- Open Access
Integrated paediatric fever management and antibiotic over-treatment in Malawi health facilities: data mining a national facility census
Malaria Journalvolume 15, Article number: 396 (2016)
There are growing concerns about irrational antibiotic prescription practices in the era of test-based malaria case management. This study assessed integrated paediatric fever management using malaria rapid diagnostic tests (RDT) and Integrated Management of Childhood Illness (IMCI) guidelines, including the relationship between RDT-negative results and antibiotic over-treatment in Malawi health facilities in 2013–2014.
A Malawi national facility census included 1981 observed sick children aged 2–59 months with fever complaints. Weighted frequencies were tabulated for other complaints, assessments and prescriptions for RDT-confirmed malaria, IMCI-classified non-severe pneumonia, and clinical diarrhoea. Classification trees using model-based recursive partitioning estimated the association between RDT results and antibiotic over-treatment and learned the influence of 38 other input variables at patient-, provider- and facility-levels.
Among 1981 clients, 72 % were tested or referred for malaria diagnosis and 85 % with RDT-confirmed malaria were prescribed first-line anti-malarials. Twenty-eight percent with IMCI-pneumonia were not prescribed antibiotics (under-treatment) and 59 % ‘without antibiotic need’ were prescribed antibiotics (over-treatment). Few clients had respiratory rates counted to identify antibiotic need for IMCI-pneumonia (18 %). RDT-negative children had 16.8 (95 % CI 8.6–32.7) times higher antibiotic over-treatment odds compared to RDT-positive cases conditioned by cough or difficult breathing complaints.
Integrated paediatric fever management was sub-optimal for completed assessments and antibiotic targeting despite common compliance to malaria treatment guidelines. RDT-negative results were strongly associated with antibiotic over-treatment conditioned by cough or difficult breathing complaints. A shift from malaria-focused ‘test and treat’ strategies toward ‘IMCI with testing’ is needed to improve quality fever care and rational use of both anti-malarials and antibiotics in line with recent global commitments to combat resistance.
Since the 1990s, the World Health Organization (WHO) and United Nations Children’s Fund (UNICEF) have promoted the Integrated Management of Childhood Illness (IMCI) strategy in low- and middle-income countries to effectively manage the most common causes of child morbidity and mortality in an integrated manner . It is well recognized that integrated protocols are critical for optimally managing the sick child in order to address co-morbidities and to differentiate among illnesses with overlapping symptoms .
Fever is a common symptom of many childhood illnesses in sub-Saharan Africa. For many years however, the IMCI strategy promoted presumptive malaria treatment of paediatric fevers in malaria-endemic African settings given high malaria mortality rates and the lack of other defining features for clinical management. While additional IMCI algorithms have been available to clinically differentiate other fever causes, the presumption of all fevers as ‘malaria’ has often impeded probing for these other conditions .
In 2010, WHO revised malaria treatment guidelines to recommend diagnosis of all suspected malaria cases prior to treatment given the increasing availability of malaria rapid diagnostic tests (RDT) . This policy shift has great potential to improve rational drug use and quality fever care , although studies indicate common inappropriate treatment of RDT-negative patients with anti-malarial or antibiotic drugs . These findings suggest poor integration of RDT into the IMCI framework, although few studies have explicitly examined integrated paediatric fever management and available evidence is largely derived from limited hospital settings [7–12]. There is also limited understanding of factors associated with non-adherence to clinical guidelines, notably antibiotic over-treatment, which is a particular concern in the era of test-based malaria case management . This concern reflects studies showing widespread antibiotic prescriptions for test-negative cases and not according to established clinical guidelines [1, 6, 8].
Malawi recently adopted the ‘test and treat’ strategy in its National Malaria Strategic Plan and began nationwide RDT deployment in July 2011 . A national facility study conducted prior to RDT implementation showed low availability of functional microscopy for malaria diagnosis , and common non-compliance to negative blood smear results . Similar evidence is needed from the post-RDT implementation period in Malawi with an expanded analysis of how RDT and IMCI are used together during outpatient consultations.
In this paper, a national facility census conducted in Malawi in 2013–2014 was analyzed to examine integrated paediatric fever management using RDT and IMCI, including other presenting complaints, completed assessments, diagnoses/classifications, and treatment prescriptions . The large number of facilities audited coupled with the broad data collection scope provides a unique opportunity to investigate the association between RDT results and antibiotic over-treatment. Classification trees are well suited for such an analysis since numerous influences on different levels may shape the complex nature of the clinical encounter and there may be complicated inter-relationships among variables that are not well defined in advance or easily detected using standard statistical methods . Indeed, traditional regression models assume a uniform influence of the exposure on an outcome unless an interaction is specified, which is unrealistic in real-life contexts and complicates results interpretation. It is also a challenging situation to model if numerous variables may influence an examined relationship and there is limited a priori knowledge of these potentially complex interactions in order to define a clear hypothesis for statistical testing.
Malaria is endemic in most parts of Malawi with peak transmission in November–April, although transmission has declined in recent years. Malawi’s health system is primarily comprised of government-run facilities and publicly supported facilities run by the Christian Health Association of Malawi (CHAM) . This system contains three main tiers: regional hospitals, district hospitals and health centres. The primary tier is the health centre, which provides essential services, including family planning, antenatal care and other outpatient services. The secondary tier is the district hospital, which are referral facilities that also provide in-patient care, laboratory diagnostics and maternity care. The tertiary level is the central or regional hospital, which are teaching and research hospitals that provide specialized medical care. Community-based sick child services are also provided in Malawi but are not included in this facility-based assessment.
The Malawi Service Provision Assessment (SPA) was conducted in June 2013–February 2014 by the Ministry of Health and the Demographic and Health Survey (DHS) programme, which includes facility and laboratory audits, observed consultations, patient exit interviews, and health worker interviews. Survey methods are described elsewhere . Briefly, Malawi SPA 2013–2014 was designed as a census of all formal public and private facilities in the country. At each facility, clients attending the facility on the interview date were systematically selected for observation. The expected patient load for outpatient sick child curative services on that date was estimated in advance and every Nth client attending the facility on that date was selected for observation in order to yield no more than 15 observations per facility. Clients were eligible to participate if they were under 5 years of age and presented with an illness complaint and not an exclusive injury or non-disease condition. Sick child observations aim to assess clinical practices according to Malawi IMCI guidelines . During the exit interview, a limited re-examination protocol was conducted by clinicians, nurses or nurse midwives specifically trained to take a 60-s respiratory rate count and temperature reading by thermometer.
Ethical approval for collection of these data was obtained by the DHS programme from the Department of Health and Human Services Institutional Review Board (IRB) and the host country IRB, which includes authorization to distribute unrestricted survey files for secondary analysis purposes upon receipt of a research proposal. Written informed consent was obtained separately from health workers and caregivers prior to participation in the observation, exit interview and re-examination .
Children aged 2 months–5 years attending an observed outpatient consultation as a first-time visit for an illness were included if they had a fever complaint and provided consent for the observed consultation and exit interview (Fig. 1). The antibiotic over-treatment analysis applied only to those clients ‘without antibiotic need’ as defined below.
Integrated paediatric fever management
Table 1 defines key measures of integrated paediatric fever management reported in this paper, including other complaints, completed assessments, classifications/diagnoses and drug prescriptions for RDT-confirmed malaria, IMCI-pneumonia, and clinical diarrhoea.
Antibiotic over-treatment or any antibiotic prescription ‘without antibiotic need’ is defined as a IMCI-pneumonia negative classification based on re-examination and additionally excluding the following diagnoses recorded during the consultation: sepsis, acute ear infection, mastoiditis, dysentery, abscess, or severe malnutrition. Any antibiotic prescription includes any antibiotic injection (benzyl penicillin or other) or antibiotic capsule, syrup or tablet (amoxicillin, cotrimoxazole or other) that the provider reported was prescribed during the consultation. Table 1 defines the main predictors: RDT conducted (yes or no) and RDT result (positive or negative). Table 2 defines the other 38 input variables at different levels in the analysis, which includes malaria risk (infection prevalence) values for 2013–2014 linked to datasets through geocoded facility locations, and transmission season estimates derived from facility locations and interview dates [21, 22].
Visual content mapping depicted the potential inter-relationships of input variables on clinical treatment decisions using the Visual Understanding Environment 3.3.0 (Tufts University, Somerville, MA, USA) . Frequencies and cross-tabulations were calculated using weights to account for the unequal probabilities of selection due to differing client volumes at facilities on the interview date. Standard error estimation accounted for clustering of client observations within facilities. The level of statistical significance was set to 0.05. Stata 13.1 (Stata Corp., College Station, TX, USA) was used for analyses.
Classification trees were used to learn the relative importance of main predictors (RDT conducted and RDT result) and their inter-relationships with other input variables on the binary outcome of antibiotic over-treatment. A model-based recursive partitioning approach  was used in this analysis that embeds a parametric model into a recursive partitioning algorithm in order to identify sub-groups within the dataset where there may be different patterns of association between the main predictor and outcome. The model is subsequently re-fit to identified sub-groups, known as nodes, in order to describe different and complex relationships among variables with respect to an outcome across these sub-groups.
In this analysis, a mixed-effects logistic regression model was initially fit to estimate the relationship between the RDT result (or RDT conducted) and antibiotic over-treatment, with observations nested within facility identifiers. The potential influence of 38 other variables on this relationship was learned through recursive partitioning that allowed for detection of sub-group interactions and estimation of random effects parameters . Parameter instability was repeatedly assessed over the set of 38 potential partitioning variables using a Bonferroni-corrected significance level of 0.05. Nodes were split according to the variable, resulting in highest instability, known as a significant classifier. This process was repeated for each resulting sub-group until the minimal node size of 20 observations was reached or no additional significant classifiers were identified. This approach yields a tree fitted to models associated with each terminal node along with estimated odds ratios or other coefficients for the effect of the main predictor on an outcome in each resulting sub-group. R version 3.2.2 and the ‘partykit’ package was used for this analysis [26, 27].
The Malawi Service Provision Assessment 2013–2014 included 977 facilities out of 1060 on the Ministry of Health master facility list with non-response due to refusal (3 %), closure (2 %), inaccessibility (2 %), or other issue (1 %). A total of 2950 sick child clients met inclusion criteria and 1981 reported fever complaints (Fig. 1). Additional files 1, 2 describe characteristics of febrile clients with RDT results and receiving antibiotic over-treatment respectively.
Among 1981 eligible clients, 1436 (72 %) also reported cough or difficult breathing (CDB) complaints; 569 (29 %) had diarrhoea complaints; 359 (18 %) reported other complaints including skin problems, eye problems, ear problems, stomach problems, injuries or other issues; 1021 (52 %) reported any danger sign (lethargy, inability to drink or breastfeed, convulsions or vomits everything); 117 (6 %) reported fever alone with no other complaint or danger sign (Fig. 2).
Among 1981 eligible clients, 1684 (85.0 %) either spontaneously mentioned the fever complaint or were asked about fever by the provider during the consultation; 1386 (70.0 %) had their temperature taken or body felt for hotness; 1426 (72.0 %) had a malaria RDT done prior to the consultation or were referred for malaria diagnosis; 44 (2.2 %) had neck checked for stiffness; 524 (26.5 %) had palm pallor checked; 185 (9.3 %) had the inside of their mouth checked; and 563 (28.4 %) were undressed for examination (up to shoulders/down to ankles). Among 1436 clients with both fever and CDB complaints, 256 (17.8 %) had respiratory rates counted for 60 s. Among 569 clients with fever and diarrhoea complaints, 98 (17.3 %) had skin turgor checked for dehydration (Table 3).
Among 1981 eligible clients, 746 (37.7 %) had malaria RDT conducted prior to the consultation with results reported. Among 312 with reported RDT-positive results, 265 (85.1 %) received first-line anti-malarial prescriptions; 22 (7.0 %) received second-line anti-malarial prescriptions; and 25 (7.9 %) received no anti-malarial prescription (anti-malarial under-treatment). Among 434 with reported RDT-negative results, 44 (10.2 %) received any anti-malarial prescription (anti-malarial over-treatment) (Table 4).
Among 1981 eligible clients, 1367 (70.3 %) were assessed for IMCI pneumonia with results reported. Among 376 with non-severe IMCI-pneumonia from re-examination, 148 (39.4 %) received first-line antibiotic prescriptions; 123 (32.7 %) received second-line antibiotic prescriptions; and 105 (27.9 %) received no antibiotic prescription (antibiotic under-treatment). There were 917 with a negative IMCI-pneumonia classification and a total of 1411 were further categorized as ‘without antibiotic need’. Among 1411 clients ‘without antibiotic need’, 830 (58.8 %) received any antibiotic prescription (antibiotic over-treatment) (Table 4).
Oral rehydration solution and zinc prescriptions
Among 1981 eligible clients, 260 (13.1 %) were given diagnoses of dehydration or intestinal/digestive issue. Among 260 with these diagnoses, 187 (72.1 %) received oral rehydration solution (ORS) and 148 (56.9 %) received both ORS and zinc prescriptions.
Among the sub-set of 526 clients ‘without antibiotic need’ and reported RDT results, RDT-negative clients had 16.8 (95 % CI 8.6–32.7) times higher antibiotic over-treatment odds compared to RDT-positive clients in the crude mixed-effects logistic regression model. CDB complaint was a statistically significant classifier of this relationship learned through recursive partitioning (p < 0.001). Figure 3a depicts all observations ‘without antibiotic need’ and the dark grey bars indicate those receiving any antibiotic prescription, or antibiotic over-treatment. This figure indicates that the split by CDB complaint is largely driven by a difference in the underlying risk of antibiotic over-treatment across groups rather than changing patterns of association between RDT results and antibiotic over-treatment. The lowest risk of antibiotic over-treatment was found among clients without CDB complaint and a positive RDT result. This risk significantly increased with the negative RDT result (Node 2: OR: 8.9; n = 97). In contrast, clients with CDB complaint already had relatively high underlying risk of antibiotic over-treatment irrespective of the RDT result and this risk similarly increased with a negative result (Node 3: OR: 5.6, n = 188). Indeed, the highest risk of antibiotic over-treatment was among clients with CDB complaint and a negative RDT result. In this group, 82 % of clients were inappropriately prescribed antibiotics according to the study definition.
Figure 3b depicts the relationship between RDT conducted prior to consultation and antibiotic over-treatment among the subset of 1411 clients ‘without antibiotic need’ either tested or not for malaria. Antibiotic over-treatment odds were reduced among clients tested compared to untested in the crude mixed-effects logistic regression model (OR: 0.48, 95 % CI 0.35–0.64). CDB complaint was a statistically significant classifier in this analysis, and testing was differently associated with the outcome if this complaint was reported (node 2 OR: 0.5, n = 227; node 3: OR: 1.0, n = 513). Conducting RDT prior to the consultation reduced antibiotic over-treatment odds among clients without CDB complaints compared to those untested, but this effect was negligible among those with this complaint.
Integrated paediatric fever management was sub-optimal in terms of fever assessments completed and poor antibiotic targeting, although findings suggest common compliance to malaria treatment guidelines. The RDT negative result was strongly associated with antibiotic over-treatment conditioned by CDB complaints.
In this study, only 6 % of clients with fever complaints had no other complaint or danger sign underscoring the critical need for integrated protocols to manage sick children [1–3]. It was further shown that most clients with fever complaints received a malaria test or were referred for diagnosis, and RDT-guided malaria treatment seemed common according to provider reports. This finding suggests compliance with new malaria treatment guidelines in Malawi in 2013–2014 and contrasts with previous studies showing poor adherence to negative blood smear readings prior to nationwide RDT deployment . This result should be viewed in light of data limitations described later in this section and additional studies are needed to corroborate this finding.
Yet general fever assessments were less commonly conducted despite being essential for differential diagnosis. This finding is consistent with other research showing poor IMCI implementation in Malawi and other settings [28, 29]. There was also poor antibiotic targeting with both under- and over-treatment that is in part due to poorly assessing clients to identify antibiotic need. Poor antibiotic targeting in low-income settings has been documented in other research [7–12]. This study, however, is the first to our knowledge to provide large-scale evidence of integrated paediatric fever management using RDT and IMCI during outpatient consultations, including the important relationship between RDT results and antibiotic over-treatment.
These findings demonstrate the strong influence of RDT-negative results on antibiotic over-treatment and an inter-relationship with CDB complaints, which reinforces the primary importance of patient symptoms and diagnostic test results on clinical treatment decisions. This is consistent with research showing widespread antibiotic prescriptions for RDT-negative cases [8–11], and cough complaint as a main predictor of incorrect malaria treatment in Malawi . The relatively small sample size may have limited detection of other inter-relationships, notably over-treatment previously documented among urban clients . Nevertheless, data mining tasks are well suited to discover inter-relationships among variables with respect to an outcome, particularly if there is limited a priori knowledge of these associations . These methods have been widely used in business and biomedical research but their application in global health research has been limited and should be increasingly considered where appropriate [31–35].
Taken together, these results underscore growing concerns about irrational antibiotic prescription practices in the era of test-based malaria case management , particularly given recent research showing viral disease is a far more common cause of paediatric fevers in various African settings compared to bacterial or parasitic infections [36, 37]. A recent World Health Assembly resolution urges countries to develop action plans to combat antibiotic resistance in the coming years . A main focus for low-income countries will be to extend the reach of health systems to expand access to life-saving medicines while simultaneously strengthening quality care at facilities that could in turn improve antibiotic targeting . These results highlight the need to implement IMCI and RDT together to strengthen integrated paediatric fever management and rational use of both anti-malarial and antibiotic medicines.
To this end, the IMCI algorithm has been adapted to reflect test-based malaria treatment guidance, and there are efforts to further strengthen these guidelines based on recent etiology studies [36, 37], and in recognition of its poor implementation to date . However, this new IMCI adaptation lacks clarity on antibiotic indications in the fever algorithm that could in turn inadvertently promote antibiotic over-treatment , which has been demonstrated in other recent research . It is critical that IMCI guidelines clearly indicate when antibiotics are (or are not) recommended for sick children, particularly for RDT-negative cases, and additional review of these guidelines from this perspective may be needed.
These results should be viewed in light of data limitations. First, client selection is based on sick child attendance on the interview date and do not represent clients visiting facilities on different dates/seasons, nor all sick children in Malawi. Second, providers may perform better during observations than in routine conditions biasing results towards better practices, including RDT compliance . Third, assessments recorded do not include all IMCI fever assessments, notably asking about fever duration or measles history. There is also no recording of assessment quality or clinical findings. Fourth, it may be difficult for observers to recognize that certain assessments were conducted, such as checking for neck stiffness, which could underestimate results. Fifth, the re-examination was a limited protocol that only assessed the sick child for a raised respiratory rate, signs of anaemia and fever presence based on a thermometer reading. Other assessments that could potentially indicate pneumonia or antibiotic need were not assessed in the re-examination, such as chest indrawing or hypoxia.
Measurement limitations for main predictors and the outcome should also be highlighted. First, RDT results are based on provider reports without supporting documentation. The provider reports RDT results after providing information on diagnoses and prescriptions. Some providers may misreport a negative result as positive if anti-malarial medicines were prescribed to seem in compliance with guidelines. Misclassification of the positive result as negative seems less likely in this scenario. This could potentially explain common RDT compliance found in this assessment and these results should be corroborated by additional studies. Second, RDT compliance estimates are only for clients diagnosed by the consultation time and do not include blood smear or RDT results not available by the initial consultation. Facilities conducting RDT prior to the consultation may be systematically different from other facilities in ways that influence compliance, such as larger facilities with more staff and better quality care. Third, antibiotic over-treatment is notoriously difficult to measure in settings without diagnostics to differentiate bacterial from other pathogenic causes. This paper defines ‘need’ according to IMCI antibiotic indications for pneumonia and provider reported diagnostic categories requiring antibiotics: sepsis, dysentery, mastoiditis, acute ear infection, abscess, or severe malnutrition. Urinary tract infection is not a diagnostic category and is not included in this definition. Clients assigned these diagnoses may not have the underlying condition and may not need antibiotics. The ‘without antibiotic need’ definition in this study therefore underestimates true lack of need. Fourth, IMCI pneumonia can be difficult to assess even by a trained provider leading to some misclassification in either direction .
Based on 977 facilities and 1981 eligible clients, study findings demonstrate sub-optimal integrated paediatric fever management practices in Malawi health facilities in 2013–2014. While malaria-specific assessments and RDT-guided treatment seemed common, other fever assessments were not often completed and poor antibiotic targeting was demonstrated. RDT-negative results were strongly associated with antibiotic over-treatment conditioned by CDB complaints. These results suggest moving beyond malaria-focused ‘test and treat’ strategies toward ‘IMCI with testing’ to improve quality fever care and rational use of both anti-malarial and antibiotic medicines. Integrated paediatric fever management using RDT and IMCI together is critical to improve antibiotic targeting in line with recent commitments to combat antibiotic resistance, and should be considered for inclusion in national action plans developed by malaria-endemic African countries in the next year.
cough or difficult breathing
Christian Health Association of Malawi
Demographic and Health Survey
Integrated Management of Childhood Illness
Institutional Review Board
oral rehydration solution
malaria rapid diagnostic test
Service Provision Assessment
Gove S. Integrated management of childhood illness by outpatient health workers: technical basis and overview. Bull World Health Organ. 1997;75:7–24.
WHO. Informal consultation on fever management in peripheral care settings—a global review of evidence and practice. Geneva: World Health Organization; 2013.
WHO. The overlap in the clinical presentation and treatment of malaria and pneumonia in children: report of a meeting. Geneva: World Health Organization; 1991.
WHO. Guidelines for the treatment of malaria. Geneva: World Health Organization; 2010.
WHO. T3: Test. Treat. Track. Scaling up diagnostic testing, treatment and surveillance for malaria. Geneva: World Health Organization; 2012.
Bastiaens GJH, Bousema T, Leslie T. Scale-up of malaria rapid diagnostic tests and artemisinin-based combination therapy: challenges and perspectives in sub-Saharan Africa. PLoS Med. 2014;11:e1001590.
Batwala V, Magnussen P, Nuwaha F. Antibiotic use among patients with febrile illness in a low malaria endemicity setting in Uganda. Malar J. 2011;10:377.
Shakely D, Elfving K, Aydin-Schmidt B, Msellem M, Morris U, Omar R, et al. The usefulness of rapid diagnostic tests in the new context of low malaria transmission in Zanzibar. PLoS One. 2013;8:e72912.
Shao AF, Rambaud-Althaus C, Samaka J, Faustine AF, Perri-Moore S, Swai N, et al. New algorithm for managing childhood illness using mobile technology (ALMANACH): a controlled non-inferiority study on clinical outcome and antibiotic use in Tanzania. PLoS One. 2015;10:e0132316.
Senn N, Rarau P, Salib M, Manong D, Siba P, Rogerson S, et al. Use of antibiotics within the IMCI guidelines in outpatient settings in Papua New Guinean children: an observational and effectiveness study. PLoS One. 2014;9:e90990.
Baiden F, Webster J, Tivura M, Delimini R, Berko Y, Amenga-Etego S, et al. Accuracy of rapid tests for malaria and treatment outcomes for malaria and non-malaria cases among under-five children in rural Ghana. PLoS One. 2012;7:e34073.
Baiden F, Owusu-Agyei S, Bawah S, Bruce J, Tivura M, Delmini R, et al. An evaluation of the clinical assessments of under-five febrile children presenting to primary health facilities in rural Ghana. PLoS One. 2011;6:e28944.
Baiden F, Webster J, Owusu-Agyei S, Chandramohan D. Would rational use of antibiotics be compromised in the era of test-based management of malaria? Trop Med Int Health. 2011;16:142–4.
Government of Malawi. Malaria strategic plan 2011–2013: towards universal access. Lilongwe: National Malaria Control Programme; 2011.
Steinhardt L, Chinkhumba J, Wolkon A, Luka M, Luhanga M, Sande J, et al. Quality of malaria case management in Malawi: results from a nationally representative health facility survey. PLoS One. 2014;9:e89050.
Steinhardt L, Chinkhumba J, Wolkon A, Luka M, Luhanga M, Sande J, et al. Patient-, health worker-, and health facility-level determinants of correct malaria case management at publicly funded health facilities in Malawi: results from a nationally representative health facility survey. Malar J. 2014;13:64.
Government of Malawi, Ministry of Health and ICF International. Malawi Service Provision Assessment 2013–2014. Lilongwe and Rockville: Government of Malawi and ICF International; 2015.
Tan PN. Introduction to data mining. 2nd ed. New York: Pearson Education Inc.; 2006.
Government of Malawi, Ministry of Health. Health sector strategic plan 2011–2016. Lilongwe: Government of Malawi; 2011.
Government of Malawi, Ministry of Health. Malawi integrated management of childhood illness. Lilongwe: Government of Malawi; 2013.
Bhatt S, Weiss DJ, Cameron E, Bisanzio D, Mappin B, Dalrymple U, et al. The effect of malaria control on Plasmodium falciparum in Africa between 2000 and 2015. Nature. 2015;526:207–11.
Mapping malaria risk in Africa (MARA). Basel: Swiss Tropical and Public Health Institute; 2015. http://www.mara-database.org. Accessed 1 Dec 2015.
Visual understanding environment (VUE) v3.3.0. Medford: Tufts University; 2015. http://vue.tufts.edu. Accessed 1 Dec 2015.
Zeileis A, Horhorn T, Hornik K. Model-based recursive partitioning. J Comput Graph Stat. 2008;17:492–514.
Fokkema M, Smits N, Zeileis A, Hothorn T, Kelderman H. Detecting treatment sub-group interactions in clustered data with generalized linear mixed-effects model trees. In: Working paper in economics and statistics. Innsbruck: University of Innsbruck; 2015. ftp://repec.org/opt/ReDIF/RePEc/inn/wpaper/2015-10.pdfAccessed 1 Dec 2015.
R Core Team. R: a language and environment for statistical computing. Vienna: R Foundation for Statistical Computing; 2013.
Hothorn T, Zeileis A. Partykit. A modular toolkit for recursive partytioning in R. Journal of Machine Learning Research. http://EconPapers.RePEc.org/RePEc:inn:wpaper:2014-10 (2015) Accessed 1 Dec 2015.
Bjornstad E, Preidis GA, Lufesi N, Olson D, Kamthunzi P, Hosseinipour MC, et al. Determining the quality of IMCI pneumonia care in Malawian children. Paediatr Int Child Health. 2014;34:29–36.
Chopra M, Binkin NJ, Mason E, Wolfheim. Integrated management of childhood illness: what have we learned and how can it be improved? Arch Dis Child. 2012;97:350–4.
Oyekale AS. Assessment of Malawian mothers’ malaria knowledge, healthcare preferences and timeliness of seeking fever treatment for children under five. Int J Environ Res Publ Health. 2015;12:521–40.
Chandran TM, Berkvens D, Chikobvu P, Nostlinger C, Colebunders R, Williams BG, et al. Predictors of condom use and refusal among the population of Free State province in South Africa. BMC Publ Health. 2012;12:381.
Vinnemeier CD, Schwarz NG, May J. Predictive value of fever and palm pallor for P. falciparum parasitemia in children from an endemic area. PLoS One. 2012;7:e36678.
Thang ND, Erhart A, Speybroeck N, Hung LX, Thuan LK, Hung CT, et al. Malaria in central Viet Nam: analysis of risk factors by multivariate analysis and classification tree models. Malar J. 2008;7:28.
Protopopoff N, Van Bortel W, Speybroeck N, Van Geertruyden JP, Baza D, D’Alessandro U, et al. Ranking malaria risk factors to guide malaria control efforts in African highlands. PLoS One. 2009;4:e8022.
Thanh PV, Hing NV, Van NV, Van Malderen C, Obsomer V, Rosanas-Urgell A, et al. Epidemiology of forest malaria in Central Viet Nam: the hidden parasite reservoir. Malar J. 2015;14:86.
D’Acremont V, Kilowoko M, Kyungu E, Philipina S, Sangu W, Kahama-Maro J, et al. Beyond malaria—causes of fever in outpatient Tanzanian children. N Eng J Med. 2014;370:809–17.
Hildenwall H, Amos B, Mtove G, Muro F, Cederlund K, Reyburn H. Causes of non-malarial febrile illness in outpatients in Tanzania. Trop Med Int Health. 2015;21:149–56.
WHO. World Health Assembly resolution A68/20—antimicrobial resistance, draft global action plan on antimicrobial resistance. Geneva: World Health Organization; 2015.
Laxminarayan R, Matsoso P, Pant S, Brower C, Rottningen JA, Klugman K, et al. Access to effective antimicrobials: a worldwide challenge. Lancet. 2015;387:168–75.
WHO. Integrated management of childhood illness—chart booklet. Geneva: World Health Organization; 2014.
Leonard K, Masatu MC. Outpatient process quality evaluation and the Hawthorne Effect. Soc Sci Med. 2006;63:2330–40.
Muro F, Mtove G, Nosha N, Wangai H, Harrison N, Hildenwall H, et al. Effect of context on respiratory rate measurement in identifying non-severe pneumonia in African children. Trop Med Int Health. 2015;20:757–65.
EWJ, KES, SSP, and HH designed and conceptualized the study. EWJ compiled, prepared, analysed, and interpreted data. EWJ, KES, MP, SSP, and HH contributed to data analysis. PWG and BM analysed and modelled malaria risk populations and extracted transmission season estimates. EWJ, SSP, HH, and HN contributed to interpretation of findings. EWJ wrote the first draft of the paper. EWJ, KES, HN, BM, PWG, MP, SSP, and HH reviewed, revised and contributed writing to the paper. All authors read and approved the final manuscript.
The authors sincerely thank Professor Achim Zeileis at Universität Innsbruck for his guidance using model-based recursive partitioning in this manuscript.
The authors declare that they have no competing interests.
Availability of data and materials
The dataset supporting the conclusions of this article is found at: https://figshare.com/s/daa18fc49316b200bcf7.
Ethics approval and consent to participate
This study is based on secondary analysis of public datasets that are made available to researchers after submitting a project description. Procedures for obtaining ethical approval and participant consent are described in the final survey report and summarized in this manuscript.
This study uses public datasets with no associated cost for secondary analysis. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Uppsala University provides salary support for SSP, KES and also funds EWJ. Karolinska Institutet provides salary support for SSP. HH receives funding from the Swedish Research Council for Health, Working Life and Welfare/the European Commission under a COFAS Marie Curie Post-Doctoral Fellowship. Salary support for HN is from the Malawi Ministry of Health. Salary support for MP is from University of Gothenburg. PWG is a Career Development Fellow (#K00669X) jointly funded by the UK Medical Research Council (MRC) and the UK Department for International Development (DFID) under the MRC/DFID Concordat agreement, also part of the EDCTP2 programme supported by the European Union, and receives support from the Bill and Melinda Gates Foundation (#OPP1068048, #OPP1106023). These grants also supported BM.
Stefan Swartling Peterson and Helena Hildenwall contributed equally to this work as final authors