Socioeconomic status and elderly adult mortality in rural Ghana: evidence from the Navrongo Dss. 1

Elderly adult health and issues affecting them in Africa have not been adequately addressed by research. This study explored the relationship between socioeconomic status and elderly adult mortality in the Kassena-Nnakana District (KND) of northern Ghana using data from the Navrongo Health and Demographic Surveillance System (HDSS) in 2005-2006. 15,030 adults aged 60 years and over were included in the study, of whom 1315 died. Using Cox proportional hazards regression, we found that socioeconomic status (SES) was not a determinant of elderly mortality. Compared to the lowest SES quintile, the adjusted hazards ratios were: 0.94 (95%CI: 0.79–1.12) for second quintile, 0.91 (95%CI: 0.76– 1.08) for third quintile, 0.89 (95%CI: 0.75–1.07) for fourth quintile and 1.02 (95%CI: 0.86–1.21) for the highest income quintile. However, living without a spouse [HR=1.98, 95%CI: 1.74–2.25], being male [HR=1.80, 95%CI: 1.59– 2.04] and age [HR=1.05, 95%CI: 1.04–1.05] were significant factors for elderly adult mortality. This shows that companionship and social/family ties are of more importance than household socioeconomic status in determining elderly adult mortality. Efforts should therefore be made to introduce programs and policies to support the elderly, especially those living alone.


Introduction
Elderly adult health in developing countries is presently a matter of concern due to the various issues that directly affect their well being, yet little is being done to resolve them (Robinson et al. 2006).Modernization, and continued westernization of African societies has resulted in the collapse of social and cultural ties that communities had with the extended families such that the elderly are being left to fend for themselves.This in conjunction with increased rural urban migration among the young and middle-ages, the burden of caring for AIDS orphans and poor social security among others, has increased the economic burden on the elderly adults.This in turn affects their health and the struggle to barely survive overrides their health concerns, resulting in avoidable excess mortality (Oppong 2006).
Higher socio-economic status has been shown to reduce exposure factors that lead to morbidity, disability and eventually mortality.This is because higher socio-economic status leads to better health care, comfortable living conditions, less exposure to hazards and better diets (Rogers, Hummer, & Krueger 2005).Some studies even suggest that people in the higher SES, over the years have been known to practice healthier lifestyles and behaviour (smoking, heavy drinking, sedentary lifestyles) while those in lower SES levels are increasingly embracing them and hence have a higher risk of mortality (House et al. 1990).
Socio-economic stratification and its effect on elderly adult mortality is therefore, an important issue in public health and epidemiology.Elderly adults are very dependent on the middle age and working population in many African countries where health systems are not well managed and social security is nonexistent or limited to a few individuals (Oppong 2006).These socio-economic disparities in health outcomes among the elderly adults if addressed will go a long way in reducing the burden of disease and mortality among people in the lower socio-economic strata at the same time increase longevity.The increase in elderly populations also brings into focus the issue of dependency ratio in the developing world compounded with the emerging issues of rural -urban migration, break in family ties, less care for the elderly, poor social security and economic hardships, which as stated by Oppong (2006) all make life harder for elderly adults.Many studies elsewhere have found different findings on elderly adult mortality.However they all show the importance of socioeconomic status as a factor for elderly mortality (Liang et al. 2003;Murata et al. 2005;Sudore et al. 2006;Vrbova et al. 2005;Zimmer & Kwong 2004).Ghana, with an increasing elderly population therefore needs to look at ways of addressing the socio-economic disparities in old age.In the region where this study was carried out (Kasssena-Nankan District of Northern Ghana), many households have similar type of houses, lifestyles, face the same environmental and human hazards and asset possession is almost the same.
Lack of data on elderly adult mortality is another big problem facing researchers who would like to engage in useful analysis of elderly adult health in Africa.One way in which this can be achieved is through demographic surveillance sites (DSS) which collect vital demographic event data including mortality information about people in a defined geographical location.Data from these DSS sites can act as a guide for developing countries to look at ways of addressing issues on elderly adult mortality and hence provide proper information that can help in alleviating excess mortality (Baiden, Hogson, & Binka 2006;Chandramohan et al. 2008).
In many rural areas of the developing world and Sub-Saharan Africa in particular, the socioeconomic status of many people can be measured using household wealth and possessions, since information on income and expenditure is hard to come by (Bawah & Zuberi 2004).Use of household assets to measure socioeconomic status has been used extensively in many health related studies in less developed countries and their practicability proven (Filmer & Pritchett 2001;Morris et al. 2007;Somi et al. 2008;Vyas & Kumaranayake 2006).
In many sub-Saharan countries like Ghana, despite the projected growth of the older generation, there is little data on the health of elderly people nor clear policies by the governments about the health of this ageing population (Robinson, Mosha, Grainge, & Madeley 2006).Studies have shown a distinct difference in mortality rates among different groups with different income status (an indicator of socio-economic status) among the elderly.They suggested that decrease in mortality in these elderly adults is not just explained by conditions in early life but by their present status (Catalano 2002;Zimmer 2006).Thus the need to put more emphasis on the prevailing factors affecting mortality among the elderly and in particular their socio-economic status.
The study was based on the concept from Roger et al ( 2005) that proposed socioeconomic status, as a distal factor for mortality acting together with social relations, geographical, human and environmental factors.These distal factors affect mortality through proximate factors like nutrition, human behaviour, living conditions, psychological and physiological variables.Using data from the Navrongo Demographic Surveillance System (NDSS), this study was aimed at investigating the association between household socioeconomic status and individual elderly adult mortality in a harsh climatic and economically deprived region of northern Ghana, where poverty is widespread and asset ownership is almost homogeneous.The second objective was to assess the elderly adult death rates in the same region during a two year period (2005 -2006).

DATA
The study was based on longitudinal data collected from the Navrongo Demographic Surveillance System (NDSS), which is run by the Navrongo Health Research Centre (NHRC).It is a continuous surveillance of all the residents of the Kassena-Nankana District of northern Ghana.The NDSS was started in July 1993 to monitor demographic dynamics in the district and serve as a platform for launching research on morbidity, mortality and fertility.It started with a baseline census of all the individuals in the district and continuously monitors their vital demographic events (migration, death, pregnancies, and births) after every four months at the household level.Apart from the individual data, household and compound information is also collected and stored in a relational database for easy updates and linkages.More information on the NDSS is published elsewhere (Baiden, Hogson, & Binka 2006;Bawah et al. 2003;Bawah et al. 2006;Debpuur et al. 2005;Nyarko et al. 2002).
The stable registered population of the Navrongo DSS was about 145,000 as at June 2006 from a total of over 30,300 households; The Elderly adults (defined as all adults over the age of 60 years) were slightly over 6% of this population.However, since the population is monitored continuously and keeps changing, the total number of elderly adult residents aged 60 years and above included in the study between 01/01/2005 and 31/12/2006 was 15030 residing in 12475 households, 27803 person years and 1315 deaths.1 Socio-economic status (SES) was derived from data collected by the NDSS using a standardized questionnaire, that recorded household assets (electronics, car, motorbikes, bicycles, livestock,), water use, waste disposal methods, house type (ownership, building material used), fuel/lighting type, land ownership and availability of food.These were used as the independent variables in construction of the SES index.The variables were first re-coded into binary format as 0 or 1 to denote the absence or presence of the asset respectively.Using the Principal component analysis technique, as recommended by Filmer and Pritchett (1999) and used in a number of studies (Filmer & Pritchett 2001;Kahn et al. 2005;Mwageni et al. 2005;Nguyen et al. 2005;Vyas & Kumaranayake 2006).We used the first principal component (which explains the most variability in the data) to rank the household into five different quintiles from the lowest to the highest/richest (poorest, poorer, poor, less poor and least poor).The component analysis was based on a formula where each component is a linear weighted combination of all the initial variables such that for a set of components X 1 to X n ; PC 1 = a 11 X 1 + a 12 X 2 + ...... + a 1n X n PC m = a m1 X 1 + a m2 X 2 + ...... + a mn X n Where; X n is the n th variable out of n variables a mn is the weight of the m th principal component and the n th variable (Source; Vyas & Kumaranayake 2006) It is worth noting that the SES index was based on the households that were occupied by the elderly adults and not all the households in the region.This was done to reduce the possibility of having a biased distribution of the elderly adult population towards one side of the socioeconomic strata probably due to selective mortality in earlier ages.This was done by first dropping those households that did not have a person over the age of 60 years present.
The analysis used Cox hazards proportional analysis with person time as the underlying denominator to calculate death rates, draw Kaplan-Meier survival graphs and investigate the effect of SES on elderly adult mortality while controlling for the other confounding distal factors.
Elderly adult mortality rates in the KND reduced considerably, from a high of 40 deaths per 1000 PY and 140 deaths per 1000 PY for the ages 60-64 and 80-84 respectively in earlier reports, to a low of 28 deaths per 1000 PY and 93 deaths per 1000 PY for the same age groups in the two year observation period.
There was a distinct difference between males and females during the two-year observation period, such that females had a lower overall death rate of 39.5 deaths per 1000 PY compared to males, who had an overall death rate of 58.1 deaths per 1000 PY.In the age specific death rates by sex, males have a higher death rate than females, but as they age, there is a crossover at 75-79 and 80-84 age groups, with the females having a higher rate than males.(Fig 1) A higher death rate was observed among elderly adults who were not living with a spouse (59.2 deaths per 1000 PY) compared to those who were living with a spouse (35.9 deaths per 1000 PY).There was a general reduction in the death rates across all the explanatory variables mentioned earlier from 2005 to 2006; with the overall yearly rates dropping from 50.2 deaths per 1000 PY in 2005 to 44.1 deaths per 1000 PY in 2006.

DISTRIBUTION OF HOUSEHOLD WEALTH
The elderly adults were grouped in five quintiles after including household wealth, housing type, food availability, water usage, toilet facilities and waste disposal.A total of 41 principal components were used, and the first component accounted for 16.4% of the total variance, the second component accounted for 8.5% and the third accounted for 5.1% of the total variance of all the variables used.We therefore used the first component to score the weights of each asset and eventually come up with the SES index.Using electricity for lighting and ownership of electrical gadgets were the most important variables in determining the levels of SES among the elderly adults in the KND (Appendix 3).
Comparison of the different quintiles and how the variables were allocated to each quintile shows that there were more assets in the upper quintiles than the lower quintiles; the lower quintiles showed little ownership of almost all the variables included in the analysis while the numbers were higher for the upper SES levels, except for land (for both farming and building), house ownership and goats which were more in households in the lower SES strata.The composition of the different variables allocated to each quintile is shown in Appendix 2. There was a notable lack of sanitation and waste disposal facilities, with over 99% of the household where elderly adults resided lacking toilet facilities, and 94% of these households also lacked proper access to water.
Among the different SES quintiles, there were a relatively higher proportion of the elderly adults living alone being placed in the poorest category (42.0%) in comparison to those who were in households that had more people (18.6%).( See A relatively higher proportion of elderly adults with at least primary education were placed in the Least poor category of SES (51.7%) compared to those who had no education (18.3%), while in the poorest category there was also a relatively higher proportion of adults with no education (20.6%) compared to those with some education (8.6%).(Fig 3 .2)Elderly adults with no education were spread almost uniformly in proportions across the quintiles, but for those with primary level of education and above, there was a general increase in proportion with increase in SES such that the higher the SES, the higher the proportion of those who had some education.There were also a relatively higher proportion of women living in single person households (7.6%) than males (5.1%).The different SES strata had slightly varying mortality rates, with the poorest having a death rate of 51.7 deaths per 1000 PY, to the less poor having an overall death rate of 43.2 deaths per 1000, however, the trend reversed in the uppermost SES strata with a death rate of 50.3 deaths per 1000 PY although the difference between the different groups was not statistically significant.However, the poorest group still had the lowest survival probability of all the groups, showing that those with less assets experience higher mortality in this region.This also shows that in the KND, there is little difference in overall mortality among the different SES groups.A concentration index of -0.01 (almost zero) and a poorest-poor ratio of 1.03 shows low levels of inequality for mortality among the elderly adults in the KND.( On plotting a concentration index curve to check the level of health inequality (measured by mortality rates) among the elderly adults by SES, there was no significant differences in death rates among the different quintiles, with the concentration curve (showing the death rates by quintile) and the line of equity (showing an ideal situation of no difference between the quintiles) lying on top of each other with a small of inequality especially among the poorest.(Fig 3) A big area between the two curves would normally indicate a high inequality between the poorest and the least poor in the given community, but in our case it is almost negligible (same as table 1 above).

HOUSEHOLD SOCIO-ECONOMIC STATUS AND MORTALITY
Cox proportional hazards regression modelling was used to investigating the relationship between SES and elderly adult mortality while adjusting for the other independent variables.The following results were obtained for both univariate and multivariate analysis.(Table 2) There was no association between SES and elderly adult mortality in the Kassena-Nankana District.The hazards ratios for all the different SES quintiles (poorer, poor, less poor and least poor) show minimal deviation from the comparison category (poorest); this was especially true between the poorest and least poor groups where the HR was equal to 1, meaning that elderly adults in the KND have almost equal risk of mortality irrespective of SES.It was found that those elderly adults in the four upper categories of SES had lower hazard ratios for mortality compared to those in the poorest category.The hazards ratios for mortality reduced with increase in SES for the least poor compared to the poorest.(Table 2) A look at some of the results in (unadjusted HR) showed that the poorer were 0.1 times less likely to die compared to the poorest; the poor were 0.14 times less likely to die compared to the poorest; the less poor were 0.16 times less likely to die compared to the poorest and the least poor were 0.03 times less likely to die compared to the poorest.However, these differences were very small and not statistically significant at the 95% confidence interval.The same trend in the univariate analysis was observed when we adjusted for the other variables in the other two models.In multivariate modelling, SES was found not to be a predictor of mortality in the elderly since the hazards ratios among the different levels did not vary from the comparison group in a big way and were also not statistically significant.Interestingly when SES was included in the first and second model during the multivariate analysis, there was absolutely no difference between the poorest and the least poor categories Presence of a spouse was found to be more important as a predictor of mortality than even the presence of other household members since having other members of the household was not found to be significant in both the unadjusted and adjusted regression.The results showed that those who were living with other members of the household had a marginal advantage over those living alone (about 8 % less likely to experience mortality), although this was not statistically significant.Mostafa & Van Ginneken in Bangladesh also observed that the presence of a spouse was more important than just living with other members of the household.Other findings and papers show that social relations and family support are an integral part of the wellbeing of elderly adults elsewhere (Johnson et al. 2000;Murata et al. 2005;Oppong 2006).

DISCUSSION
The reduction in death rates in the KND over the years, according to a report by the NHRC, could be due to the mortality diminishing interventions that have been introduced in the district (Bawah et al. 2003).However these mortality rates are still quite high compared to other areas like Filabavi in Vietnam which had rates of 9.37 deaths per 1000 PY and 57.24 in the same age groups of 60-64 and 80-84 age groups (Nguyen et 2005).
Household wealth was found not to be an important predictor of elderly adult mortality in region.There are other studies which also showed lack of a relation between SES and elderly adult mortality (Hoffmann 2005;Vrbova et al. 2005).However, the differences between the settings and context of those studies makes it hard for proper comparisons with this study since they were carried out in more developed countries and from hospital data, while this study was done from continuous community surveillance.In a study done in the same region of the KND, Debpuur et al. (2005) found that SES does not predict child mortality.Although it is not entirely comparable to this study given the age difference of the participants in both studies, it suggests that SES differentials are minimal or SES is not a predictor of mortality in the KND.Similar findings were also observed in a Swedish study, showed that in middle adulthood, the effect of SES on mortality was quite high but among the elderly adults (Merlo et al. 2003).
Our findings were however contradictory to other studies in similar environments which showed an inverse SES-mortality influence among the elderly.In rural Cambodia, rural China and Bangladesh, which have comparatively homogeneous socio-economic regions like the KND, it was found that higher SES was a protective factor for morbidity and mortality they used education, household wealth or income.Those in the well off categories always had less risk of mortality compared to the lowest comparison group of SES for all ages of the elderly adults (Mostafa & Van Ginneken 2000;Zhu & Xie 2007;Zimmer 2006;Zimmer & Kwong 2004).Our findings were different from our hypothesised influence of household SES on elderly adult mortality as stated in the theoretical framework.
When we compared the mortality rates among different groups using the concentration index, the poorest-least poor ratio stood at 1.02, showing little difference between those elderly adults who were in the higher SES quintile and those in the lower quintile.The concentration index, which was used to indicate the levels of inequality depending on how big the value is from 0, was quite minimal (-0.01).From our results, the p-values for the SES hazard ratios were very high in all the models, and all the confidence intervals included unity.We therefore rule out weak association or chance in the lack of any relationship between SES and elderly adult mortality.This alludes to the convergence school of thought where the SESmortality relationship converges in the elderly population.However, there is need to do a similar study for the whole population in the region and come up with a more concrete comparison for this theory to hold.
The lack of SES effect on mortality in the KND could be due to several hypothesised reasons; first the general homogeneity in terms of SES, such that household wealth and material possessions do not accurately bring out differences in this area.Secondly, in the elderly adults, biological and physiological determinants may be independent of the SES gradient and hence lack of effect.Thirdly, the SES effect wears off with more deaths in the lower and middle age groups such that the effect becomes minimal in old age (House et al. 1990).
In the context of this study, it was hard to explain the mechanisms by which the presence of a spouse affects mortality even though its importance as a predictor of mortality among the elderly was noteworthy.Lund et al (2002) stated that the presence of a spouse is more important in determining elderly adult mortality than just being married or living with other people in the same household.In trying to explain the effect of spouses on the elderly, one study noted that elderly adults have unique dietary requirements, require specialised care and drug taking assistance.Companionship therefore serves as a buffer against poor diet and appetite as they will aid in feeding and related activities which in effect ensure good health (McIntosh, Shifflett, & Picou 1989), in addition the psychological boost married spouses have on each other (Davis et al. 1992) has been noted to be of importance to survival.
Even though there was no indication of SES being a major determinant of elderly mortality, there is still need to improve the lives of elderly adults in the KND to ensure a more dignified and comfortable ageing, especially among those with very little resources (poorest).This is due to the fact that among those households that had a single occupant (indicating elderly adults living alone) a bigger proportion was in the poorest category.Even though SES as a factor for mortality has been investigated in many studies, the way the relationship works in the older members of our society should be researched on further, not in an isolated manner, but in relation to the whole population in the region.Principal component analysis has been used in many health related studies in the developing world and found to be adequate (Filmer & Pritchett 2001), hence we considered it appropriate for this study.However, there is need for another mode of measuring SES, such as use of the real value of the assets to be tried in the KND and hence offer a comparative measure which would consequently confirm if indeed SES does not affect mortality.Although the findings in this study point to the convergence school of thought in the SES mortality relationship, there is need to carry out similar studies in the whole population.This will allow researchers to have a better comparison of all the age groups and the transition from the lower age groups to the older groups and conclusively support it.Further research on the effect of SES on cause specific mortality would be quite helpful in shedding more light on the real situation and help to zero in on the areas of importance to elderly adult survival.

Figure 1 :
Figure 1: Plot of Age specific death rates by sex

Fig
Figure 3.1

Figure
Figure 3.2

Figure 2 -
Figure 2 -Kaplan-Meier survival curve of elderly adults in the NDSS by SES

Fig 3 :
Fig 3: Plot of the concentration curve for mortality among the elderly adults in the NDSS

Table 1
Distribution of death rates across the SES quintiles

Table 2 :
Cox proportional hazards regression analysis of mortality risk for elderly adults in the NDSS