Season of death and birth predict patterns of mortality in Burkina Faso

Abstract

Introduction

Mortality patterns in developing countries are usually characterized by high infant and childhood mortality and a specific distribution of causes of death.¹ However, any estimates of overall mortality are subject to error and bias because, especially for sub-Saharan Africa (SSA), mortality statistics covering the total population often lack reliability.² Therefore, statistics on mortality rates or number of dead persons are usually to be taken with caution.

Many countries in Africa are now undertaking steps towards providing a base for reliable information to support health policy and planning. An increasing number of demographic surveillance systems (DSS) are being established, mainly in rural areas (http://www.indepth-network.org/). These field sites continuously monitor vital events in populations living in well-defined geographical areas. In order to coordinate the activities of these sites, the INDEPTH network (International network of field sites with continuous demographic evaluation of populations and their health in developing countries) was established in 1998.³ The Nouna DSS, conducted in the research zone of the Centre de Recherche en Santé de Nouna (CRSN) in the Nouna Health District in north-western Burkina Faso, is a member of INDEPTH.⁴

Relatively few publications report adult mortality patterns in African populations. Kaufmann et al.⁵ noted the scarcity of information on causes of death among adults in SSA. Although, little is known about the level and causes of adult mortality (15–59 years) in SSA, Kitange et al.⁶ reported that survivors of childhood in Tanzania show high rates of mortality throughout adult life. However, data on seasonal patterns are rarely available with the exception of a few studies in West Africa, mainly on childhood mortality.⁷⁻¹⁰ In recent years, there has been an interest in the effects of season of birth on mortality. Previous findings from Gambia that young adults (mean age 25) born in the rainy season had higher mortality rates support the thesis that being born in the rainy season and shortly after is a risk.¹¹ But recently Moore et al.¹² and Simondon et al.¹³ found that season of birth is not associated with risk of early adult death in Bangladesh and Senegal, respectively, and Jaffar et al.¹⁴ found no marked difference in the death rate between children (<5 years) born in the rainy season and those born in the dry season in The Gambia.

Kynast-Wolf et al.¹⁵ analysed the Nouna DSS data for the period 1993–98. The population and age distribution and the mortality rates reflect a typical pattern in rural Africa: high childhood mortality. Mortality patterns of adults and older people were also described by Sankoh et al.¹⁶ Risk factors for childhood mortality in this population have been investigated by Becher et al.¹⁷

We present here, our findings on seasonal mortality patterns from a rural population in north-western Burkina Faso surveyed over the period 1993–2001. The objectives of this study were to analyse (i) the overall and age-specific effects of season on mortality and (ii) the effects of month of birth on overall and age-specific mortality.

Materials and methods

Study area and the DSS data

The CRSN is located in the Nouna Health District in north-western Burkina Faso, West Africa, a country with an estimated population of ∼12 million¹⁸, 80% of which live in rural areas.

The Nouna study area is a dry orchard savannah populated almost exclusively by subsistence farmers and cattle keepers. The region has a sahelian climate with a mean annual rainfall of 796 mm (range 483–1083) over the past five decades. The rainy season usually lasts from June to October, and the dry season from November to May. The period November to February is considered the cold dry season while the period March to May is the hot dry season. Malaria is holoendemic but highly seasonal and has been demonstrated to be the main cause of morbidity and mortality in young children in the study area.¹⁹,²⁰ Health services in the area comprise the District Hospital in Nouna town and four rural health centres in the surrounding villages.

The DSS of the CRSN has collected data from 39 villages surrounding Nouna town, since it was established in 1992.²¹ The DSS area has been augmented recently, but in this paper only data from these 39 villages are considered. The village populations range from 120 to about 2350 persons.

Vital events registration (VER) system started in 1992 after an initial census. The VERs were carried out by visits of trained interviewers to each village, who asked key informants whether any vital events had occurred since their previous visit. Two control censuses were carried out in 1994 and 1998. VER interviews are undertaken every 3 months: specific interviewers visit each household and ask about members previously registered or presently living in the household. The system is able to identify new vital events. Registered variables include births, deaths and migration with respective data. For some individuals, in particular in the early years of the DSS, only year of birth or year of death was available. These individuals were excluded or specially considered during the respective analyses (see Results section).

Statistical methods

Mortality by month of death

Poisson regression²² was used for absolute and relative mortality rate estimation and for modelling mortality by month on a continuous scale. For this analysis, we assumed a constant mortality rate over the observation period 1993–2001. This appears sufficiently true with an age-adjusted mortality of 14.6/1000 in 1993 and 14.0/1000 in 2001. Detailed data are available.²³ The analyses were done adjusted by sex and age group (<1 year, 1–4, 5–14, 15–59, and 60+ years). Some analyses were done separately by age group. The sequence of analysis was as follows: Poisson regression analyses were carried out using the SAS-procedure PROC GENMOD.²⁵

• Monthly mortality rates were estimated using the model

  \[\mathrm{ln}\left[D_{ij}\mathrm{MM}_{ij}\right]\ =\ {\beta}_{1}{\times}\mathrm{sex}\ +\ {\beta}_{2i}{\times}\mathrm{age}\ +\ {\beta}_{2j}{\times}\mathrm{month}\]

  (I)

  Monthly death rates per K individuals (K = 1000) were estimated by age group using D_ij as the number of deaths in age group i = 1,2, … ,5, and month j = 1, … ,12, and likewise MM_ij as the mid-month population in age group i and month j. Rates were multiplied by 12 to allow comparison to yearly rates. All rates were scaled by an age-specific factor c = (total deaths/deaths with known exact date of death) to account for missing information on day of death. This procedure is valid under the assumption that the month of death is missing at random.
• The mean mortality rates (adjusted for sex and age) were estimated by

  \[\mathrm{ln}\left[D_{i}\mathrm{MM}_{i}\right]\ =\ {\beta}_{0}\ +\ {\beta}_{1}{\times}\mathrm{sex}\ +\ {\beta}_{2i}{\times}\mathrm{age}\]

  (II)

  To assess the relative effect of each month on mortality we used the floating absolute risk method.²⁴ The relative effect is defined as the difference β_3j(I) − β_0,(II) of the monthly effect of model (I) and the overall effect of model (II).
• To get a continuous description of the seasonal effect we fitted a sine function of the form g(month) = sin[(month + k) × π/6], where k can take the value 1,…,6 (assuming a period of 12 months) resulting in the following model

  \[\mathrm{ln}\left[D_{ij}\mathrm{MM}_{ij}\right]\ =\ {\beta}_{1}{\times}\mathrm{sex}\ +\ {\beta}_{2i}{\times}\mathrm{age}\ +\ {\beta}_{3}{\times}g\left(\mathrm{month}\right)\]

  (III)

• To investigate whether the monthly mortality rates differ between age groups, we investigated the above mentioned age groups separately in a similar fashion as described in (i)–(iii).

Month of birth and mortality

To investigate the effect of month of birth as a predictor for mortality, we applied a method proposed by Le et al.²⁶, where the day of birth is transformed to a continuous variable on the unit circle, represented by an angle 𝛉, or a pair of coordinates (x, y). Thus, January 1 is represented by 𝛉 = 0°, February 1 by 𝛉 = 30°, etc. A mean date of birth of a group of individuals is obtained by calculating the arithmetic mean of all pairs (x_i, y_i) separately for x and y and projecting the resulting point to the unit circle with (̅x̅,̅y̅)̅ = (x̅,  y̅)

We compared mean birth dates of infant deaths to all other individuals who survived infancy in the observation period. The variation of these means was assessed with bootstrap methods,²⁷ using 1000 bootstrap samples from each group. Each sample provides a bootstrap mean for date of birth among deaths and among survivors. We projected the bootstrap means to the unit circle. This yields empirical bootstrap distribution functions for both groups, and 95% CIs using the uncorrected percentile method as the 2.5 and 97.5% quantiles.

Results

Basic demographic parameters

The population size rose steadily from 28 315 persons in 1993 to 35 549 persons in 2001 with an average yearly population increase of 2.9%.

A total number of 4112 deaths were recorded during the observation period January 1, 1993 to December 31, 2001, of which 11 with missing age and 3 with missing sex were excluded, so that the analyses include 4098 deaths. Crude mortality rates were 14.5/1000 (95% CI 14.1–15.0) for the overall population, 51.3/1000 (95% CI 47.5–55.4) for infants, and 29.9/1000 (95% CI 28.2–31) for children aged 1–4 years. The under five mortality rate is 34.9/1000 (95% CI 33.3–36.5).

Seasonal mortality based on date of death

For 872 (21.3%) of 4098 persons who died during the observation period the year of death, but not the month of death, was recorded.

The overall (adjusted according to the procedure given in Methods section) death rates for each month for the entire observation period are shown in Figure 1 with 95% CI. It shows a clear seasonal trend with high overall mortality rates in the months of November to May and low overall mortality rates in the months of June to October.

Figure 1

Open in new tabDownload slide

Crude death rates by month of death (adjusted for sex and age), rural Burkina Faso, 1993–2001

Figure 2

Open in new tabDownload slide

Rate ratios by month of death (adjusted for sex) using Poisson model for all age groups and overall, rural Burkina Faso, 1993–2001. No. of deaths per age groups: infants: n = 638, children: n = 1223, youths: n = 428, adults: n = 874, older people: n = 935

The results from modelling a sine function using Poisson regression support a highly significant (P < 0.001) seasonal variation of mortality in the study area in all age groups. Table 2 summarizes the functions describing the best fit of the RRs and the corresponding dates of minimum and maximum rates.

Figure 2 illustrates the RRs for each month using the floating absolute risk method and the sine function describing the RRs depending on month for the five age groups and overall.

Seasonal mortality by date of birth

Seasonal variation in mortality depending on month of birth by age groups

This analysis requires accurate dates of birth, so that 1345 (33% of 4098) individuals, for whom only the year of birth was known, had to be excluded. Figure 3 shows the number of monthly births, which displays a more or less regular peak towards the end of the year.

Figure 3

Open in new tabDownload slide

Percentage of births by month, rural Burkina Faso, 1993–2001

For infants, mortality rates showed a significant heterogeneity by month of birth, with highest rates for the months September, October and December. For all other age groups, the mortality pattern by month of birth appears randomly distributed (Table 3).

Seasonal variation of mortality by month of birth for infants

Detailed analyses were performed for 611 of all 638 infants who died during the observation period. 11 556 individuals who survived infancy in the same period were included as a comparison group.

The mean day of birth of deceased children was October 24 (95% CI: October 6–November 15), and it was November 25 (95% CI: November 12–December 9) for survivors. The difference between both means is statistically significant (P < 0.001). Figure 4 shows both means, surrounded by the ‘clouds’ of bootstrap samples used to compute CIs. Both means are clearly distinct from the null value, which indicates that the distribution of date of birth is not random over the year.

Figure 4

Open in new tabDownload slide

Mean date of birth for infants who died during infancy compared with those surviving their first year of life, rural Burkina Faso, 1993–2001

Discussion

This study demonstrates a strong association between mortality and season in a rural area of Burkina Faso. Based on the analysis of a large number of deaths over a period of 9 years, overall mortality was found to be consistently higher during the dry season compared with the rainy season. When looking more specifically at the dynamics of mortality by age group, the overall pattern with a mortality peak in the early dry season was observed in all age groups except in infants. In infants, highest mortality rates were around the end of the rainy season. Children aged 1–4 years showed an intermediate pattern with highest mortality rates also around the end of the rainy season but in addition also during the early dry season. Moreover, in infants we found a strong association between high mortality and being born during the time period September to February.

There are not much data available on the influence of season on mortality rates in SSA. In two studies from The Gambia, high infant and childhood mortality rates were clearly associated with the time around the end of the rainy season.^7,9 This is similar to our findings and supports other results from SSA on malaria, being the major causes of observed excess mortality during and around the end of the rainy season.⁷,⁹,¹⁰,^20,28 Malaria has thus been shown to be the greatest contributor to childhood morbidity and mortality in both The Gambia and in Burkina Faso.^7,9,¹⁹,²⁰,²⁹,³⁰ However, owing to differences in malaria endemicity the results from The Gambia, which is a mesoendemic area, are not directly comparable with the results from Burkina Faso, which is a holoendemic area.²⁰,²⁹ The observed high mortality in children aged 1–4 years during the cold dry season in our study may be explained by a higher prevalence of other infections such as pneumonia, measles and meningitis in these months. As vaccination coverage is high in Burkina Faso and the region and as measles is only rarely seen in these days in the study area, this disease is not likely to play a major role.³¹,³² In contrast, meningitis is well known to greatly contribute to childhood mortality in Burkina Faso, which belongs to the African meningitis belt.³³,³⁴ In these areas, regular meningitis outbreaks during the dry seasons are probably explained by a low absolute humidity and the dusty harmattan wind, which damages the local mucosal defences.³³ However, we believe that the mortality peak in 1- to 4- year old children in February, which is the end of the cold dry season, is mainly explained by pneumonia known to be the major cause of mortality in young children in developing countries.³⁵ An availability of cause-specific mortality data from existing verbal autopsy questionnaires would not clarify the situation very much as this method lacks sensitivity and specificity particularly regarding the diagnosis of malaria in endemic areas.³⁶,³⁷

The observed infant and childhood excess mortality from August until December in our study is thus mainly to be explained by the effects of malaria around the end of the rainy season.¹⁹,³⁸ This may also explain the observed effect of month of birth on mortality in infants, as children living in areas with high malaria transmission intensity are most susceptible for severe malaria during the second half of infancy.³⁹ Our findings are consistent with the results of Becher et al.¹⁷ who found a slight association between being born in the rainy season and mortality, but only with mortality during the first year of live.

One assumption of Le et al.'s²⁶ method is that the distribution of dates has only one yearly peak. If, for example, a clustering would occur in January and July, this would yield a ‘mean date’ in the origin of the unit circle, as for an uniform distribution. This is not the case for our data, so our use of the Le et al.²⁶ method appears appropriate.

The observed excess mortality in the early dry season in the older age groups is not easy to explain. Possible causes include meningitis in older children and young adults as well as acute respiratory infections in adults. Pneumonia is known to be often fatal in adults of rural Burkina Faso, as health centres are often not accessible and antibiotics are usually not available in villages.³⁰ The possible impact of other risk factors associated with high adult mobility during this time period (e.g. accidents) also needs to be considered, but specific data are currently not available. As very few data are available on cause-specific mortality in SSA adults, only further careful analyses of verbal autopsy data will shed light on this.³⁶,⁴⁰

Previous findings on young adults in SSA born in the rainy season having higher mortality rates had led to the hypothesis that season of birth is a significant risk factor for mortality in later years.¹¹ However, other data have not supported this hypothesis¹²⁻¹⁴ and neither did our findings.

We cannot totally exclude a reporting bias, as interviews during the rainy season are more difficult than during the dry season for many reasons such as weather conditions and availability of household members owing to agricultural work. Moreover, we have not yet included data on cause-specific mortality into our analysis, as there has been some delay in the processing of the verbal autopsy data from the Nouna DSS site.

In conclusion, in rural Burkina Faso we have shown a largely unexplained excess mortality in older children and in adults during the early dry season, and an excess mortality in young children at around the end of the rainy season, which is probably explained by malaria deaths. Moreover, being born around the end of the rainy season has been shown to be a risk factor for mortality particularly during infancy. Finally, the recently observed increase in childhood mortality could be caused by the increasing resistance against the malaria first-line drug chloroquine in the area.⁴¹,⁴² As a consequence, a careful analysis of cause-specific mortality in older children and in adults is needed and malaria control measures in areas of high transmission intensity should focus more on very young children.

Open in new tabDownload slide

Village during the rainy season in Burkina Faso 2001. Photograph by Olaf Müller.

This work was supported by the collaborative research grant ‘SFB 544’ of the German Research Foundation (DFG). The authors would like to thank the field staff in Nouna/Burkina Faso and Gabriele Stieglbauer for effective data management.

References