Linking data sources for measurement of effective coverage in maternal and newborn health: What do we learn from individual- vs ecological-linking methods?

Background Improving maternal and newborn health requires improvements in the quality of facility-based care. This is challenging to measure: routine data may be unreliable; respondents in population surveys may be unable to accurately report on quality indicators; and facility assessments lack population level denominators. We explored methods for linking access to skilled birth attendance (SBA) from household surveys to data on provision of care from facility surveys with the aim of estimating population level effective coverage reflecting access to quality care. Methods We used data from Mayuge District, Uganda. Data from household surveys on access to SBA were linked to health facility assessment census data on readiness to provide basic emergency obstetric and newborn care (BEmONC) in the same district. One individual- and two ecological-linking methods were applied. All methods used household survey reports on where care at birth was accessed. The individual- linking method linked this to data about facility readiness from the specific facility where each woman delivered. The first ecological- linking approach used a district-wide mean estimate of facility readiness. The second used an estimate of facility readiness adjusted by level of health facility accessed. Absolute differences between estimates derived from the different linking methods were calculated, and agreement examined using Lin’s concordance correlation coefficient. Results A total of 1177 women resident in Mayuge reported a birth during 2012-13. Of these, 664 took place in facilities within Mayuge, and were eligible for linking to the census of the district’s 38 facilities. 55% were assisted by an SBA in a facility. Using the individual-linking method, effective coverage of births that took place with a SBA in a facility ready to provide BEmONC was just 10% (95% confidence interval CI 3-17). The absolute difference between the individual- and ecological- level linking method adjusting for facility level was one percentage point (11%), and tests suggested good agreement. The ecological method using the district-wide estimate demonstrated poor agreement. Conclusions The proportion of women accessing appropriately equipped facilities for care at birth is far lower than the coverage of facility delivery. To realise the life-saving potential of health services, countries need evidence to inform actions that address gaps in the provision of quality care. Linking household and facility-based information provides a simple but innovative method for estimating quality of care at the population level. These encouraging findings suggest that linking data sets can result in meaningful evidence even when the exact location of care seeking is not known.

The EQUIP study was a non-randomised quality improvement intervention implemented in one district of eastern Uganda (Mayuge) [23]. Quality improvement is a strategy to improve implementation levels for evidence-based essential interventions. In the EQUIP study collaborative quality improvement teams tested self-identified strategies to support the implementation of essential maternal and newborn interventions recommended by the WHO. Throughout the study, the teams had access to locally-generated high quality health data from a continuous household survey, repeat health facility censuses, complemented by routine data from health facilities. Mayuge district has a population of approximately 400 000 people, is predominantly rural, and has an estimated maternal mortality ratio of 438 per 100 000 live births and an estimated neonatal mortality rate of 23 per 1000 livebirths, based on data from the 2011 DHS [24]. In 2014 there were 38 government owned health facilities in the district and no private birthing facilities (Figure 1). At the time of the EQUIP study all facilities were conducting births, but level II facilities had only recently been upgraded to conduct births due to increases in demand for facility-based delivery in the locality. Map of Mayuge district showing location of household clusters and health facilities included in this analysis. (A) Household clusters. Yellow dot – household cluster included in household survey. (B) Health facilities. Red cross – health centre II, Green cross – health centre III, Blue cross – health centre IV, Purple cross – hospital. Full details of the data collection protocol for the project have been reported elsewhere [22]. In brief, a continuous population level cluster survey designed to represent the district at multiple time points, and six repeat health facility censuses in the district were implemented between November 2011 and April 2014 (Figure 2). Questionnaires were adapted from Demographic and Health Surveys (DHS) tools [25]. The household survey comprised of a household module capturing information on household characteristics and residents, and a women’s module addressed to all female residents aged 15-49 years. Women aged 15-49 years who reported a live birth in the two years prior to survey were also asked a detailed set of questions about the antenatal, intrapartum, and postnatal care they and their infant had received. The repeat facility census included a modular check-list type questionnaire including staff employed, drugs, supplies and equipment. Sample selection. For this analysis a sub-set of EQUIP data were analysed. From the continuous household survey data sets inclusion criteria were as follows: women aged 15-49 years who reported a live birth in fixed reference period (24-month period 1 January 2012 and 31 December 2013); and women who delivered in the district of residence (to maximise the potential to link household data on place of birth to facility readiness data) (Figure 2). The reported name and level of facility (Health Centre II, III or IV, and hospital) for each included birth were identified and cross-checked against the facility census list. Where inconclusive, reported facility names and levels were returned to the survey field team for final verification. In total six facility census data sets were available from the EQUIP study. For this analysis, the EQUIP health facility census round three (1 November 2012 – 28 February 2013) was selected as that representing measures of the service environment at the mid-point of the household survey reference period. Variability of quality of care indicators across the six censuses was examined to consider the stability of the quality estimates over time. Currently, there are not standardised agreed metrics for measuring the of quality of care for maternal and newborn health [8]. In this study we adapted measures from signal functions to provide routine and emergency obstetric and newborn care proposed by Gabrysch and colleagues [18]. Gabrysch and colleagues suggest four dimensions of facility care including general requirements, routine care, basic and comprehensive emergency care. Signal functions refer to health workforce availability and skills as well as availability of commodities to deliver care, in addition to life-saving behaviours [18]. The measure used in our study included the availability of commodities to provide routine care and basic emergency obstetric and newborn care (BEmONC). These were categorised into a binary indicator as to whether the facility had all of these commodities available or not. These components are listed in Box 1. These included availability of: infrastructure (electricity and running water); infection prevention measures; commodities to monitor and manage labour; essential medicines; commodities to provide clean cord care; and commodities to carry out neonatal resuscitation (Box 1). These six components were combined using equal weighting to represent one binary indicator of ‘facility readiness to provide basic emergency obstetric and newborn care’. These indicators were then linked to household observations on place of birth to estimate effective coverage of skilled birth attendance in a health facility ready to provide BEmONC. Facility readiness components include: Infrastructure – had a source of electricity and running water 24hr/day Infection prevention – had commodities for infection prevention available (disinfectant, disposable gloves, soap, sharps box, sterilizer) Monitoring labour – had commodities to monitor and manage labour available (blood pressure cuff, timer, urine protein dipstick, fetal stethoscope, thermometer) Essential drug – had essential drugs for management of complications in mothers and babies available (parenteral antibiotics for maternal infection and newborn sepsis, parenteral anticonvulsants, parenteral oxytocics for haemorrhage and uterotonics for active management of the third stage of labour, AMSTL) Neonatal resuscitation – had commodities for neonatal resuscitation available (bag and mask) Clean cord care – had commodities for hygienic core care available (sterile cord cutter and cord tie) All components – had all commodities for all six indicators available The household data set included a variable on skilled birth attendance (SBA), constructed using standard definitions. Each of the recent births attended by an SBA was assigned a facility readiness score based on three link methods as described below. The individual-linking method was considered as the gold standard as this linked the participant’s information from household surveys to the precise health facility at which they sought care. For individual household observations reporting skilled birth attendance, the facility readiness to provide BEmONC (Box 1) was merged in by matching name of health facility between household and facility data sets. Home births were coded as having no facility readiness. Population level tabulations of effective coverage were then made. Ecological-linking was carried out for a mean facility readiness score at the district level – thus not accounting for the number of service users (volume of births) or readiness at different levels in the health system. Using the same household and facility data sets as method A, each facility birth from the household data set was assigned the mean facility readiness status for the district as a whole (all health facilities combined); home births were again coded as having no facility readiness, and population level tabulations of effective coverage were made. In the second ecological-linking method, linking was carried out by level of facility because different levels in the health system were not equally well-equipped and had different numbers of service users. The same household survey sample of women was included as in the individual-linking method. The facility data set was collapsed by level of facility and readiness indicators tabulated for each level (level II, III, IV or hospital). For each individual household observation with a delivery attended by an SBA, the facility readiness status for the reported level of facility was merged in. Home births again were coded as having no facility readiness, and population level tabulations of effective coverage made. For each of the three linked data sets, “effective coverage of skilled birth attendance in facilities providing basic emergency obstetric and newborn care” was calculated as the product of (i) the prevalence of attendance by an SBA in a health facility within Mayuge District and (ii) the prevalence of facility readiness for each quality of care indicator. Confidence intervals surrounding the estimate of effective coverage for each quality of care indicator were calculated using the Delta method [26]. The absolute differences between effective coverage estimates from the three linking methods were examined. Agreement between linking methods was examined using Lin’s concordance correlation coefficient and Bland and Altman plots to investigate the existence of any systematic difference between the measurements (ie, fixed bias) and to identify possible outliers [27]. Ethical clearance for the EQUIP study was obtained from the Uganda National Council of Science and Technology, Makerere University School of Public Health, and the London School of Hygiene and Tropical Medicine (LSHTM). This study underwent human subjects review process at CDC, Atlanta and was approved as not being engaged in human subjects’ research. Advocacy and sensitization meetings with district and sub-district authorities were held at the start of the EQUIP study. Communities and health facilities were informed about the survey by a survey team member one day prior to interview, using information sheets in the local languages. Written, informed consent to participate in the surveys was obtained from household heads, women, facility in-charge, and health staff interviewed. In the case of illiterate participants, the translated informed consent sheet was read aloud to the participant in the presence of a literate neighbourhood witness who confirmed the content of the consent sheet, and informed consent was obtained by means of thumb print from the illiterate participant and signature from the literate neighbourhood witness.

N/A

• Dima Health