Menu
Menu
Menu
Menu
Background: Seasonal patterns of mortality have been identified in Sub-Saharan Africa but their changes over time are not well documented. Objective: Based on death notification data from Antananarivo, the capital city of Madagascar, this study assesses seasonal patterns of all-cause and cause-specific mortality by age groups and evaluates how these patterns changed over the period 1976–2015. Methods: Monthly numbers of deaths by cause were obtained from death registers maintained by the Municipal Hygiene Office in charge of verifying deaths before the issuance of burial permits. Generalized Additive Mixed regression models (GAMM) were used to test for seasonality in mortality and its changes over the last four decades, controlling for long-term trends in mortality. Results: Among children, risks of dying were the highest during the hot and rainy season, but seasonality in child mortality has significantly declined since the mid-1970s, as a result of declines in the burden of infectious diseases and nutritional deficiencies. In adults aged 60 and above, all-cause mortality rates are the highest in the dry and cold season, due to peaks in cardiovascular diseases, with little change over time. Overall, changes in the seasonality of all-cause mortality have been driven by shifts in the hierarchy of causes of death, while changes in the seasonality within broad categories of causes of death have been modest. Conclusion: Shifts in disease patterns brought about by the epidemiological transition, rather than changes in seasonal variation in cause-specific mortality, are the main drivers of trends in the seasonality of all-cause mortality.
Antananarivo is located in the central highlands of Madagascar and culminates at an altitude of 1280 m. It has a subtropical climate with a cold and dry season from May to October (with average minimal temperature around 11°C) and a hot and rainy season from November to April (with average maximal temperature around 27°C) (Figure 1). December, January and February are the 3 months with the highest rainfall and they also correspond to the lean season. Rice is by far most consumed staple food in Madagascar; it furnishes more than 50% of the average calorie ration of the country. The cropping calendar varies greatly according to rice species and climate conditions of the regions, but about 70% of the rice produced in the country is harvested between April and June. Because of its predominance in the agriculture and diet, the seasonal production of rice drives seasonal movements in food prices and overall food consumption [25]. Monthly means of daily maximum/minimum temperatures and rainfall in Ivato station (over the period 1976–2015). Sc: DGHCN/daily Since the 1950s, climate change has led to a rise in temperatures in Madagascar, particularly in the dry season [26]. The rainy season is also being delayed [27]. The frequency of extreme events such as cyclones, floods and droughts is increasing. All these changes are likely to modify seasonal patterns of mortality. Antananarivo has undergone a major epidemiological transition in the last decades, only interrupted by a mortality crisis in the mid-1980s caused by the combination of the resurgence of malaria and food shortages [28,29]. Life expectancy first declined from 56 in 1976 to 47 in 1986, before increasing steadily to reach 64 years in 2015 (Figure S1). This progress was mostly driven by a decline in under-five mortality, which fell to 34 deaths per thousand live births in 2015, from 116‰ in 1976 (a 70% decline). By comparison, there has been virtually no improvement in survival changes in adults: the risk of dying between ages 15 and 60 was 280 deaths per thousand in 1976, peaked at 473‰ in 1986, at the height of the crisis, and declined again to 283‰ in 2015. The contrast between child and adult mortality suggests that survival gains were for the most part achieved through public health interventions targeting diarrheal and vaccine-preventable diseases. Demographic and Health Surveys show increases in the percentage of children who received all 8 basic vaccinations (BCG, DPT1-3, Polio 1–3 and measles), from 43% in 1992 to 62% in 2008–2009 in the country [30]. Chronic malnutrition has been slightly reduced; the prevalence of stunting in children under age 5 declined from 60% in 1992 to 50% in 2008–2009 (47% in the capital). In contrast, there has been little progress in skilled attendance at birth, access to improved water sources and sanitation. According to the 2008–2009 DHS, only 14% of the population of the capital lived in households with improved, non-shared toilet facilities [30]. Seventy percent of the population had access to water from public taps or standpipes. Overall, the country’s health situation remains exceptionally fragile, as illustrated by recent outbreaks of plague (in 2014 and 2017) and measles (in 2019). Madagascar has one of the lowest levels of per capita health spending in the world, and more than three quarters of the population live in extreme poverty [31]. Still, the epidemiological transition is well underway. As a result of population ageing and changes in risk factors, the distribution of causes of death has changed considerably. In the period 1976–1980, 54% of deaths registered in Antananarivo were due to communicable, maternal, neonatal and nutritional conditions, but this proportion had dropped to 21% in 2011–2015 (Figures S2 and S3). This study is based on data on 249 421 residents of Antananarivo-city who died between 1976 and 2015. This corresponds to the central administrative sector of Antananarivo-Renivohitra, with a population estimated at 1.28 million inhabitants in 2018 (5% of the national population) [32]. It was not possible to reconstitute an individual database for the period before 1976 because the registers are lost. All deaths that occur within this area should be reported to the BMH. About 60% of deaths occur at home. Relatives of the deceased contact the BMH and a physician is sent to the house of the deceased to establish a cause, based on the information provided by the family on the symptoms and circumstances preceding the death, as well as available medical documents. This is equivalent to medical certification and is different from verbal autopsy methods that have been primarily developed to identify the probable cause of death in the absence of a physician. For facility deaths, the reports are filled in by medical personnel and transmitted to the BMH by relatives. The completeness of reporting of deaths among adults (with a physician-certified cause of death) was higher than 90% in the intercensal period 1975–1993. Estimates of completeness will be updated when the detailed population counts from the 2018 census become available. In recent years, under-five mortality rates inferred from the BMH are aligned with trends derived from Demographic and Health Surveys [24]. Cause-specific mortality fractions derived from the registers are also consistent with epidemiological models [24]. The team in charge of certifying deaths currently consists of eight physicians, all of whom have been trained on the application of the International Classification of Diseases (ICD). For home deaths, an ICD-10 code is assigned based on the cause noted in plain text in the information sheet used in post-mortem interviews. For health facility deaths, the ICD code is based on the cause mentioned in the death certificate, which can come in various formats (directly with an ICD-10 code, a code from a previous revision of the ICD or reported in plain text). In the past, not all deaths in the registers had an ICD code but one physician with special training ensured that all deaths were coded in ICD-9 when registers were digitalized. To group causes of deaths in broad categories, we used the hierarchical cause-of-death list established by the Global Burden of Disease (GBD) Study 2016 [33]. This list has four levels. The first level distinguishes between (a) communicable, maternal, neonatal, and nutritional diseases, (b) non-communicable diseases, and (c) injuries. The second level refers to 21 broad categories, such as, for example, diarrhea, lower respiratory and other common infectious diseases among one group of causes. The third and fourth levels refer to more detailed causes of death, such as, for example, intestinal infectious diseases (level 3) and typhoid fever (level 4). For this study, we considered only the second level of the GBD hierarchy. All ICD 9 codes were mapped to a GBD cause of death. Some ICD codes were considered ‘garbage codes’. This refers not only to causes identified as ‘undefined’ in the specific ICD chapters, but also deaths attributed to causes which should not be considered as initial causes, such as dehydration or septicemia. We used a simplified redistribution algorithm to map these codes to acceptable GBD causes. For example, ill-defined cardiovascular diseases were redistributed to ischemic heart disease and other cardiovascular and circulatory diseases. The redistribution is summarized in the Appendix and described in detail elsewhere [24]. We conducted a sensitivity analysis using ICD-9 chapters before any redistribution of garbage codes, and results were similar to those obtained with the GBD cause categories (Appendix). Mortality rate ratios associated with months are obtained from a Generalized Additive Mixed Model using a Negative Binomial distribution, a generalization of the Poisson distribution that accounts for overdispersion. The model includes a penalized regression spline [34] to model long-term trends in mortality and avoid over-fitting, month as a random effect to model seasonality and year as a random slope to assess any change in seasonality, after stratifying by age groups. Age groups consisted of infants (less than 1 year old), young children (1 to 4 years old), older children and young adults (5- to 59-year-olds) and people aged 60 or above. We used the Bayesian (BIC) Information Criteria to choose the best model among a model with penalized splines for the trend only (model 1); a model with random intercepts for months (model 2); and a model with random slopes and random intercepts (model 3). Retaining models that minimized the BIC with a difference of more than 10 [35] allowed us to characterize first, if seasonality was present (model 2) and second if it was changing over time (model 3). Because of the unequal number of days in a month, we multiplied each monthly death count by 30.4 and divided by the number of days in each month. The full model (model 3) can be expressed as follows: where yt are monthly counts of deaths, s() is a penalized spline, j ∈ {Jan., Feb., …, Dec.}, Yeari∈−20,−19,…0,1,…19,20 where 0 represents 1996, the middle of the analysis period and t ∈ {1, 2, …, 480} is a continuous variable reflecting the count of month. We evaluated goodness-of-fit by visual inspection of the deviance residuals, considering that a good fit was obtained if 95% of the deviance residuals were between −2 and 2 standard deviations and no large outliers were present. All analyses were conducted using R statistical software.
