Does hospital variation in intrapartum-related perinatal mortality among caesarean births reflect differences in quality of care? Cross-sectional study in 21 hospitals in Burkina Faso

Objectives To examine hospital variation in crude and risk-adjusted rates of intrapartum-related perinatal mortality among caesarean births. Design Secondary analysis of data from the DECIDE (DECIsion for caesarean DElivery) cluster randomised trial postintervention phase. Setting 21 district and regional hospitals in Burkina Faso. Participants All 5134 women giving birth by caesarean section in a 6-month period in 2016. Primary outcome measure Intrapartum-related perinatal mortality (fresh stillbirth or neonatal death within 24 hours of birth). Results Almost 1 in 10 of 5134 women giving birth by caesarean experienced an intrapartum-related perinatal death. Crude mortality rates varied substantially from 21 to 189 per 1000 between hospitals. Variation was markedly reduced after adjusting for case mix differences (the median OR decreased from 1.9 (95% CI 1.5 to 2.5) to 1.3 (95% CI 1.2 to 1.7)). However, higher and more variable adjusted mortality persisted among hospitals performing fewer caesareans per month. Additionally, adjusting for caesarean care components did not further reduce variation (median OR=1.4 (95% CI 1.2 to 1.8)). Conclusions There is a high burden of intrapartum-related perinatal deaths among caesarean births in Burkina Faso and sub-Saharan Africa more widely. Variation in adjusted mortality rates indicates likely differences in quality of caesarean care between hospitals, particularly lower volume hospitals. Improving access to and quality of emergency obstetric and newborn care is an important priority for improving survival of babies at birth. Trial registration number ISRCTN48510263. This study is a secondary analysis of the DECIDE cluster-randomised controlled trial, which assessed the effectiveness of a multicomponent intervention including provider training, caesarean audits and SMS reminders to reduce non-medically indicated caesarean sections. The trial included three phases: 6-month preintervention, 1-year intervention and 6-month postintervention. It was conducted in all 22 regional and district hospitals in Burkina Faso performing more than 200 caesareans per year in 2012; university hospitals in Ouagadougou and Bobo-Dioulasso were excluded. Detailed trial methods are described elsewhere.18 Similar to other West African countries, the caesarean rate in Burkina Faso is below 5% (3.7% in 2010–2015),19 with large urban–rural, wealth and educational differentials.20 21 Although 85% of births take place in health facilities, 70% occur in primary care facilities without surgical capacity.22 Women who develop complications requiring a caesarean are referred to medical centres with surgical capacity (centres médicaux avec antenne chirurgicale, referred to as district hospitals hereafter) or regional hospitals. Women with severe complications may be referred onwards to tertiary university hospitals in the capital Ouagadougou and second largest city Bobo-Dioulasso. Most—but not all—district and regional hospitals have at least one obstetrician or generalist doctor trained in emergency obstetric care. Task-shifting of caesarean care has been supported in Burkina Faso through additional 3-year training of nurses and midwives as non-physician providers with surgical skills (attachés en chirurgie) and obstetrics skills (attachés en gynéco-obstétrique). Most anaesthesia care is provided by nurses with additional training in anaesthesia. More than three quarters of study hospitals did not have Doppler ultrasounds, CTG monitors or ultrasound capacity, relying on Pinard stethoscopes for assessment of fetal well-being. Fetal scalp pH was only available in one hospital.18 Emergency obstetric care has been subsidised to improve access since 2006, initially with an 80% subsidy of the cost of caesareans, which were made free to women from 2016 onwards. Hospitals are reimbursed according to the number of caesareans and vaginal births. This policy absorbed around 3.5% of total health expenditure in 2011.23 However, some costs (formal or informal) not included in the ‘free’ package continue to be borne by households and remain unaffordable for some.24 25 Women express fears around caesarean birth related primarily to poor quality of care and economic burden.26 We included all 5134 women giving birth by caesarean section in the 21 study hospitals with caesarean capacity in the postintervention phase (2 May–2 November 2016). One study hospital’s operating theatre was no longer functional in the postintervention phase. These 21 hospitals accounted for 45% of all caesarean sections performed nationally in 2016.27 Women delivering by caesarean were included regardless of gestational age, whether they were referred to the study hospital before the caesarean or referred to another hospital after birth. Patient medical records were used in the DECIDE trial, with prospective data collection in the postintervention phase using data extraction forms and standardised clinical definitions (including for labour dystocia, acute fetal distress and indications for caesarean).18 We used postintervention data to provide the most recent description for a larger sample. We defined intrapartum-related perinatal mortality as the rate of fresh stillbirths and very early neonatal deaths (within 24 hours of birth) per 1000 caesareans.28 29 Intrapatum-related mortality is recommended by the WHO as an indicator of the quality of emergency obstetric and newborn care.30 We examined two groups of risk factors for intrapartum-related mortality: individual-level clinical risk factors, and caesarean care components and hospital characteristics. We conceptualised case mix as the hospital prevalence of clinical risk factors for intrapartum-related mortality (maternal age, parity, highest educational level achieved, previous caesarean, multiple pregnancy, number of antenatal visits, birth weight, congenital malformation, referral status and distance, labour phase, diagnosis of acute fetal distress, transverse lie/brow presentation in active labour, other severe obstetric complication or maternal death and primary indication for caesarean). ‘Other severe obstetric complications’ included severe pre-eclampsia or eclampsia, retroplacental haematoma, uterine (pre-)rupture and placenta praevia in active labour. Uterine prerupture was defined as women with severe dystocia and signs of prerupture, such as Bandl’s ring. Acute fetal distress was defined as fetal heart rate <120 or >160 bpm, either persistent after oxygen administration and lateral decubitus position, or with IUGR, placental abruption, prolonged labour, maternal fever or meconium-stained amniotic fluid. Some women diagnosed with acute fetal distress had a primary indication for caesarean other than ‘fetal distress’ (eg, pre-eclampsia), while some women had a caesarean with ‘fetal distress’ recorded as the primary indication despite not having met the diagnostic criteria for acute fetal distress. We conceptualised components of caesarean care (provider cadre deciding and performing the caesarean, decision-to-incision interval, anaesthesia type, skin/uterine incision type and antibiotic prophylaxis administration) and hospital characteristics (hospital type and monthly caesarean volume) as potential indicators of quality of patient care. Monthly caesarean volume was calculated as the mean number of caesareans performed per month in the study period, per hospital. We used these risk factors to derive two sets of risk-adjusted mortality rates per hospital: adjusting for case mix only, and additionally adjusting for components of care and hospital characteristics, because some of these variables might capture unmeasured differences in case mix. For example, women receiving general anaesthesia are more likely to have complications requiring urgent surgery. Including these additional variables also allowed us to identify whether any care components (eg, decision-to-incision interval) were strongly associated with mortality. We included care components prior to delivery as risk factors even when they were not hypothesised to causally affect perinatal mortality, since they may be proxies for quality of care. Data were complete for the outcome and nine risk factors, including multiple gestation, indication for caesarean and referral status (online supplemental table 1). Eleven risk factors had <5% missing values; six risk factors had >5% missing data, including decision–incision interval (24%) and timing of antibiotic administration (23%). Overall, 68% of women had at least one risk factor missing, and 4% had at least four risk factors missing (online supplemental table 2). Missing information on previous caesarean was assumed to indicate no previous caesarean (n=40), and missing deciding provider cadre was imputed as the hospital mode for seven women. bmjopen-2021-055241supp001.pdf Multiple imputation by chained equations was used for other variables to avoid a loss in efficiency, because missing values were likely to be missing at random given known risk factors, including referral status and severe obstetric complication.31 Five imputed datasets were created using the mi package in Stata V.14.2, including all risk factors and intrapartum-related mortality in the imputation model. The same model was used for all hospitals, with hospital type included as a risk factor. Missing values for continuous risk factors (age, parity, number of antenatal care visits, referral distance, birth weight and decision-to-incision interval) were imputed from linear regression models, missing values for binary risk factors (acute fetal distress, antibiotic prophylaxis, incision type, anaesthesia type, congenital malformation and neonatal resuscitation) were imputed from logistic regression models and categorical risk factors (education, provider cadre performing the caesarean, and timing of antibiotic administration) were imputed from multinomial regression models. Gestational age at birth had >50% missing data; it was not considered as a risk factor in the analysis model, since it is highly correlated with low birth weight, which was more complete and likely to be more accurate in a setting without routine ultrasound in the first trimester. However, we included gestational age at birth in the imputation model to improve the prediction of birth weight. Distributions of imputed values were compared with observed values for variables with >5% missing data. First, we calculated crude hospital intrapartum-related mortality rates with 95% CIs and described perinatal outcomes according to hospital type. Differences in hospital case mix were assessed by describing the prevalence of clinical risk factor for intrapartum-related mortality among women giving birth by caesarean, stratified by hospital and hospital type. We similarly described differences in components of care received. χ2 tests accounted for clustering of women by hospital using the svyset package in Stata. Next, we built two multivariable models for intrapartum-related death among caesarean births using multilevel logistic regression models of women, nested in hospitals to account for clustering. The first model (model 1) adjusted for case mix only and included all individual-level clinical risk factors for intrapartum-related mortality with Wald test p value ≤0.25 in bivariate associations, using manual backward selection to retain only variables with p values <0.1. The second model (model 2) built on model 1 by additionally including all care components and hospital characteristics with bivariate Wald test p value ≤0.25 and similarly using backward selection to retain only p values <0.1. Multicollinearity was examined by reviewing Spearman correlations and model SEs. In building model 2, provider cadre deciding the caesarean met the criteria for inclusion; however, its inclusion reduced the hospital-level estimate almost to zero, indicating that this variable acted as a proxy for broader differences between hospitals. Further inspection showed that deciding providing cadre was highly clustered within hospitals, with one category accounting for >90% of women in 13 of 21 hospitals. We therefore removed it from risk factors considered for model 2. We calculated the median OR for models 1 and 2 as a measure of interhospital variation in mortality that is not explained by the model covariates, expressed on the OR scale (see formula in online supplemental figure 1).32 For a multilevel model, the median OR is defined as the median of the ORs that could be calculated by comparing two patients with identical individual-level characteristics from two, randomly chosen, different hospitals.33 34 Risk-adjusted mortality enables comparisons in hospital outcomes taking into account differences in case mix.15–17 Risk-adjusted intrapartum-related mortality rates were calculated for each hospital by multiplying the intrapartum-related mortality rate across the study sample by the ratio of the number of observed deaths to predicted deaths based on models 1 and 2 in each hospital. Bootstrapping with 1000 iterations was used to calculate 95% CIs around both sets of risk-adjusted hospital mortality rates and found to produce stable estimates. We used the Boot MI percentile method to produce CIs with nominal coverage.35 We constructed graphs showing risk-adjusted mortality and CIs for each hospital, according to the mean monthly number of caesareans in each hospital, to visually assess any associations between risk-adjusted mortality and caesarean volume (figure 1A–C). Crude and risk-adjusted hospital intrapartum-related mortality rates among women giving birth by caesarean section in 21 hospitals, according to mean monthly number of caesareans – Burkina Faso, 2016. The DECIDE trial found a reduction in avoidable caesareans,36 suggesting changes in caesarean decision making that may affect intrapartum-related mortality. As a secondary analysis, we added trial group as a risk factor to model 2 to determine whether it was associated with mortality after adjusting for other covariates. No patients were involved in the design, conducting, reporting or dissemination of this study.