Labour and delivery ward register data availability, quality, and utility - Every Newborn - birth indicators research tracking in hospitals (EN-BIRTH) study baseline analysis in three countries

Improving quality of care at birth could save an estimated 3 million lives per year [1, 2]. To drive progress, accurate data are essential, however, the majority of deaths around the time of birth occur in settings with the least information on these deaths, the “inverse data law” [3]. Improving impact and coverage data for action is central to the Sustainable Development Goal (SDG) aspiration of “no-one left behind” [4], the United Nation’s Global Strategy for Women’s Children’s and Adolescents’ Health [5], and The Every Newborn Action Plan (ENAP). One of five ENAP strategic objectives is to transform metrics and use of data to improve outcomes and track progress towards ending preventable maternal and newborn deaths, including stillbirths [6].

Labour and Delivery (L&D) ward registers are routinely completed by facility health workers and used to track ward admissions and discharges in a parallel system to patient case notes. Birth outcomes, care and interventions for women and babies are also often documented in these registers. However, concerns of poor register data quality in low- and middle- income counties (LMIC), have reduced confidence in full utilization of this data source in Health Management Information Systems (HMIS). As global facility births increase, currently >81%, [7], it is important to reassess the availability and quality of this routine data to help address the current data gap around the time of birth.

Research assessing labour ward register data in LMICs provides some explanation for the scepticism surrounding programmatic use of this source. Maternal and newborn health (MNH) data elements were not consistently available in facility registers in 24 high burden countries [8]. In a rural primary health care context in north eastern Nigeria health workers documented in labour ward registers most completely for birthweight (99%) and woman’s age at delivery (97%); documentation was less complete for the composite indicator essential newborn care (82%) and preterm birth (77%) [9]. In two rural Kenyan hospitals, entire labour ward registers were missing for months, and when present many data elements were less than 80% complete; the proportion of data legible/correctly coded/appropriate/recognized ranged from 29 to 100% [10]. In one Ethiopian hospital, among the 20% of births missing from the labour ward register, 91% had received a clinical intervention, thus the register both underestimated total births and interventions [11]. However, the picture is not wholly negative and routine data can be improved. Data quality improvement efforts across 20 L&D wards in South Africa, including data collection training and monthly data reviews, demonstrated increased completeness from 26 to 64%, and accuracy from 37 to 65% [12]. In Rwanda, health system strengthening measures including performance review feedback activities, mentoring, and enhanced supervision led to increased value and ownership of data among health workers [13]. In Zanzibar, quarterly data use workshops with active engagement of data users, grew and improved the HMIS, enhancing staff capacity for information use, presentation and analysis for decision making [14].

The Every Newborn-Birth Indicators Research Tracking in Hospitals (EN-BIRTH) study aimed to assess the validity of selected newborn and maternal health care interventions indicators (coverage, content/quality, and/or safety) in hospitals [6] (Fig. 1). Our assessment of existing routine registers in EN-BIRTH study hospitals formed the baseline against which to evaluate any changes in documentation resulting from the presence of researchers in the L&D ward [18].

This is the largest multi-country study we are aware of in LMICs to assess labour ward register data availability, quality, and utility. Hospital registers are key tools used to collect individual data for aggregation and transmission up the HMIS data pyramid [6]. Data extracted from five CEmONC hospitals show that a large amount of data are being collected in labour ward registers. The calculation of MNH coverage and impact indicators require the availability of specific data elements for use as numerators and/or denominators, yet none of the labour ward registers contained all 21 selected data elements. Data for outputs, outcomes, and impact measurement were more widely available, than for intervention coverage. Only the Tanzanian registers captured most of the selected interventions and gestational age was only captured by the Nepal register and the additional register in Muhimbili TZ.

The Performance of Routine Information System Management (PRISM) framework identifies complexity and design as technical factors in routine health information systems performance [28]. The register designs were different between countries, and within country in Bangladesh. Data were captured from the additional perinatal register in Muhimbili TZ and from operation registers for babies born by Caesarean in Bangladesh, highlighting further complexity in multiple recording systems.

Whether a specific column was allotted for the data element related to completeness of recording. Across all five hospitals, a much lower proportion of non-specific columns had high levels of completeness than did specific columns (25.0% versus 85.2%). However, there were other examples of low completeness within specific columns (e.g. 0% SB type Kushtia BD) or high completeness within non-specific columns (e.g. 100% birth outcome, Pokhara NP) highlighting that technical factors alone are necessary but not sufficient for data availability.

Other factors associated with data completeness included mode of birth, e.g. in Kushtia BD 97.1% of babies for whom birth outcome was missing were delivered by Caesarean. This finding is similar to previous research in Ethiopia, where a high proportion of babies not recorded in the register had required a clinical intervention [11]. Previous studies have highlighted the low value placed on stillbirths and the resultant data gaps [29‐31] and similarly we found birthweight data completeness was lower for these babies in Bangladesh.

Incomplete count data affect indicator calculation results. When intervention coverage numerators are missing, rates will appear low, e.g. when only the outcome “unwell” was recorded and no maternal/ newborn deaths, the MMR, NMR at discharge from L&D ward and SBR may be inappropriately low - zero in Azimpur BD and Kushtia BD during this study. When denominators such as birth outcome are missing (e.g. for 69.2% of babies in Temeke TZ and 21.4% in Kushtia BD) many indicators which use live birth as the denominator will be adversely affected. For example, coverage of breastfeeding in Temeke TZ would be 292.5% had the numerator not been restricted to include only babies in the live births denominator (i.e. exluding babies with a birth outcome unknown). Calculating coverage using a total birth denominator instead of a clinical need denominator, as we have done for neonatal resuscitation, requires benchmarked rates for meaningful tracking across hospitals.

Alternatively, data completeness may be high, but if inaccurate, coverage will be falsely low or high. For example, uterotonic coverage was apparently low in the Tanzanian registers. These data are handwritten using a Swahili abbreviation “N” for “Ndiyo” (Yes) or “H” for “Hapana” (No) which can be hard to distinguish and possibly incorrectly extracted. Numbered coding systems may be helpful when the design is simple e.g. the “Helping Babies Breathe” column in Tanzania. Blank data elements in the register can mean either “incomplete” or a true zero, as in the Bangladesh register design, which if not differentiated can introduce another source of data inaccuracy [32].

Beyond data completeness, our data quality evaluation showed variable results. Birthweight rounding and heaping were substantial across all hospitals. If a baby whose true birthweight of 2470g is rounded to 2500g the LBW rate will be underestimated – in our model by up to 7.2%. Both analogue and digital scales were used for birthweight across the five hospitals which may contribute to rounding. Additional EN-BIRTH analyses are exploring accuracy and processes of birthweight measurement [33, 34]. In these high mortality burden countries, very large variation in hospital mortality rates may suggest data quality issues; Muhimbili TZ had a stillbirth rate almost eight times higher that Azimpur BD. The EN-BIRTH mixed-methods study aims to test validity of these indicator measurements against the gold standard of observation data.

Barriers and enablers to recording in routine hospital registers are being explored in the wider EN-BIRTH study [6]. Quality of register data is affected by HMIS input determinants described by the PRISM framework [28] including technical, organisational and behavioural factors. Factors known to negatively impact routine data quality include poor use of data, lack of feedback, low management support, lack of health worker confidence, low motivation, lack of competence and low perceived utility of routine recording tasks [14, 28, 35]. Health worker training and supportive supervision regarding the importance of routine recording around the time of birth could improve data quality for all babies, especially stillbirths. Innovations to increase health worker data utilization skills could also help sustain improvements in data recording as the purpose of these activities is recognised. Previous studies have demonstrated large gains from such efforts [12‐14, 35, 36].

Further research is needed to understand the effect of labour ward register design on data quality, the impact of increased reporting burden on frontline health workers, and ways to optimize the utility of register data whilst reducing duplication. Standardized and harmonized registers with inclusion of an appropriate number of selected key data elements need evaluating against registers that contain large numbers of data elements. At the five study hospitals, all documentation at the health worker/ mother and baby interface was in paper-based routine registers. As electronic platforms increase, the effect of digitization on data quality around the time of birth requires attention from the source data to the top of the data pyramid [6].

Utilizing the EN-BIRTH multi-country study hospitals, a strength of this research is the large amount of data extracted (20,075 births), providing the first in-depth and multi-country analysis of routine labour ward register data. However, EN-BIRTH study hospitals may not be entirely representative of routine recording practices in facilities at different levels of the health system nor in other LMIC settings. Some EN-BIRTH hospitals have been involved in previous research, thus routine recording may be better than typical. Conversely, staff workload in these high-volume CEmONC hospitals could reduce data quality. We were unable to assess whether all babies born on the labour ward were recorded in the register, nor the relationship between staff levels and data quality. Research in facilities at different levels of the health system is required before wider conclusions can be drawn. Furthermore, evidence of completeness and quality of register data do not necessarily correlate with accurate aggregation and reporting from the facility up the HMIS data pyramid, therefore research is required to review quality of facility-reported data used for district/national/global tracking of MNH indicators.

The Every Newborn strategic objective to transform measurement aims to increase availability and quality of data to use for action. Unless all births occur in hospitals, facility data will overestimate population coverage. However, as hospital births increase (globally now 81% [7]) this data source is increasingly valuable. Improving facility data quality would also benefit wider health system indicators (e.g. immunization coverage) which currently use census projection data. Whilst household surveys are useful to provide information on contact with MNH services at a population level in LMICs, they have been shown to be less valid for the capture of content or coverage of interventions around the time of birth, hence new strategies incorporating multiple data sources, including register data, are required [32, 37‐39]. Clarity is needed on the calculation of Caesarean section rates; the current denominator recommendation is live births, but more than 10% of stillbirths in this dataset were delivered by Caesarean [26, 27]. In our study, inclusion of stillbirths in the denominator as well as the numerator increased Caesarean rate by nearly 6%. We propose Caesarean rates be calculated using hospital total births and stratified for live births and stillbirths.

Findings presented here could be used now by decision makers at various levels of the health system. In the hospital for quality improvement e.g. if no fresh stillbirths are being resuscitated this could lead to review of guidelines, practice and/or documentation. Birthweight data were readily available in all five hospitals, so LBW rate reporting, one of WHO 100 core health indicators could be improved [40]. If the LBW rate is implausibly low, hospitals might use the same data to improve quality of birthweight measurement. Using birthweight categories and birth outcome data, we found differences between live births and stillbirths, e.g. the differential growth of stillbirths where 36.8% weighed 1000-1999 g compared to 3.0% of live births. Our study showed that 50.6% of fresh stillbirths had a normal birthweight yet died, this metric could also be tracked to improve quality of care.

Changes in register recording practices during the EN-BIRTH study will be explored [6]. Importantly, the EN-BIRTH observational study will further validate indicators from labour ward register data to inform use in HMIS and areas of focus to further improve data availability and quality.

Data used for action is foundational for tracking progress towards global goals for every woman and every child to survive and thrive [4, 5]. As data is used, data quality and overall HMIS performance improves [14, 28]. As data quality improves, coverage and outcome indicators can more confidently be used for action to track progress and drive change.

This study shows that large amounts of specific MNH data elements are currently available in routine labour ward registers in five hospitals in Bangladesh, Nepal and Tanzania. Data quality varied when assessed for completeness and implausibility. There is potential to improve the quality of available data if HMIS utilization with feedback loops can be strengthened. By advancing routine health facility data for use, labour ward registers can contribute to much needed regular coverage and impact measurements around the time of birth (Fig. 5). To optimize care around the critical time of birth, labour ward register data offer huge potential to be improved and used.