INTRODUCTION
The management of acute febrile illnesses is a challenge worldwide, especially in low- and middle-income countries (LMICs). Resource-constraint settings have limited access to diagnostics, so the causative agent of fever most commonly remains unknown. Hence, algorithms based on clinical assessment have been implemented, with the aim of improving the management of febrile illnesses. However, this approach has limitations. As an example, the Integrated Management of Childhood Illness (IMCI), an algorithm for the management of childhood infections based on clinical evaluation, does not give proper indications when dealing with febrile children with no localizing symptoms.1 Therefore, in recent years, attempts to integrate the IMCI with results of diagnostic tests have been made.1
Since 2010, the deployment of malaria rapid diagnostic tests (RDTs) in endemic countries has made it possible to implement the WHO-recommended test-and-treat policy and more accurately target antimalarial treatment.2–4 However, at the same time, this has also created a treatment void for non-malaria fevers, which, because of lack of diagnostic tests, also unintentionally fueled irrational antibiotic prescriptions.5,6 Having tests that can assist case management decisions for fever not caused by malaria is therefore of paramount importance both in malaria-endemic and non-endemic areas. A test that could be deployed along the malaria RDT to help screening out non-malaria, nonbacterial infections would reduce antibiotic overprescribing and improve the management of febrile illnesses. The example of malaria RDTs highlights the importance of point-of-care tests (POCTs). These are, in general, automated, bench-top, or hand-held tests that could be used by staff with limited laboratory skills at the primary healthcare level.7 A test that could provide results quickly to the healthcare providers, without involving any specific laboratory equipment, would facilitate appropriate, prompt case management decisions.7
In particular, the ideal test to be deployed in LMICs should follow the ASSURED criteria8: it should be affordable, sensitive, specific, user friendly, rapid and robust, equipment free, and deliverable to end users. In high-income countries (HICs), C-reactive protein (CRP), procalcitonin (PCT), and white blood cells (WBCs) (total and differential counts) are variably used9 to inform rapid clinical assessment of patients with fever presenting to emergency departments. Although these markers can be altered in the presence of several noninfectious conditions (such as inflammation), they are deemed useful as first-line tests in the evaluation of febrile patients.
The impact on antibiotic prescription of CRP and other biomarkers assessed with POCTs, thoroughly reviewed mainly for acute respiratory tract infections,10 still remains uncertain. Moreover, there is scant evidence on the diagnostic cutoff to differentiate between bacterial and nonbacterial infections,11 and regulatory standards providing guidance for the interpretation of the results in clinical practice still lack.4
This work aimed at reviewing biomarker POCTs for acute non-malarial febrile illness in LMICs. Specific objectives were as follows: 1) to estimate the accuracy of such tests in differentiating fevers of bacterial from nonbacterial origin and 2) to describe both the impact of these tests on antibiotic prescription and clinical outcomes.
METHODS
A systematic review of the literature with meta-analysis was conducted. The protocol was registered with the PROSPERO international prospective registry of systematic reviews (registration n CRD42019141735). On February 10, 2020, the following databases were searched for relevant studies: PubMed, Embase, the Cochrane Library, and Bireme. The electronic search strategy was as follows: Biomarkers (Biomarker*, Diagnostic Tests, Routine, Biologic* Marker, Host biomarker*, Laboratory Marker*, Serum Marker*, Surrogate Endpoint*, Clinical Marker*, Viral Marker*, Biochemical Marker*, Immune Marker*, Immunologic Marker*, Surrogate Marker*, CPR, c reactive protein, diagnostic test, point of care) AND Fever (Fever*, Pyrexia*, Hyperthermia*, malarial febrile illness) AND LMICs (Poverty [MeSH Terms], Poverty Areas, low income, resource limited setting, middle income, low income countr*, middle income countr*, Africa [MeSH Terms], Latin America [MeSH Terms], Asia [MeSH Terms], Pacific Islands [MeSH Terms]). No restrictions were applied in relation to the language and the date of publication. We also screened the reference lists of all included studies for other potentially relevant studies and authors’ personal collections (gray literature).
Two authors, G. B. and N. R., independently screened the list of articles generated from the electronic search, using the EndNote program, version 6, 2012 (Thomson Reuters, Philadelphia, PA). G. B. and N. R. extracted the data on the basis of the inclusion/exclusion criteria (reported in the following text) and entered all information in a database created with MS Office Excel 2013 (Microsoft Corporation, Redmond, WA). In case of discrepancies in the process of inclusion of articles/data extraction, a consensus was reached through the involvement of a third author (D. B.).
The inclusion criteria included prospective or retrospective studies on host biomarkers POCTs for acute non-malarial fever in LMICs; POCTs were defined as bench-top/handheld devices providing results within 2 hours.7 The exclusion criteria included literature reviews, studies conducted in HICs, case reports or case series, overviews on the acceptability (social impact) of POCTs, POCTs for a specific condition (e.g., malnutrition and typhoid fever), or apparatus (e.g., respiratory infections). For the meta-analysis, when an article reported cases of malaria, data were extracted but excluded from the analysis. Unspecified fevers were also excluded.
Statistical analysis and data synthesis.
To characterize the study populations, demographic and clinical data were summarized with descriptive statistics using SAS software version 9.4 (SAS Inst., Cary, NC). When sensitivities and specificities of the biomarkers were not reported in the published article, we either contacted the authors or, when possible, extrapolated the data. We used RevMan version 5.3 (Review Manager, Copenhagen, Denmark) to produce coupled forest plots for these parameters.
C-reactive protein data were meta-analyzed using the hierarchical summary receiver operating characteristic (HSROC) model12,13 because CRP thresholds varied among studies—a variable cutoff was added to the model as covariate. To aid in data visualization, we present the coupled forest plot and the summary ROC (SROC) estimated by the HSROC model. We do not report summary estimates of sensitivity and specificity because studies used different CRP cutoffs (as per Rutter and Gatsonis,12 the SROC is drawn restricted to the range of specificities and sensitivities of the included studies to avoid extrapolation beyond the data). Meta-analysis was performed in SAS software version 9.4 using the METADAS v1.3 macro obtained from Cochrane website 13 and Stata software version 14 (StataCorp, College Station, TX). Parameters estimations are reported with calculated 95% CIs. Small sample size and high heterogeneity precluded fitting a HSROC model for PCT and meta-analyzing WBCs.
Quality assessment.
For the qualitative assessment of the articles included in the meta-analysis, we entered data in RevMan version 5.3 and used the QUADAS-2 tool (University of Bristol, Bristol, UK).
RESULTS
The electronic search identified 2,192 articles; the study flow from electronic search to inclusion of articles is described in the flowchart (Figure 1). Eventually, eight articles were included in the review.
Study flowchart.
Citation: The American Journal of Tropical Medicine and Hygiene 103, 5; 10.4269/ajtmh.19-0935
All studies were published from 2015 onward. Seven studies were prospective, two of which were randomized controlled trials; one study was retrospective. Table 1 reports the main characteristics of the included articles.
Main characteristics of the included articles
| Reference | Country | Study design | Studied population | Sample size | Host biomarker | Setting |
|---|---|---|---|---|---|---|
| Althaus et al.14 | Thailand and Myanmar | Multicentric open RCT* | Adult and children > 1 year | 2,410 | CRP: NycoCard Reader II (Axis Shield, Oslo, Norway)† | OPD |
| Wangrangsimakul et al.15 | Thailand | Monocentric prospective observational study | Adults | 200 | CRP: NycoCard Reader II (Axis Shield)† | IPD |
| PCT: ELISA-based VIDAS PCT (BioMeÂrieux, Marcy-L’Etoile, France)† | ||||||
| WBC* and ANC* | ||||||
| Mahende et al.16 | Tanzania | Monocentric retrospective | Children (2–59 months) | 867 | CRP: Cobas c111 biochemistry analyzer (Roche diagnostics, Indianapolis, IN)† | OPD |
| WBC and ANC Melet schloesing MS9-5 (Diamond Diagnostics, Holliston, MA)† | ||||||
| Keitel et al.1 | Tanzania | Multicentric RCT* | Children (2–59 months) | 3,192 | CRP: semi-quantitative assay (bioNexia CRPplus, BiomeÂrieux)†‡ | OPD |
| PCT: ELISA-based VIDAS PCT (BioMeÂrieux)† | ||||||
| Hildenwall et al.17 | Tanzania | Monocentric prospective | Children (3 months–5 years) | 428 | CRP: Afinion AS100 Analyzer™ (Axis-Shield)† | OPD |
| WBC (HemoCue, Angelholm, Sweden)† | ||||||
| Phommasone et al.18 | Laos | Multicentric prospective | All population | 837 | CRP: DTS233 (Creative Diagnostics, New York, NY)†‡ | OPD |
| CRP: WD-23 (Assure Tech Co., Ltd., Hangzhou, China)†‡ | ||||||
| CRP: semi-quantitative assay (bioNexia CRPplus, BiomeÂrieux)†‡ | ||||||
| Lubell et al.19 | Laos | Multicentric prospective | 5–49 years | 1,083 | CRP: NycoCard Reader II (Axis Shield)† | OPD |
| Lubell et al.20 | Cambodia Laos Thailand | Multicentric prospective | All population | 1,372 | CRP: NycoCard Reader II (Axis Shield)† | OPD |
| PCT: ELISA-based VIDAS PCT (BioMeÂrieux)† |
ANC = absolute neutrophil count; CRP = C-reactive protein; IPD = inpatient department; OPD = outpatient department; PCT = procalcitonin; RCT = randomized controlled trial; WBC = white blood cell.
The article did not report the type of assay.
POCT = point-of-care test.
RDT = rapid diagnostic test.
Five studies were conducted in Southeast Asia (SEA: Thailand, Laos, Myanmar, and Cambodia), and three in Africa (Tanzania).
The three studies from Africa enrolled children aged from 1 to 59 months; of the five studies from Southeast Asia, four included patients of all ages and one adult only. Seven studies included data from outpatients. The sample size varied from a minimum of 200 to a maximum of 3,192 individuals. Six studies had a sample size larger than 500 patients.
Main biomarkers studied.
As shown in Table 1, only three studies evaluated a single host biomarker (namely, CRP), whereas most articles analyzed different host biomarkers. Overall, CRP was investigated in all studies, PCT in three, WBCs in three, and absolute neutrophil counts (ANCs) in two.
C-reactive protein.
Different CRP POCTs were evaluated in the reviewed studies, the most frequent (4/8 studies) being NycoCard Reader II (Axis Shield, Oslo, Norway). Only two studies evaluated an RDT (Phommasone et al.18 and Keitel et al.1). The main aim of Phommasone et al.18 was to assess the accuracy of RDTs based on CRP in LMICs. They found that three commercially available CRP-based RDTs (one qualitative and two semi-quantitative tests) provided reliable results when compared with quantitative tests such as the NycoCard Reader II. This study did not analyze the accuracy of the RDT in relation to the etiology of the disease (nonbacterial/bacterial infection).
Only four studies15–17,20 reported data on the accuracy of the POCTs in discriminating bacterial from nonbacterial fevers, but only three of them, based on CRP, were included in the meta-analysis; Mahende et al.’s16 study was excluded because it did not provide etiological diagnosis. The quality analysis of the three studies included in the meta-analysis is shown in Figure 2; the risk of bias was acceptable (low/unclear) for all considered domains; two studies (Lubell et al.20 and Wangrangsimakul et al.15) had only one unclear domain each, whereas Hildenwall et al.17 had four unclear domains because of incomplete information reported in the article.
Risk of bias and applicability concerns summary.
Citation: The American Journal of Tropical Medicine and Hygiene 103, 5; 10.4269/ajtmh.19-0935
The meta-analysis of the accuracy of CRP for the identification of bacterial infections was carried out on six datasets, retrieved from the three included studies (Figure 3). Five different CRP cutoff values were included in the analysis. In general, the higher the cutoff, the lower the sensitivity and the higher the specificity for the values tested within the same study (Lubell et al.20 and Wangrangsimakul et al.15), but only in the latter, CIs between the 10 and 20 mg/L cutoff did not overlap.
Forest plot for studies on the accuracy of C-reactive protein.
Citation: The American Journal of Tropical Medicine and Hygiene 103, 5; 10.4269/ajtmh.19-0935
The SROC obtained from the HSROC model, with covariate cutoff, is shown in Figure 4. The SROC represents a prediction of the accuracy of CRP obtained by including in the model the results of different studies, with different CRP cutoff values, which is why the accuracy reported by each study does not lie on the summary curve. Each circle represents one dataset, and their areas are proportional to the sample size of each study. Estimations from Hildenwall et al.17 with a sensitivity of 44% and Wangrangsimakul et al.15 (cutoff = 36) fall outside the 95% CI, whereas the only studies with estimation entirely within the CI are Lubell et al.20 (cutoff = 20) and Wangrangsimakul et al.15 (cutoff = 20). The area under the summary curve (AUC) is 0.77 (CI: 0.73–0.81), which indicates good accuracy of the test, although the best cutoff to obtain this performance could not be assessed.
Summary receiver operating characteristic displaying the accuracy of C-reactive protein.
Citation: The American Journal of Tropical Medicine and Hygiene 103, 5; 10.4269/ajtmh.19-0935
Procalcitonin.
Two studies1,20 reported data on the diagnostic accuracy of PCT-based POCTs for the identification of bacterial infections. For three studies,1,15,20 different PCT cutoff values (0.1, 0.25, and 0.5 ng/mL) were retrieved from four datasets. Sensitivities and specificities were extrapolated from data reported in each article, and presented in the forest plot (Figure 5). The cutoff 0.5 ng/mL leaded to different accuracy of PCT, with a higher specificity and sensitivity in Wangrangsimakul et al.15 Both Lubell et al.20 and Wangrangsimakul et al.15 conducted their works in Southeast Asia, but the former20 involved people of any age, and the latter only adults.
Forest plot for studies on the accuracy of procalcitonin.
Citation: The American Journal of Tropical Medicine and Hygiene 103, 5; 10.4269/ajtmh.19-0935
White blood cell (WBC).
The studies evaluating the association between WBC total counts and bacterial versus nonbacterial infections had heterogeneous results. Hildenwall et al.17 found no difference in WBC counts between children with and without signs of bacterial infections; Mahende et al.16 found only a weak association between WBC counts > 15 × 103/mm3 and a positive blood culture; and Wangrangsimakul et al.15 found that WBC count < 7.9 × 103/mm3 was predictor for viral infection.
Only two articles analyzed ANCs,15,16 and none of them found conclusive results.
Antibiotic prescriptions and clinical outcomes following the introduction of POCTs.
Three studies1,14,19 examined the effect of POCTs on antibiotic prescription. One modeling study19 estimated 80% correct antibiotic prescriptions with CRP-based POCTs compared with 52% in routine practice. The other two studies reported the overall impact on antibiotic prescription by testing strategy. Althaus et al.14 found that the use of POCTs resulted in a significant decrease in antibiotic prescription for low CRP values, and increased targeted prescription for high CRP values. Keitel et al.1 compared antibiotic prescription in three different arms: routine care; the ALMANACH (an electronic algorithm derived from the IMCI, including urine dipstick test and a clinical predictor or rapid test for typhoid fever); and ePOCT (an innovative electronic algorithm including CRP and PCT-based POCTs). They observed that the proportion of children who were prescribed an antibiotic was 95%, 29.7%, and 11.5% in the routine care, in the ALMANACH, and in the ePOCT arms, respectively.
Althaus et al.14 attempted to address the potential impact of POCT–CRP testing on clinical outcome. Despite the improved use of antibiotics, the reported proportion of patients who recovered at days 5 and 14 of follow-up, comparing two CRP testing groups, with thresholds set at 20 mg/L (group A) or 40 mg/L (group B), with a control group (routine practice), was similar across the study groups.
On the other hand, Keitel et al.1 observed significantly fewer clinical failures by day 7 (49%) when the experimental electronic algorithm was used.
DISCUSSION
In LMICs, the evaluation and deployment of POCTs aimed to improve the accuracy of diagnosis and management of acute fevers are challenging. Little is known about the causes of fever locally, and excluding malaria RDTs, there are few effective pathogen-specific POCTs, which leaves healthcare providers with just their clinical judgment where malaria is not endemic or after excluding malaria. This sustains “just-in-case” antibiotic prescriptions, which is one of the root causes of antimicrobial resistance and inappropriate fever case management.5,21 In the absence of pathogen-specific POCTs other than malaria RDTs, the main question is how healthcare providers can be assisted in deciding whether antibiotic treatment is warranted or not. One possibility is using biomarkers that can orient toward a bacterial or nonbacterial cause of infection.6
However, there is a series of caveats to consider when interpreting findings and projecting the use of POCTs in clinical practice, including how age and patterns of prevailing infections differ in studies conducted in different geographical areas. Many infections are seasonal, and studies should cover a period long enough to capture these fluctuations; we know for malaria that the positive and the negative predictive values of RDTs differ in high- and low-transmission seasons.22 Furthermore, concomitant conditions like HIV infection and malnutrition could influence the expression of biomarkers, for example, lowering CRP concentrations, thus potentially causing bacterial infections being underdiagnosed and undertreated.23 Conversely, some biomarkers, CRP in particular, can be elevated in case of malaria,24 which makes it impossible to rule out or rule in other infections in patients with concomitant malaria infection.
We found a limited number of studies to support the use of POCT biomarkers, and we could reasonably draw conclusions only for CRP. These studies show, based on few good-quality studies, that CRP is reasonably accurate in distinguishing bacterial from nonbacterial infections (HSROC AUC 0.77). Although it was not possible to identify the best cutoff, 20 mg/L was the only cutoff lying within the 0.73–0.81 CIs of the AUC. Of note, using the HSROC model with covariate cutoff allowed to include test accuracy data obtained with different thresholds, although this meant not having a summary point for sensitivity and specificity. Limitations mainly relate the heterogeneity of the included studies, which could have influenced the final interpretation. Indeed, the studies were carried out in different geographical/epidemiological contexts, where different degree of coinfections (including malaria) might affect the results of the tests. Moreover, the methodologies, including study design, age-groups of studied populations, and case definition, for bacterial versus nonbacterial infections differed between studies, partially affecting the comparison.
The question then was whether applying different cutoff values for CRP plasma levels impacted antibiotic prescription and clinical outcome. Only three studies addressed these questions: two1,19 for antibiotic prescription and one14 for clinical outcomes. Results were not entirely consistent, possibly because different populations were studied (adults in Southeast Asia and children in Africa) and different clinical decision algorithms were applied. Because available studies did not address simultaneously the effects of different CRP cutoffs on diagnostic accuracy, antibiotic prescriptions, and clinical outcomes, it is difficult to draw conclusions as to whether using CRP (or possibly other markers) translates into better case management decisions, reducing unnecessary antibiotic prescription while not denying patients’ antibiotics when needed. When the decision to prescribe or not an antibiotic is based on CRP alone (using cutoffs at 20 or 40 mg/L),14 clinical outcome was not affected as compared with a management decision made on clinical ground. The best prospects in providing better care are likely when biomarkers (CRP and PCT) are used in the context of improved electronic algorithms, but so far this evidence comes from only one non-inferiority study.1
Another question that still remains unanswered is if the deployment of algorithms combining clinical signs (e.g., fever, rash, and cough) associated with host biomarkers and locally relevant pathogen-specific tests (e.g., malaria, arbovirosis, HIV, influenza, and typhus) have a clinical impact on the management of fever of unknown origin. Although some electronic algorithms have shown a positive influence on antibiotic prescription, it is not clear if they can also have an impact on clinical outcome. Moreover, the deployment of an electronic device to run such algorithms might not be feasible in remote settings, where technologies requiring minimal skills and equipment are needed.
Indeed, only two of the POCTs studied were RDTs; all the others required readers and personnel training which may not be available at peripheral health centers in LMICs. Semi-quantitative CRP-based RDTs could be a reasonable compromise, although reading and interpretation might not be immediately obvious.18
Overall, the available data on CRP performance are enough to justify further studies to validate the best threshold for bacterial versus nonbacterial fever diagnostic triage. In parallel, research should go on to identify a battery of markers and pathogen-specific diagnostic tests, and develop diagnostic algorithms adapted to different levels of the health system.
REFERENCES
- 1.↑
Keitel K et al. 2017. A novel electronic algorithm using host biomarker point-of-care tests for the management of febrile illnesses in Tanzanian children (e-POCT): a randomized, controlled non-inferiority trial. PLoS Med 14: e1002411.
- 2.↑
World Health Organization, 2015. Guidelines for the Treatment of Malaria. Geneva, Switzerland: WHO Press.
- 3.
World Health Organization, 2013. WHO Informal Consultation on Fever Management in Peripheral Health Care Settings: A Global Review of Evidence and Practice. Geneva, Switzerland: WHO Press.
- 4.↑
Escadafal C, Nsanzabana C, Archer J, Chihota V, Rodriguez W, Dittrich S, 2017. New biomarkers and diagnostic tools for the management of fever in low- and middle-income countries: an overview of the challenges. Diagnostics (Basel) 7: 44.
- 5.↑
World Health Organization, 2014. Antimicrobial Resistance: Global Report on Surveillance 2014. Geneva, Switzerland: WHO Press.
- 6.↑
Dupuy AM et al. 2013. Role of biomarkers in the management of antibiotic therapy: an expert panel review: I–currently available biomarkers for clinical use in acute infections. Ann Intensive Care 3: 22.
- 8.↑
Kosack CS, Page AL, Klatser PR, 2017. A guide to aid the selection of diagnostic tests. Bull World Health Organ 95: 639–645.
- 9.↑
Kapasi AJ, Dittrich S, Gonzalez IJ, Rodwell TC, 2016. Host biomarkers for distinguishing bacterial from non-bacterial causes of acute febrile illness: a comprehensive review. PLoS One 11: e0160278.
- 10.↑
Aabenhus R, Jensen J, Jørgensen K, Hróbjartsson A, Bjerrum L, 2014. Biomarkers as point‐of‐care tests to guide prescription of antibiotics in patients with acute respiratory infections in primary care. Cochrane Database Syst Rev 11: CD010130.
- 11.↑
Van den Bruel A, Thompson MJ, Haj-Hassan T, Stevens R, Moll H, Lakhanpaul M, Mant D, 2011. Diagnostic value of laboratory tests in identifying serious infections in febrile children: systematic review. BMJ 342: d3082.
- 12.↑
Rutter CM, Gatsonis CA, 2001. A hierarchical regression approach to meta-analysis of diagnostic test accuracy evaluations. Stat Med 20: 2865–2884.
- 13.↑
Rutter CM, Gatsonis CA, 1995. Regression methods for meta-analysis of diagnostic test data. Acad Radiol 2 (Suppl 1): S48–S56 ; discussion S65–S67, S70–S71 pas.
- 14.↑
Althaus T et al. 2019. Effect of point-of-care C-reactive protein testing on antibiotic prescription in febrile patients attending primary care in Thailand and Myanmar: an open-label, randomised, controlled trial. Lancet Glob Health 7: e119–e131.
- 15.↑
Wangrangsimakul T, Althaus T, Mukaka M, Kantipong P, Wuthiekanun V, Chierakul W, Blacksell SD, Day NP, Laongnualpanich A, Paris DH, 2018. Causes of acute undifferentiated fever and the utility of biomarkers in Chiangrai, northern Thailand. PLoS Negl Trop Dis 12: e0006477.
- 16.↑
Mahende C, Ngasala B, Lusingu J, Martensson T, Lushino P, Lemnge M, Mmbando B, Premji Z, 2017. Profile of C-reactive protein, white cells and neutrophil populations in febrile children from rural north-eastern Tanzania. Pan Afr Med J 26: 51.
- 17.↑
Hildenwall H, Muro F, Jansson J, Mtove G, Reyburn H, Amos B, 2017. Point-of-care assessment of C-reactive protein and white blood cell count to identify bacterial aetiologies in malaria-negative paediatric fevers in Tanzania. Trop Med Int Health 22: 286–293.
- 18.↑
Phommasone K et al. 2016. Accuracy of commercially available C-reactive protein rapid tests in the context of undifferentiated fevers in rural Laos. BMC Infect Dis 16: 61.
- 19.↑
Lubell Y, Althaus T, Blacksell SD, Paris DH, Mayxay M, Pan-Ngum W, White LJ, Day NP, Newton PN, 2016. Modelling the impact and cost-effectiveness of biomarker tests as compared with pathogen-specific diagnostics in the management of undifferentiated fever in remote tropical settings. PLoS One 11: e0152420.
- 20.↑
Lubell Y et al. 2015. Performance of C-reactive protein and procalcitonin to distinguish viral from bacterial and malarial causes of fever in southeast Asia. BMC Infect Dis 15: 511.
- 21.↑
Dittrich S et al. 2016. Target product profile for a diagnostic assay to differentiate between bacterial and non-bacterial infections and reduce antimicrobial overuse in resource-limited settings: an expert consensus. PLoS One 11: e0161721.
- 22.↑
Bisoffi Z et al. 2010. Accuracy of a rapid diagnostic test on the diagnosis of malaria infection and of malaria-attributable fever during low and high transmission season in Burkina Faso. Malar J 9: 192.
- 23.↑
Duran-Chavez C, Sanchez-Herrera G, Cañedo-Solares I, Perez-Ortiz B, 1994. C-reactive protein in severely malnourished children. Nutr Res 14: 967–975.
- 24.↑
Kortz TB, Nyirenda J, Tembo D, Elfving K, Baltzell K, Bandawe G, Rosenthal PJ, Macfarlane SB, Mandala W, Nyirenda TS, 2019. Distinct biomarker profiles distinguish Malawian children with malarial and non-malarial sepsis. Am J Trop Med Hyg 101: 1424–1433.