Returning home to die or leaving home to look for care? Location of - - PDF document

1

Returning home to die or leaving home to look for care? Location of death of urban and rural residents in West Africa. Bruno Lankoandé(1), Géraldine Duthé (2), Gilles Pison(3)(2), Abdramane Soura(4)

(1)

Université catholique de Louvain, Louvain-la-Neuve, Belgique

(2)

Institut national d’études démographiques, Paris, France

(3)

Museum national d’histoire naturelle, Paris, France

(4)

Institut supérieur des sciences de la population, Université de Ouagadougou, Ouagadougou, Burkina Faso

Short abstract In Sub-Saharan Africa, place of death remains an indicator of access to healthcare. But, the focus on place of death may hide how population develop coping strategies to meet some needs at the end of their life. In this paper, we take advantage of data collected in Health Demographic Surveillance System (HDSS) sites located in urban and rural areas of West Africa to document short-term mobility of adults before their death. In rural areas, a non- negligible group of adults leaves their community to seek formal care in surrounding towns and die there. The magnitude of mobility depends on the supply of health facilities. It tends to be higher in rural remote areas and age is the strongest determinant. Younger adults are more mobile before their death whatever the types of diseases they are suffering from. Urban residents also move before their death. Particularly, rural-migrants and those suffering from non-communicable diseases go back to their community to benefit from supportive care and

• die. Of public health importance, the flows of deaths between urban and rural residents can

affect the estimation of urban/rural mortality differentials.

2

• 1. Background

According to Gu and her colleagues (2007), place of death goes through three evolutionary stages in societies. During the first stage, most people die at home because of poor access to healthcare resources. In a second stage, deaths in hospitals become more common due to medical advances. In the last stage, the emphasis is put on the quality of care at the end of life. The availability of home based care increases to meet people preferences to die at home. During the different stages, cultural norms play a role in determining one place of death. In developed countries, the debate in public health is particularly centered on place of death as an indicator of quality of end of life care with most people preferring to die at home compared to hospitals or care homes (Gomez et al., 2012). In some countries such as Japan, Belgium, a great share of deaths continue to occur at hospitals while in USA and Canada, death at home

• r in nursing homes have replaced death in hospitals (Houttekier et al., 2011; Gomes et al.,

2008; Anezaki, 2008; Flory et al., 2004; Wilson et al., 2002).This variation in place of death between developed countries seems to reflect cultural preferences towards death and most importantly differences in availability of out-of-hospitals care resources for dying persons. Referring to the transition model of Gu and al. (2007), we assume that less developed countries are still in stage 1 or in transition from stage 1 to 2. In sub-Saharan Africa (SSA) particularly, place of death reflects more cultural preferences and access to healthcare than quality of end of life care. Population still face serious health issues including the rise of non- communicable diseases in a context of persisting infectious diseases and poverty (Remais et al., 2012). Thus, prevention and cure of diseases remain the priority. Terminal care is quasi inexistent except in countries which were very affected by the HIV epidemic such as Zimbabwe, Uganda, South Africa, Kenya (Gysels et al., 2011). It is believed that most deaths

• ccur at home even though the lack of complete vital registration systems hampers extensive

research on this issue. A few population based studies have examined place of death in recent

• years. Among adults, proportion of death occurring in hospitals vary from 28% at Addis

Ababa to 50% in Zambia (Anteneh et al., 2013; Chisumpa et al., 2017). This figure was estimated at 50% for all ages in a semi-urban setting of Burkina Faso (Bado et al; 2016). In Botswana where the completeness of the vital registration system is acceptable, 60% of adults death occur in hospitals (Lazenby et al., 2010). It is recognized that socio-demographic, clinical, and ecological factors have an impact on place of death (Chisumpa et al., 2017; Houttekier et al., 2009; Cardenas-Turanzas et al., 2011). According to the Behavioral Model of Health Services Utilization developed by Phillips et al. (1998), socio-demographic factors such as age, gender, marital status and education impacts on the ability to look for care. For instance in the African context, aging may be considered as a cause of death and this prevent elderly (or the relatives) to seek care (Duthé et al., 2010). As found in Addis Abeba and Botswana, elderly are more likely to die at home compared to young adults (Anteneh et al., 2013, Lazenby et al.,2010). The clinical component in the analysis of place of death refers generally to the cause of death. In the model developed by Phillips et al. (1998), it contribute to define the need to look for care. The hypothesis is that some diseases such as cancer, renal failure, cardiovascular diseases are difficult to manage at home and will probably lead to hospitalization. Contrasting evidence on the association between cause of death and place of death appears from the limited research in sub-Saharan Africa. While cardio-vascular diseases tend to occur in health facility compared to HIV/AIDS in Addis Ababa, no association between place of death and cause of death was found in Zambia except with injuries and HIV/AIDS. Finally, place of residence and living standard are commonly used as a proxy of enabling characteristics to look for care. Place of

3

residence determines mostly the availability of healthcare resources. Home deaths is more common in rural areas than in urban settings of SSA (Chisumpa et al., 2017; Bado et al., 2016; Lazenby et al., 2010). This underscores rural-urban inequality in access to healthcare resources in favor of urban areas. Indeed, despite the various investments made in the provision of health facilities in rural areas

• f SSA since the Alma Ata conference in 1978, the gap in access to health care is still very

large between urban and rural areas (Nwakeze et al., 2011; Ujoh et al., 2014). Health facilities as well as skilled health workers (doctors, nurses, midwives) are concentrated in cities (Lemière et al., 2010). Furthermore, they offer a variety of services including the care for some non-communicable diseases such as diabetes and hypertension that may not be available in rural remote areas. For example, in Senegal, it is estimated that the ratio of a doctor per 1000 people is around 0.07 in rural areas and 2.33 in urban areas (Lemière et al., 2010). In these conditions, rural residents in need of care will have to travel to urban areas to seek for treatments while urban residents don’t necessary have to move to benefit from health care. However, in SSA, some chronic diseases can’t be really efficiently treated even in the largest hospitals as treatments are too costly for both individuals and the state. Patients suffering from chronic diseases enter much faster and at a younger age in the end of life process than in most developed countries. Among them, many will have the desire to die in their home community to benefit from supportive care. In this situation, rural natives that live in urban areas may decide to go back to the village for dying. For example, in South Africa, the return of dying rural-urban migrants from HIV/AIDS in their communities highlights that their social networks is stronger there (Clark et al., 2007; Collinson et al., 2014). In addition to the need

• f supportive care, death far from home could be considered culturally as a "Bad" death as

documented in rural Ghana (Geest, 2004). Most people would like to do die in their community surrounded by their relatives. For example, in South Africa, some terminally-ill Zimbabwean migrants tend to return home before their death (Nunez et al., 2012). In sum, this literature suggests that in African countries, there are potentially mobility flows

• f sick persons at the end of their life. On one hand, some travel from rural to urban areas in
• rder to look for care and may die there. On the other hand, some urban patients in need of

supportive care move to rural areas to die at home. The objective of this paper is to take advantage of data collected in Health and Demographic Surveillance Systems (HDSS) to document the magnitude of short-term mobility before death among adult residents in rural and urban contexts of West Africa. Underlying motivations and associated factors of these mobility flows are also analyzed. We hypothesis that mobility before death of rural residents are linked to health seeking behavior while mobility of urban

• nes are due to the need of supportive care at the end of life. Particularly, non native of urban

areas will be likely to return to their home community before death. Covariates shown to impact place of death, are expected to influence short term mobility before death but with more selectivity since mobility is taken into account. The existing literature is extended by integrating mobility in the analysis of place of death in

• SSA. The simple dichotomy of place of death, health facility versus home death, found in the

literature hide how population develop strategies to meet some needs before death. The study bears the capacity to inform on population needs in case of diseases and their coping

• strategies. It also has implications in the measurements of health inequalities precisely

mortality indicators. In large operations of data collection, a death of an adult may be wrongly classified as an urban or rural death.

4

• 2. Data and methods

2.1 HDSS sites In the context of sub-Saharan Africa, data collected in censuses and common household-based surveys such as Demographic and Health Surveys (DHS) do not allow a detailed analysis of rural-urban migration in relation to health in adults (Bocquier et al., 2016). First, these surveys put little attention on migration flows between urban and rural areas. Second, the focus is on child health rather than on adult health. Finally, the few data available on migration refers to long term mobility leading to change in place of residence. So, the capture

• f short term mobility and its relation to health is even more challenging.

Thus, this analysis is based on data collected in three Health and Demographic Surveillance Systems (HDSS) located in West Africa. These HDSSs will serve as case studies to explore mobility before death in different contexts of West Africa. The first one is the Ouagadougou HDSS in Burkina Faso which is the only fully urban HDSS of West Africa. Second, the Kaya HDSS located in a semi-rural area will help to compare patterns of mobility before death in a rural context within Burkina Faso even though it is covered by a regional hospital. Finally, the Mlomp HDSS located in rural Senegal will offer a contrasting context compared to Kaya because the site is covered by a primary healthcare only. Furthermore, it has the particularity to collect over the years, the information whether any mobility is related to health reasons or not, which allows a better analysis of mobility before death in connection to care seeking. The three HDSS sites are members of the International Network for Demographic Evaluation

• f Populations and Their Health (INDEPTH) and share similarities in terms of methodology.

Following an initial census in the area under surveillance, fieldworkers conducted regular household update rounds, registering vital events (births and deaths, migrations and marriages). In case of death, a verbal autopsy (VA) questionnaire is filled with the next of kin to determine the circumstances that led to the death, including history of the illness and the specific symptoms that preceded death. VA data are then interpreted by a team of physicians

• r by the InterVA-4 software to determine the probable causes of death (Fottrell et al., 2010;

Byass et al., 2012). The Ouagadougou HDSS has been established in late 2008 in five neighborhoods at the northern periphery of the capital city of Burkina Faso (Rossier et al., 2012). Two of these neighborhoods (Kilwin and Tanghin) are formal neighborhoods with full access to public services, while the other three (Nonghin, Polesgo and Nioko 2) are unplanned peri-urban settlements (like slums) without access to such services (Rossier et al, 2012). People living in the Ouagadougou HDSS are mostly from the Mossi ethnic group (90%), which is actually the majority ethnic group in the country. More than half of active adults work in the commerce and the construction sector (Rossier et al., 2011). The population under surveillance by the Ouagadougou HDSS totaled about 90,000 residents in 2015 and periodic household update rounds are conducted with an average periodicity of 10 months. Unlike the Mlomp and Kaya HDSSs, VAs data are entered into InterVA-4 software to determine the probable causes of

• death. The health care provision in the city of Ouagadougou is better than in any other

location in the country with a private sector representing two thirds of the offer. In addition, large teaching hospitals that are the highest reference levels of the health system are located in the city (Ministry of health, 2010). The Kaya HDSS was established in late 2007 in the North Central region of Burkina Faso, 100 kilometres from the capital city, Ouagadougou. The site covers the town of Kaya and 18 villages (Kouanda et al., 2013). The follow up population was estimated at 70,000 inhabitants

5

in 2015. This population lives in semi-urban (70%) and rural (30%) areas. The site is easily accessible from Ouagadougou and is covered by seven health facilities and one regional

• hospital. Residents are mostly from the Mossi ethnic group and are of Muslim faith. Only half
• f the population went to school and the main economic activities are rudimentary agriculture

and livestock breeding. In recent years, gold mining in the neighbouring villages of the HDSS is becoming also important. Even though the site is semi-urban, health indicators and fertility levels are typical of a rural area of Burkina Faso. Life expectancy was estimated at 54 years in 2013 and the total fertility rate was estimated at 7 children per woman during the same year. Visits in households are undertaken every 6 months. In case of death, causes of deaths are certified by a team of physicians based on information available in VA questionnaires. During the period considered in this study (2012-2015), data on causes of death are not available but summary information on the circumstances of the death was collected. The Mlomp HDSS was set up in 1985 at the Southwest of Senegal in the administrative region of Ziguinchor, nearly at 500 kilometres from Dakar, the capital city (Pison et al., 2002). The site covers eleven villages. The population under surveillance belongs to the Diola ethnic group and is mostly animist or catholic. Rice cultivation is the main activity in the area but during the dry season, seasonal migration flows are intense among adults. Young women are prone to be employed as domestic servant in Dakar or in Banjul (the capital cities of Senegal and Gambia which is close) before they get married. Regarding men, they are engaged in harvesting of wine palm and fishing in other regions. As a rural setting of Senegal, education level is relatively high in Mlomp with, in 2000, 55 % of women aged 15-49 years who have attended school compared to 14% on average in the rural area of Senegal. Health indicators are also encouraging thanks to a very dynamic private health center opened in 1961 by French catholic nurses. However, to see a physician, patients need to be referred to the local hospital of Oussouye, 10 kilometres away from Mlomp. Advanced medical care including surgery is only available in the more important regional hospital of Ziguinchor, 50 kilometres away from Mlomp. The follow-up population was estimated at 9,000 inhabitants in 2015 and update of vital events is carried out on annual basis. In a case of death, a team of physicians is in charge of interpreting the filled questionnaires of VAs to assign a probable cause of death. 2.2 Data Adult deaths that occurred after age 15 among residents are analysed in this study. The analysis covers the time period 2012-2015 in Ouagadougou and Kaya. As the population of Mlomp is much smaller, the analysis has been extended to the deaths that occurred during the period 2000-2015. In Ouagadougou and Kaya, residents are excluded from the follow-up after 6 months of absence in the area. In Mlomp, due to a high volume of circular migrations, residents are excluded from the follow-up only after two successive years of absence. In order to approximate the same residency criteria in Mlomp as in the two other sites, the date of the last presence of the residents who died was compared to the date of their death. If missing, it has been estimated on the basis of the information on the presence during the two last successive dry and rainy seasons. The deaths of individuals who were reported as absent since seven months or more from the area were excluded from the analysis in the Mlomp HDSS. As the study aims to highlight mobility before death for health reasons, injuries and violent deaths are excluded from the analysis (8.4% of deaths in Ouagadougou, 4.5% in Kaya and 9.5% in Mlomp). In the site of Kaya where causes of death were not available for the period considered, external deaths were discarded on the basis of causes of deaths reported by the deceased relatives. To sum up, the analysis includes 536 deaths in Ouagadougou, 695 deaths

6

in Kaya, and 708 deaths in Mlomp. In Mlomp, out of 809 deaths, 101 were discarded because the persons have been away from the site more than 6 months before their death. 2.3 Methods To capture mobility before death, the dependent variable is defined as the location of death, if it has occurred within the area or outside. In the Ouagadougou HDSS, to ignore mobility within the city, location of death is defined according to the entire city. This means that individuals are classified into two categories, those dead in Ouagadougou and those dead

• elsewhere. In the Kaya and Mlomp HDSS, location of death is captured throughout a variable

that indicates if the death has occurred within the HDSS boundaries or elsewhere. In each site, in a first step, in order to investigate motivations behind the ultimate mobility, the net effect of the location of death (within the site/ elsewhere) on the place of death (in health facilities compared to home) is assessed using a multinomial logistic regression. However, the main analysis consists in testing the effects of socio-demographic (sex, age group at death, marital status, education), and clinical (group of causes of death) factors on the location of death using a logistic regression. For the Ouagadougou HDSS, place of birth is also included in the model to examine the effects of migration status on location of death in an urban area.

7

• 3. Results

3.1 Mobility and place of death Once deaths from external causes are excluded, the proportion of adults who died outside their place of residence is rather low and quite similar in Ouagadougou (5.9%) and Kaya (5.3%). In the latter site, available information shows that among deaths that occurred outside, 73.0% took place outside the district where the site is located. So these deaths occurred quite far from the HDSS. The situation is quite different in Mlomp where 20.6% of the deaths

• ccurred outside of the area.

Regarding the place of death, it appears that different pictures emerge according to the area of residence and the location of death. In the urban setting of Ouagadougou, 45.7% of deaths

• ccurring in the city took place in health facilities compared to 19.4% for deaths that occurred
• utside the city. The reversed situation is observed in the site of Kaya with 34.2% of deaths

registered in health facilities among deaths that have occurred in the area compared to 70.3% for deaths that occurred outside of the area. In Mlomp, the difference between the two groups is much larger: 82.0% of the adults who died outside Mlomp died in health facilities against

• nly 7.1% among the others.

Table 1. Proportion of deaths over 15 that occurred in health facilities according to the location of the death in the three HDSS sites

Urban population Semi-rural population Rural population Ouagadougou HDSS 2012-2015 Kaya HDSS 2012-2015 Mlomp HDSS 2000-2015 Location of the death In the area 45.7% 34.2% 6.4% Outside the area 19.4% 70.3% 82.0% Proportion of deaths occurring in health facilities 44.1% 36.1% 20.6%

Looking at the probability of dying in a health facility compared to home (controlling for sex, group age at death, education, marital status, birthplace for Ouagadougou and group of causes

• f death for Ouagadougou and Mlomp), we can confirm the strong association between the

location of death and the fact that death occurs in a health facility. Moreover, differences are important regarding the site (Table 2). Interestingly for our purpose, in semi-rural or rural settings, adults who died outside the village have a significantly greater chance to die in a health facility compared to those who died at home, all things being equal. These results suggest that such mobility is clearly motivated by health care seeking. This assumption is reinforced by the fact that in Mlomp, 91.1% of those who died outside the site have left the village before their death for health reasons. A contrasting picture appears in urban areas: those who left the city before their death have on the contrary a higher chance to die at home than in a health facility, all things being equal. This highlights the fact that adults of Ouagadougou who moved before their death were probably not looking for medical care. Such mobility is more probably connected with the need to find social support from the native community when the hope of recovery is limited, and/or to go back to the native community for dying.

8

Table 2. Relative risk ratios of the multinomial logistic regression1 on the probability of an adult dying in a health facility compared to home.

Ouagadougou HDSS Kaya HDSS Mlomp HDSS Died outside of the area No 1 1 1 Yes 0.3** 4.3*** 75.6***

(a) Controlled by the other covariates: sex, age group at death, education, marital status, birthplace for

Ouagadougou and cause of death for Ouagadougou and Mlomp.

*** p<0.01; ** p<0.05; * p<0.1

3.2 Mobility and socio-economic characteristics In the urban population, as already mentioned, the migration status is important to take into

• account. The share of non-natives among the inhabitants followed up by the HDSS is indeed

large (72.5%) and most of them are from rural areas (71.2%). As expected, the proportion of adults who died outside the city is twice higher among non-natives (6.9%) than among natives (3.0%) of the city. In the semi-rural and rural sites, a higher proportion of deaths occurred

• utside their location of residence among males than among females. This is not the case in

Ouagadougou where 7.6% of women died outside the town while this proportion is only 4.5% in men. In the three sites, the older the person, the lower the proportion of deaths that

• ccurred outside the site. Compared to uneducated adults, those who have at least one year of

schooling are in proportion much more likely to die outside of the site in rural or semi-rural areas (17.1% vs 1.9% in Kaya; 37.1% vs 19.4% in Mlomp). But in Ouagadougou, proportions

• f deaths occurring outside the city do not vary much by level of education. Regarding the

marital status, differences between single and married adults are more noticeable in Kaya and Mlomp, with a higher share of death occurring outside among single than married adults. Finally, in the sites of Mlomp and Ouagadougou where the information on causes of death is available, the proportion of deaths occurring outside each site is higher when people died from a non-communicable diseases compared to communicable diseases.

1 As expected, a Hausman test shows that the Independence of Irrelevant Alternatives (IIA) assumption is not

• violated. Alternatives considered here are not similar.

9

Table 3. Proportion of deaths over age 15 that occurred outside of the area according to different characteristics in the three HDSS sites

Ouagadougou HDSS 2012-2015 Kaya HDSS 2012- 2015 Mlomp HDSS 2000- 2015 Total 5.9% 5.3% 18.80% Birthplace Native from the area 3.0% Native from elsewhere 6.9% Sex Male 4.5% 6.2% 21.2% Female 7.6% 4.4% 15.5% Age group at death 15-39 8.9% 17.2% 45.5% 40-59 7.2% 5.3% 33.3% 60-79 3.6% 2.0% 18.4% 80+ 4.3% 1.9% 6.3% Education None 5.8% 1.9% 15.5% Primary + 5.8% 17.1% 37.1% Unknown 12.5% 10.3% 19.4% Marital status Married 5.5% 2.5% 24.2% Single 7.6% 17.9% 28.1% Divorced/Widowed 6.2% 2.2% 11.6% Unknown 11.1% 16.7% Group causes of death2 Communicable diseases 3.2% 16.7% Non-communicable diseases 7.2% 20.2% Ill-defined 8.7% 16.9% Number of deaths 536 695 708

The logistic regression models presented in Table 4 help assess the net effects of the different independent variables on the location of death. For the particular case of Mlomp where mobility motivations are known, results taking into account reasons of the mobility are presented in appendix (Table A1). Whether “death outside the village due to health reasons”,

• r simply “location of death” is used as the outcome variable, results are quite similar. These

results combined with the fact that in Kaya and Mlomp, adults who move before their death tend to die in health facilities confirm that in these settings, mobility before death can be interpreted as a strategy to have access to medical care. In the urban HDSS of Ouagadougou, non-natives of the city are almost three times likely to die outside the city, all things being equal. Adult deaths due to a non-communicable diseases are also three times more likely to happen outside the city. The same goes for deaths that are

2 Previous research show that causes of death generated from the inter-VA software and diagnosed by physicians

are consistent (Bauni et al., 2011). Furthermore, in the analysis the focus is on group causes of death so that we can interpret in the same ways result found in Ouagadougou and Mlomp.

10

ill-defined for which we can suppose there is an association between being dead outside of the area and having less information through VAs. However, in the rural site of Mlomp, there is no difference related to broad group of causes of death. The age group appears to have the strongest association with the probability of dying elsewhere than within the community. Whatever the site, urban, semi-rural or rural, the elderly are much less mobile before their death. In Kaya and Mlomp, there is a marked difference between people who were less than 40 years old at their death and the older ones. In the case of Ouagadougou, age is significant only among those aged 60 and more. In the three HDSS sites, the sex of the person is not significant, though females tend to die more than males outside in Ouagadougou. The marital status also appears to have a low impact on the probability of dying outside of the area. In Kaya and Ouagadougou, adults who are not engaged in a relationship (single and divorced or widowed) tend to die elsewhere even though the results are not significant. By contrast, in Mlomp, single adults have a significantly lower probability of dying outside the village. Education does not affect location of death in Ouagadougou as well as in Mlomp. By contrast, a strong effect is observed in Kaya: adults who attended school are five times more likely to die outside of their community.

11

Table 4: Odd ratios of the logistic regression on the probability of dying elsewhere than within the community in Ouagadougou, Kaya and Mlomp

Characterisics Ouagadougou HDSS Kaya HDSS Mlomp HDSS3 Birthplace In Ouagadougou 1 Elsewhere 2.9* Sex Male 1 1 1 Female 2.0 1.0 0.9 Age group at death 15-39 1 1 1 40-59 0.7 0.4* 0.3** 60-79 0.2** 0.2*** 0.1*** 80+ 0.2* 0.2** 0.0*** Education None 1 1 1 Primary + 0.7 5.3*** 1.2 Unknown 1.7 0.0 1.0 Marital status Married 1 1 1 Single 1.2 1.6 0.3** Divorced/Widowed 1.1 2.3 0.6* Unknown(a)

• 0.5

Group causes of death Communicable diseases 1 1 Non-communicable diseases 3.4** 1.2 ill-defined 3.6* 1.1 Number of observations 526 695 708 (a) In Kaya, people with unknown education and marital status are similar

*** p<0.01; ** p<0.05; * p<0.1

3 In the site of Mlomp, the inclusion of a dummy variable in the model to control the effect of the period 2012-

2015 do not change the results. The period do not have a significant effect at the threshold of 10%.

12

• 4. Discussion

In this paper, we document the magnitude of short-term mobility before death in rural and urban contexts of West Africa. The underlying motivations and associated factors are also

• examined. The analysis is based on recent data on deceased adults (15 years and more)

collected in three Health and Demographic Surveillance Systems (HDSS) sites located in Burkina Faso and Senegal (Ouagadougou, Kaya, Mlomp). To the best of our knowledge, it is the first time that a such an quantitative analysis is conducted in SSA. It extends the existing research on place of death in SSA by showing how population move out of their usual place

• f residence to have access to medical or supportive care at the end of their life.

First of all, results indicate that location of death is a good indicator for identifying mobility to have access to medical care in rural and semi-rural areas. Adults who moved outside their place of residence shortly before their death were clearly looking for medical care. Nonetheless, the magnitude of mobility depends on the supply of health facilities in each site. In Kaya, the offer of health facilities is acceptable and includes a regional hospital, which is a referral center for surrounding villages. This explains why the proportion of adults dying

• utside the area in rather low. It is likely that those who moved before their death were

seriously ill to be treated in Kaya and were looking for specialized care in a city such as Ouagadougou which is not far. In contrast, in the site of Mlomp, the supply of health services consists only in a primary health center and a maternity clinic. Yet, since the focus is put on child and maternal health, it is not surprising that adults look for care outside the village in case of severe illness. The nearest hospitals in the region are in the small city of Oussouye, located at 10 km from the HDSS area, and in the larger city of Ziguinchor located at 50 km from the HDSS area. But, it is possible that some people travel to furthest cities, especially Dakar, where they have relatives to benefit from the familial environment and look for medical care. Findings also show that adults do move before their death in urban areas despite the relative availability of healthcare. For example, the magnitude of mobility found in the site of Kaya and Ouagadougou is similar. But the reasons behind these mobility flows tend to be different. As shown by the analysis on their place of death, in urban areas the mobility was motivated by the need to find social support. It is likely that those who undertake this kind of mobility have desperately used modern services and were willing to find relief and comfort among their relatives. In rural and semi-rural areas, age is strongly associated with mobility before death, with younger adults being more mobile than the elderly. This is in line with the fact that functional limitations tend to increase with age and play against elderly access to medical care (Debpuur et al., 2010; Van Rooy et al., 2015). Their mobility may be also compromised by the lack of support from close relatives, especially the helping-hand of young adults who are generally involved in rural-urban migration. On top of disability issues, cultural barriers also play a key role in explaining the decrease of mobility to have access to care with age. Previous studies in Mlomp found that elderly are generally more distant from health facilities and prefer to rely

• n traditional medicine (Duthé et al., 2010). Furthermore, aging per se is rapidly

conceptualized as a cause of death in a context of poverty. The community and family members are convinced that any treatment in favor of the elderly will be ineffective. Sometimes, they prefer to invest more money in their burial and funeral in order that their soul rest in peace (Sauerborn et al., 1996). Finally, elderly consider themselves as unproductive and as a burden for their close relatives. For example, it is reported in Senegal

13

and Burkina Faso, that after a certain age, some prefer to die to free themselves (Sauerborn et al., 1996; Duthé et al., 2010). Regarding the cause of death, while people who died from non-communicable diseases are more likely to die outside of Ouagadougou, in the rural setting of Mlomp, we have not found significant association between mobility before death and group causes of death. This finding emphasizes the complexity of the health care utilization in sub-Saharan Africa. In some settings, it is shown that in case of chronic diseases, and despite concurrent healing options (self-medication, traditional medicine, healers or marabous), patients tend to use healthcare

• services. For example in Lubumbashi, a city of the Democratic Republic of Congo, chronic

conditions were found to be strongly associated with the use of formal medical care (Chenge et al., 2013). In the same line, health care visits were higher among terminally ill adults who suffered from lasting diseases in Addis Abeba (Reniers et al., 2009). But in Dakar, the lowest level of health care utilization was found in patients suffering from non-communicable diseases (Duboz et al., 2015). In Mlomp, several mechanisms can contribute to explain the absence of differences in mobility before death between the defined group causes of death. First, the concept of non communicable diseases is not well integrated into rural settings of Africa (Baldé, 2007). Patients are used to deal with infectious diseases and may continue to resort on the same therapeutic itinerary even though they suffer from non-communicable

• diseases. A striking example is the case of patients suffering from hypertension and diabetes.

It is reported that among them little are aware of their disease and very few of those who are aware looks for care because of the lack of symptoms and of high costs of treatment (Kayima et al., 2013; Hall et al., 2011). In this analysis, it is not excluded that health care seeking be seriously compromised by the lack of income. Patients need to have sufficient resources to undertake the difficult journey of care seeking outside their place of residence. In addition to age and group causes of death, other variables such as education and marital status play a role in doing a mobility before death in rural and semi-rural areas. While in Mlomp single adults are less mobile before their death than the married ones, no effect is

• bserved in Kaya. The effect of marital status in Mlomp highlights the protective effect of
• marriage. Consistent with some previous findings in SSA (Anteneh et al., 2013; Chisumpa et

al., 2017 ), married adults could rely on the support of their partner to have access to medical

• care. But the effect of marital status may be blurred by the lack of information on standards of
• living. Married adults have greater chance to be wealthier than single adults. The confusion

between marital status and standard of living may be higher in Kaya. In this site, one can expect mobility before death to be more selective than in Mlomp. The availability of regional hospital means that it may be the elite (educated and wealthy residents) who tend to travel to city such as Ouagadougou to have access to care. This is confirmed by the fact that adults who went to school tend to be more mobile before their death in Kaya but level of education has no impact in Mlomp. Results found in the urban setting of Ouagadougou confirm the hypothesis that non-natives are more prone to go back to their native community in case of severe illness. It is quite probable that they went back to rural settings. In the Ouagadougou HDSS, around 70% of non-native adults of 15 years and more were born in rural areas (Rossier et al., 2011). As revealed by the analysis on their place of death, their mobility was motivated by the need to find social support. Furthermore, adults suffering from non-communicable diseases were more likely to undertake mobility. Taking together, these results suggest that even in cities access to care in cases of chronic diseases remains problematic. Either patients do not use health services properly (cultural barriers, cost,...) or the services offered are still inadequate.

14

Health systems in SSA continue to focus on infectious diseases despite the global warning on the rise of non communicable diseases. It is crucial to point out the different limitations of this study. The analysis is restricted to mobility undertaken by adults residents, at the latest 6 months before the event of death. We then fail to capture some deaths of migrants that could have left the area a few years before their death and are not anymore counted in the population of the different HDSSs. Also, in rural and semi-rural areas, a distinction between mobility initiated by the patient himself with a move in another household and constrained mobility such as emergency referral to another hospital without any stop between the home and the hospital, would refine the analysis. However, the motivation of the mobility reported in Mlomp for all causes of deaths that

• ccurred outside the area let us think that the second category is a minority. As usual,

representativeness of HDSSs is at stake in this study. It is not sure that the different sites included represent the diversity of urban and rural settings of West Africa. Finally, the effects

• f some variables on the mobility before death may be disturbed by the lack of information on

the standard of living of adults and for rural residents particularly, the presence of close relatives in cities. Despite these limitations, this study paves the way for future research on rural-urban mobility before death in West Africa. In rural areas, a non-negligible group of adults leaves their community to seek formal care in surrounding towns and die there. The magnitude of mobility depends on the supply of health facilities. It tends to be higher in rural remote areas and age is the strongest determinant. Younger adults are more mobile before their death whatever the types of diseases they are suffering from. Urban residents also move before their

• death. Particularly rural-migrants and those suffering from non-communicable diseases go

back to their community to benefit from supportive care and die. These findings shed the light

• n the poor access to health care in rural areas but also the coping strategies of adults that tend

to be in disfavor of the elderly. In urban areas, the issue faced by population is probably the quality of care and perhaps financial access, in cases of chronic diseases. Also of public health importance, the flows of deaths between urban and rural residents can affect the estimation of urban/rural mortality differentials. The death of an adult in mobility may be wrongly classified as an urban or rural death. This specifically true in large operations of data collection such as census which focus on the number of deaths occurring in each household during the last twelve months.

15

Appendix Table A1. Logistic regression on the probability of dying elsewhere than at Mlomp due to health reasons.

Characterisics Odds ratio Sex Male 1 Female 0.9 Age group at death 15-39 1 40-59 0.3** 60-79 0.2*** 80+ 0.1*** Education None 1 Primary + 1.1 DK 1.0 Marital status Married 1 Single 0.3** Divorced/Widowed Unknown(a) 0.7 0.2 Group causes of death CDs 1 NCDs 1.2 Indeterminate 1.0 Number of observations 708 *** p<0.01; ** p<0.05; * p<0.1

(a)In Kaya, people with unknown education and marital status are similar

16

References Anezaki, H. (2008). Transition of the place of death and total number of death in Japan. Asian Pacific Journal of Disease Management, 2(4), 97-101. Bado, A. R., Kouanda, S., & Haddad, S. (2016). Lieu du décès au Burkina: influence des caractéristiques sociodémographiques et environnementales. African Population Studies, 30(1). Baldé, N. M. (2007). Ampleur du diabète en Guinée: défis et initiatives locales. Médecine des maladies Métaboliques, 1(3), 99-103. Bauni, E., Ndila, C., Mochamah, G., Nyutu, G., Matata, L., Ondieki, C., ... & Etyang, A. (2011). Validating physician-certified verbal autopsy and probabilistic modeling (InterVA) approaches to verbal autopsy interpretation using hospital causes of adult

• deaths. Population health metrics, 9(1), 49.

Bocquier P. (2016), «Migration Analysis using Demographic Surveys and Surveillance Systems», M. White (ed), International Handbook of Migration and Population Distribution, Springer, Dordrecht, pp. 205-224. Byass, P., Chandramohan, D., Clark, S. J., D'Ambruoso, L., Fottrell, E., Graham, W. J., ... & Krishnan, A. (2012). Strengthening standardised interpretation of verbal autopsy data: the new InterVA-4 tool. Global health action, 5. Cárdenas-Turanzas, M., Torres-Vigil, I., Tovalín-Ahumada, H., & Nates, J. L. (2011). Hospital versus home death: results from the Mexican Health and Aging Study. Journal of pain and symptom management, 41(5), 880-892. Chenge, M. F., Van der Vennet, J., Luboya, N. O., Vanlerberghe, V., Mapatano, M. A., & Criel, B. (2014). Health-seeking behaviour in the city of Lubumbashi, Democratic Republic of the Congo: results from a cross-sectional household survey. BMC health services research, 14(1), 173. Chisumpa, V. H., Odimegwu, C. O., & De Wet, N. (2017). Adult mortality in sub-saharan Africa, Zambia: Where do adults die?. SSM-Population Health, 3, 227-235. Clark Samuel J., Collinson Mark A., Kahn Kathleen, Drullinger Kyle,Tollman Stephen M., 2007, «Returning home to die: Circular labour migration and mortality in South Africa 1», Scandinavian Journal of Public Health, 35(69 suppl), 35-44. Collinson Mark A., White Mickael J., Bocquier Philippe, Mcgarvey Stephen T., Afolabi Sulaimon A., Clark Samuel J., et al., 2014, «Migration and the epidemiological transition: insights from the Agincourt sub-district of northeast South Africa», Global health action, 7. Debpuur, C., Welaga, P., Wak, G., & Hodgson, A. (2010). Self-reported health and functional limitations among older people in the Kassena-Nankana District, Ghana. Global health action, 3(1), 2151. Duboz, P., Gueye, L., Boëtsch, G., & Macia, E. (2015). Accés aux soins à Dakar (Sénégal): fréquence, type de soignants et maladies chroniques non transmissibles. Médecine et Santé Tropicales, 25(2), 165-171. Duthé, G., Pison, G., & Laurent, R. (2010). Situation sanitaire et parcours de soins des personnes âgées en milieu rural africain Une étude à partir des données du suivi de population de Mlomp (Sénégal). Autrepart, (1), 167-187. Flory, J., Young-Xu, Y., Gurol, I., Levinsky, N., Ash, A., & Emanuel, E. (2004). Place of death: US trends since 1980. Health Affairs, 23(3), 194-200. Fottrell, E., & Byass, P. (2010). Verbal autopsy: methods in transition. Epidemiologic reviews, vol.32, DOI: 10.1093/epirev/mxq003. Gomes, B., Higginson, I. J., Calanzani, N., Cohen, J., Deliens, L., Daveson, B. A., ... & Meñaca, A. (2012). Preferences for place of death if faced with advanced cancer: a

17

population survey in England, Flanders, Germany, Italy, the Netherlands, Portugal and

• Spain. Annals of oncology, 23(8), 2006-2015.

Gu, D., Liu, G., Vlosky, D. A., & Yi, Z. (2007). Factors associated with place of death among the Chinese oldest old. Journal of Applied Gerontology, 26(1), 34-57. Gysels, M., Pell, C., Straus, L., & Pool, R. (2011). End of life care in sub-Saharan Africa: a systematic review of the qualitative literature. BMC Palliative Care, 10(1), 6. Hall, V., Thomsen, R. W., Henriksen, O., & Lohse, N. (2011). Diabetes in Sub Saharan Africa 1999-2011: epidemiology and public health implications. A systematic review. BMC public health, 11(1), 564. Kayima, J., Wanyenze, R. K., Katamba, A., Leontsini, E., & Nuwaha, F. (2013). Hypertension awareness, treatment and control in Africa: a systematic review. BMC cardiovascular disorders, 13(1), 54. Kouanda, S., Bado, A., Yaméogo, M., Nitièma, J., Yaméogo, G., Bocoum, F., ... & Sondo, B. (2013). The Kaya HDSS, Burkina Faso: a platform for epidemiological studies and health programme evaluation. International journal of epidemiology, 42(3), 741-749. Lemiere, C., Herbst, C., Jahanshahi, N., Smith, E., & Soucat, A. (2011). Reducing geographical imbalances of health workers in Sub-Sahara Africa. Washington, DC: The World Bank, 11. Ministère de la santé, Direction génerale de l'information et des statistiques sanitaires (2010). Carographie de l'offre de santé, région du centre. Nwakeze, N. M., & Kandala, N. B. (2011). The spatial distribution of health establishments in

• Nigeria. African Population Studies, 25(2).

Houttekier, D., Cohen, J., Bilsen, J., Addington-Hall, J., Onwuteaka-Philipsen, B., & Deliens,

• L. (2010). Place of death in metropolitan regions: metropolitan versus non-

metropolitan variation in place of death in Belgium, The Netherlands and England. Health & place, 16(1), 132-139. Houttekier, D., Cohen, J., Surkyn, J., & Deliens, L. (2011). Study of recent and future trends in place of death in Belgium using death certificate data: a shift from hospitals to care

• homes. BMC public health, 11(1), 228.

Lazenby, M., Ma, T., Moffat, H. J., Funk, M., Knobf, M. T., & McCorkle, R. (2010). Influences on place of death in Botswana. Palliative & supportive care, 8(2), 177-185. Lemiere, C., Herbst, C., Jahanshahi, N., Smith, E., & Soucat, A. (2011). Reducing geographical imbalances of health workers in Sub-Sahara Africa. Washington, DC: The World Bank, 11. Nwakeze, N. M., & Kandala, N. B. (2011). The spatial distribution of health establishments in

• Nigeria. African Population Studies, 25(2).

Phillips, K. A., Morrison, K. R., Andersen, R., & Aday, L. A. (1998). Understanding the context of healthcare utilization: assessing environmental and provider-related variables in the behavioral model of utilization. Health services research, 33(3 Pt 1), 571. Pison G, Wade A, Gabadinho A & Enel C. (2002) Mlomp DSS, Senegal, in Indepth network. Population and Health in Developing Countries (Volume 1). Ottawa: International development research centre, pp. 271-278. Sauerborn, R., Berman, P., & Nougtara, A. (1996). Age bias, but no gender bias, in the intrahousehold resource allocation for health care in rural Burkina Faso. Health Transition Review, 131-145. Remais, J. V., Zeng, G., Li, G., Tian, L., & Engelgau, M. M. (2012). Convergence of non- communicable and infectious diseases in low-and middle-income countries. International journal of epidemiology, 42(1), 221-227.

18

Reniers, G., & Tesfai, R. (2009). Health services utilization during terminal illness in Addis Ababa, Ethiopia. Health policy and planning, 24(4), 312-319. Rossier C, Soura A, Baya B. et al. 2012. Profile: The Ouagadougou Health and Demographic Surveillance System. International Journal of Epidemiology. 41(3):658-666. Rossier C., Soura A., Lankoande B., Millogo R., 2011, Observatoire de Population de

• Ouagadougou. Données du R0, R1 et R2 : rapport descriptif, Ouagadougou,

ISSP/Université de Ouagadougou, 71 p. Ujoh, F., & Kwaghsende, F. (2014). Analysis of the Spatial Distribution of Health Facilities in Benue State, Nigeria. Public Health Research, 4(5), 210-218. Van Rooy, G., Mufune, P., & Amadhila, E. (2015). Experiences and perceptions of barriers to health services for elderly in rural Namibia: a qualitative study. SAGE Open, 5(3), 2158244015596049. Wilson, D. M., Smith, S. L., Anderson, M. C., Northcott, H. C., Fainsinger, R. L., Stingl, M. J., & Truman, C. D. (2002). Twentieth-century social and health-care influences on location of death in Canada. The Canadian journal of nursing research= Revue canadienne de recherche en sciences infirmieres, 34(3), 141-161.