Twins have been found to have elevated mortality rates compared to singletons during the early stages of life (Bellizzi et al., Reference Bellizzi, Sobel, Betran and Temmerman2018; Monden & Smits, Reference Monden and Smits2017; Ouedraogo, Reference Ouedraogo2020a). This is mainly attributed to a higher incidence of preterm birth and low birth weight in twins, both of which are associated with higher perinatal mortality rates (Bjerregaard-Andersen et al., Reference Bjerregaard-Andersen, Biering-Sørensen, Gomes, Bidonga, Jensen, Rodrigues, Christensen, Aaby, Beck-Nielsen, Benn and Sodemann2014; Blondel, Reference Blondel2009; Oger et al., Reference Oger, Robillard, Barau, Randrianaivo, Bonsante, Iacobelli and Boukerrou2013). Furthermore, twin deliveries are more prone to complications than singleton deliveries, which further exacerbates the mortality risk for twin newborns (Santana et al., Reference Santana, Silveira, Costa, Souza, Surita, Souza, Mazhar, Jayaratne, Qureshi, Sousa, Vogel and Cecatti2018). Consequently, the under-5 mortality rates (U5MR) for twins are estimated to be two to three times greater than those for singletons (Bjerregaard-Andersen et al., Reference Bjerregaard-Andersen, Lund, Jepsen, Camala, Gomes, Christensen, Christiansen, Jensen, Aaby, Beck-Nielsen, Benn and Sodemann2012; Blondel, Reference Blondel2009).
In the context of developing countries, particularly in sub-Saharan Africa (SSA), the prevalence of home births, along with an underdeveloped perinatal health infrastructure, contributes to persistently high child mortality rates. For instance, data from the World Health Organization (WHO, 2018) indicates that in 2017, half of all deaths among children under 5 years of age occurred in SSA. This situation poses a significant risk to the survival of twins, compounding existing health challenges, especially as this region also has the highest rates of twin births in the world (Smits & Monden, Reference Smits and Monden2011). A comprehensive international study conducted by Bellizzi et al. (Reference Bellizzi, Sobel, Betran and Temmerman2018) indicated that at the time of birth, twins in Africa experience mortality risks that are nearly eight times greater than those of singletons. Furthermore, although U5MR showed a marked decrease in both twins and singletons in SSA between 1995 and 2014, the excess mortality rate for twins compared to singletons increased over the same period, rising from 2.5 (328 vs. 129 per 1000) to 3.2 (213 vs. 66 per 1000) (Monden & Smits, Reference Monden and Smits2017).
However, the above studies have rarely analyzed trends in twin mortality by age, including cross-country, subregional and period comparisons. This analysis aims to investigate excess mortality among twins in SSA, examining variations by country, period, and child age.
Factors that have been identified as influencing excess twin mortality include low birth weight, prematurity, inadequate breastfeeding, financial burden, and unfavorable social status within the community (Ouedraogo, Reference Ouedraogo2020b; Pison, Reference Pison, Van de Walle, Pison and Sala-Diakanda1992). In this analysis, we will also examine the factors associated with under-5 mortality among twins in SSA.
In summary, the primary objectives of this study are as follows: first, to estimate the level and trends of U5MR in each of the 42 SSA countries under study, comparing twins and singletons. Second, the study will pool the data to estimate the level, age distribution and temporal trends of U5MR comparing twins and singletons on a regional scale within SSA. Third, the investigation will encompass the factors contributing to the excess mortality risks among under-5 twin children in SSA, with particular interest in the influence of cultural perceptions surrounding twin births.
Methods
Data Source
This retrospective cross-sectional study used secondary data from the Demographic and Health Surveys (DHS) and the Multiple Indicator Cluster Surveys (MICS). DHS and MICS design and sampling methods are described elsewhere (https://dhsprogram.com/Methodology/index.cfm).
We used data from 140 DHS and 16 MICS carried out between 1986 and 2016 in 42 SSA countriesFootnote 1 . Figure 1 shows the geographical coverage of the study. The list of countries and surveys is available in Appendix A
Figure 1. Geographical coverage of the study.
We used the Birth file in each survey to identify all births reported retrospectively by the mothers interviewed, their characteristics, and any deaths.
Data Selection and Use
First, we computed the U5MR for twins compared to singletons across each survey and country. This facilitates an analysis of under-5 excess mortality among twins within each country over time.
Second, we analyzed the under-5 excess mortality among twins over time by pooling data from all surveys conducted in the 1990s (39 surveys), the 2000s (52 surveys), and the 2010s (54 surveys). Eleven surveys from the 1980s were excluded from this analysis. The data have been weighted to allow comparisons between decades.
Third, we pooled data from each country that had at least one survey in each of the three decades (1990s, 2000s, and 2010s). We then calculated mortality rates for ages 0 to 5 among twins and singletons in each decade (1990s, 2000s, 2010s). If a country had multiple surveys per decade, we selected the survey closest to the midpoint of the decade. Consequently, only 24 countries (72 surveys: one survey per country for each of the three decades) were included in this analysis of 0−5 age-specific mortality.
Fourth, to analyze the factors associated with excess under-5 mortality among twins, we selected one survey per country, ideally the most recent survey or the one with the most relevant variables needed to analyse child mortality. We then constructed a subsample that included only children born in the 5 years prior to each survey, as DHS and MICS only collected socio-demographic and biomedical variables for children born during this period.
It is important to note that the analysis does not explicitly include triplets or higher order births. However, a few notes on their number and the proportion of under-5 deaths among them have been included in the brief description of the data. In addition, for information and ease of comparison, we have included in Appendix A estimates of the U5MR for triplet or higher births by survey and country, where possible.
Statistical Analysis
We calculated the U5MRs using the method recommended by Measure DHS (Rutstein, Reference Rutstein1984). We utilized the SAS macro DHS_U5M developed for this purpose (Atwood & Thomson, Reference Atwood and Thomson2012). We have used the actuarial life table method to construct 0−5 age-specific mortality rates. We used the Cox semiparametric regression to analyze the factors associated with excess mortality in twins under 5 years of age. Our dependent variable was whether the child died before the 5th birthday. The choice of covariates was based on a thorough literature review (Ouedraogo, Reference Ouedraogo2020a), considering the constraints associated with our data. The independent variables included twinning status (the primary variable of interest), birth weight, cesarian section, antenatal care, breastfeeding, maternal age, birth order, child’s gender, mother’s marital status, African subregion, child’s year of birth, wealth quintile, urban/rural area of residence, socio-cultural status of twins, mother’s education level, and desired pregnancy.
Among the covariates, the twinning (twin/singleton) variable was first introduced in the model. Then, by sequentially adding the other covariates or groups of covariates (forward selection), we demonstrated their effects on the differential mortality risk ratio (Hazard Ratio, HR) between twins and singletons.
Results
Temporal Variations in the U5MR, According to the Twinning Status, in Each Country With Available Data
Birth history data extracted from the 156 surveys led to a sample of 2,425,072 children, of which 2,339,534 (96.5%) were born from singleton pregnancies, 84,047 (3.5%) from twin pregnancies, and 1491 (0.06%) were triplets or more. Overall, 265,060 children (10.9%) died before the age of 5: 239,933 singletons, accounting for 10.3% of all singletons; 24,248 twins, representing 28.9% of all twins; and 879 triplets or more, comprising 60.0% of all triplets or more.
Figure 2 presents, for each country, the U5MRs estimated for both single and twin births in each survey, listed in chronological order. These results show that the levels of U5MRs tend to decrease over time. However, the excess mortality among twins remains significant.
Figure 2. Temporal variations in U5MR derived from the 156 surveys in 42 sub-Saharan African countries.✸
Note: ✸The unique points concern countries with only one survey.
Source: Demographic and Health Surveys (DHS) and Multiple Indicator Cluster Surveys (MICS); authors’ construction.
Temporal Variations in the U5MR According to the Twinning Status After Pooling Surveys by Decades (1990s, 2000s and 2010s)
After pooling the 145 surveys performed between the decades 1990 and 2010, we observed a decreasing trend in U5MR both for singletons and for twins: it decreased from 159 (in 1990s) to 87 (in 2010s) per 1000 among singletons and from 365 (in 1990s) to 244 (in 2010s) per 1000 among twins (Figure 3). We also made estimates indicating that the ratio between the U5MRs of twins and singletons (i.e., twins’ excess mortality) did not decrease; in fact, it even increased from 2.3 (365/159 = 2.3; 95% CI [2.2, 2.4]) in the 1990s to 2.8 (244/87; 95% CI [2.7, 2.9]) in the 2010s.
Figure 3. U5MR variation from the 1990s to the 2010s in U5MR in SSA — aggregated data from 42 countries.
Source: 145 Demographic and Health Surveys (DHS) and Multiple Indicator Cluster Surveys (MICS) performed between the decades 1990 and 2010; authors’ construction.
Age-Specific Under-5 Mortality Rates for Twins and Singletons
The results (Figure 4) show that the differences in mortality rates between twins and singletons are considerable at the early years but decrease significantly with age. After the second year of age, the differences are no longer statistically significant. From one decade to the next, the mortality level decreases at all ages for both twins and singletons.
Figure 4. Age-specific mortality between ages 0 and 5 in sub-Saharan Africa: Comparison between twins and singletons.
Source: Demographic and Health Surveys (DHS) and Multiple Indicator Cluster Surveys (MICS) from 72 surveys in 24 countries: One survey per country for each of the three decades); authors’ construction.
Factors Associated With Under-5 Excess Mortality Among Twins
The univariable, bivariable and multivariable results related to the excess mortality factors for twins are presented in Table 1.
Table 1. Factors associated with twins’ excess mortality. Results of the univariate, bivariate and multivariate analyses
Note: HR, hazard ratio; CI, confidence interval; ***1 per 1000; **1% and *5 %; keep, missing values retained as a modality of the variable; sup, removed missing values. Bold type in row 1 of the table indicates the number and proportion of twins (column 2), followed by (from column 3 onwards) the hazard ratios of twins’ mortality (compared to the reference group, which is singletons) and their corresponding confidence intervals for each of the Cox models.
Source: Demographic and Health Surveys (DHS) and Multiple Indicator Cluster Surveys (MICS); authors’ calculation
Descriptive Results on Excess Mortality Factors for Twins Under 5 in SSA (Sample Description)
The subsample we constructed to analyze the factors associated with excess under-5 mortality among twins consisted of 278,706 children (including 268,032 singletons [96%] and 10,674 twins [4%]), from 25 surveys in 25 different countries for which maternal ethnicity and some biomedical variables (in particular birth weight) were available. A total of 19,502 children (7%) had died: 2,348 among twins (22% of all twins) and 17,154 among singletons (6.4% of all singletons). The characteristics of the children are presented below.
Biomedical and nutritional covariates. Birth-weight data were available for 41.4% of the neonates: 99% had a birth weight above 2.5 kg, and 1% had a birth weight below 2.5kg. Cesarean section data were available for 83% of the neonates: 5% were born by cesarean section and 95% by vaginal delivery. Antenatal care data were available for 64% of the newborns: 7% had no antenatal medical visits, 13% had 1 or 2 visits, 80% had at least 3 visits.
Finally, 97% of the children were breastfed, while 3% were not.
Demographic covariates in children. 43% were children whose mothers were under 25 years old; 45% were of birth order higher than 3; 49% were female; 88% were children of women in union. 4% of the children were from Southern Africa, 22% from Central Africa, 22% from East Africa and 52% from West Africa.
Socio-economic covariates. Some 48% of children lived in very low or low-income households, and 70% lived in rural areas.
Socio-cultural covariates . Data on socio-cultural perception of twins related to the ethnic groups were available for 96% of the neonates: 20% of the children had mothers who belonged to ethnic groups that worshipped twins, 5% from ethnic groups that traditionally abhorred twins, and 75% from ethnic groups that used to be ambivalent about twins (accepting them with great fear). 41% of the children were from mothers with no education. Data on pregnancy intention were available for 88% of the newborns: 76% of the children were desired, 19% were desired later, and 6% were unwanted.
Results of the Cox Model on Twins’ Under-5 Excess Mortality Factors in SSA
Child mortality risk factors (twins as well as singletons). The multivariable analysis shows the following factors that are independently associated with significant risks of under-5 death in SSA: being a twin (Adjusted hazard ratio [HRa] = 3.1; p < .001); birth weight under 2.5 kg (HRa = 3.0; p < .001); cesarean delivery (HRa = 1.4; p < .001); no antenatal visits (HRa = 1.13; p < .01); no breastfeeding (HRa=6.7; p < .001); young maternal age (HRa of mothers under 20 years of age is at least 20% higher than that of mothers in other age groups; p < .001); high birth order (HRa of order 6 or more is 28% higher than that of order 1; p < .001); male sex (HRa = 1.2; p < .001); single mother (HRa = 1.13; p < .01) or divorced/widowed (HRa = 1.27; p < .001); living in West Africa (HRa is at least 10% higher); living in a very low-income household (HRa = 1.11; p < .001); maternal primary education level or less (HRa = 1.13; p < .001).
Factors associated with excess mortality in twins in SSA. The Cox model also shows that in SSA, twin children had a very high mortality risk compared to singletons. Indeed, the hazard ratio was 3.9 times higher for twins than for singletons (95% CI [3.7, 4.0], p < .001). After controlling for the biomedical and nutritional variables (birth weight, cesarean section, antenatal visits and breastfeeding), this ratio changed significantly, dropping to 3.15 (95% CI [2.9, 3.3]). After a gradual (forward selection) inclusion of the other groups of covariates in the model, the risk of mortality for the twins was no longer significant. Consequently, biomedical and nutritional covariates are the predominant factors associated with excess mortality among twins.
The interaction effects introduced in the model (between the twinning variable and the other covariates) show that within the biomedical and nutritional covariates, the absence of cesarean section and breastfeeding are particularly and independently associated with twins’ under-5 excess mortality. When the delivery is not by cesarean section, the HRa for twins is 3 times higher than for singletons (95% CI [2.9, 3.3]; p < .001). But if the delivery is by cesarean section, it drops to 1.75 (95% CI [1.5, 2.1]; p < .001).
Regarding breastfeeding, the HRa of nonbreastfed twins is 4 times higher than that of nonbreastfed singletons (95% CI [3.6, 4.3]; p < .001). However, if the children were breastfed, this risk decreased to 3.12 (95% CI [2.95 - 3.28]; p < .001).
Furthermore, two aspects of the effect of socio-cultural beliefs on twins’ excess mortality should be noted. First, considering the twins’ socio-ethnic status in the Cox model does not contribute to variation in the HRa of twins. Second, we did not find socio-ethnic disparities in mortality risks among twins (see Appendix C), which would depend on how these ethnic groups perceive them (worshipped, hated, or ambivalent).
Discussion
Unequal Decrease in U5MR for Twins and Singletons
A sub-Saharan twin has an overall U5MR three times higher than a singleton. From the 1990s to the 2010s, the U5MRs decreased significantly for all children in SSA countries. However, the decrease was less marked for twins than for singletons. The high twin mortality rates we found here are consistent with the findings of the study by Monden and Smits (Reference Monden and Smits2017). They showed that the ratio of U5MR of twins to singletons was 3.2 in SSA in 2014. The high levels of infant mortality and excess twin mortality in SSA are undoubtedly related to the low income position of this region. This suggests that the health system is insufficiently equipped (with obstetric and paediatric services) to deal with the complications associated with twin births (Ouedraogo et al., Reference Ouédraogo, Pison, Abdramane, Le Coeur, Delaunay and Kassoum2021a).
The excess mortality of twins compared to singletons is observed at all ages below 5. It is mainly concentrated at the beginning of life. This finding is similar to that of Bellizzi et al. (Reference Bellizzi, Sobel, Betran and Temmerman2018). In 60 developing countries, including SSA, they found that twins’ mortality risks were almost 8 times higher than those of singletons during the first week of life. Similarly, although to a lesser extent, our results showed that the risk of death in the first year of life is almost 5 times higher for twins than for singletons. The high prematurity rate of twins and their frequent low birth weight would be the factors that maintain this high excess mortality in early life. The sharp decline in twin mortality after the age of 2 is expected, as observed, worldwide in children. There are two main reasons for the sharp decline in twin mortality after the age of 2. First, a trivial one, is the increase in the chances of survival with age. The second reason would be a ‘natural selection’ (a ‘survival bias’) within twins. Indeed, with age, the twins who survive are those who are (or become) as ‘resistant’ as the singletons (Pongou et al., Reference Pongou, Shapiro and Tenikue2019; Ouédraogo et al., Reference Ouédraogo, Pison, Abdramane, Le Coeur, Delaunay and Kassoum2021a).
Biomedical and Nutritional Factors Explain Twins’ Under-5 Excess Mortality
Biomedical and nutritional factors in general, and noncesarean section delivery and the absence of breastfeeding, are the main factors associated with excess mortality risk among twins. These results demonstrate the ‘fragile’ nature of twin births.
Regarding the role of birth weight in explaining the survival difference between twins and singletons, Pongou (Reference Pongou2013) demonstrated that ‘the higher mortality of twins is mainly attributed to low birth weight’ (p. 428). Monden and Smits (Reference Monden and Smits2017) have shown that in SSA, while the effect of birth weight is undeniable, it is also associated with a lack of medical assistance at delivery. However, our results may have underestimated the role of low birth weight in explaining twin excess mortality due to the high proportion (59%) of missing or inadequate birth-weight values.
We found that cesarean section was associated with a higher risk of death when all children (twins and singletons) are considered. However, in twins, we found that it reduces the risk of death. This interesting result could be explained by the fact that in nontwin pregnancies, cesarean sections are usually performed because of delivery complications. On the other hand, in the case of twin pregnancies (at least those with medical assistance), the choice of cesarean delivery is a planned and preventive practice to avoid any complications that vaginal delivery might cause (Ouedraogo, Reference Ouedraogo2020a; Ouedraogo et al., Reference Ouédraogo, Pison, Le Coeur and Soura2021b).
We also found that the absence of breastfeeding was associated with a higher risk of death for all children, for twins and for singletons. In fact, twins weakened by low birth weight or prematurity may not have the strength to breastfeed. In addition, difficult deliveries affect the ability of mothers of twins to breastfeed or to breastfeed two infants at the same time. This lack of adequate breastfeeding deprives them of a vital supply of antibodies and nutrients (van der Pol, Reference van der Pol1989).
Socio-Cultural Beliefs Surrounding Twins No Longer Explain Their Excess Mortality
The factors associated with twins’ under-5 excess mortality showed that their socio-cultural status has no independent effect on their excess mortality risks. But are there any differences in mortality among twins depending on their social status? Our results did not find disparities in mortality risks among twins related to how they are welcomed (worshipped, hated, or ambivalent). All these elements seem to indicate that the socio-ethnic status of twins in SSA does not (or no longer) influence their mortality or their excess mortality compared to singletons. This is thought to be due to an improvement in the social status of twins. Contrary to long-standing practices, twins on the continent are increasingly not seen as ‘evil’ children (Ouedraogo, Reference Ouedraogo2020b). Urbanization, the spread of modern medicine, and education have contributed to these changes (Ouedraogo, Reference Ouedraogo2020a). However, we must be cautious in this statement because this finding may be the result of deliberate omission of twin deaths among ethnic groups traditionally unfavorable to twins.
Strengths and Limitations
This study is not without limitations, including data quality issues, the heterogeneity of the countries studied, and the overrepresentation of some countries or sub-regions in the sample of surveys examined. First, ages reported in the national surveys may be inaccurate. This could have an impact on the robustness of the estimated mortality rates.
Second, the heterogeneity of the demographic contexts of the countries included in our study constitutes another limitation. Undeniably, the demographic contexts of sub-Saharan countries are diverse, ranging from east to west and from centre to south. Therefore, the interpretation of some of our results as being at a ‘SSA level’ obscures disparities between these countries.
Third, we note that the over-representation of West African countries in the multivariate analysis can affect the HR estimated for the geographical area and socio-ethnic covariates. Nevertheless, weighting the data contributed to correcting this limitation. There were also several missing values for some covariates, which may not be random. For example, missing birth weight may indicate a lack of a weighing scale, such as in a home birth or a birth with limited medical assistance. This may be a confounding factor.
The study’s extensive geographical coverage, its analysis of excess mortality among twins by age between 0 and 5 years, and its examination of the impact of the perception of twinning on twin survival represent significant contributions to the scientific literature.
Conclusion
Infant mortality remains high in SSA, and this is even more pronounced for twins. Although infant mortality has declined on the continent, excess mortality among twins remains very high. Biomedical and nutritional factors, such as low birth weight, lack of breastfeeding, and non-cesarean births, are mainly associated with higher mortality risks for twins than for singletons. This study confirms that the health of twin children in SSA is a major challenge. Targeted interventions are therefore needed. These actions should concern the medical field, in particular the development of obstetric and pediatric services needed to cope with the frequent complications associated with twin births. They should also involve social services, which need to implement measures to counteract possible hidden social risks that could still affect the health of twins.
Acknowledgements
We acknowledge the DHS program (https://www.dhsprogram.com/) for free access to the surveys used in this study.
Funding statement
None.
Declaration of interest statement
This article is based on the doctoral research of Adama OUEDRAOGO, conducted between October 2017 and September 2020 at the French Institute for Demographic Studies (INED) and the Research Centre of Paris 1 Pantheon-Sorbonne University’s Institute of Demography (CRIDUP).
Appendix A. Under-5 mortality rate from 42 sub-Saharan African countriesa
Note: aCountries not included because lack of data: Botswana, Cape Verde, Djibouti, Equatorial Guinea, Eritrea, Mauritius, and Seychelles. Bold italic type indicates, for each survey, twins’ under-5 excess mortality ratio, that is, the U5MR of twins divided by that of singletons. Italic type indicates a 95% confidence interval of that ratio.
Standard DHS, standard Demographic and Health Surveys; MIS, Malaria Indicators Survey; AIS, AIDS Indicator Survey; MICS, Multiple Indicator Cluster Surveys; Continuous, Continuous DHS.
Source: Demographic and Health Surveys (DHS) and Multiple Indicator Cluster Surveys (MICS); authors’ calculation.
Appendix B. Socio-ethnic status of twins in sub-Saharan Africa
Source: Pison (Reference Pison1987). Reproduced with permission.
Appendix C. Analysis of the effect of the socio-ethnic status of twins on their under-5 mortality risk
Note: ***1 per 1000, **1% and *5%; keep, missing values retained as a modality of the variable; sup, removed missing values.
Source: Demographic and Health Surveys (DHS) and Multiple Indicator Cluster Surveys (MICS); authors’ calculation.
Open access
