A prospective study of bloodstream infections as cause of fever in Malawi: clinical predictors and implications for management


Remco P. H. Peters, Current affiliation: VU University Medical Center, Department of Internal Medicine, PO Box 7057, 1007MB Amsterdam, the Netherlands. E-mail: r.peters@vumc.nl
Ed E. Zijlstra (corresponding author), Maarten J. Schijffelen, John J. Kumwenda and David K. Lewis, Department of Medicine, College of Medicine, University of Malawi, Private Bag 360, Chichiri, Blantyre 3, Malawi. Tel.: +265 1 670202; Fax: +265 1 673933; E-mail: eezijlstra@malawi.net, schijffelen@hotmail.com, jonnykumwenda@hotmail.com, mariadavid@callaghanlewis.fsnet.co.uk
Amanda L. Walsh, James G. Kublin, Malcolm E. Molyneux, Malawi-Liverpool-Wellcome Trust Research Programme, Private Bag 396, Blantyre, Malawi. Tel.: +265 1 676444; Fax: +265 1 675774; E-mail: amanda.walsh@hpa.org.uk, james_kublin@merck.com, mmolyneux@malawi.net
George Joaki, Department of Microbiology, College of Medicine, University of Malawi, Private Bag 360, Chichiri, Blantyre 3, Malawi. E-mail: joaki@med.unc.edu


Objective  To determine the contribution of a blood culture service to the diagnosis of fever in a resource-poor setting and to identify clinical predictors of specific bloodstream infections (BSI).

Methods  In a descriptive, prospective study at the Medical Wards at Queen Elizabeth Central Hospital, Blantyre, Malawi, we tried to identify a specific cause of fever in febrile patients, comparing the use of routinely available diagnostic methods with the same methods plus blood culture. Clinical predictors of specific BSIs were sought.

Results  A total of 352 patients admitted with fever (axillary temperature ≥37.4 °C) or a history of fever within the last 4 days were enrolled. Tuberculosis (TB) was the diagnosis most commonly suspected initially on clinical grounds (28%), followed by lower respiratory tract infection (16%), malaria (12%) and gastroenteritis (5%). Blood cultures were positive in 128 patients (36%); Mycobacterium tuberculosis was the most commonly isolated organism (57 patients). In most cases the diagnosis of TB had already been made using routinely available diagnostic methods, including chest radiography. In all 16 cases of Streptococcus pneumoniae bacteraemia, infection with this agent was clinically suspected, usually on the basis of pulmonary symptoms and signs. In contrast, in 30 of 65 patients (65%) with non-typhi salmonellae (NTS) bacteraemia, there were no symptoms or signs specifically suggestive of this diagnosis. Fever ≥39 °C and splenomegaly predicted NTS bacteraemia with an odds ratio of 8.4 (95% confidence interval 3.4–20.6, P < 0.001).

Conclusion  BSIs are common among patients admitted with fever. While BSI with mycobacteraemia and S. pneumoniae can usually be predicted on clinical grounds and with routinely available diagnostic methods, NTS bacteraemia often presents as a primary BSI without localizing symptoms and signs. Splenomegaly in this population indicates NTS bacteraemia rather than malaria.


Febrile illness is a common clinical presentation in patients admitted to medical wards in Sub-Saharan Africa. Although infectious diseases have always been a major cause of admission (Odio et al. 1982; Harries et al. 1990; Brown 1975), ranking of the most likely diagnoses may have changed during the last two decades as a result of the HIV-pandemic. In a recent study, 70% (88% in age group 30–40 years) of all admissions to the medical wards at Queen Elizabeth Central Hospital (QECH), Blantyre, were HIV seropositive (Lewis et al. 2003).

The HIV pandemic has a major effect on the global burden of tuberculosis (TB). Increased frequencies of clinical malaria and parasitaemia have also been reported in HIV-infected individuals, especially in those with advanced immunosuppression (Whitworth et al. 2000; French et al. 2001). Bacteraemias, particularly by Streptococcus pneumoniae and non-typhi salmonellae (NTS), are a major cause of morbidity and mortality in HIV-infected patients in African medical wards (Gilks et al. 1990; Lucas et al. 1993; Grant et al. 1997). The diagnosis of these infections can only be confirmed by blood culture, which is routinely available in very few hospitals in tropical Africa. Finding clinical markers of bacteraemia in patients presenting with fever would improve empirical management, including the use of appropriate antibiotics. We studied how blood culture results influence diagnosis and to what extent clinical parameters can predict the outcome of blood cultures in patients admitted with fever to the medical wards of QECH in Blantyre, Malawi.


Study design

This prospective study was conducted between February and May 2000 in QECH, the largest government hospital in Malawi and the main provider of secondary health care in Blantyre and surrounding districts. On two random days per week all adults (age ≥14 years) admitted to the medical wards were considered for the study. Patients were eligible if they had an axillary temperature ≥37.4 °C, or were in shock (pulse ≥120 beats/min and systolic blood pressure <100 mmHg) or if they reported a history of fever in the last 4 days. After counselling and informed consent, a full history was taken and a physical examination conducted. Before any antibiotic treatment was started, a venous blood sample was collected for full blood count, thick film for malaria parasites, HIV serology and culture for bacteria and mycobacteria. A chest X-ray was taken for all patients within 1 day of admission. From those with cough three sputum samples were collected and examined for acid-fast bacilli (AFB). Additional investigations were undertaken when clinically indicated, and treatment was given according to standard practice in our department. Patients were reviewed daily.

Initially patients were diagnosed according to the routinely available diagnostic tools and a list of most important diagnoses was made. Diagnoses were later revised according to blood culture results.

Clinical definitions

Patients were defined as having AIDS on admission on the basis of the expanded WHO surveillance criteria for AIDS in Africa (World Health Organisation 1994). Clinical malaria was defined as a documented febrile illness with parasites in the blood film, no other identified aetiology (including no growth on blood culture) and resolution of fever on antimalarial treatment (normally sulphadoxine–pyrimethamine). Patients were diagnosed as having bacteraemia after isolation of a known pathogen in the blood culture. Patients who had a positive blood culture only, without clinical or radiological signs of focal disease, were designated as having a primary bloodstream infection (BSI).

Patients were classified as smear-positive, smear-negative or extrapulmonary TB according to guidelines of the National Tuberculosis Control Programme of Malawi. Pneumonia was defined as an episode of acute cough (<1 week), accompanied by consolidation on a chest X-ray and sputum-smears negative for AFB; patients presenting with similar criteria but without radiological evidence of consolidation were diagnosed as other lower respiratory tract infections. Patients were diagnosed as having Pneumocystis carinii pneumonia (PCP) if the clinical picture was suggestive (dry cough, breathlessness, paucity of chest findings, typical chest X-ray and response to treatment with high dose cotrimoxazole) (Malin et al. 1995). Meningitis was diagnosed on the basis of cerebrospinal fluid characteristics. Gastroenteritis was diagnosed when the patient complained of diarrhoea and vomiting. Patients with febrile illness of other origin were classified as other febrile illness; if no focus could be identified the fever was regarded as of unknown aetiology.

Ethical approval for the study was obtained from the University of Malawi College of Medicine Research and Ethics Committee.

Laboratory procedures

A full blood count was performed using a Coulter® Onyx Analyzer. Blood samples were tested for HIV using Abbot Determine (Dainabot Company, Tokyo, Japan) and confirmed by the Unigold test (Trinity Biotech, Bray, Ireland). Thick films for malaria parasites were stained with Field's stain and malaria parasites were counted against 200 leucocytes. Final calculation of the number of parasites per microlitre assumed an average leucocyte count of 10 000 leucocytes/μl. All slides were read by two independent microscopists blinded to clinical data; discrepancies of >10% were checked by a third microscopist.

For bacterial culture 5 ml blood were added to a single blood culture bottle (50 ml Brain Heart Infusion Broth containing sodium polyanethol sulphonate, E&O Laboratories, UK) and incubated overnight at 37 °C before venting. Cultures were examined macroscopically every day, followed by Gram staining if turbid or haemolysed. Routine subcultures on to sheep blood agar were performed for all bottles after 18–24, 36–48 h and finally after 7 days. Isolates were identified according to standard techniques, including sero agglutination (Pro-Lab Diagnostics, Merseyside, UK and Murex Biotech, Dartford UK) and biochemical tests (API strips, BioMerieux UK Ltd, Basingstoke, UK). Mycobacterial blood cultures were performed as described by Lewis et al. (2002).

Statistical analysis

Data were double-entered into EPI-INFO version 6.04b, validated and analysed using SPSS 10.07. Categorical data were compared with the chi-square test with Yates’ correction, and Fisher's exact test, where appropriate. Continuous variables were compared with the t-test or Wilcoxon two-sample test. Using logistic regression, indicators of bacteraemia and malarial parasitaemia were identified and odds ratios (OR) with 95% confidence intervals were calculated. Subsequently, a multivariate analysis was performed adjusting for the indicators found, to identify independent clinical predictors.


During the study period 637 patients were admitted to the medical wards, of whom 367 met the inclusion criteria; 352 (96% of those eligible) gave consent and were enrolled in the study; 44% were men. On admission, all patients were febrile except 17, who gave a history of fever; nine patients had shock and were febrile. The distribution of recorded temperatures was 37.4–37.9 °C in 67 patients (19%), 38.0–38.4 °C in 80 (23%), 38.5–38.9 °C in 54 (15%) and ≥39 °C in 134 (38%). A total of 291 patients (83%) were HIV-infected, of whom 162 met the criteria for AIDS on admission (56%). More women than men were HIV-infected (88%vs. 77%, P = 0.005). Respiratory tract related findings were by far the most common clinical presentation: 69% of patients presented with cough, 32% had a cough for >1 month, and 62% had shortness of breath. Chest X-rays were abnormal in 65% with infiltrative changes in 85% and consolidation in 25%. Twelve per cent of study patients died in hospital (12% of those who were HIV positive and 11% of those who were HIV negative, P = 0.90).

Table 1 shows initial diagnoses. TB was the most common with 103 cases (29%); six patients with smear-positive and five patients with smear-negative pulmonary TB also had extrapulmonary TB. Lower respiratory tract infections were the second commonest initial diagnosis (16%), followed by malaria (12%), gastroenteritis (5%) and meningitis (5%). In 33 (9%) patients another febrile illness was diagnosed, such as urinary tract infection or pelvic inflammatory disease. In 87 patients (25%) the cause of the febrile illness remained unknown; this was more common in those who were HIV negative (23% in HIV positive and 34% in HIV negative, P = 0.053). This group had a similar mean temperature as those for whom a clinical diagnosis could be made (38.6 °C vs. 38.5 °C, respectively).

Table 1.  Initial diagnosis for 352 study patients using routinely available diagnostic methods by category in numbers and according to HIV status. Only the most likely main diagnosis is listed. Odds ratios were calculated comparing the likelihood of a diagnosis in HIV-positive and -negative individuals
DiagnosisTotal (n = 352)HIV-positive (n = 291)HIV-negative (n = 61)OR (95% CI)
  1. * Not significant.

  2. † Six patients with smear-positive and five patients with smear-negative pulmonary tuberculosis were also diagnosed with extrapulmonary TB.

  3. ‡ Another three patients with tuberculosis are listed under meningitis.

  4. § These include urinary tract infection, abscesses and pelvic inflammatory disease.

Tuberculosis†100‡9464.4 (1.8, 12.8)
Lower respiratory tract infections55496*
 Pneumocystis carinii pneumonia550Undefined
 Cryptococcus neoformans440Undefined
Others§3320130.34 (0.15, 0.76)
Unknown febrile illness876621*

Pathogens were isolated from blood cultures of 128 patients (36%); 73 patients had bacteraemia (21%), 59 mycobacteraemia (17%) and one fungaemia; six patients had dual infections. Of the 59 mycobacterial isolates, M. tuberculosis was the most common with 57 positive cultures (16% of all study patients); the other two were atypical mycobacteria. NTS were isolated in 46 cases (13%) and S. pneumoniae in 16 (5%). Of those with bacteraemia 19% (14/73) died, compared with 9% (25/279) of those without (OR 2.4, 95% CI 1.1–5.2).

Table 2 shows the most commonly isolated pathogens and the corresponding initial diagnosis (if any). In six of 57 cases (11%) of M. tuberculosis mycobacteraemia, an initial diagnosis other than TB had been made; in seven (12%) others who had no focal signs, TB was strongly suspected but not confirmed before blood cultures (primary mycobacteraemia). In the remaining 44 a diagnosis of TB was made using routinely available diagnostic methods. Forty-two patients (58% of the total number of patients with bacteraemia) had a primary BSI as a single final diagnosis; in another five patients bacteraemia was a second, unsuspected diagnosis (one patient with meningitis, three with TB, one with PCP). There were 16 cases of S. pneumoniae bacteraemia; in 14 respiratory infection was suspected, one patient who also had evidence of respiratory infection presented with pneumococcal peritonitis and the remaining patient had pneumococcal meningitis. In one case there was co-infection with TB. For NTS bacteraemia the majority (65% of cases) were primary BSI; in four patients there was an unsuspected co-infection with cryptococcal meningitis, TB (two patients) or PCP, respectively.

Table 2.  Frequency of isolation of most important pathogens from blood cultures and corresponding initial diagnoses: total number, and number (%) that are HIV positive
PathogenInitial diagnosisnHIV +ve
  1. Six patients had dual infections – five with M. tuberculosis and other bacteria (two S. pneumoniae, one Salmonella species, P. aeruginosa and K. pneumoniae); one patient with S. enteritidis and S. pneumoniae.

  2. LRTI, lower respiratory tract infection.

  3. * These include: Mycobacterium species (two), K. pneumoniae (two), P. aeruginosa, S. aureus, P. mirabilis, S. pyogenes, S. typhi, C. neoformans and Flavibacterium spp.

M. tuberculosis 57 55 (96)
Tuberculosis44 42 (96)
Other LRTI5  5
Gastroenteritis1  1
Primary7  7
Non-typhi salmonellae46 44 (96)
S. typhimurium (n = 28)Pneumonia3  3
Gastroenteritis2  2
Meningitis2  2
Lung abscess1  1
Primary20 20
S. enteritidis (n = 16)Gastroenteritis4  4
Pneumonia1  1
PCP1  1
Tuberculosis1  1
Primary9  8 (89)
Salmonella species (n = 2)Tuberculosis1  1
Primary1  0
S. pneumoniae 16 13 (81)
Pneumonia9  7 (78)
Other LRTI1  1
Tuberculosis1  1
Meningitis1  1
Peritonitis1  1
Primary3  2 (67)
Escherichia coli 4  4 (100)
Primary4  4
Others* 11  8 (78)
Other LRTI2  2
Tuberculosis2  2
Meningitis1  1
Pyelonephritis1  1
Primary5  2 (40)

Primary BSI and clinical malaria (42 cases each) were the third commonest diagnosis after TB and LRTI (55 cases). Malaria parasites were seen in thick films of 69 patients (20%); all films showed Plasmodium falciparum; three also showed Plasmodium malariae. The median parasite density was 19 500/μl (range 100–657 900). This did not significantly differ between the group of 42 patients (12% of total) in whom a diagnosis of clinical malaria was made and the 27 patients with other clinical diagnoses. Nine patients with malarial parasitaemia also had positive blood cultures: five with bacteria (three with S. typhimurium and two with E. coli) and four with mycobacteria. Bacteraemia was less common in parasitaemic patients than those without parasitaemia (OR 0.25 [95% CI 0.1–0.6]). Fifty-seven patients (83%) with parasitaemia were HIV-infected; there was no significant association between HIV status and the presence or level of parasitaemia (P = 0.99 and 0.83, respectively), nor with the clinical diagnosis of malaria (P = 0.36). Patients with AIDS were less likely to have been admitted for clinical malaria (P < 0.001) or parasitaemia (P < 0.001) than patients without AIDS.

Eighteen patients (5%) had gastroenteritis as the initial diagnosis; of these six (34%) had NTS bacteraemia and seven (39%) had P. falciparum parasitaemia. Gastroenteritis was associated with S. enteritidis BSI (OR 7.7 [95% CI 7.2–26.8], P = 0.006). Six of 10 patients presenting with jaundice had bacteraemia (OR 6.3 [95% CI 1.7–22.7], Fisher's exact test P = 0.007).

While the level of fever was not associated with mycobacteraemia (P = 0.9), patients with high fever (≥39 °C) were more likely to have NTS bacteraemia (OR 2.2 [95% CI 1.2–4.0], P = 0.02), particularly in combination with splenomegaly (OR 8.4 [95% CI 3.4–20.6], P < 0.001). However, these combined clinical features were only present in 15 of 46 (32%) patients who were diagnosed as NTS bacteraemia. There was no association between level of fever and/or splenomegaly and malaria.

Indicators of bacteraemia and malarial parasitaemia are summarized in Table 3. Using logistic regression (excluding laboratory investigations), independent clinical predictors were identified (Table 4). Splenomegaly was an independent predictor for NTS bacteraemia, but not for malarial parasitaemia. A similar analysis of mycobacteraemia in the patients in this study is reported elsewhere (Lewis et al. 2002).

Table 3.  Indicators of bacteraemia and malarial parasitaemia in study patients admitted at the medical wards at QECH
IndicatorNTS bacteraemia (n = 46)SP bacteraemia (n = 16)Other bacteraemia (n = 11)All bacteraemia (n = 73)Malaria parasitaemia (n = 69)
  1. NTS, non-typhi salmonellae; SP, Streptococcus pneumoniae.

  2. * Not significant.

  3. † Defined as presence of either diarrhoea >1 month, fever >1 month, reported weight loss or impaired daily activity.

Chronically ill†****0.42 (0.25–0.72)
Fever >1 month****0.46 (0.25–0.83)
Fever ≥39 °C2.2 (1.2–4.0)**2.1 (1.2–3.5)*
Reported weight loss**0.24 (0.06–0.89)*0.40 (0.23–0.69)
Cough*1.1 (1.0–1.1)**0.25 (0.14–0.43)
Cough <3 weeks*4.3 (1.5–13)**0.55 (0.30–1.0)
Pleuritic chest pain****0.36 (0.20–0.63)
Shortness of breath*1.1 (1.0–1.1)**0.25 (0.14–0.43)
Antimalarials <2 weeks2.4 (1.2–4.5)**2.0 (1.2–3.4)*
Antibiotics <2 weeks****0.38 (0.21–0.68)
Jaundice4.7 (1.3–17)5.8 (1.1–30)*6.3 (1.7–22.7)*
Oral thrush***1.8 (1.0–3.1)0.073 (0.017–0.30)
Head-neck lymphadenopathy0.36 (0.16–0.80)***0.46 (0.25–0.86)
Chest exam abnormal*6.3 (1.7–23)*2.0 (1.2–3.5)0.30 (0.16–0.57)
Hepatomegaly****0.49 (0.24–1.0)
Splenomegaly3.8 (2.0–7.2)**2.5 (1.5–4.2)*
Gastroenteritis3.7 (1.3–10)***2.8 (1.0–7.5)
HIV-infected5.3 (1.2–2.2)****
AIDS on admission****0.31 (0.17–0.56)
Anaemia (Hb < 7)2.2 (1.1–4.5)****
Leucopenia (WBC < 5.0)***2.1 (1.0–4.4)2.3 (1.3–3.9)
Thrombocytopenia (platelets < 150)4.8 (2.3–10.3)**2.5 (1.4–4.4)4.3 (2.4–7.8)
Chest X-ray abnormal*7.3 (0.94–57)*3.8 (1.1–13)0.13 (0.071–0.25)
Table 4.  Independent clinical predictors of bacteraemia and parasitaemia in patients at QECH
 Prevalence in study population (%) OR*95% CIP-value
  1. NTS, non-typhi salmonellae; SP, Streptococcus pneumoniae.

  2. * After adjusting for indicators in Table 3.

All bacteraemia
 Oral thrush241.91.0–3.60.040
 Fever >1 month380.430.22–0.840.014
 Abnormalities on chest exam391.91.1–3.50.029
NTS bacteraemia
 Head-neck lymphadenopathy340.300.12–0.730.008
SP bacteraemia
 Chronically ill680.250.079–0.800.020
Other bacteraemia
 Reported weight loss570.130.025–0.650.013
 Oral thrush240.0430.006–0.320.002


This study confirms earlier findings of a high prevalence of HIV infection and AIDS among patients admitted to the medical wards (Lewis et al. 2003). As has been described from several other areas in Sub-Saharan Africa, infections with M. tuberculosis, S. pneumoniae and NTS are more common than the opportunistic infections encountered in the developed world (Lucas et al. 1993; Harries 1998; Floyd et al. 1999).

Tuberculosis was the most important cause of fever at our hospital, with the majority having disseminated disease as suggested by the presence of mycobacteraemia. Eighty-nine per cent of TB patients could be diagnosed using routine methods and protocols as used in our wards and identifying the presence of mycobacteraemia did not improve outcome (Lewis et al. 2002).

Bacteraemia was common and primary BSI came third in the list of diagnoses. Compared with other studies, we found a relatively high prevalence of bacteraemia (21% of enrolled patients) and a high case-fatality rate (19%) (Vugia et al. 1993; Archibald et al. 1998, 2000; Ssali et al. 1998; Arthur et al. 2001). It is likely that the case fatality rate in hospitals without blood culture facilities would even be higher. This is particularly worrying in the case of NTS bacteraemia, the most commonly isolated bacterial pathogen, as this infection usually presents as a primary infection without localizing signs. Streptococcus pneumoniae bacteraemia occurred in patients in whom this pathogen would be suspected as they presented with respiratory signs, meningitis or peritonitis, confirming findings of Gilks et al. (1996). Escherichia coli bacteraemia was uncommon but not suspected in all four cases; two of these had a positive blood film for malaria.

Only 42 (61%) of patients with P. falciparum parasitaemia had clinical malaria as the sole diagnosis (12% of the total number of patients). Finding a positive blood film should not deter the clinician from pursuing or treating for another more common and potentially fatal infection. Of the independent clinical predictors, splenomegaly on physical examination seems most useful in clinical practice. Medical practitioners in Malawi commonly believe that splenomegaly primarily implies malaria. In our study of an adult population with a high prevalence of HIV infection, splenomegaly was associated with NTS bacteraemia rather than malaria, and appropriate antibiotic therapy should be instituted. Similarly, the presence of oral candidiasis implies HIV infection and makes bacteraemia more likely than malaria. For clinical practice in many hospitals in Sub-Saharan Africa the clinician should be guided by knowledge of the most prevalent diagnoses that will be largely HIV related. It is not uncommon for a patient to present with two illnesses, one of which could be a bacteraemia that cannot be diagnosed with confidence by clinical examination and basic laboratory tools.

As it is unlikely that a blood culture service will become widely available, monitoring of infection with common pathogens and determination of antimicrobial-resistance pattern in a central facility are crucial. For peripheral hospitals it may be advisable to treat any patient not responding to treatment for their initial diagnosis with broad spectrum antibiotics.


This study was funded by a grant from the Dutch Society for Tropical Medicine, the Malawi Health Support Fund of the Royal Netherlands Embassy, Lusaka (grant number MW004401/02) and the National Tuberculosis Control Programme of Malawi. We thank Ledson Mkwaira, Freda Nsamala, Mercy Mtegha, Gabriel Mateyu, Jonathan Waluza, and the laboratory technicians at the College of Medicine and Wellcome Trust Research Laboratories for their help in patient management and laboratory work.