Sensitivity of C-reactive protein for the identification of patients with laboratory-confirmed bacterial infections in northern Tanzania

Abstract

Objective

Identifying febrile patients requiring antibacterial treatment is challenging, particularly in low-resource settings. In Southeast Asia, C-reactive protein (CRP) has been demonstrated to be highly sensitive and moderately specific in detecting bacterial infections, and to safely reduce unnecessary antibacterial prescriptions in primary care. As evidence is scant in sub-Saharan Africa, we assessed the sensitivity of CRP in identifying serious bacterial infections in Tanzania.

Methods

Samples were obtained from inpatients and outpatients in a prospective febrile illness study at two hospitals in Moshi, Tanzania, 2011–2014. Bacterial bloodstream infections (BSI) were established by blood culture, and bacterial zoonotic infections were defined by ≥4-fold rise in antibody titer between acute and convalescent sera. The sensitivity of CRP in identifying bacterial infections was estimated using thresholds of 10, 20, and 40 mg/L. Specificity was not assessed because determining false positive CRP results was limited by the lack of diagnostic testing to confirm non-bacterial etiologies and because ascertaining true negative cases was limited by the imperfect sensitivity of the diagnostic tests used to identify bacterial infections.

Results

Among 235 febrile outpatients and 569 febrile inpatients evaluated, 31 (3.9%) had a bacterial BSI and 61 (7.6%) had a bacterial zoonosis. Median (interquartile range) CRP values were 173 (80–315) mg/L in bacterial BSI, and 108 (31–208) mg/L in bacterial zoonoses. The sensitivity (95% Confidence Intervals) of CRP was 97% (83–99%), 94% (79–98%), 90% (74–97%) for identifying bacterial BSI, and 87% (76–93%), 82% (71–90%), 72% (60–82%) for bacterial zoonoses, using thresholds of 10, 20 and 40mg/L respectively.

Conclusion

CRP was moderately sensitive for bacterial zoonoses and highly sensitive for identifying BSIs. Based on these results, operational studies are warranted to assess the safety and clinical utility of CRP for the management of non-malaria febrile illness at first-level health facilities in sub-Saharan Africa.

Introduction

The management of febrile patients and the identification of those who would benefit from antibacterial treatment is challenging, particularly in low-resource settings where laboratory support is often lacking [1, 2]. The adoption of malaria rapid diagnostic tests has led to a substantial decrease in the unnecessary use of antimalarials [3]. However, once malaria is ruled out, health workers lack tools to diagnose other causes of illness, resulting in increasingly indiscriminate use of antibacterials, fueling the spread of antibacterial resistance [4–7]. Biomarkers identifying bacterial infections have been proposed as a potential means to guide health workers’ antibacterial prescription, but evidence for their utility in many low-resource settings, and in Africa in particular, is scant [8, 9].

C-reactive protein (CRP) has been shown to be a highly sensitive and moderately specific biomarker of bacterial infection across various settings and populations in Southeast Asia [10–12]. In addition, clinical trials show a safe reduction of antibacterial prescription when using CRP at point-of-care in the context of first level health facilities [13]. In Africa, most studies of CRP performance identifying bacterial infections have focused on lower respiratory tract infections [14–17]. Whether CRP testing can be used to identify bacterial infections in all febrile patients and guide antibacterial prescription in sub-Saharan Africa (SSA) has not been well established.

The first step in evaluating the utility of CRP testing to guide antibacterial prescribing requires well-characterized clinical samples with confirmed bacterial infections. If CRP is not consistently elevated among persons with confirmed bacterial infections, then it is unlikely to be a safe biomarker test for use at first-level health facilities. Here we report the serum CRP concentration levels in samples from patients with laboratory-confirmed bacterial bloodstream infections (BSI) and bacterial zoonotic infections from a cohort of febrile inpatients and outpatients in northern Tanzania. We estimate the sensitivity of CRP in identifying these patients as requiring antibacterial treatment using previously proposed thresholds of 10, 20, and 40mg/L [10, 13, 18].

Methods

Study design

Observational febrile illness surveillance was conducted at two sentinel hospitals in Moshi, Tanzania: the medical ward of Kilimanjaro Christian Medical Centre, and the medical ward, pediatric ward, and outpatient department of Mawenzi Regional Referral Hospital.

Participants

Inpatients were eligible for enrolment if they had a history of fever within the previous 72 hours or a documented fever (defined as axillary temperature >37.5˚C or a tympanic, oral or rectal temperature ≥38˚C at screening). Outpatients were only eligible if they had a documented fever. After obtaining informed consent, a standardized questionnaire and physical examination were administered by a study clinician who also recorded participant clinical outcome as discharged, transferred, or dead. Children were defined as participants under 10 years of age following WHO standards [19].

Procedures

Blood collected prospectively from febrile participants was apportioned as follows: aerobic culture (8–10 mL in BacTAlert Standard Aerobic bottle for adults, 4–5 mL in Pediatric FAN bottle for children) to identify bacterial bloodstream infection (BSI) or fungal BSI; an EDTA tube for Giemsa-stained thick and thin blood parasite films to identify malaria; and a serum fractionation tube to conduct serological assays for bacterial zoonoses. Blood cultures were incubated in a BacTAlert 3D 480 continuously monitored blood culture instrument for 5 days. Standard methods were used to identify bacterial or fungal bloodstream isolates. The bacterial zoonoses (brucellosis, leptospirosis, Q fever, and spotted fever group rickettsiosis) were diagnosed using case definitions from the US National Notifiable Diseases Surveillance System and reference standard diagnostics on acute and convalescent serum. Diagnostic serology for brucellosis and leptospirosis was performed by microscopic agglutination test at the US Centers for Disease Control and Prevention (CDC) Bacterial Special Pathogens Branch [20, 21]. Diagnostic serology for Q fever [22] and spotted fever group rickettsiosis was performed by indirect immunofluorescence assay for C. burnetii and R. africae, respectively, at the US CDC Rickettsial Zoonoses Branch. Malaria was defined as parasitemia >25 trophozoites per 200 white blood cells (approximately 1000 parasites/μL) on blood parasite smears [23].

Infections were assigned an etiological grouping as either bacterial infections, fungal BSI, malaria, or undetermined. Bacterial infections were further analyzed according to the mode of diagnosis—BSI identified by blood culture and bacterial zoonosis identified by serology. In participants with co-infections from different etiological groups, those that included a bacterial BSI were classified as bacterial BSI. Participants with bacterial zoonoses and malaria co-infection were classified in the bacterial zoonoses group.

For participants with sufficient sample volumes available, serum CRP concentrations were assayed at the Intermountain Central Laboratory (Murray, Utah, USA), using the Multigent Vario assay on the Abbott c8000 chemistry analyzer (Abbott, Abbott Park, Illinois, USA). Differences in median CRP concentrations were tested using the Wilcoxon rank-sum test for nonparametric samples.

Outcomes

The primary objective of this study was to evaluate the sensitivity of CRP thresholds in detecting all bacterial infections, including the bacterial BSI and bacterial zoonoses groups, calculated as the proportion of samples with a CRP concentration above the thresholds of 10, 20, and 40mg/L [10, 13, 18]. Higher thresholds have been described in the literature, but mainly among patients presenting with a lower respiratory tract infection from hospital-based settings in high-income countries [24–26].

Confidence intervals for the sensitivity estimates were calculated using the Wilson score method. As secondary objectives we sought to describe the CRP distribution among participants with fungal BSI and malaria; we sought to describe CRP distribution across both inpatients and outpatients; and we compared antibacterial prescription between current practices and CRP-guided prescription estimates, including broad-spectrum antibacterials. Broad-spectrum antibacterials were defined as ceftriaxone, ciprofloxacin, azithromycin, and amoxicillin/clavulanic acid [27, 28].

Research ethics

This research was approved by the Kilimanjaro Christian Medical University College Research Ethics Committee (approval #295), the Tanzania National Institute for Medical Research National Ethics Coordinating Committee (approval NIMR/HQ/R.8a/Vol.IX/1000), and an Institutional Review Board of the Duke University Hospital System (approval IRB#Pro00016134). All study participants or their guardians provided written informed consent.

Results

Participant characteristics

Among 1,753 febrile patients enrolled between September 2011 and May 2014, CRP assays were carried out on 804 (45.9%) participants with sufficient sample volume available. Participant characteristics are described in Table 1. (Age-stratified descriptions of participants having a CRP measurement are shown in Supplementary Table 1; comparisons of characteristics of participants with sufficient sample volume for CRP testing are compared to those without sufficient sample for CRP testing in Supplementary Table 2). Of 804 participants, 235 (29.2%) were outpatients and 569 (70.8%) were inpatients, in whom 174 (74.0%) and 487 (85.6%), respectively (p <0.001) were prescribed an antibacterial by their clinician after enrollment into this observational study.

Table 1.

Demographic and clinical characteristics of febrile inpatients and outpatients, Kilimanjaro Christian Medical Centre and Mawenzi Regional Referral Hospital, Tanzania, 2011–2014.

Enrolment characteristics	Outpatients (n=235)	Inpatients (n=569)	P-value
Demographic characteristics
Age in years, median (IQR)	31 (19–40)	36 (26–45)	<0.001
< 10 years of age , n (%)	38 (16.2)	13 (2.3)	<0.001
Self-reported HIV infection, n (%)	14 (6.0)	134 (23.6)	<0.001
Other healthcare facility prior to enrollment, n (%)	149 (63.4)	354 (62.2)	0.171
Fever duration in days, median (IQR)	4 (3–7)	5 (3–14)	0.019
Antibacterial intake prior to enrollment, n (%)	87 (37.0)	233 (41.0)	0.288
Antimalarial intake prior to enrollment, n (%)	63 (26.8)	205 (36.0)	0.012
Clinical characteristics
Tympanic temperature (˚C), median (IQR)	38.5 (38.2–39.0)	38.1 (37.4–38.7)	<0.001
Neurological symptoms, n (%)	163 (69.4)	425 (74.7)	0.131
Respiratory symptoms, n (%)	160 (68.1)	394 (69.2)	0.773
Gastro-intestinal symptoms, n (%)	162 (68.9)	452 (79.4)	0.002
Undifferentiated presentation, n (%)	8 (3.4)	9 (1.6)	0.101

P-value calculated using a Mann-Whitney test for non-normally distributed variables, and chi-squared test for categorical variables.

Antibacterial and antimalarial intake prior to enrollment was based upon participant or caregiver report.

Neurological symptoms include headache, convulsions, or stiff neck.

Respiratory symptoms include sore throat, dyspnea, hemoptysis, or cough.

Gastrointestinal symptoms include nausea, vomiting, diarrhoea, constipation, jaundice, bloody stools, or abdominal pain.

Undifferentiated presentation defined by the absence of any symptoms present in neurological, respiratory, nor gastrointestinal systems. Common symptoms in this group included myalgia, arthralgia, tiredness, chills, sweating, rash, bleeding, or appetite loss.

Among children (n=51), the median (interquartile range, IQR) age was 5 (3–7) years, while the median (IQR) age in adults was 36 (26–45) years.

Antibacterial prescription with respect to CRP levels

The median (IQR) CRP value in all patients was 75 (18–176) mg/L. The distribution of CRP concentrations differed significantly between outpatients and inpatients (p<0.001), with a median of 46mg/L in outpatients (IQR 9–119) and 93 mg/L in inpatients (IQR 31–199). Antibacterial prescription by non-study clinicians after study enrollment was prevalent across CRP concentrations, including broad-spectrum antibacterials (Figure 1). Among patients who were prescribed an antibacterial, 95 (40.4%) of 235 outpatients and 317 (55.7%) of 569 inpatients were prescribed broad-spectrum antibacterials, and the prescription of broad-spectrum antibacterials did not vary across CRP concentrations (p-value 0.584). Considering children, 32 (84.2%) of 38 outpatients and all of the 13 inpatients had an antibacterial prescribed, while 19 (50.0%) of 38 and 5 (38.5%) of 13 had a broad-spectrum antibacterial prescribed, respectively.

Figure 1.

Antibacterial prescription after enrolment among febrile inpatients and outpatients, Kilimanjaro Christian Medical Centre and Mawenzi Regional Referral Hospital, Tanzania, 2011–2014

Note. The overall antibacterial prescription is in green color and the broad-spectrum antibacterial prescription is in red color, overlaid with all patients (black) across CRP distributions (log scale). The left Y-axis represents antibiotic prescription (in percentage) while the right Y-axis represents the number of cases (both inpatients and outpatients in absolute numbers). The lines represent the logistic regression model between CRP-levels and: i. Prescription of overall antibacterials (green); ii. Prescription of broad-spectrum antibacterials (red); iii. Prescription of broad-spectrum antibacterials among those with an antibacterial agent prescribed (orange).

Among patients with a CRP <10 mg/L, 37 (61.7%) of 60 outpatients and 52 (67.5%) of 77 inpatients had an antibacterial prescribed; of these, 22 (59.5%) and 33 (63.5%), respectively, received a broad-spectrum antibacterial.

Among the outpatient participants, 190 (80.9%) of 235 were discharged and 41 (17.5%) were transferred to the hospital ward; outpatient disposition was not recorded for four participants. Among the 569 inpatients, 525 (92.3%) were discharged, 30 (5.3%) were transferred to other facilities, and 15 (2.6%) died in-hospital. Among the 15 inpatients who died, median CRP was 85 (IQR 39–150) mg/L, and two inpatient deaths had CRP levels below 10mg/L.

Laboratory confirmed diagnosis

A laboratory-confirmed diagnosis was assigned to 107 (13.3%) participants consisting of 34 BSIs (31 bacterial and 3 fungal), 61 bacterial zoonoses, and 12 malaria infections; co-infections were detected in five participants as detailed in Table 2.

Table 2.

Pathogens detected and etiological groupings among febrile inpatients and outpatients by age category, Kilimanjaro Christian Medical Centre and Mawenzi Regional Referral Hospital, Tanzania, 2011–2014.

Pathogen	Outpatient (n=8)		Inpatient (n=23)
Bacterial BSI (n=31)	Children (n=2)	Adults (n=6)	Children (n=1)	Adults (n=22)
E. coli	0	2	0	11
E. coli + Brucella	0	1	0	0
Enterobacteraeceae	0	0	0	2
K. pneumonia + Leptospira	0	0	0	1
Non-typhoidal Salmonella	0	0	0	2
Pseudomonas + P. falciparum	0	0	0	1
S. pneumoniae	1	0	0	0
S. pneumoniae + Brucella*	0	0	0	1
S. pyogenes	0	0	0	1
S. enterica serovar Typhi	1	3	1	3
Bacterial zoonosis (n=61)	n=0	n=17	n=2	n=42
Brucella	0	8	1	19
Brucella + P. falciparum	0	0	0	1
Brucella + R. africae	0	0	0	1
C. burnetii	0	0	1	3
C. burnetii + R. africae	0	0	0	1
Leptospira	0	6	0	9
Leptospira + Brucella	0	2	0	1
Leptospira + C. burnetii	0	0	0	1
R. africae	0	1	0	6
Fungal BSI (n=3)	n=0	n=0	n=0	n=3
C. neoformans	0	0	0	3
Malaria (n=12)	n=0	n=2	n=0	n=10
P. falciparum	0	2	0	10
Total (n=107)	n=2	n=25	n=3	n=77

Note. BSI – Bloodstream Infection . All detections of Brucella were based upon serologic testing; none of the brucellosis cases was based upon detection by aerobic blood culture.

The distribution of CRP concentrations in the different etiological groups is shown in Figure 2. The median CRP concentrations were 120 (46–224) mg/L in all bacterial infections, 173 (80–315) mg/L in bacterial BSI, 108 (31–208) mg/L in bacterial zoonoses, 117 (78–195) mg/L in malaria, and 2 (1–14) mg/L in fungal BSI. The median CRP concentration was 87 (28–191) mg/L among inpatients with undetermined cause of fever versus 32 (5–109) mg/L among outpatients with undetermined cause of fever (p<0.001). Two (6.5%) of 31 participants in the bacterial BSI group had CRP concentrations <20 mg/L: both were outpatients with Salmonella enterica serovar Typhi infection. One participant (3.2%) with a bacterial BSI had CRP concentration <40 mg/L (K. pneumoniae with leptospirosis co-infection). For the bacterial zoonotic infections, 8 (13.1%) and 11 (18.0%) of 61 had CRP concentrations <10 mg/L and <20 mg/L, respectively. Brucellosis cases accounted for 5 (62.5%) of the 8 bacterial zoonoses with CRP <10 mg/L and 8 (73%) of the 11 bacterial zoonoses with CRP <20 mg/L.

Figure 2.

Distribution of CRP concentrations by etiological group among febrile inpatients and outpatients, Kilimanjaro Christian Medical Centre and Mawenzi Regional Referral Hospital, Tanzania, 2011–2014.

Considering all bacterial infections, the sensitivity of CRP varied between 86% (75–92%) and 94% (85–98%) using a threshold ranging from 10 to 40 mg/L. The sensitivity for bacterial BSIs ranged from 97% (95% CI 84–97%) with a lower threshold of 10 mg/L to 90% (95% CI 75–97%) with a threshold of 40 mg/L. For bacterial zoonoses the sensitivity ranged from 87% (95% CI 76–93%) with a threshold of 10 mg/L to 72% (95% CI 60–82%) with a threshold of 40 mg/L. CRP sensitivity among inpatients with a bacterial zoonoses was 87% for 10 mg/L, 87% for 20 mg/L, and 73% for 40 mg/L. When excluding two participants with bacterial infections (Pseudomonas aeruginosa BSI and serologically confirmed Brucella, respectively) who were co-infected with Plasmodium falciparum, the sensitivity of the CRP thresholds for bacterial BSI ranged from 90–97%, and for bacterial zoonoses, the sensitivity of these thresholds ranged from 81–91%. The proportion of patients with CRP levels above each threshold for each etiological group is summarized in Table 3. When considering only adults (Supplementary Table 3), sensitivity reached 100% for detecting bacterial BSI using a 10 mg/L threshold, 96% (95% CI 89–100%) for a 20 mg/L threshold and 90% (95% CI 79–100%) for a 40 mg/L threshold. For bacterial zoonosis, CRP sensitivity among adults ranged from 88% (95% CI 80–97%) for 10 mg/L to 73% (95% CI 61–85%) for 40 mg/L thresholds. Sensitivity of CRP stratified by self-reported HIV infection status is shown in Supplementary Table 4.

Table 3.

Percent sensitivity (95% Confidence Intervals) of CRP at thresholds of 10, 20 and 40 mg/L by etiological group among febrile inpatients and outpatients, Kilimanjaro Christian Medical Centre and Mawenzi Regional Referral Hospital, Tanzania, 2011–2014. For each etiologic grouping, the number of febrile inpatients and outpatients exceeding the respective CRP threshold is shown as a numerator and the denominator shows the total number of febrile participants who met criteria for the etiologic grouping.

Etiologic Group	CRP >10mg/L	CRP >20mg/L	CRP >40mg/L
Bacterial BSI (n=31)	97 (83–100) 30/31	94 (79–99) 29/31	90 (74–98) 28/31
Bacterial zoonosis (n=61)	87 (76–94) 53/61	82 (70–91) 50/61	72 (59–83) 44/61
Fungal BSI (n=3)	33 (8–91) 1/3	0	0
Malaria (n=12)	92 (62–100) 11/12	92 (62–100) 11/12	83 (52–98) 10/12
Undetermined inpatients (n=489)	86 (83–89) 420/489	79 (75–82) 386/489	70 (66–74) 344/489
Undetermined outpatients (n=208)	73 (67–79) 152/208	64 (57–70) 132/208	51 (44–58) 106/208

Among the 208 outpatients with an undetermined cause of fever, 154 (74.0%) were prescribed an antibacterial. CRP-guided treatment would have resulted in the following prescribing proportions: 73.1%, using a threshold of 10 mg/L, 63.5% at 20 mg/L, and 51.0% at 40 mg/L. Furthermore, in these scenarios all patients with CRP-levels above these thresholds would (by definition) be prescribed antibacterials, as compared with current practice where 23 (21.7%) of 106 with CRP-levels >40 mg/L were not treated with any antibacterials. Among the 235 outpatients, broad-spectrum antibacterials would have been reduced from 95 (40.4%) to 73 (31.1%) using a threshold of 10 mg/L, to 63 (26.8%) at 20 mg/L, and to 50 (21.3%) at 40 mg/L.

Discussion

Our aim in this study was to assess whether CRP could reliably identify patients with bacterial infections in northern Tanzania, with view of validating CRP tests to guide antibacterial prescription in primary care settings. To obtain a high number of samples from patients with a documented bacterial infection, our study population consisted mainly of hospitalized patients. Among the 804 febrile inpatients and outpatients evaluated, a CRP threshold of 10 mg/L had high sensitivity (97%) among patients with a bacterial BSI and moderate sensitivity (87%) among patients with a bacterial zoonosis. While these findings suggest that CRP may be a safe guide for antibacterial prescription, we would not advocate use of CRP testing to withhold antibacterial treatment among hospitalized patients, who are generally severely ill. Rather, by demonstrating that CRP is elevated in more than 90% of patients with a confirmed bacterial BSI, these data from hospital-attended patients are a key step in examining the safety of CRP-guided therapy in sub-Saharan Africa. Based on these findings, further studies are warranted to examine the safety and clinical utility of CRP-guided therapy for febrile illness in first-level health facilities, the setting where point-of-care CRP testing might have the largest impact in safely reducing unnecessary antibacterial prescription.

Two recent studies evaluated the diagnostic utility of CRP in detecting bacterial infections among febrile outpatient children in Tanzania. Mahende et al. (2017) concluded that CRP could play a useful role in guiding antibacterial therapy in children, with an area under the receiver operator characteristic curve of 0.83 [29]. Hildenwall et al. (2017) also found CRP concentrations to be significantly higher in bacterial infections (median 41mg/L) than non-bacterial illnesses, but two of the six children with a positive blood culture had low CRP levels [30]. CRP levels in our outpatients were generally higher than both the latter two studies, as well as those in Keitel et al. (2017) where more than 91% of febrile children attending Tanzanian primary care clinics had CRP levels below 40 mg/L [31]. The latter study, a randomized non-inferiority trial with an endpoint of clinical failure at day 7, demonstrated that CRP-guided antibacterial treatment in an electronic care algorithm was non-inferior to an electronic care algorithm of Integrated Management of Childhood Illness without CRP testing.

Collectively, these three prior studies indicate that CRP has the potential to safely reduce unecessary antibacterial prescription in the pediatric primary care setting. As febrile adults constituted 92% of participants in our cohort, our study complements these findings among pediatric populations by demonstrating that CRP may also have a useful role in guiding antibacterial prescribing in older patients.

In our cohort, among outpatients prescribed an antibacterial, more than half were prescribed a broad-spectrum antibacterial. The likelihood of a broad-spectrum antibacterial prescription among outpatients did not increase or decrease based upon CRP concentrations, suggesting a widespread tendency among local clinicians to prescribe broad-spectrum agents. Broad-spectrum antibacterials are a major contributor to antimicrobial resistance [32], and CRP has recently been demonstrated to reduce such prescription among patients attending primary care in Southeast Asia. Among outpatient participants who were managed according to routine clinical practice, implementing a conservative prescribing strategy using a CRP-threshold of 10 mg/L would have reduced the use of broad-spectrum antibacterials by nearly a quarter, from 40% down to 31%; and a higher threshold of 40 mg/L would have reduced broad-spectrum antibacterial prescription by approximately half, from 40% down to 21%. Conversely, antibacterial coverage was missing in more than 20% of outpatients with CRP levels above 40 mg/L. This suggests that CRP may not only reduce unnecessary antibacterial prescription, but also help identify patients who warrant empiric antibacterials.

For bacterial zoonotic infections we found that CRP had moderate sensitivity for identifying participants when analyzing these infections as a collective group. These results should be interpreted with caution because of the small number of cases we identified for each of the bacterial zoonoses under study. Of note, CRP was elevated among the bacterial zoonotic cases (median 108 [IQR 31–108] mg/L), and the CRP thresholds captured most cases of leptospirosis, Q fever and spotted fever rickettsioses. Eight of the 11 bacterial zoonoses with CRP levels below 20 mg/L were brucellosis cases. Given that Brucella is an intra-cellular pathogen with mechanisms of immune-evasion, brucellosis might elicit only a mild to moderate CRP response [33]. Our finding of low CRP among brucellosis cases should be verified in other cohorts before drawing conclusions on CRP performance for this zoonotic disease. Similarly, while our findings for the other three bacterial zoonoses look promising, further study is warranted for each of those infections.

A limitation of our study is the absence of viral investigations, preventing the determination of the specificity of CRP in distinguishing between viral and bacterial febrile illness. In fever etiology studies in Southeast Asia where CRP levels in patients with bacterial infections were compared with those with viral infections, CRP had a specificity of 70% and 86% using thresholds of 20 mg/L and 40 mg/L, respectively [10, 12]. However, it is noteworthy that CRP levels in the current study were relatively high compared with similar studies done in Southeast Asia [34]. Among the children enrolled in this hospital-based febrile illness study, most had insufficient blood volume available for CRP testing. Compared to the children who had sufficient blood volume for CRP-testing, the children with insufficient blood volume had comparable characteristics except for being significantly younger and less likely to present with neurological complaints. Given the relatively small number of children included, conclusions drawn from this analysis are most relevant to febrile adults who comprised over 92% of the participants with CRP results. Participant HIV infection status was not confirmed with laboratory testing and the self-reported prevalance of 23.9% among adult inpatients is below the typical HIV prevalence of 32–39% among febrile adult inpatients at these study sites [7]; so we were limited in our ability to assess CRP performance among patients living with HIV and our results for this sub-group should be interpreted with caution. A prior study in Malawi demonstrated that CRP does rise in HIV-infected persons presenting with bacterial, mycobacterial, or fungal infections. In our analysis, CRP sensitivity at all three proposed thresholds was 100% among HIV-infected persons with a bacterial BSI, and sensitivity was 100% for those with a bacterial zoonotic infection, if using CRP thresholds of 10 or 20 mg/L. While this suggests that CRP can provide useful information to risk-stratify febrile HIV-infected persons in sub-Saharan Africa, further studies are warranted to evaluate cut-off thresholds in this patient population.

Conclusions

As in Southeast Asia, CRP was found to be a highly sensitive biomarker for bacterial bloodstream infections and moderately sensitive for bacterial zoonoses. Prospective data are needed from sub-Saharan Africa, particularly from first-level health facilities where CRP distributions are likely to be lower than in this hospital attended cohort, and where CRP testing might be most appropriately used to guide antibacterial prescription decisions.