Socio-economic, physical and health-related determinants of causes of death among women in the Kintampo districts of Ghana

Abstract

This study examined the socio-economic, physical and health-related determinants of causes of death among women of reproductive age (WRA) in the Kintampo North Municipality and Kintampo South District of Ghana. Longitudinal data from the Kintampo Health and Demographic Surveillance System (HDSS) was used. Causes of death data from 2005 to 2014 for 846 WRA aged 15–49 were categorized into three broad groups: maternal, infectious and non-communicable diseases. Three hierarchical multinomial logistic regression models were used to examine the determinants of causes of death, with the maternal causes of death as the reference category. Distal, intermediate and proximate factors were entered cumulatively one after the other in Models 1, 2 and 3, respectively, to account for their separate effects on the outcome variable. Across all three models, ever-married (RRR = 0.12; p < 0.001) WRA were significantly less likely to die from infectious or NCD than maternal causes compared to those who were never-married. At the adjusted level (Model 3), infectious causes of deaths differed from the maternal causes of deaths by age at death, marital status, land ownership, district of residence, year of death, season of death, place of death, admission in the last 12 months, surgical operation in the last 24 months and sudden death. Marital status is a key determinant of causes of death among WRA.

Keywords:

• verbal autopsy
• women of reproductive age
• causes of death
• determinants
• Kintampo health and demographic surveillance system

PUBLIC INTEREST STATEMENT

The nexus between socio-economic factors and health has been given less attention in low- and middle-income countries and much less among women. Nonetheless, the burden of ill health disproportionately affects women due to biological, gender and other socio-economic factors. This study examined the socio-economic, physical and health-related determinants of causes of death among women of reproductive age (WRA) in the Kintampo area of Ghana. Causes of death data from 2005 to 2014 for 846 WRA aged 15–49 were categorized into three broad groups: maternal, infectious and non-communicable diseases. Three hierarchical multinomial logistic regression models were used to examine the determinants of causes of death. Across all three models, ever-married (RRR = 0.12; p < 0.001) WRA were significantly less likely to die from infectious or NCD than maternal causes compared to those who were never-married whilst older women were more likely to die of infectious or NCD causes than maternal causes relative to younger women.

1. Background

The physical and socio-economic environment in which people live has been found to influence their health through shaping their living conditions and quality of life (Bahadori et al., 2015). The socio-economic determinants of death have been given less attention in low- and middle-income countries (LMICs) and much less among women, especially those in the reproductive age (Mane et al., 2013). Yet, the burden of ill health disproportionately affects women of reproductive age due to biological, gender and other socio-economic determinants (Langer et al., 2015).

Addressing the burden of ill-health, especially among women, remains a major priority in the Sustainable Development Goals (SDGs) just like in the Millennium Development Goals (MDGs) (UNDP, 2009). SDG 3 aims to “ensure healthy lives and promote wellbeing for all at all ages”. This increased global attention, particularly on maternal and child health, has resulted in considerable success as maternal, perinatal and nutritional disorders reduced by about 20% between 2000 and 2013 (Langer et al., 2015; Scrafford & Tielsch, 2016). Nevertheless, poor access to healthcare services and poor quality of care that influence the level of mortality still persist particularly among women in many low- and middle-income countries and in many disadvantaged communities in high-income countries (HICs) (Barros et al., 2012). There is a need to understand the determinants of causes of death among women in LMICs and settings with a high-burden of ill-health in order to fashion out policies and programmes to resolve the challenges.

Gender inequalities tend to worsen the adverse effects of low socio-economic status on health and mortality for women. Differential access to and control over health resources, both within and outside families are influenced by gendered norms and values. In many LMICs, many agricultural workers are women, and many of them are not paid since their labour is regarded as part of their role within the family (World Bank, 2018). Studies have shown that gender inequalities in the allocation of resources, for example, income, education, health care, nutrition and political voice, are strongly associated with morbidity and mortality. Consequently, women face exposures and vulnerabilities that tend to influence what they die from. Therefore, the primary aim of this study was to investigate the determinants of maternal, infectious and NCD causes of death among women of reproductive age (WRA) in the Kintampo North Municipality and Kintampo South District.

2. Methods

2.1. Study area

The study location is part of the catchment area of the Kintampo Health Research Centre (KHRC) that administers the Kintampo HDSS. The Kintampo HDSS covers both the Kintampo North Municipality and the Kintampo South District of the Bono East Region of Ghana (Abubakari et al., 2015). The two districts are mainly rural, and their capitals, Kintampo and Jema, are semi-urban. Together, the two districts recorded a population of 156,145 as of December 2016 (Kintampo Health Research Centre, 2017).

2.2. Study population

A total of 1,259 deaths and 329,505 person-years of observation (PYO) were recorded among WRA aged 15–49 years during the 10-year study period. However, 162 (12.9%) cases of death had no respondents providing information on them. The two main reasons accounting for this were: (i) the difficulty in getting a family member that will be able to provide the required information; and (ii) the refusal to either take part in or to complete a verbal autopsy (VA) interview. Some of the VA interviews were done, but there was insufficient information or specific information were missing for 196 (15.6%) of them. In view of this challenge, the Physicians could not assign any cause of death to them or the cause of death was unknown. Accidents contributed 55 (4.4%) cases which were excluded because they fell outside the scope of the current study. The remaining 846 (67.2%) were used as the population for all analyses in the present study.

2.3. Study design

The current study used a quantitative research design. The study sample consisted of a prospective open cohort of women aged between 15 and 49 years who lived in the Kintampo HDSS area. The study was designed to analyse the determinants of causes of death among WRA in the Kintampo districts of Ghana from 2005 to 2014, using verbal autopsy data from the Kintampo Health and Demographic Surveillance System (Kintampo HDSS). Details about the study area, design and operations of the Kintampo HDSS have been cited in an earlier publication (Abubakari et al., 2019).

2.4. Data source

Data for this study came from the Kintampo HDSS of the KHRC. Kintampo HDSS is made up of field and computing operations to manage the longitudinal follow-up of persons and their households as well as residential units and all their demographic and health characteristics within the Kintampo area (Nettey et al., 2010; Owusu-Agyei et al., 2012).

2.5. Key variables and measures

The selection of both the outcome and explanatory variables for this study was guided by a review of the literature, and the multilevel eco-epidemiological lifecycle framework adapted for this study (Figure 1), and sometimes to allow for enough data points especially at the multivariate level of analyses.

2.5.1. Outcome variables

The outcome variables for this study were deaths due to maternal, infectious or non-communicable causes among WRA in the two Kintampo districts from 2005 to 2014. Maternal causes of death were operationally defined to include only direct (obstetric) deaths as the death of a woman while pregnant or within 42 days of termination of pregnancy, irrespective of the duration and site of the pregnancy. Infectious causes of death were also defined to include deaths from all infections and parasitic diseases whilst non-communicable causes of death were operationally defined as non-maternal and non-infectious and non-external or injury causes of death.

2.5.2. Explanatory variables

Age was measured as a categorical variable, with women grouped into five-year bands namely: 15–19, 20–24, 25–29, 30–34, 35–39, 40–44 and 45–49 years. This division made it possible to explore age-related differences associated with maternal, infectious and non-communicable causes of death including comparing adolescents and their older counterparts in the analysis.

Parity, which is defined as the number of children a woman had, was categorised into three (3) groups: 0 (no child), 1–3 children and 4 children and above.

Education: The highest educational level attained by the deceased was classified into three—never had any formal education as “none”, those who attained up to primary school level as “primary” and those who got up to middle or junior high school or senior high school or tertiary as “JHS and above”.

Ethnicity: we used migration status as a proxy to distinguish between indigenous ethnic groups and from ethnic groups that migrated from elsewhere to settle in the study area. Theories of migration suggest that migrant group tend to be disadvantaged relative to the indigenous people in access to health and other social services.

We measured religion using the affiliation of women to the major religious groups. Women were classified into three broad religious groups: Christians, Muslims and Other. Occupation was put into two categories: unemployed and employed.

In order to explore the cultural dimension of mortality further, this study considered a measure of female level of independence in the community. The reviewed literature discusses autonomy as very important as far as adult female mortality is concerned. This is because several studies have linked autonomous women with low mortality and vice versa. In view of this observation in the reviewed literature, farmland ownership was used as a proxy to measure the level of female autonomy in the community. This is quite appropriate because culturally, the head of the family who is mostly a male owns the land in trust for all. Therefore, a woman owning a farmland is an indication of how independent she is in the community. Land ownership was categorised into two in terms of whether the deceased woman owned land or not (yes or no).

The physical and climatic factors, including the household level factors, were also considered in this study. The place of residence variable is a household characteristic, and it was determined by the location of the compound in which the household is found. The household is either rural or urban. The study area is predominantly rural. There are only three communities, namely, Kintampo, Jema and Babato, with a population of over 5,000, and therefore classified as urban based on Ghana Statistical Service criteria (Ghana Statistical Service, 2022). The Kintampo HDSS covers two administrative districts. Based on this, the district was also categorised into 2, namely, Kintampo North and Kintampo South. Another physical and climatic factor considered in this analysis was the season of death which was classified into two: wet and dry whilst the year of death was categorised into ten, from 2005 to 2014.

In addition, factors related to the health system were considered to reflect the conceptual framework adapted for this study. Distance is one key factor identified in the literature as affecting health service utilisation. Several studies have established a positive correlation between health facility use and a 5-km radius. In contrast, studies have shown that health facilities beyond 5-km radius tend to reduce health service utilisation. This has the tendency to affect the health status and incidence of death eventually. Therefore, distance to the health facility was dichotomized into 5 km or less and more than 5 km by using the GIS data. Another health-related factor considered is the place where the deceased died. This variable is categorised into two: hospital and other to represent WRA who died in or outside a health facility. The health facilities have the clinicians and skill birth attendants as well as the equipment needed during emergency situations but these facilities are not available outside the hospital. Therefore, the “place of death” is used in this study as a proxy measure for health service utilisation.

Intermediate level factors in this analysis were living standards and lifestyles. These factors were conceptualised to include both individual and household level variables. At the household level, the Principal Component Analysis (PCA) method was used to categorise households into three socio-economic statuses (SES) of most-poor, poor and least-poor. Household assets such as television, radio, refrigerator, telephone, lighting type, type of roofing material, type of floor material, vehicles, motorbikes and livestock were used in the PCA. Other household level variables considered in this study to reflect the importance of sanitation were source of drinking water and type of toilet facility, which were categorised separately as improved or unimproved. At the individual level, tobacco, alcohol and drug use were measured as to whether it was used or not. In the multivariate analysis, we created a composite variable labelled as “substance use” was measured (Yes or No) in terms of whether the deceased woman used any of the substance or not.

Finally, proximate level variables are considered in this current study. These variables sought to measure the health status of the deceased WRA. The proximate level variables are the most immediate factors that influence the outcome of interest, which are the risk of dying from maternal, infectious and non-communicable causes of death. The health status in this current study includes admission in the last twelve months before dying, whether or not the deceased woman had some surgical operation 24 months before dying and whether or not the deceased woman died suddenly. Responses to each of these variables were categorised into two (2): yes or no.

2.6. Analytical methods

Causes of death were categorized into three broad groups: maternal, infectious and non-communicable diseases. We fitted three hierarchical multinomial logistic regression models, with maternal causes of death as the reference group or base outcome. The choice of this method was due to the multiple outcomes. This was done to deepen our understanding of the effect of the various determinants of maternal, infectious and non-communicable causes of death in a step-wise manner.

Thus, we first introduced into the distal variables or factors into the model (Table A1) to examine their separate effect on the outcome variables. In the next stage, we introduced the intermediate variables into the model in order to examine their cumulative effect, controlling for the distal factors in model 1 (Table A2). It is important to note that intermediate level factors were conceptualised to cover standards of living as well as life styles and behaviours. Standards of living comprise household socio-economic status, source of drinking water and type of toilet facility. Life style and behaviours, on the other hand, include tobacco, alcohol and drug use.

In the final stage (model 3), we entered the proximate factors into the model in line with our conceptual framing of the causes of death in this study. The proximate variables were also entered cumulatively to investigate their effect on the outcome variables when the distal and intermediate variables were controlled for. Proximate factors were conceptualised as the health status of the deceased WRA before dying. Health status in the current study included admission in the last twelve months before dying, and if the deceased woman had a surgical operation in the last 24 months before dying as well as whether the death was sudden or not. The R² is determined at each stage of the hierarchical model. P-values of less than 0.05 were considered statistically significant.

2.7. Limitation of the study

The influence of mortality from the various causes may not have been estimated accurately since not all deaths recorded by the Kintampo HDSS had successful verbal autopsy interviews but this is expected to be random and therefore, should not have major effects on the present study. In addition, a proportion of the cases with successful interviews were coded as “cause of death not determined”. This is also expected to be random. Also, ethnicity was used as a proxy for migration status but this is not the usual convention, and same applies to the use of season to measure climatic factors. Furthermore, health system variables (Distance to health facility and Place of death) do not fully consider availability and quality of health services. Moreover, it is well established in the literature that NCDs affect older persons more. Therefore, by restricting this study to WRA, the effect of NCD causes of death may not be well accounted for. However, the analyses in this study were done based on the observation that the three categories of causes of death were recorded among the WRA. Finally, this study used secondary data and other key factors, such as pre-conceptional, conceptional, and geo-political contextual variables that may affect maternal, infectious and non-communicable causes of death, were not included in the analysis.

2.8. Ethical consideration

Written informed consent was obtained from participants involved in interviews as part of the Kintampo Health and Demographic Surveillance activities. The Kintampo Health Research Centre Institutional Ethics Committee reviewed the protocol and all instruments associated with this study as part of the activities of the Kintampo HDSS. The approval certificate details are Ref: KHRC/IEC/ICF/2010–1; FWA: 00011103; and IOR0004854.

3. Results

The adjusted model (Table A3), where all the factors were controlled for, examined the demographic socio-economic, physical, health-systems, behavioural and personal health-related determinants (distal, intermediate and proximate factors) of causes of maternal, infectious and non-communicable causes of deaths from 2005 to 2014 in the Kintampo districts. Model 3 explains about 15% of the variations. It is assumed that distal, intermediate and proximate factors, namely age at death, children ever born, marital status, highest educational level attained, place of residence, district of residence, religion, employment status, migration status, land ownership, place of death, season of death, year of death, distance from household to health facility, household socio-economic status, source of drinking water, type of toilet facility, alcohol-tobacco use, admission in the last 12 months before death, surgical operation in the last 24 months, whether death was sudden or not, affect maternal, infectious and non-communicable causes of death. The Wald χ2 (prob>chi2) of the model is significant at 99% confidence interval indicating that the model has a good fit.

From Model 3, it is observed that the age at death of WRA has significant influence on the causes of death (maternal, infectious and non-communicable). However, the age group 30–34 years which was significant in both Models 1 and 2 (Tables A1 & A2) was no longer significant in Model 3. This means that once the proximate factors were introduced, the risk of dying from infectious relative to maternal causes for women aged 30–34 was not significantly different from those aged 15–19 years. In addition, the significance levels for age groups 35–39 and 45–49 dropped from P-value < 0.01 to < 0.05 for infectious versus maternal causes of death, and from age 35 to 49 for NCD versus maternal causes of death. This means that the effect of age was diminished once proximate factors were introduced. In addition, the magnitude of the relative risk ratio increased across the age categories as was observed in Models 1 and 2 (Tables A1 & A2) but in Model 3 (Table A3), the magnitude of the RRR for age group 45–49 was lower than the preceding age group 40–44 in the case of infectious versus maternal causes of death. This means that though age is still significant, its effect as observed initially is diminished once proximate factors were accounted for. Therefore, part of the initial strength of age was due to proximate factors. Thus, provision of health services such as admission and surgical operation facilities as proximate factors diminishes the age difference in causes of death.

The results showed that WRA who were 35–39 years had an increased relative risk of dying from infectious than maternal causes by more than six times (RRR = 6.77; p < 0.05) compared to those who were 15–19 years. Likewise, WRA who were 40–44 years had an increased relative risk of dying from infectious than maternal causes by more than ten times (RRR = 10.05; p < 0.01) compared to those who were 15–19 years. Also, WRA who were 45–49 years had an increased relative risk of dying from infectious than maternal causes by about nine times (RRR = 9.980; p < 0.05) relative to those who were 15–19 years. Similarly, for NCD versus maternal, the results showed that WRA who were 35–39 years (RRR = 6.80; p < 0.05) were more than six times as likely as those who were 15–19 to die from non-communicable than maternal causes. Furthermore, those who were 40–44 years (RRR = 9.05; p < 0.05) were more than nine times as likely as those 15–19 years to die from non-communicable than maternal causes. Moreover, those who were 45–49 years (RRR = 11.12; p < 0.05) were more than eleven times as likely as WRA (15–19 years) to die from non-communicable than maternal causes.

After the introduction of the proximate factors, children ever born and migration status that had been significant in Models 1 and 2 (Tables A1 & A2) ceased to be significant in Model 3 (Table A3). However, land ownership that was not significant in neither of the two models appeared significant in Model 3. This means that children ever born and migration status lost their significant effect on the causes of death after adding the proximate factors whilst land ownership gained significance on the causes of death after introducing the proximate factors.

From Model 3, WRA who owned land at the time of death were three and a half times (RRR = 3.50; p < 0.05) as likely as those who did not own any land to die from infectious than maternal causes of death. This observation is probably because women who own land are perhaps engaged in agricultural activities which are seasonal and financially unstable since land in the study area is mainly used for agricultural purposes. This means lower income for WRA who own land while some of those who do not own land are probably engaged in other occupations that are more stable and generate better incomes. However, land ownership had no significant effect when non-communicable causes were compared to maternal causes of death.

Marital status still continued to be a significant predictor of causes of death among the WRA after introducing the proximate factors. Unlike other variables, the p-values and the magnitude of the RRR for marital status remained the same in all three models. This means that marital status maintained the same importance as a significant determinant of the causes of death after introducing the proximate factors.

From the results of Model 3, it is again observed that compared to WRA who were never-married, those who were ever-married (RRR = 0.12; p < 0.001) had 88% reduced risk of dying from infectious than maternal causes of death. This means that WRA who were ever-married were significantly less likely to die of infectious than maternal causes compared to those who were never-married. Similarly, WRA who were married (RRR = 0.12; p < 0.001) had an 88% reduced risk of dying from non-communicable than maternal causes. This also means that WRA who were ever-married were significantly less likely to die from non-communicable than maternal causes relative to those who were single at the time of death.

Furthermore, district of residence maintained its significant influence on the causes of death in the final model as it was in the first two models but its effect on the causes of death slightly increased further in Model 3. This suggests that the effect of district of residence is fairly stable even after accounting for the proximate factors. From Model 3, WRA who lived in Kintampo South District were 0.35 times (RRR = 0.35; p < 0.01) as likely as those who lived in the Kintampo North to die from infectious than maternal causes. However, unlike Models 1 and 2 where district of residence had significant effect when non-communicable causes were compared to maternal causes of death, in Model 3, district of residence had no significant effect when non-communicable causes were compared to maternal causes of death. This means that it did not matter in which district WRA lived in the study area when proximate factors are accounted for in the model.

With respect to the season of death variable, the significance level improved from P-value of < 0.05 to < 0.01 for infectious versus maternal causes of death. In addition, the magnitude of the RRR increased whilst in the NCD versus maternal causes of death that was not significant in both Models 1 and 2, was significant in Model 3. This means that season of death increased its significant influence on the causes of death in the final model when the proximate factors were added.

From the final or adjusted model, WRA who died in the dry season were more than twice (RRR = 2.26; p < 0.01) as likely as those who died in the rainy season to die from infectious than maternal causes. This appears unexpected but the result also means that WRA who died in the rainy season were significantly more likely to die from maternal than infectious. Therefore, challenges with access to maternal services during the rainy season may partly explain this result. It could partly be because infections in the rainy season linger on in the dry season.

Similarly, WRA who died in the dry season (RRR = 2.02; p < 0.05) were more than twice as likely to die from non-communicable than maternal causes compared to those who died in the rainy season. This means that WRA who died in the dry season were significantly more likely to die from NCD than maternal causes. This means that WRA who died in the dry season were significantly more likely to die from infectious or NCD than maternal causes. Alternatively, this means that WRA who died in the rainy season were significantly more likely to die from maternal than infectious or NCD causes. This finding may be because of challenges with access to maternal services as explained earlier.

Year of death significantly predicted the causes of death in Model 1 but not Model 2. However, in Model 3, it appeared again as a significant predictor of the causes of death. The introduction of the proximate factors made it to recover its significance. From the results, WRA who died in 2008 had reduced risk of dying from infectious than maternal causes by 65% (RRR = 0.35; p < 0.05) compared to those who died in 2005. Similarly, for NCD versus maternal, the results showed that WRA who died in 2008 (RRR = 0.24; p < 0.05) showed 76% reduced risk of dying from non-communicable than maternal causes. In Model 1, it was only NCD versus maternal that showed significant effect. This means that the year of death also matters for the infections versus maternal deaths once the proximate factors are accounted for.

The results from Model 3 showed that place of death continued to significantly influence the causes of death. With respect to the place of death, it was observed that the significance level reduced from P-value < 0.001 in Models 1 and 2 to < 0.01 in Model 3. The results indicated that dying in a hospital reduced the risk of dying from infectious than maternal causes by 68% (RRR = 0.32; p < 0.01). Similarly, WRA who died in hospital had a reduced risk of dying from NCD than maternal causes by 69% (RRR = 0.31; p < 0.01) compared to those who died in other places. This means that compared to WRA who died in other places, those who died in hospitals were significantly less likely to have died from infectious or NCD than maternal causes. The result suggests that despite free maternal care introduced in 2005, women who die in the hospital were less likely to die of infectious or NCD causes relative to maternal causes.

All three variables measured under the proximate factors significantly correlated with the causes of death considered in the present study. One of such variables is whether the WRA was admitted in the last 12 months before dying. From the results, WRA who were admitted in the last 12 months before dying (RRR = 3.97; p < 0.01) were more than three times likely to die from infectious than maternal causes compared to those who were not admitted. However, hospital admission in the last 12 months before death had no significant effect when NCD were compared to maternal causes of death. This observation suggests that infectious causes increase the risk of hospital admission relative to maternal compared to NCDs relative to maternal. This is expected as infectious causes which tend to be more acute are more likely to lead to hospital admissions compared to NCDs.

Another significant proximate determinant of causes of death was whether the deceased WRA had surgical operation or not in the last 24 months. From the results, WRA who had surgical operation 24 months before dying (RRR = 0.14; p < 0.001) had 86% reduced relative risks of dying from infectious than maternal causes compared to those who did not have surgical operation 24 months before dying. This means that women who had surgical operations 24 months prior to death were more likely to die from maternal causes than infectious causes. This is expected, given that surgical operations usually characterise delivery and abortions. However, surgical operation 24 months before dying did not have a significant effect on NCD and maternal causes of death.

Finally, the other significant health-related factor on causes of death was whether the deceased WRA died suddenly or not. According to the results, WRA who died suddenly had a reduced relative risks of dying from infectious than maternal causes by about 68% (RRR = 0.32; p < 0.01). This means that compared to WRA who did not die suddenly, those who died suddenly were significantly more likely to have died from maternal than infectious causes of death. This is expected because maternal deaths are more sudden given that the women do not die from a prolonged illness. It is usually during delivery or surgical operations they die which is unexpected, untimely and sudden. However, sudden death did not have a significant effect on NCD and maternal causes of death.

4. Discussion

The findings of this study show the determinants of maternal, infectious and non-communicable causes of death among WRA in the Kintampo North Municipality and Kintampo South District from 2005 to 2014. The results of all the three models indicated that age is an important determinant of causes of death. Although the literature for both Global North and South is replete with the observation that mortality, generally, positively correlates with age for various causes of death, this present study specifically found that older women are more likely to die of infectious or NCD causes than maternal causes when compared with younger women.

Furthermore, the study found that marital status maintained its importance as a predictor of causes of death among the WRA consistently for all the three models at 99.9% confidence level. It was observed in the present study that ever-married women were significantly less likely to have died from infectious and non-communicable than maternal causes relative to those who were never-married. This means that ever-married women were at a higher risk of dying from maternal mortality. This finding is consistent with that of Asamoah et al. (2011). The authors reported that married women died most (93.7%) from maternal causes of death. This observation may be because, in the Ghanaian and other sub-Saharan African setting, often, a greater proportion of married women are likely to get pregnant or experience childbirth.

However, a case-controlled study using health facility data from a hospital in Dakar, Senegal, reported that women who were never-married were one and half times more likely to die from maternal causes relative to women who had ever been married (Garenne et al., 1997). This reported protection for married persons may be partly due to the social, psychological and other support systems that those in union may benefit from. It may also be probably because those in union are less exposed to unwanted pregnancies that are likely to end in abortion. Less abortion will mean less maternal mortality since abortion is one of the leading causes of maternal deaths in Ghana (Adjei et al., 2015). In addition, Illah et al. (2013) used HDSS data from Rufiji, a rural setting in Tanzania and reported that WRA who were ever-married were 62% less likely to die from maternal mortality compared to WRA who were never married (HR = 0.38, 95% CI = 0.176–0.839). The authors further observed that the relationship continued to be significant even after adjusting for maternal age (Illah et al., 2013).

The results of the present study also showed that season of death was a significant predictor in all three models and even improved from 95% to 99% significant level at the adjusted level. WRA who died in the dry season were significantly more likely to have died from infectious causes than maternal causes of death. This probably suggests that the rainy season increases the risk of dying from maternal causes due possibly to the limited access to maternal health services in such a season. This finding from the present study is consistent with the observation by Etard et al. (2003) in Burkina Faso. They also found greater number of deaths from obstetric causes in the rainy season (Etard et al., 2003). This observation may be because it is difficult to reach the health facility or for the health officers to reach the community during the raining season. On the contrary, Hounton et al. (2008) reported greater maternal deaths in the dry season in Senegal.

The findings of this study indicated that the district of residence is an important predictor of causes of death. It was found that WRA who resided in Kintampo South District were significantly less likely to die from infectious than maternal causes compared to those who lived in the Kintampo North Municipality. This means that women who lived in the Kintampo South were at a greater risk of dying from maternal mortality. Several factors may explain this observation. One hypothesis is that Kintampo South is a relatively new district and thus relies on Kintampo North Municipality for social infrastructure including health. Another related hypothesis is the availability of both health equipment and personnel at the Kintampo North Municipality, which continues to serve as referral for the Kintampo South during emergencies including emergency obstetric care.

All the three proximate factors related to health status namely hospital admission in the last 12 months before death, surgical operation in the last 24 months and the nature of death, whether sudden or not, significantly predicted the causes of death only for infectious versus maternal causes of death. This means that the effect of the proximate factors is more pronounced for infectious causes relative to maternal than NCD relative to maternal.

WRA who were admitted in a hospital in the last 12 months before dying were less likely to have died from maternal causes than infectious causes of death. It is hypothesised that activities of the Kintampo Health Research Centre (KHRC) may have partly accounted for this observation. The study area is where the KHRC is located. The Centre since its creation in 1994 has carried out a number of health interventions involving WRA and their new-borns. The activities of the Centre in terms of supporting the health facilities with clinicians, donation of health equipment, ambulance services and ensuring maximum health care for study participants who are invariably WRA, may have partly contributed to this observation.

At the adjusted level, infectious causes of deaths differed from the maternal causes of deaths in terms of age at death, marital status, land ownership, district of residence, year, season, place of death, admission in the last 12 months, surgical operation in the last 24 months and sudden death. On the other hand, non-communicable causes of death differed from maternal causes in terms of age at death, marital status, year, season and place of death.

5. Conclusion

Determinants of causes of death among WRA were complex and cut across distal and proximate factors but not intermediate factors. There is therefore, the need to expand the scope of the intermediate variables in future studies. In addition, marital status is a key determinant of the causes of death among WRA in the Kintampo North Municipality and Kintampo South District during the study period.

Acknowledgements

We are grateful to the Director and the management of KHRC as well as the Ghana Health Service for allowing the use of the KHDSS data. The study team wishes to acknowledge useful comments from KHRC staff, and the Faculty and students of Regional Institute for Population Studies (RIPS), University of Ghana. Our appreciation goes to Prof Delali Margaret Badasu and Professor Samuel N. A. Codjoe of RIPS for their comments on earlier drafts, as well as the community for allowing KHRC to collect data from them.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

There was no funding for this study.

Notes on contributors

Sulemana Watara Abubakari

Sulemana Watara Abubakari is a Principal Research Fellow at the Kintampo Health Research Centre (KHRC), Ghana. He holds a PhD degree in Population Studies and MPhil degree in Geography & Resource Development, University of Ghana, and a Bachelor of Arts degree in Geography and Economics, Kwame Nkrumah University of Science and Technology, Ghana. Sulemana has distinguished himself from primary school where he was awarded best pupil to university where he was also awarded Vice Chancellor’s award for outstanding doctoral dissertation. His poster presentation was adjudged the best, and he was presented with the young scientist award at 2006 INDEPTH Network Conference in Ouagadougou, Burkina Faso. He headed the Kintampo Health and Demographic Surveillance System (Kintampo HDSS), and now heads the Environmental Health Research of KHRC. Currently, he coordinates a study in all regions of Ghana aimed at reducing household air pollution, and activities of malaria vaccine pilot evaluation in three regions of Ghana.

Appendix

Table A1. Multinomial logistic regression for the relationship between socio-demographic, environmental factors and causes of maternal, infectious and non-communicable deaths from 2005 to 2014 in the study area (Model 1)

Determinants of causes of death	Infectious causes of death vs maternal causes of death	NCD causes of death vs maternal causes of death
	Relative risk ratio (RRR) (standard error)	Relative risk ratio (RRR) (standard error)
Age at death
15–19 (ref)	1.00	1.00
20–24	1.74 (1.04)	1.81 (1.13)
25–29	2.29 (1.37)	1.68 (1.05)
30–34	5.33* (3.50)	4.49* (3.09)
35–39	7.61** (5.34)	7.39** (5.38)
40–44	13.58** (10.89)	11.27** (9.35)
45–49	20.60** (19.58)	19.95** (19.48)
Parity
0 (ref)	1.00	1.00
1–3	2.46* (0.96)	1.52 (0.63)
4+	1.98 (0.93)	2.08 (1.02)
Marital status
Never-married (ref)	1.00	1.00
Ever-married	0.12*** (0.07)	0.12*** (0.07)
Highest educational level attained
No Education (ref)	1.00	1.00
Primary	1.04 (0.48)	1.43 (0.68)
Middle/JHS+	0.77 (0.32)	0.85 (0.38)
Migration status
Non-migrant (ref)	1.00	1.00
Migrant	0.43* (0.16)	0.48 (0.19)
Religion
Christianity (ref)	1.00	1.00
Muslim	0.71 (0.26)	0.65 (0.26)
Other	0.63 (0.26)	0.62 (0.27)
Model 1 CONTINUATION
Employment status
Unemployed (ref)	1.00	1.00
Employed	0.86 (0.29)	0.66 (0.23)
Land ownership
No (ref)	1.00	1.00
Yes	0.80 (0.32)	0.72 (0.30)
Place of residence
Rural	1.00	1.00
Urban	0.70 (0.28)	0.90 (0.38)
District of residence
North (ref)	1.00	1.00
South	0.30** (0.10)	0.47* (0.17)
Season of death
Wet (ref)	1.00	1.00
Dry	2.04* (0.59)	1.76 (0.54)
Year
2005 (ref)	1.00	1.00
2006	1.30 (0.73)	1.14 (0.66)
2007	1.84 (1.11)	0.72 (0.46)
2008	0.50 (0.28)	0.28* (0.17)
2009	1.53 (0.93)	0.88 (0.56)
2010	1.94 (1.12)	0.74 (0.45)
2011	3.61 (2.46)	1.10 (0.78)
2012	3.68 (2.52)	1.85 (1.31)
2013	3.62 (2.47)	2.70 (1.89)
2014	7.54 (8.48)	3.39 (3.89)
Distance to facility
<5 km	1.00	1.00
5 km+	0.61 (0.20)	0.61 (0.20)
Place of death
Other (ref)	1.00	1.00
Health facility	0.24*** (0.08)	0.25*** (0.08)
Log likelihood = −657.10051	Pseudo R2 = 0.1112 Prob > chi2 < 0.001	Number of observations = 846

***p < 0.001; **p < 0.01; *p < 0.05.

Source: Kintampo HDSS (2005–2014).

Table A2. Multinomial logistic regression for the relationship between socio-demographic, environmental, behavioural factors and causes of maternal, infectious and non-communicable deaths from 2005 to 2014 in the study area (Model 2)

Determinants of causes of death	Infectious causes of death vs maternal causes of death	NCD causes of death vs maternal causes of death
	Relative risk ratio (RRR) (standard error)	Relative risk ratio (RRR) (standard error)
Age at death
15–19 (ref)	1.00	1.00
20–24	1.76 (1.05)	1.88 (1.17)
25–29	2.27 (1.37)	1.73 (1.10)
30–34	5.02* (3.32)	4.51* (3.12)
35–39	7.02** (4.98)	6.95** (5.11)
40–44	11.96** (9.61)	10.46** (8.71)
45–49	16.30** (15.56)	17.72** (17.76)
Parity
0 (ref)	1.00	1.00
1–3	2.35* (0.94)	1.51 (0.63)
4+	1.91 (0.92)	2.13 (1.07)
Marital status
Never-married (ref)	1.00	1.00
Ever-married	0.12*** (0.07)	0.12*** (0.07)
Highest educational level attained
No Education (ref)	1.00	1.00
Primary	1.02 (0.47)	1.40 (0.67)
Middle/JHS+	0.70 (0.30)	0.79 (0.35)
Migration status
Non-migrant (ref)	1.00	1.00
Migrant	0.45* (0.18)	0.49 (0.20)
Religion
Christianity (ref)	1.00	1.00
Muslim	0.79 (0.31)	0.66 (0.27)
Other	0.68 (0.29)	0.63 (0.28)
Model 2 CONTINUATION
Employment status
Unemployed (ref)	1.00	1.00
Employed	0.86 (0.29)	0.66 (0.23)
Land ownership
No (ref)	1.00	1.00
Yes	0.89 (0.37)	0.78 (0.33)
Place of residence
Rural	1.00	1.00
Urban	0.63 (0.28)	0.98 (0.45)
District of residence
North (ref)	1.00	1.00
South	0.32** (0.11)	0.47* (0.17)
Season of death
Wet (ref)	1.00	1.00
Dry	2.02* (0.59)	1.78 (0.55)
Year
2005 (ref)	1.00	1.00
2006	1.32 (0.75)	1.26 (0.74)
2007	1.98 (1.21)	0.77 (0.50)
2008	0.56 (0.32)	0.31 (0.19)
2009	1.52 (0.95)	0.97 (0.63)
2010	2.01 (1.19)	0.82 (0.51)
2011	3.53 (2,44)	1.08 (0.79)
2012	3.76 (2.60)	1.88 (1.35)
2013	3.49 (2.41)	2.66 (1.89)
2014	6.32 (7.17)	3.15 (3.64)
Distance to facility
<5 km	1.00	1.00
5 km+	0.69 (0.23)	0.64 (0.22)
Place of death
Other (ref)	1.00	1.00
Health facility	0.24*** (0.08)	0.25*** (0.08)
Model 2 CONTINUATION
Household socio-economic status
Most Poor	1.00	1.00
Poor	0.84 (0.31)	1.40 (0.55)
Least Poor	0.86 (0.35)	1.25 (0.54)
Source of drinking water
Improved	1.00	1.00
Unimproved	0.58 (0.18)	0.74 (0.24)
Type of toilet facility
Improved	1.00	1.00
Unimproved	0.83 (0.33)	1.22 (0.51)
Substance use
Yes	1.00	1.00
No	3.71 (2.89)	2.41 (1.93)
Log likelihood = −647.5523	Pseudo R2 = 0.1241 Prob > chi2 < 0.001	Number of observations = 846

***p < 0.001; **p < 0.01; *p < 0.05.

Source: Kintampo HDSS (2005–2014).

Table A3. Multinomial logistic regression for the relationship between socio-demographic, economic, physical, health-systems, behavioural and personal health-related determinants and maternal, infectious and non-communicable causes of deaths from 2005 to 2014 in the Kintampo districts (Model 3)

Determinants of causes of death	Infectious causes of death vs maternal causes of death	NCD causes of death vs maternal causes of death
	Relative risk ratio (RRR) (standard error)	Relative risk ratio (RRR) (standard error)
Age at death
15–19 (ref)	1.00	1.00
20–24	1.78 (1.12)	1.78 (1.15)
25–29	2.02 (1.30)	1.49 (1.00)
30–34	3.77 (2.64)	3.43 (2.49)
35–39	6.77* (5.12)	6.80* (5.29)
40–44	10.05** (8.60)	9.05* (7.96)
45–49	9.80* (9.75)	11.12* (11.32)
Parity
0 (ref)	1.00	1.00
1–3	2.36 (1.04)	1.59 (0.73)
4+	1.78 (0.93)	2.04 (1.10)
Marital status
Never-married (ref)	1.00	1.00
Ever-married	0.12*** (0.07)	0.12*** (0.07)
Highest educational level attained
No Education (ref)	1.00	1.00
Primary	1.21 (0.59)	1.74 (0.87)
Middle/JHS+	0.78 (0.35)	0.88 (0.41)
Migration status
Non-migrant (ref)	1.00	1.00
Migrant	0.45 (0.19)	0.48 (0.20)
Religion
Christianity (ref)	1.00	1.00
Muslim	0.70 (0.29)	0.59 (0.26)
Other	0.61 (0.28)	0.55 (0.26)
Model 3 CONTINUATION
Employment status
Unemployed (ref)	1.00	1.00
Employed	0.80 (0.29)	0.61 (0.23)
Land ownership
No (ref)	1.00	1.00
Yes	3.50* (2.11)	2.45 (1.51)
Place of residence
Rural	1.00	1.00
Urban	0.69 (0.33)	1.05 (0.51)
District of residence
North (ref)	1.00	1.00
South	0.35** (0.13)	0.51 (0.20)
Season of death
Wet (ref)	1.00	1.00
Dry	2.26** (0.71)	2.02* (0.65)
Year
2005 (ref)	1.00	1.00
2006	1.52 (0.89)	1.40 (0.84)
2007	1.72 (1.17)	0.83 (0.59)
2008	0.35* (0.22)	0.24* (0.16)
2009	1.06 (0.71)	0.80 (0.56)
2010	1.55 (1.01)	0.74 (0.51)
2011	2.77 (2.06)	0.96 (0.74)
2012	3.17 (2.38)	1.74 (1.34)
2013	2.71 (2.01)	2.23 (1.69)
2014	3.39 (3.94)	1.87 (2.21)
Distance to facility
<5 km	1.00	1.00
5 km+	0.74 (0.26)	0.68 (0.25)
Place of death
Other (ref)	1.00	1.00
Health facility	0.32** (0.11)	0.31** (0.11)
Model 3 CONTINUATION
Household socio-economic status
Most Poor	1.00	1.00
Poor	0.91 (0.35)	1.52 (0.62)
Least Poor	0.89 (0.37)	1.34 (0.59)
Drinking water source
Improved	1.00	1.00
Unimproved	0.57 (0.18)	0.74 (0.25)
Toilet facility type
Improved	1.00	1.00
Unimproved	0.90 (0.39)	1.32 (0.59)
Admission in last 12 months before death
No (ref)	1.00	1.00
Yes	3.97** (1.85)	4.48 (2.12)
Surgical operation in last 24 months before death
No (ref)	1.00	1.00
Yes	0.14*** (0.07)	0.28 (0.14)
Sudden death
No (ref)	1.00	1.00
Yes	0.32** (0.14)	0.42 (0.19)
Log likelihood = −626.28364	Pseudo R2 = 0.1529 Prob > chi2 < 0.001	Number of observations = 846

***p < 0.001; **p < 0.01; *p < 0.05.

Source: Kintampo HDSS (2005–2014).

Figure 1. Conceptual framework of factors associated with dying from maternal, infectious or non-communicable causes of death.

Source: Adapted from Defo (2014).