Working with what you have: How the East Africa Preterm Birth Initiative used gestational age data from facility maternity registers

Objective Preterm birth is the primary driver of neonatal mortality worldwide, but it is defined by gestational age (GA) which is challenging to accurately assess in low-resource settings. In a commitment to reducing preterm birth while reinforcing and strengthening facility data sources, the East Africa Preterm Birth Initiative (PTBi-EA) chose eligibility criteria that combined GA and birth weight. This analysis evaluated the quality of the GA data as recorded in maternity registers in PTBi-EA study facilities and the strength of the PTBi-EA eligibility criteria. Methods We conducted a retrospective analysis of maternity register data from March–September 2016. GA data from 23 study facilities in Migori, Kenya and the Busoga Region of Uganda were evaluated for completeness (variable present), consistency (recorded versus calculated GA), and plausibility (falling within the 3rd and 97th birth weight percentiles for GA of the INTERGROWTH-21st Newborn Birth Weight Standards). Preterm birth rates were calculated using: 1) recorded GA <37 weeks, 2) recorded GA <37 weeks, excluding implausible GAs, 3) birth weight <2500g, and 4) PTBi-EA eligibility criteria of <2500g and between 2500g and 3000g if the recorded GA is <37 weeks. Results In both countries, GA was the least recorded variable in the maternity register (77.6%). Recorded and calculated GA (Kenya only) were consistent in 29.5% of births. Implausible GAs accounted for 11.7% of births. The four preterm birth rates were 1) 14.5%, 2) 10.6%, 3) 9.6%, 4) 13.4%. Conclusions Maternity register GA data presented quality concerns in PTBi-EA study sites. The PTBi-EA eligibility criteria of <2500g and between 2500g and 3000g if the recorded GA is <37 weeks accommodated these concerns by using both birth weight and GA, balancing issues of accuracy and completeness with practical applicability.

In a collaboration between the University of California, San Francisco (UCSF), the Kenya Medical Research Institute, and Makerere University in Uganda, PTBi-EA implemented a package of intrapartum and immediate postpartum interventions aimed at improving the quality of maternity and newborn care. The study was a cluster randomized control trial (CRCT) targeting healthcare workers in the maternity and newborn units of 10 intervention facilities, 10 control facilities, and 3 referral facilities that received the intervention but were not included in the primary analysis. The study facilities were mostly public, government hospitals and healthcare centers, staffed predominantly by nurse-midwives, nurses, and clinical officers. Success was measured by the comparison of fresh stillbirth and neonatal mortality among preterm babies born at the intervention versus control facilities. Results of the CRCT saw a 34% decreased odds in neonatal mortality in the intervention sites among eligible infants and are published elsewhere [25]. The intervention package included data strengthening, introduction of a modified version of the WHO Safe Childbirth Checklist (mSCC), a quality improvement (QI) collaborative, and PRONTO simulation and team training. To address data quality concerns, all facilities received the mSCC and on-going data strengthening support which included a 2-day training during the baseline data collection period in an effort to improve baseline data for more accurate comparisons with the intervention data. The importance of GA documentation was emphasized during this training and did result in an increase in maternity register GA recordings [24]. Early data collection revealed GA quality and accuracy concerns, therefore PTBi-EA senior staff from both countries convened to agree on the CRCT 28-day follow-up eligibility criteria. These criteria were: all babies less than 2500g and babies between 2500g and 3000g if the GA is reported as less than 37 weeks. These criteria were chosen for ease of use, and because they included all low birth weight (LBW) babies, likely to be preterm using INTERGROWTH 21st Newborn Birth Weight Standards (IG21-NBWS) data as a reference, and babies between 2500g and 3000g only if they had a registered GA less than 37 weeks [26]. This would capture more late preterm babies and exclude the majority of large babies that are unlikely to be preterm. While growth-restricted term babies were also likely to be included in the cohort of babies less than 2500g, the distinction between preterm and growth-restricted term babies was not possible to make without early ultrasound dating. This nested analysis was a retrospective chart review evaluating the completeness, consistency, and plausibility of the GA data in the maternity registers during the baseline period (March 1, 2016 –September 30, 2016) of the PTBi-EA CRCT. Eligibility criteria included all live and fresh stillbirth babies born at the 23 study facilities during baseline, with recorded GA, birth weight, and sex, that were greater than 24 and less than 42 weeks GA, and greater than 500g and less than 6000g birth weight in order to compare the data to the IG21-NBWS [15]. Macerated stillbirths were excluded to comply with standard preterm birth rate definitions where live birth is the denominator. Fresh stillbirths, however, were included to parallel the PTBi-EA parent study in which they were included to account for early neonatal deaths misclassified as fresh stillbirths and to assess the impact of the intrapartum intervention package on fresh stillbirth rates. A team of data collectors conducted line-by-line extraction of the maternity register data from each of the 23 facilities. All births were included in the dataset and data were transcribed as they were written by health providers. The data were entered into an Open Data Kit (ODK) tool and uploaded to a server hosted at UCSF. The datasets were combined and cleaned using Structured Query Language (SQL) and analyzed using RStudio (Version 1.0.136). The GA data recorded in the maternity registers came from various sources, dependent on the individual midwife and data availability. Some were transcribed from ANC booklets provided by the mothers when they presented for labor, others were calculated from maternal-reported LMP or measured from fundal height, while others appeared to be adjusted based on informal post-partum provider assessments. It was unclear when the information was recorded in the register and seemed to vary dependent on the midwife, with some filling the information in throughout the shift and others filling it in batches from the patient charts at the end of a shift. Few women received ultrasounds during their pregnancy as they could only be obtained at private facilities through out-of-pocket expenditure and they were rarely received for the purpose of GA dating. Data completeness was calculated as a proportion of all births where GA, birth weight, sex, and birth outcome were recorded (looked at as separate variables and a combined variable for births where all four variables were complete). The consistency evaluation was conducted only for Kenyan data as their maternity registers list both a recorded GA and a separate LMP date, and the Ugandan registers listed only a recorded GA. We used Naegele’s rule to create a “calculated GA” variable from the LMP date and date of delivery [27]. The differences between the calculated and recorded GAs were compared and those with a difference of less than one week were considered to be equal. Descriptive statistics of calculated GA and the GA differences were included and a Bland-Altman plot displays the differences graphically. Plausibility of GAs were evaluated by calculating the percentage of births where the birth weight for a given GA fell within the 3rd and 97th percentiles according to the IG21-NBWS data. Any birth that fell outside of these boundaries was considered to have an implausible GA. As GAs were recorded in whole weeks the 3rd percentile of week, 0 days and the 97th percentile of week, 6 days was used. For example, a female baby with a GA of 30 had a range from 900g (3rd percentile for 30 weeks, 0 days) to 2070g (97th percentile for 30 weeks, 6 days). Finally, we calculated different approaches to estimating a preterm birth rate: The IRB committees of the University of California, San Francisco (Study ID# 16–19162), the Kenyan Medical Research Institute (Study ID# 0034/321), and Makerere University School of Public Health (Study ID# 189) reviewed and approved this study. All data were stored on encrypted computers and servers. The subset of data used in this analysis was de-identified prior to access.

• miller, lara
• wanduru, phillip
• santos, nicole
• butrick, elizabeth
• waiswa, peter
• otieno, phelgona
• walker, dilys