All (febrile and non-febrile) patients > 6 months consulting one of the study sites for any clinical problem during the study periods (24th April to 19th May and 2nd to 20th October, 2006) and giving their (or their guardians') written informed consent were consecutively submitted to a standardized medical examination, and to thick and thin film, by specifically trained research assistants. The research assistants were trained by the study investigators and by professional laboratory staff from an Italian referral centre (see below) for three days preceding each study period: training included the correct execution of malaria smears (thick and thin film) and the execution and reading of the RDT.
Exclusion criteria were: severe clinical condition needing urgent care. An axillary temperature was obtained upon recruitment for all patients using an electronic digital thermometer (accuracy ± 0.1°C, certified CE 0197). Fever was defined as an axillary temperature ≥ 37.5°C.
The reference test was malaria microscopy executed by highly experienced staff from the Centre for Tropical Diseases (CTD) of S. Cuore Hospital of Negrar, Verona, a reference centre in Italy. The thick and thin films were coded locally and transported daily to a central laboratory (Centre Muraz, Bobo Dioulasso) for Giemsa staining by local staff, supervised by two senior microscopists from the CTD. Reading was done by the senior microscopists who were masked to the result of the RDT as well as to the clinical status (febrile or non-febrile) of the patients. A number of microscopic fields corresponding to 200 WBC were read in the thick film. The parasite density was calculated (for P. falciparum only) in the conventional way according with WHO criteria. A double blind cross reading of a random sample of 300 slides (thick plus thin films) was carried out in order to check for inter-observer variability, as a double reading of all the > 5,000 slides was not feasible.
Statistical analysis
Data were double-entered at Centre Muraz, Bobo Dioulasso, with Epi Info software (EpiInfo, CDC Atlanta, version 3.3.2). Data analyses were carried out with R 2.8.0 (R Development Core Team. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing. Vienna, Austria, 2008. ISBN: 3-900051-07-0), and Stata 10.1 (StataCorp LP, College Station, TX 77845 USA) statistical packages.
The primary aim of the analysis was to estimate how RDTs predict malaria-attributable fever in the high and low transmission seasons. The main conceptual steps are summarized below, and detailed in the following paragraphs.
1. The prevalence of (falciparum) malaria infection was assessed in febrile and non-febrile patients.
2. The proportion of fevers attributable to malaria was estimated, stratifying by parasite density class and age group. The attributable fraction (AF) was defined as the proportion of fevers, among infected patients, that would not have occurred in the absence of malaria infection. Formulas used for AF calculation are reported below.
3. The diagnostic accuracy of RDTs for malaria infection by parasite density and age group was calculated.
4. The attributable fractions and the calculated diagnostic accuracy of RDT for malaria infection were combined to obtain an estimate of the sensitivity and specificity of RDT for malaria-attributable fever, accounting for both the expected increasing sensitivity of RDT for infection and the higher likelihood of a fever to be due to malaria at higher parasite densities.
All analyses were performed on the rainy and dry season data, separately.
The prevalence of falciparum malaria was estimated as the proportion of patients with a positive slide for
P. falciparum asexual forms (any parasite density) among febrile and non-febrile patients. The AF of fever to malaria infection was estimated from the odds-ratios obtained from logistic regression modelling according to methods described for case-control studies [
23] where patients with fever (axillary temperature ≥ 37.5°C) were defined as cases and patients without fever and without recent (3 days) fever history as controls. The odds-ratio (OR) of fever in each stratum of parasite density (0, 1-400, 401-4000, 4001-40,000 and > 40,000 parasites/μl, the upper limit of each stratum roughly corresponding to 1/10,000 parasites/RBC, 1/1,000, 1/100 and > 1/100) and age-group (6-11 months, 1-4 years, 5-14 years, ≥15 years) were calculated. The AF was then estimated from the ORs, as AF = (OR-1)/OR, for each cross-classification of parasite density and age (20 strata) for the rainy season. The number of fever cases and of positive malaria films was too low to determine the AF by age groups for the dry season. Consequently, for this season AFs were estimated by parasite density class only, adjusting for age, from the adjusted odds-ratios (aOR) as AF = (aOR-1)/aOR. In addition to the AF, the population attributable fraction (PAF) was also estimated, defined as the proportion of fevers attributable to malaria infection among all patients with fever, to assess the burden of disease in the whole population, and obtained by multiplying the AF by the prevalence of malaria infection among all febrile patients (PrevMal): PAF = AF(PrevMal).
RDT sensitivity, specificity, positive and negative predictive value (PPV, NPV) were estimated for malaria infection on the subset of febrile patients undergoing the RDT, and with microscopy results taken as the gold standard. As for the AF, sensitivity and specificity were calculated for each cross-classification of parasite density and age (20 strata) for the rainy season and for each parasite density stratum for the dry season. In addition, PPV and NPV of the RDT were assessed in febrile patients during the rainy and dry seasons. Confidence intervals were estimated with the Wilson's score method[
24]. To assess how the agreement between RDT and microscopy was influenced by variables other than parasite density (such as season, age and sex), a logistic regression model was used where the outcome was a dichotomous variable taking values of 1 in case of method agreement or 0 in case of disagreement.
Based on RDT performances on malaria infection at each level of parasite density and age, its accuracy was subsequently assessed on malaria-attributable fever.
The AF-based approach does not allow classifying each individual febrile case with a positive slide as having clinical malaria, or simple malaria infection with another cause of fever. However, through this approach it is possible to estimate the number of malaria-attributable fevers at each stratum of parasite density and age, by multiplying the number of febrile cases in each stratum by the respective AF.
The number of true positive RDT results was then calculated in each stratum as the product of the number of malaria-attributable fevers and the probability of a RDT positive test result for febrile patients in the stratum. The total number of RDT true positives was the sum of the RDT true positives in all strata, as in formulas reported in Table
1a. In a similar way, the number of false positives (Table
1b), false negatives (Table
1c), and true negatives (Table
1d) were estimated. The RDT sensitivity was then calculated as the ratio of true positives (Table
1a) to the sum of true positives and false negatives (Table
1a + b), and the specificity as the ratio of true negatives (Table
1d) to the sum of true negatives and false positives (Table
1c + d). Similarly, the PPV was estimated as the number of true positives (Table
1a) divided by the number of RDT positives (Table
1a + c), and the NPV as the number of true negatives (Table
1d) over the number of RDT negatives (Table
1b + d).
Table 1
Formulas used for the estimation of the RDT sensitivity and specificity on malaria - attributable fever
RDT+ | a) True Positives | c) False Positives |
| | |
RDT- | b) False Negatives | d) True Negatives |
| | |
| | |
As a sensitivity analysis, the assessment of RDT diagnostic accuracy was repeated using logistic regression models for the risk of fever (to calculate AF) and for the probability of testing positive at RDT for each individual patient, including malaria infection status (yes/no), log-parasite density, and linear and quadratic terms for age. For the rainy season analyses, age/infection status and age/parasite density interaction terms were also included. The product of the AF and probability for an RDT positive result was added up for all febrile patients to obtain the diagnostic accuracy of RDT for malaria-attributable fever, similar to the stratified analysis.
The study protocol was approved by the "Comité National d'Ethique" (National Ethical Committee) of Burkina Faso (N. 2006-011 of 7th April 2006). Written informed consent was obtained through the use of an information sheet with detailed explanation of the purpose of the study and the procedures involved. Once the clinical officer had decided that a patient was eligible for inclusion, a research assistant gave the explanation in local language, in the presence of at least one independent witness. In case of agreement, the informed consent form was signed both by the patient (or one of the parents in case of minors) and by the witness. For illiterate people the signature was replaced by the fingerprint.