Setting

This study was done at Kilimanjaro Christian Medical Centre (KCMC) in Northern Tanzania. The hospital is a zonal hospital serving more than 13 million people from 4 regions namely; Kilimanjaro, Arusha, Tanga and Manyara. We established a cohort of babies based on records from the Medical Birth Registry comprising all deliveries at the hospital from July 2000 to September 2008 and followed the cohort in a registry of neonates transferred from the labour ward to NCU. The KCMC Medical Birth Registry system was established in 1999 as a collaboration between Kilimanjaro Christian Medical College, Tumaini University and the University of Bergen, Norway. The annual number of deliveries is around 3000 of which nearly two thirds are from urban area. Approximately 10-15% of the neonates are transferred to NCU for observation and management.

A total of 26 025 births were recorded in the Medical Birth Registry from July 1^st 2000 to September 30^th 2008. We excluded multiple deliveries, stillbirths, neonatal deaths in labour ward and neonates with missing child status record after delivery (Figure 1). In order to obtain a study group that reflected the general population, we excluded deliveries where mothers residing in rural areas had been referred for delivery at KCMC for medical reasons. Women residing in Moshi urban were not excluded since they could have delivered at KCMC anyway. KCMC is located in Moshi Urban and that 50% of the deliveries at KCMC are from Moshi Urban district [11]. We finally analyzed a total of 21 206 singleton live births.

Data collection

Information from all mothers who delivered at KCMC were collected within the first 24 hours after delivery. Trained midwife nurses conducted the interviews on a daily basis with all eligible subjects using a standardized questionnaire. A verbal consent was obtained from the participants prior to the interview. Mothers also provided their antenatal visit card for more information such as date of first ANC visit, immunization history, malaria prophylaxis, drugs, illnesses recorded during follow up, weight at first ANC visit, number of ANC visits, as well as referral to ANC (self-referred or referred by health worker).

Information in the birth registry includes maternal health conditions before and during pregnancy, parents' socio-demographic characteristics, complications during labour and delivery, and information on the newborn; sex, gestational age, birth weight, Apgar score, and child status in four categories: 1) live born 2) live born transferred to NCU 3) neonatal death in labour ward, 4) stillborn.

The paediatric registry form was recorded in the NCU for all neonates who were transferred. The neonatal registry includes information on primary reasons for transfer, management, and discharge/death diagnoses. The two databases were linked using the unique child identification number, the mother's hospital registration, and the newborn's birth registration number.

Variable definition

Data analysis

Data were analyzed using Statistical Package for Social Science (SPSS) program Version 15.0 for Windows (SPSS 15.0 Chicago Inc. III, USA). Cross tabulations and generalized linear models were used to obtain relative risks (RR) and corresponding 95% confidence intervals. From the bivariate analyses we present all variables with p-value less than 0.1, which were then entered into the multivariable analysis. Three steps were involved in the multivariable analysis. In the first step (model A) all socio-demographic factors and maternal health condition before pregnancy were included. In the second step (model B) we included all variables in step one as well as pregnancy and labour-related conditions. In the third and final step (model C), we included all variables in step two as well as neonatal conditions. We used Poisson regression with robust variances to obtain a valid confidence interval when a log-binomial analysis failed to converge [17]. A priori we also considered some maternal conditions to be important and included in the final analysis, these were hypertensive conditions (preeclampsia, eclampsia and abruption placenta) and diabetes (pre-gestational or gestational).

Ethical approval

The birth registry at Kilimanjaro Christian Medical Centre obtained ethical clearance from the Tanzania Ministry of Health, Institute of Science and Technology, from the Norwegian National ethics committee and from the Kilimanjaro Christian Medical College (KCM-College) research ethics committee in 1999. The protocol for this study was approved by KCM-College research ethics committee, with certificate no. 333 of 15^th July 2010.

Socio-demographic characteristics and pre pregnancy conditions

Pre-gestational diabetes mellitus was strongly associated with neonatal transfer to NCU (RR 4.4; 95% CI: 3.3-5.8). A history of acute or chronic lung disease other than tuberculosis showed a weaker association (RR 1.2; 95% CI: 1.1-1.4).

Pregnancy, labour and delivery

Factors related to pregnancy, labour and delivery were included in the multivariable model in B. Hypertensive conditions such as eclampsia and preeclampsia (RR 2.8; 95% CI:1.7-4.4 and 2.0; 95% CI: 1.7-2.3, respectively), labour-related complications such as premature rupture of membrane and abruption placenta (RR 2.9; 95% CI: 2.6-3.4 and 2.6; 95% CI: 1.6-4.1, respectively), and other vaginal delivery (i.e. breech, vacuum or forceps) and caesarean section delivery (RR 2.9; 95% CI: 2.3-3.6 and 2.1; 95% CI: 1.9-2.3, respectively) were all associated with transfer (Table 4; model B). Gestational diabetes increased the risk of babies transfer by 40% although not statistically significant. Referral to ANC and few ANC visits were also found to be important predictors of neonatal transfer to NCU (RR 1.3; 95% CI: 1.1-1.4 and 1.3; 95% CI: 1.2-1.4), respectively.

Significant factors in model A continued to be independent predictors for neonatal transfer also in model B, except for maternal body height below 150 cm. However, addition of variables in model B slightly reduced the relative risk for most factors.

Neonatal factors

In model C, neonatal factors were added into the multivariable model. All the selected neonatal factors were significantly associated with transfer to NCU, with the highest relative risks being for birth weight above 4000 g (RR 7.2; 95% CI: 6.5-8.0) and five minutes Apgar score below 7 (RR 4.0; 95% CI: 3.4-4.6) (Table 4; model C).

After inclusion of the neonatal factors, some pre-pregnancy factors, such as women giving birth to their first babies (RR 1.4; 95% CI: 1.2-1.5), maternal age 26-35 years (RR 1.2; 95% CI: 1.1-1.3), and single marital status (RR 1.2; 95% CI: 1.0-1.3) were still significantly associated with neonatal transfer. Lack of paternal education (RR: 0.5; 95% CI: 0.3-0.9) was negatively associated with transfer to NCU. Birth to fourth or later born babies, maternal overweight or obesity, pre-gestational diabetes and epilepsy were no longer significantly associated with neonatal transfer.

Neonatal factors

The studied neonatal factors included classical risk factors for morbidity and mortality, such as birth weight, preterm delivery, Apgar score and sex, and were as expected strongly related to neonatal transfer. Although the causal effect of birth weight is controversial [19] low birth weight is a good predictor of need for neonatal care. Low birth weight has been proposed to contribute to 40-80% of neonatal morbidity and mortality [20, 21]. Preterm delivery is estimated to account for 28% of all neonatal deaths [20].

We also found a very high admission rate of newborns with a birth weight above 4000 g. Fetal macrosomia is associated with obstetric complications and neonatal morbidity such as injuries, respiratory distress and hypoglycaemia. Observation for transient or persistent hypoglycaemia is a common reason for admission of high birth weight babies to NCU [22]. At KCMC, such babies will be discharged within 24 hours if there is no risk of persistent hypoglycemia and the blood glucose level is normal. The outcome is in general good for these babies, and one may speculate whether observation without transfer to NCU for many of these babies would represent a better use of resources.

In general, male neonatal morbidity exceeds female morbidity, partly due to a higher occurrence of preterm birth and other neonatal risk factors [23]. The male-to-female ratio of transfer 1.24, declining to 1.18 in the adjusted analyses, corresponds well with the established higher risk in males, and does not indicate a difference in care according to infant sex.

Pregnancy, labour and delivery

Risk of neonatal transfer was high in mothers with preeclampsia, eclampsia and abruption placenta, however no or weak effects were observed after inclusion of neonatal factors in the model. Hypertensive conditions in pregnancy are associated with preterm birth and low birth weight [15, 24–27], and many cases of abruption placenta occur at a low gestational age, which explain the indirect association between these complications and neonatal transfer. The direct cause of transfer would be the preterm birth.

Other conditions, such as premature rupture of membrane, caesarean section and operative vaginal delivery, showed a high risk of neonatal transfer also after accounting for the neonatal condition of the baby. The high rate of transfer for babies born with mothers having PROM is similar to what is reported elsewhere [15]. Premature rupture of the membrane (PROM) is associated with preterm delivery and low birth weight [15, 27]. A previous study at KCMC reported a high prevalence (38%) of low birth weight babies after PROM [27]. Such babies are at higher risk of developing neonatal infection. Antibiotic prophylaxis given to mothers with PROM has shown to reduce risk of infection in the newborn [28, 29]. The high transfer rate after PROM in our data is likely to be explained by the fact that a majority of mothers with a history of PROM did not receive antibiotic prophylaxis prior to delivery, due to late arrival to the centre.

Mothers with less than five antenatal care visits were more likely to have their baby transferred and this association persisted after we took into account our measures of the condition of the newborn. Amount of antenatal care plays a role in neonatal outcome [30–32], and each additional ANC visit has previously been found to offer a protective effect on neonatal outcome [31]. When the mother had been referred for antenatal care, however, the risk of transfer was increased.

Pre pregnancy factors

Among diseases that the mothers had before the pregnancy, only lung disease remained significantly associated with neonatal transfer when pregnancy conditions and neonatal conditions were accounted for (Table 4, model C). Pre-gestational diabetes was strongly related to transfer in models A and B, but the association disappeared after accounting for the neonatal conditions in model C. Noteworthy, gestational diabetes had a weak and non-significant association with transfer, and the relative risk was not affected by adjustment for neonatal factors. The low risk of transfer in babies born to mothers with gestational diabetes compared to babies of mothers with pre-gestational diabetes is also reported elsewhere [5, 15, 33].

Women giving birth to their first child and single mothers were more likely to have their baby transferred to NCU, also after accounting for pregnancy conditions and neonatal conditions. Birth to a first child and single motherhood are classical risk factors for neonatal morbidity and mortality [9, 10, 12, 18, 34, 35]. However, the 40% higher risk of admission for a first born child in the fully adjusted model (model C), is higher than what one would expect according to previous knowledge on morbidity and mortality associated with first delivery. In a previous study from the same hospital, perinatal mortality was not associated with birth order except for a higher perinatal mortality in offsprings of mothers with three or more previous pregnancies [11]. To further elaborate this finding, we performed a regression analysis with a finer categorization of Apgar score. In this model, the parity effect was still statistically significant, however reduced. In a setting with limited obstetric services, the generally higher neonatal stress on first born babies might be even more evident.

In line with previous findings [36–38] we found that overweight and in particular obese mothers had a high risk of having their baby transferred to NCU. Maternal obesity is associated with some pregnancy complications [33, 36–40] and overweight or obese mothers are more likely to have high birth weight babies [37, 38, 40]. A meta-analysis review showed a lower risk of low birth weight among babies of overweight or obese mothers compared to normal weight mothers, however the risk of very low birth weight and extremely low birth weight was increased due to more induced preterm deliveries in overweight or obese mothers [41]. In our data, the association of neonatal transfer associated with maternal overweight and obesity was weakened but still statistically significant after adjustment for pregnancy conditions, however disappeared after adjustment for neonatal conditions. Hence, pregnancy conditions and neonatal conditions seem to be mediators in the association between maternal overweight and neonatal transfer. A similar pattern was seen for mothers of short stature, where an increased risk seen in model A seemed to be linked to a higher rate of pregnancy complications for these mothers.

Drinking unboiled water was one of the factors associated with neonatal transfer. Waterborne disease including diarrhoea and dysentery is prevalent in Tanzania, therefore, it is recommended to boil water for drinking including tap water. In our study 92% of the participants used tap water, however only 31% boiled water for drinking. In a study from Tanzania, lack of boiling water prior to consumption was more common in households with low income, and lack of proper knowledge on the importance of how to handle and store water safely was associated with E.coli occurrence [42]. Both ignorance and poverty might be the major barriers to boiling drinking water.

Lack of paternal education was associated with a low chance of neonatal transfer (RR = 0.5; 95% CI: 0.3-0.9) in the fully adjusted model. Although our results should be interpreted with care due to the low numbers (110 fathers with no education) and a confidence interval close to one, the findings could reflect low focus on neonatal health care in deprived families. A previous study using the same birth registry reported that paternal socio-demographic factors seemed to be more important predictors of perinatal mortality than maternal socio-demographic factors in this area [11]. However, such an interpretation is not compatible with the principle that transfer mechanisms should be unaffected by parental and family influence.