The accuracy of CRP to detect bacteraemia in Malawian children

Despite improvements in recent years, child mortality in Sub-Saharan Africa (SSA) remains alarmingly high, largely due to potentially life threathening infections such as bacteraemia and malaria. While malaria can be rapidly diagnosed using Malaria Rapid Diagnostic Tests (MRDTs), the rapid diagnosis of bacteraemia remains challenging. In high-resource settings (HRS), C-reactive protein (CRP) is considered a useful tool in guiding decisions around the start of antibiotics, but its performance in SSA may differ due to a different aetiology of fever, co-morbidities such as malaria, and late presentation. This study investigates the role of CRP as a diagnostic marker for bacteraemia in a paediatric population living in a high endemic malaria area. This study analysed data from a previously conducted case-control study on anaemia in Malawian children aged 6–60 months presenting to hospital. All children received a blood culture, malaria microscopy test, and CRP assessment. Children were excluded if CRP values were missing. We assessed if CRP was associated with bacteraemia, malaria or a combination of both and sought optimal cut-offs and diagnostics algorithms, combining malaria diagnostics and CRP. Among 703 included children 63 (9.0%) had bacteraemia and 347 (49.4%) had malaria parasitaemia, of whom 23 (3.3%) had both bacteraemia and malaria. CRP demonstrated moderate diagnostic accuracy for bacteraemia (area under the ROC-curve (AU-ROC) 0.672) and malaria (AU-ROC 0.691). Sensitivity and specificity at the 40 mg/L CRP threshold were 82.5% (52/63) and 42.6% (260/610) for bacteraemia, 75.5% (262/347) and 56.6% (201/355) for malaria. Of the four algorithms assessed, performing malaria tests and CRP using a dynamic cut off (> 40 mg/L malaria negative and > 70 mg/L malaria postive children) showed the most balanced sensitivity (79.4% (50/63)) and number needed to treat (NNT) (1:6.94) for bacteraemia. The accuracy of CRP to detect bacteraemia is reduced as compared to HRS due to malaria in this paediatric population in Malawi. CRP accuracy can be optimised using the malaria results and applying a dynamic cut-off which is more accurate than clinical symptoms alone. Not applicable