Incidence and determinants of perinatal mortality in five urban hospitals in Dar es Salaam, Tanzania: a cohort study with an embedded case–control analysis

The study was carried out in the five busiest government hospitals in Dar es Salaam, Tanzania. In these five hospitals, 60–70% of all births in the city are estimated to take place, and in each health facility (HF) between 6,000 and 10,000 women give birth per year [21]. These HFs are part of a network of 22 HFs in Dar es Salaam region, which collaborate in a maternal and newborn healthcare strengthening program established in 2010 [22]. This program is implemented through a public–private partnership between the regional health authorities and the non-governmental organization ‘Comprehensive Community Based Rehabilitation in Tanzania’ (CCBRT). Furthermore, the five hospitals are the study sites of the PartoMa project, which seeks to achieve best possible maternal and newborn care by adapting clinical guidelines and skills training to the local context [21, 23].

Three of the study HFs are Regional Referral Hospitals (RRHs): HF1, HF2 and HF3. The other two are upgraded health centers that serve as primary level maternity hospitals: HF4 and HF5. All five facilities provide comprehensive care during childbirth that includes vacuum assisted births, cesarean section (CS) and blood transfusions. While the three RRHs have neonatal high care units, the other two (HF4 and HF5) refer all sick newborns to the nearest RRH. All five hospitals receive referrals from dispensaries and health centers throughout the wider Dar es Salaam region. Due to resource-constraints, blood tests for hematology, liver function and renal function are not routinely performed for laboring women with hypertension. All five hospitals typically serve women of lower socio-economic status [24]. Dispensaries and health centers provide a wide range of preventative and first line emergency care, which includes reproductive healthcare, antenatal care (ANC) and care during birth for low-risk pregnant women. Dispensaries are expected to refer all high-risk pregnancies, women with abnormal labor progress, obstetric complications and intrauterine fetal deaths.

Figure 1 presents the definitions of stillbirths and early neonatal deaths as defined by the World Health Organization (WHO) [25]. Additional terms used in this study, including pre-facility and intra-facility stillbirth are inspired by previous studies [5, 26]. Definitions e–g were applied instead of antepartum stillbirth, fresh stillbirth and macerated stillbirths due to the low reliability of the stillborn baby’s appearance to the time of fetal death [27, 28].

For our cohort study, in order to assess incidence of perinatal deaths, the study population consisted of all births in the five hospitals from 1^st January 2020 to 31^st December 2020, including all perinatal deaths diagnosed before discharge. For the embedded case-control study, we included a subset of singleton births with birthweight ≥ 2,000 g who had their medical case records available for analysis. Newborns with major congenital malformations were excluded. Within this category, cases of perinatal deaths were sub-categorized by presence or absence of the fetal heart tones on admission to the study HF, as follows:

Controls were defined as singleton, live newborns with a birthweight ≥ 2,000 g, Apgar score of ≥ 8 at one minute and ≥ 9 at five minutes, who did not require bag and mask resuscitation and who were discharged home alive [30]. Controls were selected to match the included intra-facility perinatal deaths by month and facility of birth.

For calculation of perinatal mortality rates, we collected data on total births, livebirths, stillbirths, early neonatal deaths, CS, vacuum-assisted births and maternal deaths, which are available in the five hospitals’ routine National Health Information System (MTUHA). Additional information pertaining to perinatal deaths was extracted from the Perinatal Problem Identification (PPIP) database, which contains additional information on actual birthweight, multiple pregnancy and presence of congenital malformations [21].

For the case–control study, all perinatal deaths that met the inclusion criteria were extracted from the PPIP database and their paper-based medical records were intensively searched for. If medical records were retrieved, women were included, and data extracted from their records. Controls were retrieved from the piles of medical records (mainly partographs) of all women with livebirths, which were separated by month of birth and divided into vaginal and cesarean births. The average CS rate in the study sites (2019 data) varied from 18 to 25% [21]. The ratio of vaginal to cesarean births was approximately 4:1. In accordance with the sample size calculation presented below, and by use of a random number generator, controls were then systematically sampled by inclusion of every 10^th file, with each fifth of the included files being selected from the CS pile. When the selected controls did not meet the inclusion criteria, the next 10^th file was selected.

A data collection tool was developed on Open Data Kit (ODK) XForm using a data dictionary and deployed to an ODK app on smart devices (www.​betterevaluation​.​org). Validation codes were incorporated in the tool to limit data errors and the tool was pilot tested in each study facility by BSD. Data extraction from files was performed by midwives and medical doctors working in the study sites who had all attended a three-day data entry training, conducted by BSD. The first 50 files were double entered by TWJ to check accuracy of data entry. Any discrepancy in double entry was re-checked. BSD reviewed data entry progress and related quality every week throughout the data collection period. All files were assigned a unique identification number and personal identifiers removed. The study database was stored in Ifakara Health Institute Data center, Tanzania.

The sample size was estimated using an alpha of 0.05, power of 0.8, ratio of cases to control of 1:5 and assuming an odds of intra-facility perinatal deaths of 1.45. The estimated sample was 395 cases for intra-facility deaths. However, in this study we obtained only 287 cases for intra-facility deaths.

For the case-control study, the variables of interest are presented in Fig. 2 and Table 3. These variables were selected from the literature to broadly assess antenatal and admission risk factors as well as intrapartum quality of care provision, including surveillance and associated treatment if needed related to maternal vital signs, fetal heart rate and labor progress [31]. Notably, the selection of variables was limited to what could retrospectively be assessed in the case files.

Data cleaning and analysis was performed using Stata version 14.2 (Stata Corp, Texas, USA). Categorical data were summarized using frequencies and percentages while continuous data were summarized using mean and standard deviation. Stillbirth rate and perinatal mortality were obtained using total births from the five study HFs in the year 2020 per 1,000 total births. A Pearson chi-square test was used to determine associations between variables of interest and outcome variables. One-way ANOVA test was used to compare mean maternal age of the three groups in the case control study population (healthy women, pre-facility stillbirths and intra-facility perinatal deaths). Both bivariable and multivariable logistic regressions were performed to identify the presence of a significant association between independent variables and the dependent variable. Variables with a p-value of < 0.2 were considered for multivariable analysis. Any variable whose univariable test has a p-value < 0.25 along with all variables of known clinical importance should be included into multivariable analysis [32]. Furthermore, variables with a high proportion of missing cases above 10% were omitted from the multivariable analysis. Following fitting of the model, we assessed the importance of each covariate using p-values. Variables that did not contribute at traditional levels of statistical significance (p ≥ 0.05) were eliminated. Bivariate analyses were used for pre-facility stillbirths and multivariable analyses were used for analyses of intra-facility perinatal deaths. Crude Odds Ratio (cOR) and adjusted Odds Ratio (aOR) were calculated with 95% confidence interval (CI). Free text descriptions in women’s files on the care given in response to abnormal vital signs were manually extracted from the notes, categorized and presented. Results were presented using tables and figures.

Mean age of the women was 27.0 (SD ± 6.2). There was a small difference in mean age of women between the groups with healthy newborns and with pre-facility and intra-facility perinatal deaths, but no difference between pre-facility and intra-facility perinatal deaths groups. The history of a previous perinatal death among the 194/1,331 (14.6%) multiparous women was similar between controls (135/860; 15.7%) and pre-facility stillbirths (47/299; 15.7%). However, among multiparous women with a past history of previous perinatal death, a lower proportion of intra-facility perinatal deaths was observed (12/172; 7.0%; aOR 0.32; 95% CI: 0.13–0.81) (Supplementary Table 4).

All women in the case–control study, except eight, attended ANC with 1,627/2,153 (75.5%) attending four or more visits (Table 2). More than 90% of the women received routine testing for Human Immunodeficiency Virus (HIV) and syphilis, tetanus immunization and presumptive treatment for malaria. Of all women, 402 (18.1%) had an ultrasound examination, but only 21/402 (5.2%) before 24 weeks of gestation. Among the 1850 (83.1%) women with available hemoglobin test results, mean hemoglobin (Hb) was 10.9 g/dl ± 1.3 SD and severe anemia (Hb ≤ 8 g/dl) occurred in 45 (2.4%) women. There was no difference in mean hemoglobin in the three outcome groups (Supplementary Table 1).

Among the case control study population, self-referrals from home occurred in 1,807/2,224 (81.3%) women and 283/2,224 (12.7%) were referred from other non-study HFs (mainly health centers and dispensaries). More specifically, among pre-facility stillbirths, 109 (24.1%) were referred from lower-level HFs, and our data did not indicate if the fetal heart tone was present when the women were admitted in the primary-level HF.

While more than 86% of women had their vital signs measured on admission, very few among those with abnormal vital signs had documentation of any action taken in response. For example, a blood pressure measurement on admission was recorded for 2,060 (92.6%) of women. Among the 71 women with severe hypertension (SBP ≥ 160 and/or DBP ≥ 110 mmHg) measured on admission, 16 (22.5%) of their case-notes had specific treatment recorded.

Concerning the stage of labor on admission, 621 (30.1%) were admitted in the latent stage, 748 (36.3%) were admitted in early active labor (cervical dilatation between 4 and 6cms) and 694 (33.7%) were admitted in late active labor with cervical dilation between 7 to 10 cms. There were no statistically significant differences between the groups.

We reviewed 2,085 available partographs. All three sections of the partograph (maternal vital signs, labor progress and fetal heart rate) were completely filled in for 874/2,085 (41.9%) women.

We noted that women with fetal death prior to admission (pre-facility stillbirth) had significantly increased odds of experiencing challenges during the second stage of labor ( cOR 11.52; 95%CI: 8.63-15.36), and breech presentation with vaginal birth (cOR 9.41;95% CI: 3.02-29.36) (Table 3).

CS was performed in 533/2,224 (23.9%) women. Out of the total CSs, 469/533 (88%) were performed as emergencies. The commonest indication for CS was previous CS in 136/533 (25.5%) women, followed by obstructed labor in 103/533 (19.3%) and fetal distress in 70/533 (13.1%). The study design ensured that for the control group, the random selection included CSs that approximated the CS rate in the study hospitals, there were no statistical or clinically relevant differences in CS as mode of birth and birth outcomes. Among the 533 women with CS birth, we noted 188 (35.2%) perinatal deaths, distributed as follows: pre-facility stillbirths (74/452; 16.4%), intra-facility stillbirths (54/122; 44.3%) and intra-facility early neonatal deaths (60/165; 36.4%).

Among our study population, 1,862/2,224 (83.7%) women reported no complications during labor and childbirth, while 412 complications were extracted from the remaining 362 women, some of whom reported more than one complication. Hypertension was reported most frequently (165/412, 40.0%) followed by anemia (69/412, 16.7%) and hemorrhage (51/412, 12.3%). Within the study population, there was one maternal death due to hemorrhage from placental abruption in the intra-facility stillbirth group. Uterine rupture occurred in 18 women, including in six among women with pre-facility stillbirth, 11 women with intra-facility perinatal death, and one woman in the control group.

Compared to healthy newborns, factors associated with increased odds of pre-facility stillbirths included the following: any hypertension as a complication during intrapartum care (cOR 4.72; 95% CI: 3.30–6.76); low birth weight (newborns within the lowest included birthweight category between 2,000 and 2,500 g), (cOR 4.40; 95% CI: 3.13–6.18); presence of at least one risk factor detected during ANC (cOR 2.68; 95% CI 2.02- 3.56) and if the pregnant woman was referred from peripheral HFs (cOR 1.88; 95% CI 1.17–3.03) (Table 3).

Furthermore, a sub-analysis of maternal blood pressure measured on admission (adjusted for age of the woman, parity and newborn birthweight), revealed increased odds of pre-facility stillbirths compared to healthy newborns: mild high blood pressure measurement on admission (cOR 1.93; 95% CI: 1.42–2.62); severe high blood pressure measured on admission (cOR 3.04; 95% CI: 1.83–5.06) (Table 3 and Supplementary table 2).

Compared to healthy newborns (controls), determinants of intra-facility perinatal death included: breech birth (aOR 40.3; 95% CI: 8.75–185.61); complications in the second stage of labor (aOR 20.04; 95% CI: 12.02–33.41); vacuum-assisted birth (aOR 6.23; 95% CI: 1.6–23.55); low birth weight or late preterm birth with birthweight between 2,000 and 2,500 g (aOR 5.57; 95% CI: 2.62–11.84); cervical dilation that crossed the partograph’s action line (aOR 4.16; 95% CI:2.29–7.56); meconium liquor on admission (aOR 4.44; 95% CI:1.86–10.57); at least one risk factor detected during ANC (aOR 3.7; 95% CI 1.96–6.98) and maternal hypertension during intrapartum care (aOR 2.9; 95% CI 1.03–8.14) (Table 3).

This study includes extensive data on facility births from multiple busy public maternity hospitals that receive women with a mix of low- and high-risk pregnancies, predominantly of lower-socio economic status, and referred from the entire Dar es Salaam region, thereby providing an in-depth understanding of the urban disadvantage experienced by women during pregnancy and childbirth in Dar es Salaam. There are, however, limitations to this study.

This was a retrospective study using data from routine health information systems. Previous studies report that hospital registers may have a high sensitivity and specificity for reporting perinatal outcomes [60, 61]. Furthermore, the Dar es Salaam quality improvement initiative has invested in strengthening routine data at the study HFs since 2010 [21]. Yet, we recognize that the quality of data may not be optimal. We acknowledge that facility-based pre-discharge neonatal mortality, although high, may still be an underestimate, as we could not include neonates who may have died after discharge or referral. We recognize that stillbirths and early neonatal deaths are prone to misclassification, particularly in understaffed facilities where fetal motion, heart rate or respiration in a liveborn infant may not have been observed [3, 26].

Even though we included all eligible cases, the 287 intra-facility perinatal deaths fell short of the 395 required in the initial power calculation. (A proportion of these could not be included due to missing case files). Furthermore, concerning hypertensive disorders, our data and practices in the study HFs did not allow for clinical classification of hypertension with or without severe features. Also, mild cases of hypertension may have been under-reported due to poor documentation. Consequently, the more severe types of hypertensive disorders may more likely have been captured, resulting in selection bias. Lack of storage capacity, poor protection of case notes (particularly during the COVID-19 pandemic), deficient documentation and missing files contributed to shortfall in intra-facility perinatal deaths and may also have resulted in selection bias, which potentially could have resulted in an underestimation of the strength of the associations between the determinants and perinatal deaths. The targeted care for teenagers, grand multiparous women of advanced maternal age and women with previous adverse obstetric outcomes is part of the decade long quality improvement initiative at the five study HFs and may explain the unexpected lower odds for intra-facility deaths in these high-risk groups [22].

We used the term pre-facility stillbirth, where the point of reference was presence or absence of fetal heart sounds on admission [31]. The term pre-facility stillbirth is not entirely synonymous to ‘antepartum stillbirths’ (pre- labor) or ‘macerated stillbirths’, which is based on fetal appearance [5]. Gold et al. consider fetal appearance an unreliable surrogate marker for determining the time of fetal death [28]. These differences in terminology may make comparison of our findings with other studies challenging. Our categorization was the most feasible in the setting and provides a reliable distinction between stillbirths that occurred before or after admission to HFs during labor and birth, but requires accurate fetal heart rate documentation on admission [60].