que la CRP. En atteignant une sensibilité de 95%, la PCT pourrait réduire la nécessité d’une culture bactérienne de 49% et les coûts globaux de diagnostic de 7 à 35% [en supposant des coûts variables pour la mesure de la PCT (allant de 10 à 30 USD) et un coût fixe de 72,5 USD par hémoculture].

Conclusions:  Chez les enfants hospitalisés avec une PCS et l’absence d’un paludisme simultané, la présélection par la PCT pourrait aider à réduire le nombre d’hémocultures et les coûts de diagnostic, en ciblant spécifiquement les patients plus susceptibles à révéler des résultats positifs.

Abstract

Objetivos:  Evaluar los beneficios de utilizar la procalcitonina (PCT) y la proteína C-reactiva (PCR) como herramientas de preselección, para predecir un hemocultivo positivo en niños de Mozambique con neumonía clínica severa (PCS).

Métodos:  Se incluyeron 586 niños <5 años con PCS que no cumplían criterios para una malaria concurrente. Se determinaron la PCT y PCR a todos los niños con un cultivo bacteriano positivo (grupo BC+, n = 84) así como de un grupo de niños con bacteriemia negativa seleccionados al azar (grupo BC-, n = 246).

Resultados:  Los niveles de PCT y PCR eran mayores en el grupo BC+ que en el grupo BC- (PCT: mediana 7.73 versus 0.48 ng/ml, p < 0.001; PCR: 177.65 mg/l versus 26.5 mg/l, p < 0.001). En un análisis multivariado, la PCT era el único vaticinador independiente del grupo. Como herramienta de preselección, la PCT presentaba mayor especificidad para sensibilidades predeterminadas (≥85%) que la PCR. Alcanzando una sensibilidad del 95%, la PCT podría reducir en un 49% la necesidad de un cultivo bacteriano, y los costes totales del diagnóstico en un 7–35% [asumiendo costes variables para medir PCTt (en un rango de 10 a 30 USD) y unos costes fijos de 72.5 USD por hemocultivo].

Conclusiones:  Entre niños hospitalizados con PCS y en ausencia concurrente de malaria, la preselección con PCT podría ayudar a reducir el número de hemocultivos y los costes de diagnóstico al centrarse específicamente en aquellos pacientes con mayor posibilidad de dar resultados positivos.

Introduction

Pneumonia, the major cause of morbidity and mortality in children <5 years worldwide (Black et al. 2010), can be caused by a large number of pathogens. Identification of the causative agent in patients with bacterial pneumonia is crucial to determine the optimal antibiotic therapy and to monitor local resistance patterns. For surveillance purposes, identification of the causative agent is needed to generate knowledge of prevalent causes of pneumonia and to measure the effects of specific interventions.

To date, bacterial culture remains the gold standard for bacterial diagnosis in patients with pneumonia. However, sample collection and laboratory procedures for this technique are expensive and require specially trained personnel and adequate facilities. In developing countries, where 90% of pneumonia deaths occur (Black et al. 2010), an affordable and easy-to-apply test that predicts blood culture positivity could identify patients whose culture should be performed. Reducing the number of blood cultures to those cases that are more likely to yield positive results could improve their economical and logistical viability in resource-limited settings.

Very few hospitals in Africa have the capacity to perform bacterial culture. In ideal conditions, the expected positivity rate among bacterial cultures on the continent is approximately 10–15% of paediatric pneumonia cases (Banya et al. 1996; Norton et al. 2004; Nantanda et al. 2008; Sigauque et al. 2009; Reddy et al. 2010; Schwarz et al. 2010). Unfortunately, the proportion of contaminated samples is still unacceptably high in most of the settings, ranging between 5% and 25% (Bahwere et al. 2001; Berkley et al. 2005; Brent et al. 2006; Roca et al. 2006; Reddy et al. 2010; Schwarz et al. 2010; Talbert et al. 2010). Samples often need to be transported over long distances and storage conditions are compromised during the transport, reducing the number of positive cultures and increasing the overall cost of the diagnosis. A pre-screening test to predict blood culture positivity in African and other resource-limited countries could help overcome these inconveniences, by minimising the risk of contamination and reducing the number of samples needing to be sent to reference laboratories. In contrast to richer settings, where two blood culture samples per patient are normally collected, in Africa, this is generally restricted to a single one.

Clinical parameters have been assessed to predict bacterial culture positivity in Africa, but results are inconsistent between studies and inter-observer variability among poorly trained health workers compromises their utility (Banya et al. 1996; Bahwere et al. 2001; Norton et al. 2004). In recent years, interest in procalcitonin (PCT) and C-reactive protein (CRP) as potential markers of bacterial infections has re-emerged because both markers could be built into rapid diagnostic tests (RDTs). A recent study in Switzerland confirmed the utility of PCT to screen blood samples to be cultured among adults hospitalised with pneumonia (Muller et al. 2010).

In our setting, we have shown the capacity of PCT and CRP to differentiate viral from invasive bacterial pneumonia among hospitalised children without malaria parasites (Diez-Padrisa et al. 2010). In the current analysis, we aim to evaluate the benefits of using PCT and/or CRP as a pre-screening tool to decide which patients could benefit from blood culture. We included in our study children <5 years hospitalised with clinical severe pneumonia (CSP) in rural Mozambique. We excluded children with malaria parasites as, according to the data we previously generated (Diez-Padrisa et al. 2010), they present increased PCT and CRP levels irrespectively of bacterial infection.

Methods and materials

Study area

This study was conducted by the Centro de Investigação em Saúde da Manhiça (CISM) at Manhiça District Hospital (MDH), the referral health facility for Manhiça District, a rural malaria-endemic area in southern Mozambique (Loscertales et al. 2002).

Since 1996, CISM has been running a continuous Demographic Surveillance System (DSS). The DSS covers 500 km2 and approximately 80 000 inhabitants; 18% are children <5 years. Under-five mortality in the area was 138.6/1000 in 2005 (Nhacolo et al. 2006).

At the time of this study, Haemophilus influenzae type b and pneumococcal conjugate vaccines had not been introduced in Mozambique.

Hospital surveillance and clinical management

The MDH is a 110-bed hospital with 36 paediatric beds. Since 1997, MDH and CISM have operated round-the-clock surveillance of all paediatric visits (Loscertales et al. 2002). Finger prick blood is obtained for malaria tests and determination of packed-cell volume (PCV) in children with fever (axillary temperature ≥37.5 °C) or history of fever in the preceding 24 h. Blood cultures are performed upon hospital admission for all children <2 years or older with an axillary temperature ≥39 °C or other severity signs. Lumbar puncture is performed in hospitalised children with suspicion of meningitis, sepsis or neurological impairment (Roca et al. 2009).

CSP accounted for 16% of MDH admissions among children <2 years in 2004-2006, of which 15% presented with bacteremia and 19% with concurrent malaria (Sigauque et al. 2009). HIV prevalence among pregnant women attending the hospitals was 12.4% (Naniche et al. 2009).

Patient’s selection and sample collection

This analysis is part of a larger study designed to describe clinical and epidemiological characteristics of children <5 years admitted to the MDH with CSP between September 2006 and September 2007 (Bassat et al. 2011; O’Callaghan-Gordo et al. 2011). CSP was defined as cough and difficult breathing with increased respiratory rate according to age group (World Health Organization 2005) and at least one of the following signs: indrawing, nasal flaring, grunting or crackles. Written informed consent was obtained from all participants’ parents/legal guardians recruited for the study. The study was approved by the Mozambican National Bioethics Committee and the Institutional Review Board of the Hospital Clínic de Barcelona.

Exclusion criteria were presence of malaria parasites, contaminated blood culture [Staphylococcus epidermidis, Viridans group Streptococci and Bacillus spp. (non-anthracis)] and recall of having received antibiotics before admission. Study children were classified according to culture results from a blood sample collected early on admission. All children with bacteremia were included in the positive bacterial culture (BC+) group. A random selection of all other children was included in the negative bacterial culture (BC−) group.

As part of the study, one blood sample was collected on admission for full blood cell count and PCT/CRP determination. Blood culture and malaria determination were extended to all study participants. Children residing in the DSS area were offered voluntary HIV counselling and testing. Additional written informed consent and finger prick blood were required for HIV determination.

Laboratory methods

P. falciparum parasites were detected by microscope observation of Giemsa-stained blood films (Guinovart et al. 2008). PCV was measured using microcentrifuge and a Hawksley haematocrit reader card (Hawksley& Sons Ltd, UK). Blood cultures were performed using an automated system (BACTEC® 9050; Becton-Dickinson, Franklin Lake, NJ, USA). Positive blood cultures were examined following standard procedures (Roca et al. 2006; Valles et al. 2006).

HIV testing was performed using two RDTs: Determine® (Abbott Laboratories, North Chicago, IL, USA) and Unigold® (Trinity Biotech, Bray, Ireland). HIV-1 infection was confirmed using antigen DNA-PCR Roche HIV-1 DNA test® (Roche Molecular Systems, Branchburg, NJ, USA) for <18-month-old positive children and for those cases with discordant results from the two RDTs.

Plasma obtained from 1.5 ml of blood in EDTA after full blood cell count was stored at −20 °C until processing at HCB/Institut d’Investigacions Biomèdiques August Pi i Sunyer. PCT quantification was performed using automated immunoanalysis with Liaison® (Diasorin, Saluggia, Italy) or Kryptor Compact® (Brahms, Hennigsdorf, Germany). For CRP, an immunoturbidimetric assay with ADVIA Chemistry CRP_2® (Siemens Medical Solutions Diagnostics, Tarrytown, NY, USA) was used. The limits of detection of these techniques were 0.04 ng/ml, 0.02 ng/ml and 4 mg/l, respectively.

Data management and statistical analysis

Data were double entered using Fox Pro version 2.6 (Microsoft Corporation, Redmond, WA, USA) and analysed using STATA version 11 (Stata Corporation, College Station, TX, USA). Discrepancies in data entry were resolved by referring to the original forms.

Proportions were compared using chi-square test. Distributions of both markers were evaluated by Kruskall–Wallis test. To assess the capacity of PCT, CRP and other parameters to predict bacterial culture positivity, univariate and multivariate logistic regressions adjusted by all significant parameters were used. To perform this analysis, PCT and CRP were categorised in ranks (low/medium/high) according to tercils. Diagnostic features were determined using receiver-operating characteristic (ROC) analyses and the BC− group as reference. P-values ≤0.05 were considered significant.

Direct costs comparison estimates were performed according to data obtained from PCT cut-offs in the ROC analysis and assuming that only children identified as BC+ would need blood culture for bacterial identification. The estimated cost per blood culture was 72.5 USD, as only one set is used per patient in Africa (Craven 2004; Zwang & Albert 2006; Muller et al. 2010). The estimated current (Muller et al. 2010) and optimal costs per PCT measurement were 30 and 10 USD. Other costs like handling, storage and transportation of the samples, equipment maintenance, contamination risk, time reduction of sample processing and other administrative procedures were not taken into consideration.

Results

Study profile

During the study period, 835 children were hospitalised with CSP at the MDH; of these, 182 (22%) presented with exclusion criteria (97 had malaria parasites, 77 had contaminated blood culture results and 8 had both) (Figure 1). Of the remaining 653 cases admitted, 586 (90%) had the required samples available for testing. 84 of these children (14%) presented positive and 502 (86%) negative bacterial culture. All children with positive bacterial culture were included in the BC+ group. The most prevalent bacteria isolated in the BC+ group were Streptococcus pneumoniae (= 39), Haemophilus influenza type b (= 15), Staphylococcus aureus (= 6), Escherichia coli (= 6) and Salmonella spp (= 6). A random selection of approximately half (246/502) of the children with negative bacterial culture was included in the BC− group. No differences in age, sex and clinical variables were observed between children who were or were not included in this selection (data not shown).

Figure 1.

 Study profile of children recruited for the study.

Overall, 13% (38/298) of the children in the study groups with hospital outcome data died and 30% (64/210) with HIV data available were HIV-infected. Significantly, higher rates of mortality, HIV, anaemia and malnutrition were observed in the BC+ group (Table 1).

Table 1.   Clinical and epidemiological characteristics of the children according to the study groups
VariablesBC−
(= 246)
BC+
(= 84)
P-value
  1. Data are n (%) of patients.

  2. *Weight to age z-score <3 SDs from US reference population.

  3. †Measured with pulsioximetry.

Age   
 <1 month11 (4)6 (7)0.625
 1–<12 months115 (47)39 (46.5)
 1–<5 years120 (49)39 (46.5)
Sex   
 Male147 (60)53 (63)0.589
 Female99 (40)31 (37)
Malnutrition* (= 300)
 No190 (82)44 (66)0.006
 Yes43 (18)23 (34)
Haematocrit (= 329)
 ≥33%138 (56)30 (36)0.002
 <33%108 (44)53 (64)
HIV (= 210)
 No125 (76)21 (47)<0.001
 Yes40 (24)24 (53)
Mortality (= 298)21 (9)17 (25)0.001
Oxygen saturation† (= 327)
 ≥94%125 (51)40 (59)0.859
 90- <94%47 (19)18 (22)
 <90%73 (30)24 (29)

Distribution of PCT and CRP comparing study groups

Median PCT concentration in the BC+ group was 7.73 ng/ml, and in the BC− group, it was 0.48 ng/ml (= 76 and = 246, < 0.001) (Figure 2a). Median CRP concentration in the BC+ group was 177.65 mg/l vs. 26.5 mg/l in the BC− group (= 82 and = 246, < 0.001) (Figure 2b).

Figure 2.

 Distribution of procalcitonin (PCT) (a) and C-reactive protein (CRP) (b) according to the study groups.

Capacity of PCT, CRP and other parameters to predict the study group

PCT, CRP, HIV infection, haematocrit and malnutrition were the predictors of the study group in the univariate logistic regression (Table 2). When using all significant parameters from this regression in the multivariate model for PCT, only this marker remained associated (Table 3), whereas in the multivariate model for CRP, both CRP and HIV status remained associated.

Table 2.   Predictors of bacterial culture positivity (study group) in the univariate logistic regression analysis (= 196)
PredictorOR95% CI P-value
PCT, ng/ml
 Low (≤0.25)1  
 Median (>0.25–≤3.38)7.61.6–36.150.011
 High (<3.38)30.886.91–137.87<0.001
CRP, mg/l
 Low (≤18.5)1  
 Median (>18.5–≤97.9)16.312.07–128.550.008
 High (>97.9)48.716.32–375.37<0.001
Age
 <1 month1  
 1–<12 months0.570.1–3.170.521
 1–<5 years0.610.11–3.390.571
Sex
 Male1  
 Female0.890.44–1.820.753
HIV
 No1  
 Yes3.511.68–7.350.001
Haematocrit
 ≥33%1  
 <33%2.541.21–5.340.013
Oxygen saturation
 ≥94%1  
 90–<94%1.470.63–3.40.372
 <90%0.540.21–1.450.224
Malnutrition
 No1  
 Yes2.521.15–5.550.021
Table 3.   Predictors of bacterial culture positivity (study group) in the multivariate-adjusted logistic regression analysis (= 196)
PredictorOR95% CI P-value
Model for PCT
 PCT, ng/ml
  Low (≤0.25)1  
  Median (>0.25–≤3.38)6.721.39–32.560.018
  High (<3.38)23.625.13–108.71<0.001
 HIV
  No1  
  Yes1.940.82–4.550.130
 Haematocrit
  ≥33%1  
  <33%1.540.66-3.600.315
 Malnutrition
  No1  
  Yes1.910.74–4.910.181
Model for CRP
 CRP, mg/l
  Low (≤18.5)1  
  Median (>18.5–≤97.9)15.311.91–122.450.01
  High (>97.9)41.745.19–335.97<0.001
 HIV
  No1  
  Yes2.631.12–6.160.026
 Haematocrit
  ≥33%1  
  <33%1.240.51–3.010.64
 Malnutrition
  No1  
  Yes1.870.72–4.820.198

Diagnostic accuracy of PCT and CRP

When using PCT or CRP to predict the probability of being part of the BC+ group, PCT presented higher specificities for predetermined sensitivities (≥85%) (Table 4). For a sensitivity of 95%, we observed a specificity of 56% for PCT (cut-off ≥ 0.64 ng/ml) and 45% for CRP (cut-off  20.9 mg/ml). For a sensitivity of 85%, the specificities increased up to 62% (cut-off ≥ 1.02 ng/ml) and 57% (cut-off ≥ 38 mg/ml), respectively. The area under the curve for the prediction of BC+ was 0.80 for PCT and 0.79 for CRP (= 320, = 0.617) (Figure 3). Combination of both markers did not improve the diagnostic profile (data not shown).

Table 4.   Diagnostic accuracy of PCT and CRP to predict bacterial culture positivity among study children without malaria parasites
Cut-offSensitivity (%)Specificity (%)PPV*
(%)
NPV†
(%)
  1. *Positive predictive value.

  2. †Negative predictive value.

PCT (ng/ml)
 0.1100231759
 0.6495562772
 0.8390602768
 1.0285622765
 1.1580632661
 1.475642559
CRP (mg/l)
 110071530
 20.995452267
 27.990502364
 3885572463
 59.880662863
 7375682860
Figure 3.

 PCT and CRP ROC curves to predict bacterial culture positivity among children included in the study groups.

Economic implications derived from the use of PCT

We calculated the potential cost-savings of using PCT cut-offs as predictors of bacteremia among children meeting inclusion criteria of the study groups (= 586) (Table 5). PCT cut-offs between 95 and 75% of sensitivity (cut-offs ≥ 0.64 and ≥1.4 ng/ml, respectively) could lower by 49–59% the number of bacterial cultures required. For the same sensitivities, the cost of the diagnosis could be reduced by 7–17% when assuming a cost of 30 USD per PCT measurement and by 35–45% when assuming a cost of 10 USD.

Table 5.   Implications to use PCT as a pre-screening tool before performing bacterial culture among study children without malaria parasites
VariablesNo PCT measurementPCT measurement (ng/ml)*
≥0.64≥0.83≥1.02≥1.15≥1.4
  1. All data referred to costs are in USD.

  2. *Cut-off ≥0.1 ng/ml was not included as it increases 22% the total cost of the diagnosis when using 30 USD PCT measurement and only reduces 6% the total cost of the diagnosis when using 10 USD PCT measurement.

  3. †72.5 USD per blood culture.

  4. ‡Pre-screening with PCT performed to all samples.

  5. §Using as reference the total cost of the diagnosis when performing blood culture to all samples without using PCT pre-screening.

Blood culture, n (%)
 Performed586 (0)300 (51)275 (47)264 (45)252 (43)243 (41)
 Not performed0 (0)286 (49)311 (53)322 (55)334 (57)343 (59)
Blood cultures total cost†424852174819948191161827117587
PCT determinations total cost‡
 30 USD per measurement01758017580175801758017580
 10 USD per measurement058605860586058605860
Total cost of the diagnosis
 30 USD per PCT measurement424853932837528366963585135167
 10 USD per PCT measurement424852760825808249762413123447
Reduction in the total cost of the diagnosis, %§
 30 USD per PCT measurement0712141617
 10 USD per PCT measurement03539414345
Cost per patient
 30 USD per PCT measurement72.56764636160
 10 USD per PCT measurement72.54744434140
Missed bacteria, n (%)04 (5)9 (11)12 (14)18 (21)21 (25)

Discussion

We assessed the utility of PCT and CRP to predict blood culture positivity among children <5 years hospitalised with CSP and no malaria parasites in rural Mozambique. Our results show that PCT and CRP levels below the estimated cut-offs identify children with low risk of detectable bacteremia and, consequently, reduced the need of bacterial culture confirmation. CRP presented lower specificities for pre-established sensitivities (≥85%) than PCT and its association with the group was also influenced by HIV. Therefore, we only estimated the benefits of using PCT.

High sensitivity is essential to ensure that only a small fraction of positive cases is missed if using PCT to establish the need of bacterial culture performance. With a sensitivity of 95% (5% of positive cases missed), PCT could avoid as many as 49% of bacterial cultures performed to children meeting inclusion criteria of the study groups. This represents a potential reduction in the cost of bacterial culture diagnosis by 7–35% (PCT cost between 30 and 10 USD). These findings reinforce PCT as a potential tool to be used in Africa to: (i) restrict the use of bacterial culture to patients with higher probability of positive results and (ii) reduce costs of diagnosis associated with bacterial culture. Our results agree with those of a recent study in Switzerland, where PCT was the most reliable predictor of blood culture positivity among adult patients with pneumonia (Muller et al. 2010). Results also concur with our previous observations (Diez-Padrisa et al. 2010) and data from Malawi (Carrol et al. 2009), in which PCT and CRP levels were increased in the presence of severe bacterial infection among children with severe pneumonia.

Most African settings do not have the required microbiology facilities and skilled personnel to perform blood culture for bacterial detection. Sometimes, blood samples are sent to distant national laboratories for culture. The extra cost of storage and transportation of the samples was not taken into account in our cost estimates; hence, the potential savings of using PCT for pre-screening might be even larger. We did also not include in our estimates the time saved in sample processing and the reduction of workload in national laboratories. Our results show that PCT could reduce the cost of the diagnosis associated with bacterial culture in Africa and, more importantly, increase culture’s viability in the continent if turning PCT determination into an inexpensive RDT. Common advantages of RDTs are easy access in remote settings, minimal training and specialised equipment requirements, and low cost introduction. PCT-RDTs should be temperature stable and their results reproducible to indicate which patients are more likely to yield positive results and whose blood culture should thus be given priority when resources are limited.

In the advent of wide deployment of new bacterial vaccines in Africa, WHO reference laboratories have been designated on the continent and selected clinical samples from neighbouring countries are sent to these laboratories for bacterial culture (Centers for Disease Control & Prevention 2009, Network for Surveillance of Pneumococcal Disease in the East African Region 2012). To our understanding, in areas where resources are very scarce, pre-selecting samples with a higher probability of generating positive bacterial results could make regional surveillance more effective. As suggested previously (Madhi et al. 2005; Cheung et al. 2008), a marker capable to predict bacterial infection could also be used to improve bacterial pneumonia endpoint in vaccine effectiveness studies.

However, the use of PCT as a pre-screening tool has some limitations. On one hand, PCT cannot be applied to children with concurrent malaria parasitemia (13% of the children meeting entry criteria) as PCT levels among those children are increased, compromising the capacity of this marker to differentiate clinical groups (Diez-Padrisa et al. 2010). But data suggest that malaria endemicity is steadily decreasing in many African countries (Ceesay et al. 2008; World Health Organization 2010; Alonso et al. 2011). If so, pre-screening with PCT could be applied to a larger number of children hospitalised in the continent with CSP. Non-malaria-endemic Latin American and Asian countries could also benefit from the development of a PCT-RDT. Cost of malaria diagnosis was not included in this analysis as in malaria-endemic areas ill children must be routinely tested for malaria.

On the other hand, antibiotic pre-treatment decreases the likelihood of finding positive blood cultures (Muller et al. 2010). Although antibiotic pre-treatment data were not available for this study (only maternal recall), we anticipate that antibiotic self-medication in Manhiça is probably lower than in other African settings, where our proposed pre-screening tool could be compromised. A further limitation of our study is that we have based our estimates of blood culture and PCT costs on previous publications (Craven 2004; Zwang & Albert 2006; Muller et al. 2010), all performed in developed countries. How this affects our results is unclear. Although in Africa the cost of labour is much lower than in western countries, the cost of reagents, laboratory materials and equipment is usually very high because of the need to import them. Finally, as concentrations between stored and fresh specimens may vary, validation of the results with fresh specimens will be needed in further studies.

Overall, PCT could potentially contribute to reduce the costs of bacterial blood cultures for clinical management and epidemiological surveillance of African children with CSP. We advocate high potential public health gains by introducing this pre-screening test in resource-limited countries/settings where bacterial culture is difficult to perform and the need for a PCT-RDT.

Acknowledgments

We thank the parents and children from Manhiça who agreed to participate in this study. Our thanks also extend to the colleagues of the MDH and the CISM who directly or indirectly participated in the study. We are very grateful to Xavier Filella and Ernest Mas for performing PCT and CRP determinations and to Joan Vives for his inputs on the economical parts of the analysis. This study was supported by a grant of the Bill and Melinda Gates Foundation (45 452). The CISM received major core funding from the Spanish Agency of International Cooperation. Quique Bassat and Anna Roca were supported by grants from the Spanish Ministry of Science and Innovation (FIS: CM05/00134 and Ramón y Cajal: RYC-2008-02777, respectively).

Ancillary