SEARCH

SEARCH BY CITATION

Keywords:

  • malaria diagnosis;
  • clinical symptoms;
  • contextual factors;
  • decision-making;
  • prospective observational study;
  • Tanzania
  • diagnostic de la malaria;
  • symptômes cliniques;
  • facteurs contextuels;
  • prise de décision;
  • étude prospective d’observation;
  • Tanzanie
  • diagnóstico de malaria;
  • síntomas clínicos;
  • factores contextuales;
  • toma de decisiones;
  • estudio prospectivo observacional;
  • Tanzania

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusion
  8. Acknowledgements
  9. References

Objective  To gain a better understanding of the decision-making context in the diagnosis of malaria in order to inform behaviour change strategies, using quantitative methods.

Methods  We observed hospital outpatient and inpatient consultations in northeast Tanzania where malaria testing was routinely available, recording potential influences on testing and prescribing decisions. We analysed the effects of variables at patient, clinical context and clinician levels on three key decisions in malaria diagnosis and treatment: decision to test for malaria, presumptive treatment and treatment of test-negative patients with antimalarials.

Results  Observation of 2082 consultations took place during 120 clinics (different shifts on different days and in different departments) with 34 clinicians. Malaria tests were requested for 16.9% of patients. This decision was driven primarily by clinical symptoms. Of patients not tested for malaria, 36.0% were prescribed antimalarials, this decision being associated with both clinical and non-clinical factors. In outpatients fever was a strong predictor of presumptive treatment [adjusted odds ratio (AOR): 45.9, 95% CI: 30–73], in inpatients this was less so (AOR: 2.7, 95% CI: 0.98–7.7). Outpatient clinicians who were working alone or who had attended <2 in-service training sessions in the past year were more likely to prescribe antimalarials presumptively. The decision to prescribe antimalarials without also prescribing antibiotic treatment to 22.8% patients who tested negative for malaria was not driven by clinical symptoms but was associated with age over 5 years, lower patient load and male sex of clinician.

Conclusions  Non-clinical factors are important in the overdiagnosis of malaria. Strategies to target antimalarials and antibiotics better need to use methods that address the context of clinical decision making in addition to the dissemination of conventional clinical algorithms.

Objectif:  Obtenir une meilleure compréhension du contexte décisionnel dans le diagnostic de la malaria en vue d’informer les stratégies de changement de comportement sur base de méthodes quantitatives.

Méthodes:  Nous avons observé les consultations des patients hospitalisés et ambulatoires dans le nord-est de la Tanzanie, où les tests pour la malaria étaient disponibles en routine, en enregistrant les facteurs pouvant influencer les décisions pour les tests et les prescriptions. Nous avons analysé les effets de variables à l’échelle du patient, du contexte clinique et du clinicien sur trois niveaux clés de décisions dans le diagnostic et le traitement de la malaria: la décision de tester pour la malaria, le traitement présomptif et le traitement de patients à test négatif avec les antimalariques.

Résultats:  Les observations de 2082 consultations ont eu lieu au cours de 120 actes cliniques (différents horaires de travail sur des jours différents et dans différents départements) avec 34 cliniciens. Les tests pour la malaria ont été demandés pour 16,9% des patients. Cette décision était principalement alimentée par les symptômes cliniques. Chez les patients non testés pour la malaria, 36,0% ont été prescrits des antimalariques, cette décision étant associée à la fois à des facteurs cliniques et non cliniques. Chez les patients ambulants, la fièvre était un prédicteur important de traitement présomptif (AOR: 45,9; IC95%: 30-73), mais un prédicteur moins important chez les patients hospitalisés (AOR: 2,7; IC95%: 0,98-7,7). Les cliniciens de patients ambulatoires qui travaillaient seuls ou qui avaient assistéà moins de 2 sessions de formation sur le service l’année précédente étaient plus susceptibles de prescrire les antimalariques de façon présomptive. La décision de prescrire des antimalariques, sans prescrire un traitement antibiotique à 22,8% des patients qui étaient testés négatif pour la malaria n’était pas motivée par les symptômes cliniques, mais associée avec l’âge de plus de 5 ans, des moyens plus faible des patients et le sexe masculin pour le clinicien.

Conclusions:  Des facteurs non-cliniques sont importants dans le diagnostic en excès de la malaria. Les stratégies pour une meilleure utilisation des antimalariques et antibiotiques nécessitent des méthodes qui répondent au cadre de la prise de décision clinique en plus de la dissémination des algorithmes cliniques conventionnels.

Objetivo:  Alcanzar un mayor entendimiento del contexto de la toma de decisiones en diagnóstico de la malaria, con el fin de informar sobre estrategias de cambio de comportamiento, utilizando métodos cuantitativos.

Métodos:  Hemos realizado observaciones en consultas hospitalarias y externas en el noreste de Tanzania, en donde la prueba para malaria estaba disponible de forma rutinaria, tomando nota de influencias potenciales sobre las decisiones de prueba y prescripción. Hemos analizado los efectos de variables a nivel de pacientes, del contexto clínico y del clínico, en tres decisiones claves en el diagnóstico y el tratamiento de la malaria: la decisión de realizar la prueba diagnóstica para malaria, el tratamiento presuntivo y el tratamiento con antimaláricos dado a pacientes con una prueba diagnóstica negativa.

Resultados:  Se realizaron 2082 observaciones en 120 clínicas (diferentes turnos, en diferentes días y diferentes departamentos) con 34 clínicos. Las pruebas diagnósticas de malaria fueron requeridas a un 16.9% de los pacientes. Esta decisión fue impulsada principalmente por los síntomas clínicos. De los pacientes a los cuales no se les hizo la prueba, un 36.0% recibió prescripción de antimaláricos. Esta decisión estaba asociada tanto con factores clínicos como con los no clínicos. La fiebre en pacientes externos era un buen vaticinador del tratamiento presuntivo (AOR:45.9 95%CI:30-73), mientras que en pacientes hospitalizados este no era tanto el caso (AOR:2.7 95%CI:0.98-7.7). Los médicos en consultas externas que estaban trabajando solos o que habían atendido a <2 entrenamientos, estando de servicio, durante el último año, tenían más posibilidad de prescribir antimaláricos de forma presuntiva. La decisión de prescribir antimaláricos sin prescribir también tratamiento antibiótico a 22.8% de los pacientes que habían dado negativo para malaria no estaba impulsado por los síntomas clínicos sino que estaba asociada a que el paciente tuviese más de 5 años de edad, a una menor carga de pacientes y a que el clínico fuese un hombre.

Conclusiones:  Los factores no clínicos son importantes en el sobrediagnóstico de malaria. Para que las estrategias dirigidas a los antimaláricos y antibióticos tengan un mayor impacto necesitan utilizar métodos que tengan en cuenta el contexto del proceso de la toma de decisión clínica, además de la diseminación de algoritmos clínicos convencionales.


Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusion
  8. Acknowledgements
  9. References

Malaria overdiagnosis persists as a major public health problem in Africa with studies suggesting between 50% and 99% of those prescribed antimalarials being test-negative, depending on endemicity in the clinical setting (Font et al. 2001; Amexo et al. 2004; Reyburn et al. 2004; Talisuna & Meya 2007; Van Dillen et al. 2007). At case-management level it leads to potentially serious, treatable, alternative diagnoses being missed (Berkley et al. 2005). At population level the advent of new, effective but more expensive antimalarials means the massive wastage of drug on those without malaria, threatening the financial viability of artemisinin combination treatment (ACT) rollout. It also exposes those without disease to side-effects of the drug needlessly (Jonkman et al. 1995; WHO 2006a; Wiseman et al. 2006; Zurovac et al. 2006). Restricting antimalarials to true test-positives has been shown to be safe in areas of low endemicity (Njama-Meya et al. 2007). For several reasons it is therefore essential to address diagnostic attitudes to what remains the commonest diagnosis made in Africa. Providing accurate rapid diagnostic tests, in themselves, has limited or no impact on this (Hamer et al. 2007; Reyburn et al. 2007). An exploration of the reasons for the overdiagnosis of malaria by clinicians is therefore urgently needed in order to guide interventions to improve prescribing behaviour (Zurovac & Rowe 2006).

Clinical performance is likely to be affected by a number of both clinical and non-clinical factors which may be exacerbated in resource poor settings (Rowe et al. 2005). Predictors of health worker prescribing practice have been explored in a number of developing country settings (Rowe et al. 2000, 2003; Zurovac et al. 2004, 2005; Osterholt et al. 2006), where patient (age, complaint), consultation (time of day, duration) and health worker (level of training) factors were identified as influencing decisions to prescribe antimalarials. These studies focussed on health workers prescribing correct antimalarials to non-severe febrile paediatric outpatients and were largely conducted as short surveys in health facilities. Studies have not yet sought to identify predictors of the practice of overdiagnosing malaria, which is particularly relevant in hospital settings where laboratory facilities are available, nor have they observed the local context of these decisions over longer periods of time.

We set out to observe the decision making process for malaria diagnosis and antimalarial treatment in typical district hospitals where microscopy was routinely available, recording changes in contextual variables alongside consultation outcomes. We examined three key decision points: Why are some patients chosen to be tested for malaria and not others? If not tested, how are patients selected for antimalarial treatment? If tested, and test-negative for malaria, which patients are nevertheless given antimalarials and which antibiotics?

Methods

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusion
  8. Acknowledgements
  9. References

Study design and data collection

The study design was a prospective observational study with a case–control analysis. Data collection took place over a 6-month period in two busy district hospitals in northeast Tanzania known for significant overdiagnosis of malaria. Both hospitals treat large numbers of patients: between 2000 and 5000 paediatric admissions per year. Hospital HI was run jointly by the Catholic Church and the district council in an area of low malaria transmission, and hospital HII was run by the government via the district council in an area of high transmission. At both hospitals blood slide microscopy was the standard test for malaria supplemented by a supply of rapid diagnostic tests in paediatric wards.

All hospital clinicians involved in paediatric or outpatient care were asked to participate and all gave written consent to be observed during their working hours. Patients were informed of the research and asked to give verbal consent upon entering consultations.

Structured data collection forms were used to record consultation observations. These comprised patient data, consultation content; and clinic context data for that particular clinic occasion – defined by date, shift time (morning, evening or weekend) and department, either mother and child health clinic (MCH), outpatient department (OPD) or inpatient ward round. Consultations were selected for observation through purposive sampling to represent the proportion of consultations seen by each clinician in the past 3 months according to hospital records. Clinics were selected firstly to include each clinician in their usual department; secondly to gain a greater representation of outpatient than inpatient clinics (the majority of malaria based decisions were deemed to occur during outpatient consultations where admission also takes place); and thirdly by observing whichever clinician was available during times when only one was on duty, such as during evening and weekend clinics. To reduce biases associated with observed practice, consultations were observed by non-clinical members of the study team (CC, GB, KJ). Researchers were trained in symptom and examination definitions and, after asking for consent from the clinician at the start of the clinic and the patient at the start of each consultation, recorded data from observations of verbal interactions during consultations as well as from clinician notes or prescriptions during or after the consultation. The same process of observation, supplemented by chart review, was conducted for inpatients and outpatients. Patient numbers were taken from hospital records. Clinicians were asked to complete an enrolment questionnaire, detailing demographic and work history details. The main patient, clinic context and clinician factors recorded and examined in the models are shown in Box 1. Patient variables can be divided into ‘clinical’ factors, for example history of fever, or ‘contextual’ factors, for example sex. All non-patient variables are considered ‘contextual’.

Table Box 1.   Variables collected
Patient variablesAge, sex, insurance details, complaint
Consultation variablesStart and end time of consultation
History taken (defined as asking further questions beyond initial complaint presentation) and details
Examination(s) performed [for children, this followed definitions of examinations used in IMCI (Gove 1997)], and results of examinations (e.g. temperature, blood pressure)
Tests requested and details, and test results received (patients returning with results during a clinic had this filled on the same form)
Medication prescribed
Referral to guidelines observed during consultation (wall chart or books) and referral to other staff during consultation (in or outside of the room)
Question(s) asked by the patient
Advice or explanation given by clinician regarding diagnosis, test results, treatment or management
Diagnoses
Whether the patient was admitted, or for inpatients discharged
Clinic variablesStart and end times of clinic
Number of clinicians working
Number of patients seen in clinic overall
Number of patients seen by clinician being observed
Presence of electricity
Clinician variablesAge, sex
Qualification and year of graduation
Secondary school education details
Date started working in hospital
Originates from area around hospital
How many in-service training seminars attended in past 12 months
Salary bracket
If extra duties are held within the hospital
If top-up is received in addition to salary from extra work in the hospital

Decision models

Guidelines for hospital care in Tanzania state that malaria should be suspected in patients with fever, that tests should be conducted when they are available, as is usually the case at hospitals, and that alternative causes of disease should be treated (though these causes are not specified in the guidelines) in case of negative test results. Once alternative causes are ruled out, malaria can be treated presumptively. Three decisions (Figure 1) were identified for analysis in relation to these guidelines and on the basis of being common and important to the use of antimalarial drugs. The first decision was to request a malaria test: the guideline recommends this (where available) for all patients with clinical features suggestive of malaria, these primarily being a current or recent fever. The second decision was to presumptively treat patients with an antimalarial drug without recourse to blood slide testing. The third decision was to treat patients who had a negative test result and were presumed to have microbial infection (prescribed antibiotic, antimalarial or both) but were prescribed an antimalarial without an antibiotic: this does not follow guidelines and fails to treat bacterial causes of illness (Berkley et al. 2005). At the time of the study the national malaria treatment guideline in Tanzania (Tanzania Ministry of Health 2001) did not vary by either age or transmission intensity. Outpatient and inpatient data were analysed together and then separately. Findings are presented separately for outpatients and inpatients only where this variable was found to have a modifying effect on the results.

image

Figure 1.  Outline of the three decisions under study.

Download figure to PowerPoint

Data analysis

Analysis used multivariable logistic regression to adjust for multiple variables affecting each decision and multi-level modelling with random-intercepts to adjust for clustering at the clinic and clinician levels. The models were built within a two level cross-classified framework, in order to account for multiple patients seen within each clinic where multiple clinicians may have been working and for clinicians working in multiple clinics over time (Figure 2). A third level, hospital, was excluded due to small sample size. Variables significant in univariate analysis (< 0.1) for each outcome were added into a multivariable model one by one, and eliminating those no longer significant when adjusting for multiple explanatory variables (< 0.05). Differences between outpatient and inpatient consultations were expected a priori. Evidence of effect modification for out/inpatients was therefore sought by including interaction terms between the out/in variable and each other significant variable in the final model. Stratification was considered in the case of variables very different between outpatients and inpatients.

image

Figure 2.  Cross-classified framework for analysis.

Download figure to PowerPoint

The equation for the multi-level cross-classified logistic regression model was based on a conventional logistic regression model:

  • image(1)

where πi is the predicted probability of a particular outcome for the ith patient (and πi/(1 –πi) is the odds of the outcome), β0 is the intercept and β1, …., βp are the regression coefficients of explanatory variables x1, …., xp.

For the multilevel model, let patients, i, be nested within the cross-classification of clinics, j1, and clinicians, j2. Variation between clinics and between clinicians is adjusted for in the fixed parameters, inline image, and quantified by clinic-level residuals, inline imageand clinician-level residuals, inline image. These residuals form the random part of the model (Goldstein 1995).

  • image(2)

The variance of the residuals represents the clinic- or clinician-level variance, noted as inline image and inline image respectively. The variance was used to calculate variance partition coefficients (VPCs) for each level two variable, giving the proportion of the total residual variance of the model attributable to clinician differences and clinic differences (Goldstein et al. 2002). This was calculated by assuming a continuous unobserved variable underlying the binary response variable and using a threshold model (Snijders & Bosker 1999).

  • image(3)

The VPC is equivalent to the intraclass correlation coefficient in variance components models (Merlo et al. 2005) and is an estimate of the variation in the outcome attributable to a particular unit of clustering, such as the percentage of variation attributable to differences between clinicians or differences between clinics.

The models were fitted in MLwiN (Rasbash et al. 2005), using Markov chain Monte Carlo (MCMC) methods for Bayesian estimates (Browne 2003) rather than traditional maximum likelihood methods in order to obtain more accurate estimates of random effects (Goldstein 1995). Odds ratios were calculated by exponentiating fixed coefficients. 95% confidence intervals were taken from the quantiles of the MCMC distribution. The analysis therefore provides odds ratios adjusted for clustering at the clinic and clinician levels as well as an estimate of the residual variance attributable to each level two variable for each model (VPC).

Ethical approval

Ethical approval was granted by the National Institute for Medical Research, Tanzania and the London School of Hygiene and Tropical Medicine Ethics Committee. Research clearance was granted from the Tanzania Commission for Science and Technology.

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusion
  8. Acknowledgements
  9. References

Study sample

We observed 2082 consultations over 6 months from August 2006; of these 51% were in maternal and child health clinics (MCH) where children under 5 years and pregnant women (both eligible for free health care) are seen; 29% were of outpatient department (OPD) consultations, mostly with adult patients but also with children when the MCH was closed for the weekend or due to staff shortage; 20% were with paediatric in-patients during ward rounds (Table 1). There were no refusals to being observed either by clinicians or patients. The study was conducted during the low season for malaria at the first hospital, with a short rains falling during the study at the second hospital. However, rainfall was low and late and the period was not considered high season for malaria, with no significant peak in admissions during this time.

Table 1.   Observation study sample
 Number of consultations observedNumber of clinics† observedNumber of clinicians observed
  1. Defined as set of consultations seen in any one department on any one shift on any one day.

Inpatients (Paediatric)4093017
Outpatients (Mother and Child Health Clinic)10634520
Outpatients (Outpatient Department)6104525
Total208212034

Clinic and clinician characteristics varied between hospitals (Tables 2 and 3). At HII more children and pregnant women were seen in the MCH, power cuts were more frequent, fewer clinicians with high levels of training saw patients, more young clinicians and fewer clinicians originated from the local area. Most clinicians had attended at least one seminar.

Table 2.   Clinical contexts
 HIHII
Median number of patients seen in an MCH clinic session (range)14 (3–18)42 (22–66)
Median number of patients on a paediatric ward round (range)25 (10–34)20 (10–28)
Median number of patients attending OPD during a clinic session (range)88 (23–122)60 (11–135)
Median duration of a clinic session (range)
 Inpatient ward round90 min (27–193)90 min (45–120)
 Outpatient/MCH clinic4 h (½–6½)4 hour (2½–5½)
Number of clinicians scheduled to work daily
 Ward round, paediatric1–22–3
 MCH22
 OPD morning5–64
 OPD evening1–22
 Power cuts observedRare19% clinics
Table 3.   Clinician characteristics
 HIHII
% of 21† clinicians observed% of 13† clinicians observed
  1. †Sample size smaller by no more than five clinicians at HI and two clinicians at HII in some variables where data was missing from enrolment questionnaire.

  2. ‡Grouped due to small sample attending no seminars.

  3. §200 000 Tanzanian Shillings (TSh) is about $180.

Medical qualification
 Medical Officer (MO)4.80
 Assistant Medical Officer (AMO)19.07.7
 Clinical Officer (CO)76.292.3
Age group
 40+ years66.730.8
Sex
 Male66.776.9
Year of graduation (most recent qualification)
 Since 200038.161.5
Secondary school education
 Form 4 completed (4 years post-primary)56.269.2
 Form 6 completed (6 years post-primary)43.830.8
Number of years worked at hospital
 <2 years33.353.8
 3–9 years22.230.8
 10+ years44.515.4
Employer
 Government/District Council71.4100
 Mission28.60
Originate from area around hospital
 Yes50.015.4
Number of in-service training seminars attended in last 12 months
 1 or none‡25.038.5
 2 or more75.061.5
Extra duties conducted within hospital, e.g. TB or HIV clinic37.546.2
Salary top up received for extra duties or overtime12.516.7
Salary to take home each month
 <200 000 TSh§68.869.2
 200 000 TSh+31.230.8

Children under 5 years formed 70.1% of the sample. Of these children, 43.6% were infants (Table 4). A history beyond initial complaint presentation was taken from 81.1% of patients or care takers and examinations were conducted in 60.6% of consultations. Table 4 shows occurrence of a number of other variables recorded during consultations and clinics. 37.3% of patients had a history of fever, 27.1% were tested for malaria before or during the observed consultation and 37.8% were prescribed antimalarials. Table 5 shows that antimalarial prescription was not exclusive to febrile patients, that many febrile patients were not prescribed antimalarials and that there was some variation in testing and prescribing by age group.

Table 4.   Descriptive and univariate analysis of selected variables
 Overall (%)Univariable odds ratio (95% confidence interval)†
Decision 1: Malaria testDecision 2: No test, antimalarialDecision 3: Test-negative, antimalarial without antibiotics
n = 2082‡n = 1868n = 1517n = 101
  1. †Bold type indicates a statistically significant association with a P-value <0.1.

  2. ‡Unless otherwise stated in variable description.

  3. §Clinic and clinician variables are shown as percentages of consultations. These differ from Tables 2 and 3 due to differences in individual clinician workloads.

  4. ¶Median for patient load varied by clinic type (Table 2) so this variable comprises these different medians.

Patient variables
 Age of patient: <5 years (vs. 5+ years)70.10.81 (0.63–1.05)0.20 (0.15–0.26)5.18 (1.93–13.94)
 Age of paediatric patient: (<12 months vs. 1–5 years, n = 1460)43.61.18 (0.89–1.57)1.40 (1.09–1.86)1.41 (0.31–6.43)
 Sex of patient - female46.20.99 (0.89–1.09)1.29 (1.04–1.60)1.43 (0.52–3.92)
 Complaint - fever37.32.63 (2.07–3.33)19.27 (14.73–25.20)1.09 (0.42–2.82)
 Complaint - vomit (no fever) 5.71.79 (1.14–2.82)2.19 (1.35–3.55)2.19 (0.48–9.96)
 Complaint - diarrhoea (no fever)6.81.36 (0.87–2.11)1.72 (1.12–2.63)0.55 (0.06–4.78)
 Complaint - abdominal pain (no fever)7.20.94 (0.60–1.48)0.27 (0.16–0.46)1.52 (0.36–6.43)
 Complaint - cough or difficulty breathing (no fever)14.50.61 (0.42–0.89)0.33 (0.23–0.46)0.18 (0.02–1.41)
 Complaint - headache (no fever)3.122.86 (1.63–5.00)0.77 (0.36–1.62)6.94 (1.52–31.78)
Consultation variables
 Inpatient (vs. outpatient) 19.60.76 (0.53–1.09)1.11 (0.82–1.49)0.34 (0.12–1.01)
 History taken81.11.03 (0.76–1.38)3.60 (2.58–5.03)2.71 (0.57–12.78)
 Examination conducted60.61.50 (1.17–1.92)1.41 (1.14–1.75)0.89 (0.36–2.25)
 Temperature taken12.63.21 (2.40–4.29)6.28 (4.29–9.20)0.25 (0.03–2.03)
 Abdomen felt10.91.78 (1.28–2.48)0.98 (0.69–1.39)2.13 (0.64–7.14)
 Respiration rate counted2.62.04 (1.05–3.99)1.79 (0.85–3.79)
 Chest exposed (below nipple line)8.61.48 (1.01–2.17)1.33 (0.91–1.95)1.14 (0.21–6.09)
 Stethoscope used9.91.37 (0.96–1.96)1.25 (0.88–1.77)0.55 (0.06–4.78)
 Eyes or palms checked for anaemia14.52.13 (1.60–2.84)0.80 (0.28–2.33)
 Any tests done in addition to  malaria or haemoglobin16.411.96 (9.12–15.70)0.20 (0.12–0.33)0.61 (0.19–2.01)
 Medication other than antimalarial  prescribed including antibiotics78.60.37 (0.29–0.48)11.54 (6.97–19.08)1.68 (0.59–4.73)
 Referral to other staff during consultation8.90.79 (0.50–1.26)0.21 (0.12–0.37)0.69 (0.18–2.63)
 Question asked by the patient/carer9.90.97 (0.65–1.43)0.69 (0.41–1.00)4.67 (1.51–14.41)
 Malaria principal diagnosis36.91.03 (0.81–1.31)19.34 (14.81–25.35)2.78 (1.07–7.22)
 Consultation duration over 5 min37.31.39 (1.10–1.76)0.74 (0.59–0.93)2.38 (0.88–6.45)
Clinic variables§
 Patient load above median for clinic¶55.10.73 (0.58–0.92)1.12 (0.91–1.39)0.30 (0.12–0.79)
 Number of clinicians in clinic 2+59.71.24 (0.97–1.58)0.78 (0.63–0.97)1.78 (0.68–4.68)
 Consultation started after 10.30 am 32.41.15 (0.89–1.49)0.69 (0.53–0.89)2.43 (0.87–6.81)
 Afternoon / weekend shift vs. morning10.90.77 (0.51–1.16)1.08 (0.77–1.52)0.91 (0.23–3.60)
Clinician variables§
 Clinician aged 40+ years36.71.34 (1.06–1.70)0.53 (0.42–0.67)0.48 (0.18–1.31)
 Year of most recent graduation after 200050.00.81 (0.64–1.03)1.55 (1.26–1.92)0.73 (0.29–1.86)
 Clinician sex male68.01.23 (0.95–1.58)0.92 (0.74–1.15)0.27 (0.08–0.88)
 Originate from area around the hospital32.60.67 (0.53–0.85)0.52 (0.41–0.66)1.11 (0.43–2.83)
 Has attended 2+ seminars in past year43.71.60 (1.21–2.12)0.62 (0.47–0.81)0.32 (0.10–1.06)
Hospital
 Hospital observed HII (vs. HI)67.71.48 (1.13–1.94)7.06 (5.17–9.63)0.32 (0.11–0.89)
Table 5.   Malaria decisions by history of fever and age group
  Malaria test requested during consultation†Malaria not tested, antimalarial prescribedNegative malaria result, antimalarials prescribedNegative malaria result, antibiotics prescribedNegative malaria result, antimalarials prescribed, with no antibiotics
  1. Denominator excludes patients already tested for malaria prior to observed consultation.

 % of (n)% of (n)% of (n)% of (n)% of (n)
History of fever
 <5 years25.8 (681)77.3 (467)45.7 (35)71.4 (35)13.8 (29)
 5+ years28.9 (97)71.9 (57)50.0 (12)33.3 (12)55.6 (9)
No history of fever
 <5 years10.1 (779)20.6 (569)36.4 (55)63.6 (55)12.5 (40)
 5+ years13.0 (525)6.6 (424)38.5 (39)35.9 (39)39.1 (23)
Total16.9 (1868)36.0 (1517)40.4 (141)55.3 (141)22.8 (101)

Figure 1 outlines the samples analysed in the three decisions under study. Of 1868 patients observed during their initial consultation, for whom data were available regarding the context of the decision to test, 351 (19%) were tested for malaria. Of the remaining 1517 patients who were not tested for malaria, 547 (36%) were treated with antimalarials presumptively. Patients were reviewed with malaria test results in 194 consultations, and 53 (27%) of these patients were positive for malaria and were prescribed antimalarials. Of the 141 patients who tested negative for malaria, 101 were prescribed antimicrobial (antibiotic or antimalarial) drugs; 25 (25%) of these patients were prescribed antimalarials only. Whilst univariate analyses (Table 4) for all three decision models showed statistically significant associations with patient, consultation, clinic, clinician and hospital variables, many became non-significant when entered into multivariable multilevel models. A diagnosis of malaria was not included in the multivariate analysis as it is considered on the causal pathway.

Decision to request a test for malaria

The decision to request a malaria test (in 96% of cases for blood slide microscopy, 4% rapid test) for 19% of patients appeared mainly to follow clinical reasoning and symptoms. History of fever, headaches and vomiting positively predicted a malaria test request in the final multivariate model (Table 6), and there was an association between requesting a malaria test and requesting other non-malarial tests. Patients were more likely to be sent for malaria tests at HII than HI. Of the variation in the decision to request a malaria test, variation partition coefficient (VPC) estimates suggest around 4.6% of this was attributable to variation between clinicians and around 8.7 to differences between clinics occasions.

Table 6.   Decision to request a test for malaria, multivariable model (n = 1868)
ParameterEstimate (SE†)Adjusted OR‡ (95% CI)
  1. †Standard error.

  2. ‡Odds ratio.

Fixed part:
 Fever1.111 (0.168)3.04 (2.18–4.20)
 Vomit0.843 (0.192)2.32 (1.60–3.40)
 Headache1.848 (0.344)6.35 (3.18–12.33)
 Tests other than malaria or haemoglobin requested or done2.831 (0.173)16.96 (12.13–23.97)
 Hospital (HII vs. HI)0.752 (0.326)2.12 (1.11–4.04)
Random part:Variance partition coefficient (VPC):
 inline image clinician variance0.159 (0.141)4.6%
 inline image clinic variance0.314 (0.168)8.7%

Decision to prescribe antimalarials presumptively

Clinical reasoning appeared to combine with contextual factors in deciding to prescribe antimalarials to 36% patients who had not been tested for malaria. Table 7 presents results stratified by outpatients and inpatients. Patient symptoms predicted an antimalarial prescription. This was more strongly evident for outpatients than inpatients, with the adjusted odds-ratio of fever being associated with antimalarial prescription being 45.92 (95% CI 30.0–73.3) in outpatients compared to 2.68 (0.98–7.7) for inpatients. History of vomiting or diarrhoea also predicted presumptive treatment in outpatients. Non-symptom factors were also important in both models: 17.0% of outpatients and 80.5% of inpatients had no current or recent history of fever. If a history was taken beyond the initial presentation by the patient or carer, antimalarials were more likely to be prescribed and if other tests were done, antimalarials were less likely to be prescribed. For outpatients the additional prescription of other medication increased the odds of an antimalarial prescription and presumptive treatment was less likely if more clinicians were working in that clinic or if the treating clinician had attended two or more seminars on any topic in the past year. For inpatients, antimalarials were more likely to be prescribed presumptively at HII than HI. There was more variation between clinicians (VPC = 12.4%) and between clinics (VPC = 8.2%) in the decision to treat outpatients with antimalarials presumptively compared to the same for inpatients (VPC = 1.5% and 1.7% respectively).

Table 7.   Decision to prescribe antimalarials presumptively, multivariable model (n = 1517)
Parameter Estimate (SE†)Adjusted OR‡ (95% CI)
  1. †Standard error.

  2. ‡Odds ratio.

Outpatient model
Fixed part:
 Fever3.827 (0.231)45.92 (29.973–73.33)
 Vomit1.627 (0.322)5.09 (2.75–9.62)
 Diarrhoea1.134 (0.297)3.11 (1.76–5.65)
 History taken (beyond initial presentation)0.583 (0.282)1.79 (1.04–3.18)
 Tests other than malaria or haemoglobin requested or done−1.042 (0.466)0.35 (0.14–0.86)
 Other medication prescribed (including antibiotics)1.287 (0.407)3.62 (1.66–8.17)
 2+ clinicians working in clinic−0.638 (0.238)0.53 (0.33–0.84)
 2+ seminars attended in past year−0.758 (0.343)0.47 (0.23–0.92)
Random part:Variance partition coefficient (VPC):
 inline image clinician variance0.295 (0.195)12.4%
 inline image clinic variance0.465 (0.320)8.2%
Inpatient model
Fixed part:
 Fever0.984 (0.522)2.68 (0.98–7.66)
 History taken (beyond initial presentation)1.855 (0.557)6.39 (2.23–20.17)
 Tests other than malaria or haemoglobin requested or done−1.974 (0.715)0.14 (0.03–0.52)
 Hospital (HII vs. HI)2.751 (0.453)15.66 (6.67–38.78)
Random part:Variance partition coefficient (VPC):
 inline image clinician variance0.056 (0.100)1.5%
 inline image clinic variance0.051 (0.095)1.7%

Decision to prescribe test-negative patients antimalarials without antibiotics

The prescription of antimalarials without antibiotics to 25% of patients with a negative test suspected for microbial infection (given antimalarial and/or antibiotic) did not appear to follow clinical reasoning. As shown in Table 5, prescribing antimalarials without antibiotics was not associated with a history of fever. It was more likely in patients over 5 years and less likely when the patient load was higher or if the consulting clinician was female (Table 8). Differences between clinicians and clinic sessions were not statistically significant, although this may reflect the limitation of sample size and also the skewed nature of the variance components. We have presented the variation partition coefficients to show the potential effect variables at these levels had over the outcome. The difference between clinicians in prescribing antimalarials to test-negative patients was greater (VPC = 15.1%) than in the models for malaria testing or presumptive treatment, and the difference between clinic contexts was even greater (VPC = 39.3%). However, these statistics should be interpreted with caution due to the small size of the sample and the large standard errors around the point estimates.

Table 8.   Decision to prescribe test-negative patients antimalarials without antibiotics, multivariable model (n = 101)
ParameterEstimate (SE†)Adjusted OR‡ (95% CI)
  1. †Standard error.

  2. ‡Odds ratio.

Fixed part:
 Age over 51.845 (0.870)6.33 (1.56–55.26)
 Patient load above median−2.059 (1.1164)0.13 (0.006–0.60)
 Clinician sex male1.776 (1.066)5.91 (1.04–71.16)
Random part:Variance partition coefficient (VPC):
 inline image clinician variance0.583 (1.192)15.1%
 inline image clinic variance2.126 (5.923)39.3%

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusion
  8. Acknowledgements
  9. References

Understanding the reasons for the overdiagnosis of malaria is essential if antimalarials and antibiotics are to be better targeted, and with the advent of newer but more expensive antimalarials this is an increasing priority. This study demonstrates that the decisions of who to test for malaria (decision 1), and who to treat with antimalarials presumptively when not tested (decision 2), were strongly determined by clinical factors such as fever and headache, as would be expected if guidelines were followed, but also that contextual factors played a role. The decision to prescribe antimalarials without antibiotics to 25% of slide negative patients who were considered to have microbial infection and given an antimicrobial (decision 3) appeared, however, to be determined by no obvious patient symptoms but solely by contextual factors. The lack of overt clinical logic in this decision means changing this practice may be more difficult by conventional training methods.

The finding that the probability of requesting a malaria test was greater if the patient presented with fever or vomiting supports findings from Zambia (Barat et al. 1999) and, together with headache, is consistent with national guidelines (Tanzania Ministry of Health 2001). Requesting a malaria test more frequently when other tests are also requested may reflect a tendency to add malaria tests to a battery of tests when considering alternative causes of disease, together with a reluctance to test malaria when it is deemed the most likely cause of disease. This reluctance could be due to experience of laboratory delays or lack of trust in test results both of which may be addressed through increasing laboratory manpower at peak testing times, introducing rapid tests at the outpatient department, and through quality assurance feedback to clinicians. Differences at the hospital level in deciding to test for malaria may also reflect differences in laboratory factors together with norms amongst colleagues at that hospital.

The prescription of antimalarials to patients not tested for malaria also followed clinical reasoning: fever, vomiting and diarrhoea complaints increased the probability of receiving an antimalarial, although the apparent assumption by clinicians that vomiting and diarrhoea indicate malaria is not supported by clinical findings in true cases of malaria. It is, however, consistent with national guidelines, although testing is recommended if facilities are available as they were in these hospitals (Tanzania Ministry of Health 2001). Other studies have found that symptoms associated with malaria, particularly fever or history of fever, lead to presumptive antimalarial treatment although many of these studies were in settings where microscopy is not available, which was not true in this study (Rowe et al. 2000, 2001; Zurovac et al. 2004; Osterholt et al. 2006). Interestingly, we found that the association between fever and presumptive treatment was significantly weaker for inpatients, indicating an additional context-specific decision. But drawing comparisons between these two groups should be approached with caution, given the small sample of inpatients. A shared outpatient workload decreased the odds of presumptive antimalarial treatment for outpatients; when clinicians worked alone they were more likely to prescribe antimalarials to patients who had not been tested for malaria, which may reflect pressure from their workloads. Clinicians who had attended two or more seminars on any topic in the past year were less likely to prescribe antimalarials presumptively, supporting some findings from elsewhere in Africa that in-service training improves adherence to guidelines for the treatment of malaria and other childhood diseases (Rowe et al. 2001; Zurovac et al. 2004) in contrast to findings of other similar studies which did not find an association (Rowe et al. 2000, 2003; Zurovac et al. 2005; Osterholt et al. 2006).

In contrast to the decision to test for malaria or to treat presumptively, decisions to prescribe antimalarial treatment without antibiotics for patients with negative malaria tests could not be predicted based on clinical symptoms. Both WHO (2006b) and Tanzanian guidelines (Tanzania Ministry of Health 2006) now make it clear that negative malaria tests should exclude malaria in adults, whilst leaving open the possibility to prescribe in children, yet the results show age had no impact on this decision. Prescribing antimalarials without antibiotics to test-negative cases became less common as patient load increased. This may relate to a tendency for physicians to overprescribe drugs under time pressure which has been observed in some settings (Davidson et al. 1995; Bjerrum et al. 1999). Male clinicians in our study gave antimalarials without antibiotics to test-negative patients significantly more often than female clinicians, irrespective of age and length of training. This echoes findings from Morocco where female health workers were more likely to adhere to Integrated Management of Childhood Illness guidelines than male health workers (Naimoli et al. 2006), although other studies have found no association between adherence to guidelines and health worker sex (Rowe et al. 2001, 2003; Zurovac et al. 2004). Decisions on antibiotic prescribing were more evidently rational than antimalarial prescribing. Antibiotics were more commonly prescribed to patients under 5 years, which recognises the most at-risk group and is consistent with other studies which have found that younger children are relatively protected from the treatment of malaria with non-recommended antimalarials (Zurovac et al. 2004) and from the treatment of non-severe with an ineffective antibiotic (Rowe et al. 2001).

The reason clinicians decide to request a test and then ignore the negative result in prescribing antimalarials without other drugs may in part be because guidelines have been ambiguous on this point (D’Acremont et al. 2007). The overall decision to request a test that will then be ignored is not obviously rational from a solely clinical perspective, and the lack of association with any syndrome suggests there is no obvious clinical reasoning behind the decision. It is particularly difficult to understand in adults who in this setting are very unlikely to develop severe malaria. However, non-clinical factors are clearly important in decision making, as shown by the variables significant in the model as well as the variation attributable to clinician and clinic levels, although the latter variables were not significant in this model, potentially due to small sample size. In addition, on the hospital, and community level, there may be pressures for the use of specific tests or prescriptions. Social pressures have been found to influence (non-malarial) prescribing practice elsewhere, in both rich (Hemminki 1975; Schwartz et al. 1989; Avorn & Solomon 2000) and poor settings (Paredes et al. 1996; Howteerakul et al. 2003). Exploration of these factors is essential to inform intervention strategies to successfully change prescribing behaviour (Veninga et al. 2000; Rowe et al. 2005). A qualitative study concurrent with the study presented here suggests a number of organisational and social influences lead to malaria overdiagnosis as an easier and more acceptable diagnosis than alternative causes of disease (Chandler et al. 2008).

A strength of our study is the use of multilevel modelling which both adjusts for clustering at the clinic and clinician levels and enables examination of variance in decision making attributable to differences between individual clinics or clinicians through an estimate of the variance partition coefficient (VPC). The long duration of observation in the study sites enabled this analysis of contextual factors, however the resulting small number of hospitals reduces our ability to look at variance on this level and potentially limits the generalizability of our results. The longer period in two settings may have reduced observer bias: observation of consultations has been found to have a small positive or no effect on performance, at least according to health worker opinions (Rowe et al. 2002), but any effects have also been observed to reduce over periods of observation (Leonard & Masatu 2005). Inter-observer differences in data recording are possible. Around 5% of consultations were observed by two observers (CC and GB or CC and KJ) to informally check for this but discrepancies were few due to the simple nature of the data collection forms and the presence of definitions of examination, testing and treatment categories clearly listed on observation forms.

Conclusion

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusion
  8. Acknowledgements
  9. References

Changing ingrained clinical behaviour is difficult (Haynes & Haines 2002), particularly when the reasons are not based on perceived or real clinical logic (Armstrong 2002). Our results demonstrate antimalarial prescribing practice is contingent upon more than clinical symptoms or algorithms, especially when prescribing for patients with negative malaria tests. The identification of contextual factors affecting clinical decision making may help explain why dissemination of new clinical guidelines has often had limited success in changing practice (Ross-Degnan et al. 1997),and may contribute to the design of interventions to change behaviour more effectively. We identified some contextual factors that may be amenable to change, such as patient load and the number of clinicians available during a clinic session. In addition, variation was found between clinicians and between clinic sessions in the decisions explored, even when the variables measured in this study were taken into account. Differences were also found between hospitals. This suggests that factors operate at these levels to influence practice but they have either not been measured or are unmeasurable quantitatively. Unmeasurable factors may be social – revolving both around organisational culture and medical culture - and need to be identified and characterised. Social context may be a barrier to correct performance, but social factors can also be used to effect change in behaviour (for example in commercial advertising). Our study suggests that interventions to improve clinical decision making for malaria should take into account contextual factors specific to local settings and that further study of social influences on practice are needed to guide the design of such interventions. With the advent of more expensive antimalarials and increasing recognition of the burden of bacterial disease, designing interventions for behaviour change which take these factors into account has become a priority; current prescribing practices are both wasteful and potentially dangerous.

Acknowledgements

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusion
  8. Acknowledgements
  9. References

The authors thank the clinicians and patients who agreed to be observed in the study. We thank the management at participating hospitals, for support from the Joint Malaria Programme of Northeast Tanzania and also to the following researchers and data entry team: Frank Mtei, Beatrice Siteyian, Frida Temba and Clara Lweno.

This study was funded by an Economic and Social Research Council and Medical Research Council Interdisciplinary studentship (CC), by the Sir Halley Stewart Trust (CC), the Gates Malaria Partnership (CW) and the European Union (HR).

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusion
  8. Acknowledgements
  9. References
  • Amexo M, Tolhurst R, Barnish G & Bates I (2004) Malaria misdiagnosis: effects on the poor and vulnerable. Lancet 364, 18961898.
  • Armstrong D (2002) Clinical autonomy, individual and collective: the problem of changing doctors’ behaviour. Social Science & Medicine 55, 17711777.
  • Avorn J & Solomon DH (2000) Cultural and economic factors that (mis)shape antibiotic use: the nonpharmacologic basis of therapeutics. Annals of International Medicine 133, 128135.
  • Barat L, Chipipa J, Kolczak M & Sukwa T (1999) Does the availability of blood slide microscopy for malaria at health centers improve the management of persons with fever in Zambia? The American Journal of Tropical Medicine & Hygiene 60, 10241030.
  • Berkley JA, Lowe BS, Mwangi I et al. (2005) Bacteremia among children admitted to a rural hospital in Kenya. The New England Journal Med 352, 3947.
  • Bjerrum L, Sogaard J, Hallas J & Kragstrup J (1999) Polypharmacy in general practice: differences between practitioners. The British Journal of General Practice 49, 195198.
  • Browne W J (2003) MCMC estimation in MLwiN, Institute of Education, London.
  • Chandler CI, Jones C, Boniface G et al. (2008) Guidelines and mindlines: why do clinical staff over-diagnose malaria in Tanzania? A qualitative study. Malar Journal 7, 53.
  • D’Acremont V, Lengeler C & Genton B (2007) Stop ambiguous messages on malaria diagnosis. British Medical Journal 334, 489.
  • Davidson W, Molloy D W & Bedard M (1995) Physician characteristics and prescribing for elderly people in New Brunswick: relation to patient outcomes. Canadian Medical Association Journal 152, 12271234.
  • Font F, Alonso Gonzalez M, Nathan R et al. (2001) Diagnostic accuracy and case management of clinical malaria in the primary health services of a rural area in south-eastern Tanzania. Tropical Medicine & International Health 6, 423428.
  • Gove S (1997) Integrated management of childhood illness by outpatient health workers: technical basis and overview. The WHO Working Group on Guidelines for Integrated Management of the Sick. Bulletin of the World Health Organization 75 (Suppl. 1), 724.
  • Goldstein H (1995) Multilevel Statistical Models, 2nd edn. Edward Arnold, London.
  • Goldstein H, Browne W & Rasbash J (2002) Partitioning variation in multilevel models. Understanding Statistics, 1, 223232.
  • Hamer DH, Ndhlovu M, Zurovac D et al. (2007) Improved diagnostic testing and malaria treatment practices in Zambia. Journal of American Medical Association 297, 22272231.
  • Haynes B & Haines A (2002) Barriers and bridges to evidence-based clinical practice. In: Getting Research Findings into Practice, 2nd edn (eds AHaines & ADonald), BMJ Books, London, pp. 115122.
  • Hemminki E (1975) Review of literature on the factors affecting drug prescribing. Social Science and Medicine 9, 111116.
  • Howteerakul N, Higginbotham N, Freeman S & Dibley MJ (2003) ORS is never enough: physician rationales for altering standard treatment guidelines when managing childhood diarrhoea in Thailand. Social Science & Medicine 57, 10311044.
  • Jonkman A, Chibwe RA, Khoromana CO et al. (1995) Cost-saving through microscopy-based versus presumptive diagnosis of malaria in adult outpatients in Malawi. Bulletin of World Health Organization 73, 223227.
  • Leonard KL & Masatu MC (2005) The use of direct clinician observation and vignettes for health services quality evaluation in developing countries. Social Science & Medicine 61, 19441951.
  • Merlo J, Chaix B, Yang M, Lynch J & Rastam L (2005) A brief conceptual tutorial of multilevel analysis in social epidemiology: linking the statistical concept of clustering to the idea of contextual phenomenon. Journal of Epidemiology and Community Health. 59, 443449.
  • Naimoli JF, Rowe AK, Lyaghfouri A, Larbi R & Lamrani LA (2006) Effect of the Integrated Management of Childhood Illness strategy on health care quality in Morocco. International Journal for Quality in Health 18, 134144.
  • Njama-Meya D, Clark TD, Nzarubara B et al. (2007) Treatment of malaria restricted to laboratory-confirmed cases: a prospective cohort study in Ugandan children. Malar Journal 6, 7.
  • Osterholt DM, Rowe AK, Hamel MJ et al. (2006) Predictors of treatment error for children with uncomplicated malaria seen as outpatients in Blantyre district, Malawi. Tropical Medicine & International Health 11, 11471156.
  • Paredes P, De la Pena M, Flores-Guerra E, Diaz J & Trostle J (1996) Factors influencing physicians’ prescribing behaviour in the treatment of childhood diarrhoea: knowledge may not be the clue. Social Science & Medicine 42, 11411153.
  • Rasbash J, Steele F, Browne W & Prosser B (2005) A User’s Guide to MLwiN. Centre for Multilevel Modelling, University of Bristol.
  • Reyburn H, Mbatia R, Drakeley C et al. (2004) Overdiagnosis of malaria in patients with severe febrile illness in Tanzania: a prospective study. British Medical Journal 329, 1212.
  • Reyburn H, Mbakilwa H, Mwangi R et al. (2007) Rapid diagnostic tests compared with malaria microscopy for guiding outpatient treatment of febrile illness in Tanzania: randomised trial. British Medical Journal 334, 403.
  • Ross-Degnan D, Laing R, Santoso B et al. (1997) Improving Pharmaceutical Use in Primary Care in Developing Counties: A Critical Review of Experience and Lack of Experience. Chiang Mai, April 1997.
  • Rowe AK, Hamel MJ, Flanders WD et al. (2000) Predictors of correct treatment of children with fever seen at outpatient health facilities in the Central African Republic. American Journal of Epidemiology 151, 10291035.
  • Rowe A, Onikpo F, Lama M, Cokou F & Deming M S (2001) Management of Childhood illness at Health Facilities in Benin: Problems and their Causes. American Journal of Public Health 91, 16251635.
  • Rowe AK, Lama M, Onikpo F & Deming M S (2002) Health worker perceptions of how being observed influences their practices during consultations with ill children. Tropical Doctor 32, 166167.
  • Rowe A K, Onikpo F, Lama M & Deming M S (2003) Risk and protective factors for two types of error in the treatment of children with fever at outpatient health facilities in Benin. International Journal of Epidemiology 32, 296303.
  • Rowe AK, De Savigny D, Lanata CF & Victora CG (2005) How can we achieve and maintain high-quality performance of health workers in low-resource settings? Lancet 366, 10261035.
  • Schwartz R K, Soumerai S B & Avorn J (1989) Physician motivations for nonscientific drug prescribing. Social Science & Medicine 28, 577582.
  • Snijders TAB & Bosker RJ (1999) Multilevel Analysis. Sage, Newbury Park.
  • Talisuna AO & Meya DN (2007) Diagnosis and treatment of malaria. British Medical Journal 334, 375376.
  • Tanzania Ministry of Health (2001) Diagnosis and Treatment of Malaria. A Handbook for Clinicians. National Malaria Control Programme, Dar es Salaam.
  • Tanzania Ministry of Health (2006) Malaria Guidelines. National Malaria Control Programme, Dar es Salaam.
  • Van Dillen J, De Jager AJ, De Jong I & Wendte JF (2007) Overdiagnosis of malaria in hospitalized patients in Namibia. Tropical Doctor 37, 18586.
  • Veninga CC, Denig P, Zwaagstra R & Haaijer-Ruskamp FM (2000) Improving drug treatment in general practice. Journal of Clinical Epidemiology 53, 762772.
  • WHO (2006a) Facts on ACTS. WHO, Geneva. http://www.searo.who.int/LinkFiles/Drug_Policy_RBMInfosheet_9.pdf Accessed 7 August 2007 .
  • WHO (2006b), Guidelines for the Treatment of Malaria. WHO, Geneva (WHO/HTM/MAL/2006.1108).
  • Wiseman V, Kim M, Mutabingwa TK & Whitty CJ (2006) Cost-Effectiveness Study of Three Antimalarial Drug Combinations in Tanzania. PLoS Medicine 3, e373.
  • Zurovac D & Rowe A K (2006) Quality of treatment for febrile illness among children at outpatient facilities in sub-Saharan Africa. Annals of Tropical Medicine and Parasitology 100, 283296.
  • Zurovac D, Rowe A, KOchola SA et al. (2004) Predictors of the quality of health worker treatment practices for uncomplicated malaria at government health facilities in Kenya. International Journal of Epidemiology 33, 10801091.
  • Zurovac D, Ndhlovu M, Rowe A K et al. (2005) Treatment of paediatric malaria during a period of drug transition to artemether-lumefantrine in Zambia: cross sectional study. British Medical Journal 331, 734.
  • Zurovac D, Larson B A, Akhwale W & Snow R W (2006) The financial and clinical implications of adult malaria diagnosis using microscopy in Kenya. Tropical Medicine & International Health 11, 11851194.