Ethics statement

The study protocol was reviewed and approved by the institutional ethics committee of the Institut National de Recherche en Santé Publique, Bamako, Mali. Parents or guardians were informed about the study and were asked for their written consent, which was given for all cases included in the study.

Study site

The study was conducted in Sikasso Hospital, in South-East Mali. Sikasso is the second largest town in Mali, and the hospital covers a population of approximately 2 million people, predominantly rural. Malaria is endemic in this area with seasonal transmission during the rainy season (July – November). One study in this area showed that 87% of febrile patients in the rainy season were positive for malaria parasites [13].

An internal audit in 2002 revealed that the inpatient case fatality rate in the paediatric ward was 24.3% (151 deaths out of 621 admissions), with 64 deaths (42.4%) attributed to malaria. Changes were then made, to have a stock of essential medications (including artemether and 10% glucose) on the ward to enable immediate treatment and to provide blood transfusions as rapidly as possible. During the present study, all inpatients with presumed severe malaria were treated according to local consensus guidelines [14]. The first-line parenteral antimalarial was intramuscular artemether, followed by artesunate/amodiaquine once the patient was able to swallow. This is because staffing levels were insufficient to permit adequate supervision of 8-hourly quinine infusions. Blood glucose was evaluated with a portable glucometer on admission and hypoglycaemia was treated. Intravenous ceftriaxone was prescribed for all comatose patients, and all patients with repeated convulsions underwent a lumbar puncture to investigate for meningitis. Staff were advised to avoid non-essential and non-evidence-based medications such as vitamins, steroids, and intravenous paracetamol. Emergency medications and medical equipment were made available on the paediatric ward so that treatment could start immediately on arrival of the patient.

Staff were motivated through participation in the development and implementation of the guidelines, regular meetings to discuss progress and difficult cases; and small bonus payments to compensate for extra work. User fees were waived for participants during the study period. The care provided to the patients in hospital was not influenced by patients' ability to pay. Therefore we could explore how differences in outcome were influenced by differences in pre-hospital care, controlling for differences in severity at the time of admission.

Participants

All children presenting to the Paediatric Department of Sikasso Hospital with a clinical diagnosis of severe malaria (according to the admitting physician) from 28^th September to 30^th November 2006 were invited to participate. This was a prospective consecutive sample during the peak malaria season. Patients with no parasites seen on blood films were not excluded, so as not to bias the sample (because pre-admission antimalarials may make it harder to find parasites). We were careful to collect information on all patients, even those who died within a short time of admission.

Data collection

Patients were clinically assessed, treated and stabilised as soon as possible after admission. As soon as the patient was stable, one of the interns interviewed the parent(s) or guardian of the patient using a structured questionnaire. Parents were asked about the duration of the illness before presenting to hospital; whether the patient had received any pre-hospital treatment before or after the malaria became severe, who had given this, and what the treatment was. Patients were also shown a photograph and sample of two plants commonly used for the treatment of malaria to ask whether they recognised them and whether they had used them to treat this illness. We had to rely on information given by parents because in most cases there were no written records. Patients were then followed prospectively until discharge from hospital or death.

Statistical analysis

Data were entered using the customised data entry facility of Epi-info 6.04 (CDC, Atlanta, GA, USA) and analysed with Stata version 10.0 (Stata Corporation, College Station, TX, USA) and R statistical software (http://www.r-project.org). Descriptive statistics were computed for demographic characteristics, clinical features, pre-hospitalization malaria treatments and subsequent interventions. We examined the risk factors for death according to univariate analyses and to a logistic regression model. The model was adjusted for variables which were thought to be associated with mortality, including demographics, clinical baseline features, indicators of severe malaria and treatments. All variables were thus kept in the logistic regression, including those which were not statistically significant in univariate analyses.

A survival analysis was then carried out to test the robustness of our findings. Survival curves were estimated using the Kaplan-Meier method and differences in the curves were explored using the log-rank test. The inter-relationship of possible pre-hospital risk factors and survival was analysed using a Cox Proportional Hazards model, focusing on drug treatments received prior to hospital admission, patient demographic factors, duration of illness, and measures of the severity of the illness on arrival at the hospital. Interactions were included for weight and age, and modern and traditional treatment. Weight, age, the duration of illness and the time taken to travel to the hospital were log transformed to improve model fit. Multivariate selection of key risk factors was carried out using a stepwise procedure. For the purposes of the analytic model, patients who were discharged “cured” by the doctors were assumed to have survived for the duration of the study period. There were 12 patients who self-discharged without consent from a doctor and their survival was only assumed up to the point of discharge. In a sensitivity analysis, we verified that similar results were obtained (data not shown) when survival of those discharged “cured” was only assumed up to time of discharge. The assumption of proportional hazards over time was checked for all covariates.

Principle findings

As 86% of deaths occurred within 24 hours of admission, pre-hospital treatment is likely to have the biggest impact on reducing case fatality. The commonest pre-hospital treatments were traditional (n = 174), paracetamol (144), chloroquine (76) and quinine (68). These were all widely available at the time of the study. However none of the pre-hospital treatments had an effect which reached statistical significance in the adjusted logistic regression model. It is possible that some of these effects would become statistically significant with a larger sample size, as the number of patients having taken any particular treatment was relatively small. Furthermore no information is available about dose, duration, compliance, sequence and quality of the treatments. Traditional treatments did not significantly alter the effect of modern treatments (OR 0.88 “modern only” versus 0.77 “both treatments” as compared to “no treatment”). The extreme diversity of different treatments and combinations (modern-modern, traditional-modern) shows that official treatment recommendations were not followed most of the time. Although chloroquine is no longer recommended as a treatment for malaria in Mali, it seemed to be more protective than amodiaquine, which has been increasingly used as monotherapy when ACTs are out of stock. This may have led to an increase in amodiaquine resistance which may explain why this treatment alone seemed the least effective. It is interesting that co-trimoxazole seemed protective, as it has both antimalarial and antibiotic properties.

The major factor associated with survival was sex: the case fatality rate in girls was twice that in boys, even after adjusting for all the other risk factors we could think of. Data on duration of illness prior to hospitalisation and time to death imply that girls deteriorated faster than boys. Retrospective data from hospital records did not reveal the same discrepancy between the sexes. Hospital policy was to exclude from their statistics those patients who died before receiving treatment. In our cohort we were particularly careful to include all children who arrived in the paediatric department with severe febrile illness (even if they died very soon after arrival), and of these, twice as many girls as boys were excluded from the official statistics. Other studies which have not attempted to include all patients may therefore have missed a difference in mortality between the sexes [15]. Demographic and Health Survey data on infant mortality in Mali found that male infants were at greater risk of death from malaria than female infants [16]; but this trend may be reversed for older children. Sociological factors are the most obvious explanation for the difference in survival, for example girls might be less well nourished, and have less resources spent on them (for treatment) than boys in a male-dominated society. Indeed the fact that this cohort included significantly more boys than girls (55.6% vs 44.4%) suggests that families are more likely to bring boys to hospital than girls, because there is no reason to believe that incidence of severe malaria is greater in boys than in girls. Girls may only be brought to hospital when they are more seriously ill than is the case for boys. The only significant differences we could find between management of boys and girls in our dataset was that boys were slightly more likely to have received a pre-hospital treatment (although the type of treatment did not differ), and more likely to receive a transfusion when it was prescribed. These are important findings and should be investigated further, but they do not fully explain the large observed difference in case fatality.

There was no significant association in any of the models between case fatality and reported pre-hospital duration of illness, or treatment provider. Traditional treatments did not significantly worsen prognosis, and may indeed have improved prognosis when taken together with chloroquine. It would be interesting to research this further, as there are indications that some herbal medicines can reverse resistance to chloroquine [17]. These findings differ from those of other studies; in Sudan it was found that delay in seeking treatment was a risk factor for mortality [7]. A study in Nigeria found that chloroquine was associated with a worse prognosis [6]. Some studies claim that that patients consulting traditional healers or taking traditional medicines have a worse outcome [18] but this is not borne out by our findings. Since most patients (64%) presented at the hospital after a disease duration of three days or more, it was not statistically possible to assess the possible protective effect of an early presentation. The odds ratio for death in children presenting three or more days after the start of illness had a wide confidence interval (OR 2.26, 95% CI 0.48–10.65). Furthermore living in Sikasso town (relatively close to the hospital) and a short time of travel to the hospital were associated with a worse prognosis, implying that patients with rapidly progressive illness from further afield die before arriving at the hospital.

Mortality was greatest for patients with respiratory distress, hypoglycaemia or coma. The effect of hypoglycaemia is discussed in more detail in a separate paper [14]. These findings agree with the results of other studies [5], [19], [20]. We investigated the treatment of hypoglycaemia with sublingual sugar in a separate study [21], but did not implement the findings in this study (infusions of intravenous glucose were used as per the consensus guidelines). It is possible that pre-hospital use of sublingual sugar on a routine basis could further reduce mortality in patients with hypoglycaemia.

Strengths and weaknesses of the study

This is one of the first studies to examine the impact of pre-hospital treatment on case fatality for children admitted with severe malaria. We made every effort to implement a best-practice standard protocol for inpatient management of severe malaria, and treatment was given free of charge. This may otherwise have been an important confounding factor as poorer patients may not have been able to afford the full treatment. Comparison with hospital statistics from previous years suggests that this intervention in itself reduced case fatality below the previous level (although the inclusion criteria were not directly comparable). Another study has shown that an intervention to improve quality of care in the hospital, and to make care free of charge, can reduce inpatient case fatality [22]. However we were unable to completely standardise blood transfusion, because no patient can receive a transfusion until a relative or friend has donated blood and our data suggests that it was harder to find blood donors for girls than for boys.

The fact that we recruited only hospitalised patients means that the sample is biased, because the majority of patients with severe malaria do not present to hospital [8]. However we believe that more patients presented than usual, because fees for treatment were waived. The number of malaria admissions was greater than in the previous malaria season. All those included represent a failure of pre-hospital treatment, which should aim to prevent not only death but also severe malaria. Therefore community studies would be better for estimating the effectiveness of treatments in the whole population. However such studies would have to be extremely large to gather information on the relatively small proportion of deaths. Even in this study of 425 patients, the number in each treatment sub-category was small because of the high diversity of treatments used, so there was insufficient power to detect differences between individual treatments with confidence and statistical significance.

The diagnosis was left to the admitting physician, without defining precise criteria for “severe malaria”. Almost every severe febrile illness was labelled as “severe malaria”, and the diagnosis at discharge was still “severe malaria” in all but 4 patients. A study in the community at the same time found that 87% of patients presenting with symptoms of fever had a blood film positive for malaria [13]. Although the population is not strictly comparable, the rate of over-diagnosis was probably <15%.

There is a risk that answers to the questionnaire may not have been accurate, because the patient's relatives may have feared criticism, or because they were not fully aware of the treatment taken, or simply because their memory was imperfect. The history was taken at admission, although for some cases it was not possible to take the history before the patient died (as 34% died less than one hour after arrival). In 16.2% of the cases, the respondent was not one of the parents of the child. The fact that the study took place in a hospital, and the questionnaire was administered by an intern, may have influenced the replies. We relied on the memory and honesty of parents and did not verify the drug history through urine testing, so we have no information on compliance or quality of the drugs taken. It is possible that some of them may have been of poor quality, out of date, or counterfeit. As this is an observational study, “confounding by indication” cannot be excluded, even while controlling for many clinical factors. It is also impossible to accurately standardise the time from symptom onset to arrival at hospital.

Meaning of the study

These findings have several implications for public health policies. In spite of the improvements we made to hospital care, the majority of deaths occurred within a short time of admission. It seems that, like meningitis in European settings, the infection can progress very rapidly from a seemingly innocuous prodrome, and can unexpectedly become severe. We did not ask care-givers how long after the start of the illness they gave the first treatment, and this would be interesting to include in future studies. However it seems logical that the earlier an effective treatment is given, the better the outcome should be.

As the first source of treatment was the family in over 55% of cases, parents and families need training on when and how to administer “home-based management of malaria”. This could be delivered alongside interventions to prevent childhood diseases, such as vaccination and distribution of insecticide-treated bednets. Efficacy alone is not sufficient to prevent death; effective treatments must be widely available, for girls as well as for boys.

Unanswered questions and future research

Further research is needed to show whether a concerted and integrated intervention in home-based prevention and management of childhood febrile illness can indeed reduce case fatality, and how all treatment providers can collaborate to make this a reality. There is a lot of scope for further research on traditional medicines, and for their improvement. As hypoglycaemia is also an important risk factor for death, early administration of sugar may also be important [14]. Further studies are needed to investigate compliance with different treatments, determinants of choice, and treatment sequence. It would also be useful to collect information on severity of symptoms before hospitalisation.

Further research is needed to clarify why the case fatality was twice as high in girls as in boys, whether this finding is more widely generalisable to other settings, and what interventions could be envisaged to address this inequality. Higher case fatality in girls with severe malaria has also been noticed in a similar study in Gabon [1] and in Ghana it has been reported that girls are more likely to be treated at home, whereas boys are more likely to be taken to a health facility [23].

Although improvements in quality of care at the hospital can reduce inpatient case fatality, the real challenge is to investigate how these improvements can be sustained beyond the period of the research study, and how capacity can be increased to meet the increased demand for hospital treatment. This will probably require continued motivation of staff through regular audit or “significant event analysis”, and possibly a degree of external supervision, as well as additional financial and human resources.

Conclusion

Aside from well-recognised markers of severity, the main risk factor for death in this study was female sex, which could be a proxy for other socio-cultural determinants. Differences in pre-hospital treatments were not associated with case fatality in this study.