Sixty countries were included in this analysis; 33 countries from AFR, 8 from EUR, 7 from AMR, 6 from SEAR, 4 from EMR and 2 from WPR ( Table 1 ). The study population included 536 179 livebirths (521 867 singletons and 14 312 twins) and ranged from 1 208 newborns in the 2007 Ukraine DHS to 50 026 in the 2005/06 India DHS. Africa contributed the largest number of newborns (56% of the total sample size in this analysis) followed by SEAR with 18%.

Table 2 shows by country the number of livebirths (singletons and twins) contributing to the analysis, and the early neonatal mortality in each group of newborns with regional averages. The lowest country percentages of twin-pairs were registered in the Bolivia, Moldova and Nepal surveys with rates of 6 per thousand births; the highest rates were found in Africa: Benin, Cameroon, Comoros, Congo (Brazzaville), Cote d’Ivoire, Ghana, Sao Tome and Principe and Togo, which reported 20 or more twin-pairs per thousand births ( Table 2 ). By WHO region, highest rates of twin pregnancies were found in AFR (1.65%) followed by EMR and EUR with 0.9% and 0.85%, respectively.

This data set for 60 countries presented an early newborn mortality (number of deaths per 100 livebirths) of 2% in singletons and of 12.1% in twins. Based on available DHS data from selected countries, SEAR and AFR were the WHO regions with the highest mortality for both singletons (2.2 early neonatal deaths for 100 livebirths in both regions) and twins (14.6 and 12.8 deaths for 100 livebirths, respectively). On the other hand, AMR and EUR reported the lowest mortality for singletons (1.2 and 1.3 early neonatal deaths for 100 livebirths, respectively) while WPR reported the lowest mortality for twins (7.1 deaths for 100 livebirths) ( Table 2 ).

There is a significant higher percentage of twin pregnancies in older mother (35 years or older) and those with higher parity ( Table 3 ).

Table 4 shows crude and adjusted ORs for early neonatal mortality among twins when compared to singletons. The pooled adjusted OR (95% CI) was found to be 7.6 (7.0-8.3). All countries except Azerbaijan, Guyana, and Ukraine had statistically significant associations. In addition, Armenia and Moldova had no deaths among twins rendering it impossible to determine OR.

In the subgroup analysis for the 20 countries with less than 20% missing data on birth weight, was associated with low birth weight significantly increased risk (aOR = 2.8; 95% CI = 2.2-3.5) of early newborn mortality of twins vs singletons. Low birth weight twins and singletons had 4.0 (2.0-8.0) and 7.6 (6.0-9.5) times the OR of early mortality, respectively, compared with twins and singletons of normal birth weight.

The overall proportion of twin births in health facilities was almost 70%. This proportion was highest in EUR (91.2%) and lowest in SEAR (56.8%) ( Table 5 ). The average rate of caesarean section in twins was 17.5%; highest in AMR (55%) and EMR (51.7%) and lowest in WPR where only 11.6% of twin deliveries were through CS.

Overall, sixteen countries had high rates of cesarean sections (>15%), 36 middle (5%-15%) and eight low (<5%). For twins 22 countries had high rates of cesarean sections (defined for this analysis as those >15%), 34 had middle rates (5%-15%) and four had low rates (<5%), ( Table 5 ).

After adjusting for maternal age, maternal education, household wealth, rural\urban residence, sex, birth spacing and desired pregnancy, twins delivered vaginally in health facility had a statistically significantly increased risk for early newborn mortality (pooled aOR = 2.1; CI = 1.5-3.1) compared to twins delivered by caesarean section ( Table 6 ). This statistically significant association became more pronounced (aOR = 4.8; 95% CI = 2.4-9.4) for countries with high rates of CS deliveries but lost significance for middle and low-rates countries ( Table 6 ).

The same adjusted logistic regression performed among singleton deliveries showed a pooled (aOR = 0.7; CI = 0.6-0.8) protective effect of vaginal birth in health facility compared with a caesarean section; this beneficial effect held true for across the three rates-categories ( Table 6 ).

In twin pregnancies, home delivery without SBA was associated with increased mortality both compared with delivering at home with SBA (aOR = 1.3; CI = 1.0-1.8) and with vaginal birth in health facility (aOR = 1.7; CI = 1.4-2.0). Home deliveries with SBA were at increased risk of early mortality compared with vaginal birth in health facility (aOR = 1.5; CI = 1.0-2.3) ( Table 7 ).

Individual level data analysis from 60 low- and middle-income countries using population-based nationally-representative surveys revealed twins had around 3 times increased odds of death compared to singletons after adjusting for birth weight. CS was associated with a significant protective effect in twins (OR = 2.1; 95% CI = 1.5-3.1), however, when stratifying by national CS rates, the significance remained only in the countries where the rate of CS was 15% or higher. In contrast, delivery by CS was associated with increased risk of mortality among singletons regardless of section rates.

These associations need cautious interpretation. The DHS surveys do not provide information about factors that may lead to a CS (eg, obstructed labour, fetal distress or pre-eclampsia/eclampsia) and thus we could not control for other morbidity associated beyond the increased inherent risk of a multiple pregnancy. Determining whether the CS was the cause or the consequence of the morbidity or mortality suffered by the mother or the newborn is a chronic challenge in this type of analysis.

This study has several limitations. No information was available regarding medical conditions (chronic hypertension, malaria), obstetric antenatal conditions (prelabour rupture of membranes, pregnancy-induced hypertension, pre-eclampsia, eclampsia, vaginal bleeding in the second half of pregnancy) and fetal presentation which have an impact in the mode of delivery and the outcome of pregnancy. Similarly, no data on race and smoking status were available as well as lack of information on the use of antenatal ultrasound with early detection of twins and its impact on the outcome.

No information was available on who the first and who was the second twin was, which would have an impact on the interpretation regarding potential mortality in retained second twin. One additional limitation resulted from reporting bias and the lack of information on women who died during delivery, likely associated with increased risk of early neonatal deaths. However, given the large number of neonatal deaths per every maternal death, the overall impact maternal deaths is small.

We lack data on chorionicity which is it well known to have impact on neonatal outcomes [22]. We were not able either to measure the influence of twin birth interval [23,24] nor the influence of difference in weight between first and second twins [25].

Lastly, in consideration of the cross-sectional nature of DHS, there is the possibility of underreporting and recall bias. The chance of underreporting of live births, recognized problem even in industrializes countries [26], would be higher for one of twins than for singleton, which translates in an underestimation of the excess mortality of twins [27].

Literature has provided discordant conclusions so far on the safest mode of delivery for twins: several studies generated significant protective results on perinatal outcomes with CS in case of elective CS vs vaginal delivery or emergency CS [28,29]. However, a recent systematic review [30] of two randomized trials found no clear evidence of benefits from planned CS for term twin pregnancies with leading cephalic presentation.

Most previous studies were conducted in high-income countries where access to quality care during CS is almost universal. However, our results lead to the same conclusions of the few studies conducted in low- and middle-income countries [30], which supports the need for timely access to safe caesarean births for twin pregnancies [31].

Despite the significant overall beneficial effect on early neonatal mortality when twins are delivered by CS in this data set, when stratifying countries according to CS rates in twins, this effect was maintained only in countries with rates of CS in twins of more than 15%. This may be explained by an important reduction of adverse outcomes for retained second twins, which can occur in up to one fourth twin deliveries [32], as well as fewer deaths in case of non-vertex first twin presentation [33]; it could also reflect a more widespread availability of health facilities performing CS for women experiencing obstetrical complications such as obstructed labor or eclampsia, which would reduce fetal distress.

In countries with lower CS rates (<15%), the lack of benefit may be due to a limited timely-access to emergency CS at population level. This may reflect not only the insufficient number of facilities performing sections but also the lack of skills among the health care providers, poor infrastructure and equipment, as well as routines such as early discharge from health facility after birth. The fewer CS conducted could be those in desperate situations, dangerous conditions without appropriate means rendering almost impossible to achieve a good outcome and thus limiting the impact of CS on early mortality.

That singleton deliveries by CS had an associated higher risk of mortality is discordant with findings from all-deliveries ecological studies [34-37], which show that newborn mortality decrease as CS increases up to 10%-15% [38], in line with the World Health Organization recommendations [39-40].

However, we should highlight the methodological bias [41] of the above-mentioned studies since no individual level data was used and only ecological associations were measured.

A similar study [15] using data from 46 Demographic and Health Surveys to assess the association between CS and newborn mortality showed significant increased OR of neonatal mortality among singletons delivered by CS in countries with low or middle CS rates. The same study speculated on the possible negative role of poor infrastructure in health facilities, lack of surgical and poor neonatal care underscoring the need to avoid unnecessary CS especially in settings that lack the facilities and/or capacity to properly conduct safe surgery and treat surgical complications, as recommended by the 2015 WHO statement [39-40].

Home delivery of twins with or without SBA was associated with increased risk of early neonatal mortality compared with delivering in a health facility through vaginal delivery aOR = 1.5 (95% CI = 1.0-2.3) and aOR = 1.7 (1.4-2.0), respectively. Home births without an SBA were associated with increased early neonatal mortality (OR 1.3; 95% CI = 1.0-1.8) compared with home births attended by SBA. This beneficial effect of institutional delivery on twins regardless of the availability of a SBA at home to give birth is not surprising for a wide range of factors which makes twins pregnancies more prone to adverse outcome, thus requiring higher capacity to manage complications at birth. The lesser benefit of SBA in home births when compared to non-SBA births is not surprising.

Twin pregnancies are associated with 7.6 (95% CI = 7.0-8.3) increased early newborn mortality compared to singleton pregnancies. This association was found in 55 out of 60 countries-analyzed. After adjusting for birthweight, the association decreased to 2.8 (95% CI = 2.2-3.5). Our findings on increased overall early neonatal mortality in twins compared to singleton births is in line with previous findings [42,43] and it is due to increased rates obstetric and perinatal complications [7-9,44] and several innate characteristics such as premature separation of placenta [45]. Other possible reasons include feeding and other cultural practices, especially in Africa and in Asia [46,47], whereby one of the twin babies could get preferential care. Concerning the differential mortality due to low birthweight between twins and singleton births we may infer a confounding role played by gestational age: twins are LBW because of constricted growth and at any given stage their organs are more mature than similarly LBW singletons likely because at higher gestational age.

Our findings suggest the need to identify twin pregnancies prior to labor to centralize care at birth; there is also a need to improve the access to safe CS for twin pregnancies in low- and middle-income countries where skills and capacity for the care and management of twin pregnancies as well as the potential complications may not exist. This is particular important for women in rural areas of countries of sub-Saharan Africa due to the enormous gap in the availability of skilled workers able to perform surgery [48]. In the absence of access to safe CS and/or capacity to treat and manage its complications or those of inherent to twin pregnancies, risks and benefits must be carefully balanced due to the negative potential consequences of unsafe surgery.

These medical aspects should be explored together with cultural beliefs, ie, related to birth order and gender, which may play an important role in the twins’ survival.