Comparison of all-cause and malaria-specific mortality from two West African countries with different malaria transmission patterns

Background Malaria is a leading cause of death in children below five years of age in sub-Saharan Africa. All-cause and malaria-specific mortality rates for children under-five years old in a mesoendemic malaria area (The Gambia) were compared with those from a hyper/holoendemic area (Burkina Faso). Methods Information on observed person-years (PY), deaths and cause of death was extracted from online search, using key words: "Africa, The Gambia, Burkina Faso, malaria, Plasmodium falciparum, mortality, child survival, morbidity". Missing person-years were estimated and all-cause and malaria-specific mortality were calculated as rates per 1,000 PY. Studies were classified as longitudinal/clinical studies or surveys/censuses. Linear regression was used to investigate mortality trends. Results Overall, 39 and 18 longitudinal/clinical studies plus 10 and 15 surveys and censuses were identified for The Gambia and Burkina Faso respectively (1960–2004). Model-based estimates for under-five all-cause mortality rates show a decline from 1960 to 2000 in both countries (Burkina Faso: from 71.8 to 39.0), but more markedly in The Gambia (from 104.5 to 28.4). The weighted-average malaria-specific mortality rate per 1000 person-years for Burkina Faso (15.4, 95% CI: 13.0–18.3) was higher than that in The Gambia (9.5, 95% CI: 9.1–10.1). Malaria mortality rates did not decline over time in either country. Conclusion Child mortality in both countries declined significantly in the period 1960 to 2004, possibly due to socio-economic development, improved health services and specific intervention projects. However, there was little decline in malaria mortality suggesting that there had been no major impact of malaria control programmes during this period. The difference in malaria mortality rates across countries points to significant differences in national disease control policies and/or disease transmission patterns.


Background
Mortality rates in sub-Saharan Africa (SSA) remain very high, especially when compared to other regions of the world [1,2]. While mortality in many parts of the world has declined markedly in the last century, in SSA reductions have been less, although marked differences in trends between urban and rural areas are still observed with trends declining faster in urban areas [1,[3][4][5]. A recent report from the World Health Organization estimates that each year there are about ten million deaths worldwide in children below five years of age [1]. The leading causes of death among children in this age-group are pneumonia, diarrhoea, malaria, and perinatal causes.
Approximately 90% of all malaria deaths in the world occur in Africa [1,6]. Major factors that contribute to this high mortality from malaria in Africa include poor access to health services, low quality of health services, and the increased resistance of malaria parasites to affordable first-line drugs such as chloroquine and sulphadoxinepyrimethamine [7,8]. There is some evidence for a positive correlation between Plasmodium falciparum transmission intensity and malaria mortality but there are few data on this point and also such analysis is hampered by the difficulties in assessing causes of death [9][10][11]. Moreover, in regions where malaria is endemic, an association between observed all-cause mortality rates and malariaspecific mortality rates, especially among the children below five years of age would be expected. However, it is difficult to substantiate this association due to the scarcity of quality data from Africa on all-cause and in particular cause-specific mortality. Many surveys and a few censuses have been conducted in Africa since the early 1950s and these provide information on all-cause mortality rate but not on cause-specific mortality [12]. To address this gap, several demographic surveillance systems (DSS) have been established in developing countries where overall mortality, cause-specific mortality determined using the post-mortem questionnaire technique, morbidity, socioeconomic status and migration can be studied [13].
In this study, data from surveys and censuses, as well as DSS data obtained from The Gambia and Burkina Faso were used to systematically compare all-cause and causespecific mortality rates from under five children. Burkina Faso has four DSS sites located in Nouna, Ouagadougou, Oubritenga and Sapone, while Gambia has one in Farafenni [14]. The climate of both countries is similar with a short and intense rainy season occurring between June and October. Malaria transmission intensity is generally more marked in Burkina Faso[15], than in The Gambia [16].

Methods
A literature search was done on papers containing information on mortality and malaria among children below five years of age from each country for the years since 1950. The online search was done in Medline, Embase, and Popline, using key words: "Africa, child deaths, The Gambia, Burkina Faso, malaria, Plasmodium falciparum, mortality, child survival and morbidity". Subsequent literature searches were accomplished by reviewing the literature cited in each of the publications obtained. In addition, research groups and experts with publications related to studies conducted in these countries were contacted for further information. Each of the papers obtained was indexed and thoroughly reviewed to extract all relevant information. The following information was obtained: study period for the data presented (the year the study started and ended), duration of the study, number of children less than five years included, overall number of deaths during the study, number of deaths due to malaria, person-years of follow-up, all-cause mortality, and malaria-specific mortality. The information in the papers was cross-checked twice to ensure that the information extracted was consistent with the information in the original paper.

Mortality information
Mortality information was summarized as infant mortality rate (IMR: <1 years), child mortality (CMR: 1-4 years) and under-five child mortality (U5MR: 0-4 years) per 1,000 person-years. Mortality information was recorded as all-cause mortality and malaria-specific mortality for all the above age categories with the exception of studies for which such detailed information was not available.
For malaria mortality estimations, studies that were conducted among young infants only (0-3 months) were excluded because many studies have shown protection of children in this age group in malaria endemic areas through acquired immunity and other mechanisms [17]. Similarly, studies from which only seasonal rates could be obtained were also excluded. For clinical trials, mortality rates were calculated in both the intervention and control groups combined and in the control group alone. all children who died during the study lived on average for half of the total observation time. An overall person-year based mortality rate was computed by dividing the observed total deaths by the sum of all person-years.

Cause of death ascertainment
In nearly all of the studies reviewed, cause of death was ascertained by using the verbal autopsy method [18,19]. This method involves interviewing a relative of the deceased child on symptoms and signs of the illness before death. By reviewing the circumstances surrounding the time of death as described in the interviews and by studying any available clinical records, clinicians allocate a potential cause of death. Details of the verbal autopsy procedures are available elsewhere [19].

Mortality rate calculations
After compiling information on number of deaths and observed/estimated person years, all-cause mortality rates (MR) per 1,000 person-years were calculated as;

MR = (D/PY)* 1000
where D is number of all observed deaths during the study period and PY is the observed or estimated total personyears. For the malaria mortality rate, D was replaced with the observed number of all malaria-specific deaths. If applicable, mortality for the control arm was recorded separately.
For the surveys and censuses (see Table 1 and Table 2) which are cross-sectional studies, mortality information generated from collecting birth histories and child survival from mothers who were interviewed was used to estimate mortality. The mortality information was given as the probability of dying before reaching the nth birthday. To enable comparisons with other studies, the respective probabilities of dying in the different age-groups were converted to mortality rates per 1,000 per year using the following method [20]; where λ x is estimated mortality rate for the age group x, q x is the probability of dying before reaching the nth birthday and ∆ x is the length in years of the age interval.

Modelling
To model the under five malaria and overall mortality rate weights to account for study size, with different functions of f to investigate a possible non-linear relation between rate and calendar year, and with a factor to assess a possi- ble interaction between country and year. All studies described in Tables 1 to 5 were used in the modelling. Since for survey/censuses the person-years were unknown, the weights were chosen equal to that of the largest cohort study. The effect of the different weights was investigated with a sensitivity analysis. For clinical trials, only results from the control arms were used.

Results
The literature review identified thirty nine longitudinal (clinical, epidemiological, cohort) studies from The Gambia in which malaria had been studied among children below five years of age. Twenty-one of the 39 studies had information on mortality (seventeen of these studies contained cause-specific mortality information). In addition, ten surveys and censuses were identified for which only all-cause mortality information was available.
Similarly, eighteen longitudinal studies on malaria where identified for Burkina Faso, with eight studies containing mortality information. Three of these studies contained cause-specific mortality information. In addition, fifteen surveys and censuses were identified for which all-cause mortality information was available.

Trends of mortality in The Gambia
Mortality rates based on surveys and censuses Table 1 shows under-five year all-cause mortality rates for The Gambia following a review of studies based on surveys and censuses conducted between 1973 and 2004. The information is summarized as infant mortality rate (IMR), child mortality rate (CMR) and under-five mortality rate (U5MR). Two major types of survey have been conducted repeatedly (contraceptive prevalence and fertility determinants surveys and UNICEF multiple indicator cluster surveys). Three population censuses were conducted in 1973, 1983 and 1993. Mortality rates obtained from the 10-year periodical census show that IMR and U5MR declined significantly from 243 and 67.8 in 1973 to 87.7 and 27.6 in 1993 respectively ( Table 1). Estimates of mortality based on surveys show similar declines in mortality, however, these are not as pronounced as census based estimates. In contrast, results reported by Pacque-Mangolis and colleagues [27] from a survey conducted in nearly the same time period provided much lower rates  3 Estimate of children under five year from census population. 4 Results based on combination of many surveys.

Mortality rates based on longitudinal and follow-up data
Information on child mortality rates obtained from longitudinal or follow-up studies conducted in The Gambia between 1946 and 1999 is summarized in Table 3. Variations in the estimates obtained in these studies were observed, even when the reference period or year was the same. A number of studies reported mortality estimations over a period of several years [25,28,29]. In 1946, the IMR was estimated at 531.6 per 1,000 live births in three Man-dinka villages in The Gambia [30]. Studies conducted in Keneba and Manduar villages [28] for the period 1951 to 1975 showed different mortality rates (sometimes with big outliers) between villages although, overall, no clear improvements in mortality were noticeable over this time period. Longitudinal studies in the same area conducted after 1975 showed a declining mortality rate [31]. Hill and colleagues [25] report similar patterns for the period 1982 to 1992 in the Farafenni areas (see Table 3). Rayco-Solon and colleagues [32] provide estimates for three-time windows between 1949 and 1997 and show clearly that mortality has declined over time in The Gambia.   3 Whole population and not under-5 year population. 4 Malaria diagnosis laboratory confirmed.   Under-five mortality rates 1950 1954 1958 1962 1966 1970 1974 1978 1982 1986 1990 1994 1998 2002 Years of study

Malaria mortality
Malaria mortality rates show no clear change for the period 1981 to the late 1990s (see Table 4 and Figure 3).

Mortality rates based on survey and census data
A summary of all-cause mortality rates based on surveys undertaken in Burkina Faso is given in Table 2. Overall, IMR and U5MR show a declining trend for the period 1960 to 2004 (Figure 1 and 2

Mortality rates based on longitudinal data
The results for all-cause and malaria-specific mortality rates obtained from longitudinal studies conducted in Burkina Faso are summarized in   Infant mortality rate 1950 1954 1958 1962 1966 1970 1974 1978 1982 1986 1990 1994 1998 2002 2006 Years of study data were restricted to the control group (Table 5). These patterns for U5MR are shown in Figure 1. There are a few studies with results for IMR, but among these an overall decline in the rates for the period 1993 to 2000 can be seen (Table 5 and Figure 2).

Malaria mortality
The weighted malaria mortality rate for the period 1999-2003 averaged 15.4 and 15.6 per 1,000 person-years with all studies including combined arms, and with all studies including control arms only, respectively (Table 5 and Figure 3). The malaria-specific mortality rates reported here come mostly from one study site (Nouna) in Burkina Faso and are within the range of previously estimated figures from other countries in West Africa [33]. Similar estimates for children below one year are shown in Figure 2. For malaria, there is no significant change in the rate over the observation period (Figure 3), however the difference between both countries is significant (p = 0.02). The estimated weighted average malaria mortality rates by country, for the observation period are displayed in Figure 3. The sensitivity analysis in which an up to tenfold weight was assigned to census/survey data to account for the larger population sizes in those studies showed only marginal variations in the model estimates.

Discussion
This study has reviewed, summarized and compared allcause and malaria-specific mortality rates for The Gambia and Burkina Faso. Many studies have been conducted in The Gambia to investigate different aspects of malaria  in contrast to Burkina Faso where only a few studies have been conducted since the late 1980s to early 2000s. Findings based on censuses, surveys and longitudinal studies show that child mortality in the Gambia has declined significantly. Reasons for this decline are likely to include socio-economic development, improved health services including the establishment of primary health care and immunisation programmes and specific intervention projects [25,29,34,35]. However, since 1990, this decline in mortality has slowed down. In parts of Africa where HIV/AIDS is a major problem, mother-to-child transmission has been a major explanation for this rebound in child mortality [36]. However, The Gambia and Burkina Faso have not been severely affected by the African HIV/AIDS epidemic and seroprevalence rates in adults remain less than 4% in both countries [37].
Estimates of mortality rates from nationally representative surveys conducted from 1960 to 2003 show that child mortality has also declined in Burkina Faso. However, this trend is not clearly discernable in the few studies for which mortality was estimated from DSS longitudinal studies because of coverage of only a few years. Nevertheless, results from DSS studies from the two countries during the period 1992 to 1999 yield very comparable estimates for under-five mortality rates, although results from surveys and censuses for under-five mortality rates are slightly different for the later years. This observation suggests the need for caution in generalising results from DSS longitudinal studies to represent national figures. Differences in observed mortality rates from nationally representative surveys could be a reflection of differences in national health policies and programmes in the two countries whereas the comparable rates obtained from the DSS based studies could be due to the effects of interventions (such as insecticide treated bednets, vaccination programmes, primary health care service delivery etc) that are routinely implemented among the DSS populations [38,39].
It is not easy to fully interpret the data provided in Figures  1, 2 and 3. However, overall, all-cause mortality rates fell in both countries during the study period covered, Est. malaria mortality rates 1 9 8 0 1 9 8 3 1 9 8 6 1 9 8 9 1 9 9 2 1 9 9 5 1 9 9 8 2 0 0 1 Years of study Burkina Faso Gambia Gambia weighted ave. mort.rate Burkina Faso weighted ave. mort.rate although some peaks occurred in the early 1990's in The Gambia. Malaria mortality rates did not show a similar decline. This could be partly due to parasite resistance to chloroquine, the first-line drug for treatment of malaria during those years, as well as to a break down in community based health services due to some under-funding by the national government during the years that followed 1990 [40][41][42][43]. There is some evidence that in The Gambia the incidence of clinical malaria is now beginning to decline (Cessay, personal communication).
Mortality differences across age groups are evident in the two countries. The observed differences in age-mortality patterns are reminiscent of mortality in other countries in SSA where the mortality rates for children in the age-group 1 and 4 have been noted to be lower than the infant mortality rate [44]. In this review, the ratio of infant mortality rate to child mortality rate has been shown to have increased over time. The overall risk of dying in the first year has declined over time, but not to the same extent as the decline in the risk of dying between the ages of one and four years. Possible explanations for these results include introduction of the expanded immunization programmes (EPI) which might have had a bigger effect on CMR than on U5MR. For instance, measles is no longer a major cause of death in children between one and four years of age in many parts of Africa where effective EPI programmes are well established [45]. Other explanations might include improvements in nutrition with deaths due to malnutrition in older children being averted [46].
The estimated malaria mortality rates of 9.5 (The Gambia) and 15.4 (Burkina Faso) per 1,000 person-years differ considerably but are within the previous estimates for SSA [47]. Numerous interventions have been undertaken in The Gambia that may have had an effect on malaria including the introduction of village-based health services and insecticide-treated bednets (ITNs) [34,43,48]. Previous studies have shown variable evidence for the link between malaria mortality rates and malaria transmission intensities [49,50]. In Burkina Faso, the annual average entomological inoculation rate (EIR) is above 80 while in The Gambia an annual average EIR is less than 50 infectious bites per person per year [16]. Therefore, differences in estimated malaria mortality rates might be a reflection of differences in malaria transmission patterns, although this is controversial [11,51].
This study has several limitations. It is possible that other relevant literature might have been left out, especially literature that was not published online or in the English language. However, it is likely that the mortality trends observed are not significantly biased due to this limitation. The fact that different studies were designed to answer a given research question makes comparison of studies difficult. The sampling methods and aims of the surveys were different and although cross-comparison of child mortality rates between studies is possible, such comparisons needs to be treated with caution. On the other hand, population censuses should provide a complete enumeration and mortality rates estimated from the census should be more realistic without major sampling errors. For instance, mortality rates resulting from a bednet intervention study cannot easily be compared to mortality rates resulting from a study that evaluated the performance of primary health care. The latter is most likely to give rates that are lower due to the direct protection provided by the mosquito net which lowers malaria transmission rates. Furthermore, the gross differences in the years covered by the malaria-related studies conducted in the two countries made comparisons of estimates difficult. In The Gambia, studies have been conducted over a longer period than in Burkina Faso where most studies were conducted after the year 1990.
Efforts were made to standardize rates by either computing person years or presenting results in the same units for the same age-group categories. For some studies it was difficult to achieve such standardized rates especially when computing malaria-specific mortality rates. In many papers, desired details were not available and it was not possible to make estimates for all age-specific mortality rates.
Despite all these limitations, this review demonstrates that with available data, mortality patterns can be compared across countries in Africa. However, caution is needed in extrapolating the results obtained in DSS surveys to give national rates. Further work will include investigating the linkage between malaria transmission parameters and child mortality. This needs to be done locally using DSS based data since estimates cannot be readily inferred from national mortality figures.
MJ substantially participated in the conception in particular with regard to the studies in The Gambia and writingup of the manuscript.
AS substantially participated in the conception in particular with regard to the studies in Burkina Faso and writingup of the manuscript.
BK substantially participated in the conception in particular with regard to the studies in Burkina Faso and writingup of the manuscript.
BG substantially participated in the conception in particular with regard to the studies from The Gambia, in the interpretation of the results, formulating the discussion and writing-up of the manuscript.
OM substantially participated in the conception, analysis and interpretation of the results, formulating the discussion and writing-up of the manuscript.