A global reference for caesarean section rates (C‐Model): a multicountry cross‐sectional study

Objective To generate a global reference for caesarean section (CS) rates at health facilities. Design Cross‐sectional study. Setting Health facilities from 43 countries. Population/Sample Thirty eight thousand three hundred and twenty‐four women giving birth from 22 countries for model building and 10 045 875 women giving birth from 43 countries for model testing. Methods We hypothesised that mathematical models could determine the relationship between clinical‐obstetric characteristics and CS. These models generated probabilities of CS that could be compared with the observed CS rates. We devised a three‐step approach to generate the global benchmark of CS rates at health facilities: creation of a multi‐country reference population, building mathematical models, and testing these models. Main outcome measures Area under the ROC curves, diagnostic odds ratio, expected CS rate, observed CS rate. Results According to the different versions of the model, areas under the ROC curves suggested a good discriminatory capacity of C‐Model, with summary estimates ranging from 0.832 to 0.844. The C‐Model was able to generate expected CS rates adjusted for the case‐mix of the obstetric population. We have also prepared an e‐calculator to facilitate use of C‐Model (www.who.int/reproductivehealth/publications/maternal_perinatal_health/c-model/en/). Conclusions This article describes the development of a global reference for CS rates. Based on maternal characteristics, this tool was able to generate an individualised expected CS rate for health facilities or groups of health facilities. With C‐Model, obstetric teams, health system managers, health facilities, health insurance companies, and governments can produce a customised reference CS rate for assessing use (and overuse) of CS. Tweetable abstract The C‐Model provides a customized benchmark for caesarean section rates in health facilities and systems.


Group A1
Group A2 * These coefficients were calculated using the R package by an independent statistician and are very similar to those calculated using Stata (reported in the main manuscript).

Lower bound Upper bound
USA (2012) Table S8. Sensitivity analysis including only databases with complete data for all C-Model versions. (summary estimates of areas under the ROC curves with 95% confidence intervals; random effects meta-analyses).

AUC Lower bound Upper bound
Sensitivity Analysis

National Perinatal Data Collection (NPDC)
The National Perinatal Data Collection (NPDC) is a national population-based cross sectional data collection of pregnancy and childbirth. The data are based on births reported to the perinatal data collection in each state and territory in Australia. Midwives and other staff, using information obtained from mothers and from hospital or other records, complete notification forms for each birth. Information is included in the NPDC on both live births and stillbirths of at least 400 grams birthweight or at least 20 weeks gestation. The NPDC is compiled annually by the National Perinatal Epidemiology and Statistics Unit. https://npesu.unsw.edu.au/data-collection/national-perinatal-data-collection-npdc England Information on the Hospital Episodes Statistics: HES is a data 'warehouse' that includes records of all inpatient admissions and day cases in English NHS trusts, with the data being extracted from local patient administration systems. In HES, each record contains data on the patient demographics (for example, age, sex, ethnicity, postcode), the episode of care (for example, hospital name, date of admission and discharge) and clinical information. Diagnoses for each patient are recorded using the International Classification of Diseases, 10th edition (ICD-10) . Procedures performed during an episode are coded using the Office of Population, Censuses and Surveys Classification of Surgical Operations and Procedures, 4th revision (OPCS) (Health and Social Care Information Centre, 2014). In addition, each episode related to the delivery of a baby can capture details about the labour and delivery (for example, parity, mode of delivery, gestational age, birthweight) in supplementary data fields known as the HES 'maternity tail'. Over 96% of all deliveries in England occur in NHS hospitals and are therefore captured by HES (Birthplace in England Collaborative Group, 2011).

Germany
The German data are taken from a standard nationwide perinatal data set defined solely for the purposes of monitoring quality of care in German hospitals. For the present analysis the data were restricted to the state of Bavaria, constituting approximately 14% of all German births. The Bavarian data may be considered as representative for the entire German data with little regional variation. The data are not registry data compiled for the sake of generating official national reports. However, the data checks are rigorous and the perinatal data set has remained stable for more than 10 years. The quality of care in German hospitals is reported annually for many other fields of care apart from obstetric care.

Mali & Senegal
Our perinatal database was developed for the QUARITE Trial in 46 referral hospitals in Mali and Senagal. QUARITE is an international, multi-centre, controlled clusterrandomized trial of a complex intervention. Inclusion criteria for hospital are: functional operating rooms and more than 800 deliveries annually. Exclusion criteria are: private health care facility, already had a structured program for carrying out maternal death audits, written consent not provided by local authorities. Inclusion criteria for women in the QUARITE study are 1) being a patient who delivered in one of the participating facilities, 2) between September 1, 2007, and August 31, 2011. Exclusion criteria are 1) having delivered at home or 2) in another centre, with postnatal transfer. The database is based on the WHO global survey on maternal and perinatal health, which considers clinical data at the individual level and organizational data at the facility level. All deliveries carried out in the participating centres are registered by local collectors (nurses or midwives trained to do this). These collectors complete a standard form for each eligible patient that includes information on maternal characteristics, prenatal care, labour and delivery, diagnosed complications, and the vital status of both mother and child at discharge from hospital.  (1). Data collection was conducted between 2006 and 2007 in 5 hospitals with these provider types. Data was collected by direct observation of deliveries and review of medical records of women delivering at participating hospitals. Data collection was focused on evidence-based recommended obstetric care at admission, active phase of delivery and postpartum. All but one hospital had the capacity to perform c-section 24 hours a day. Information was available for all variables of C-model, except chronic hypertension. Data was insufficient to completely differentiate among breech and other non-cephalic presentations. Preterm birth was recorded as gestational age <36 weeks. Data on admission to ICU was not available. Organ dysfunction was defined as the presence of any of the following: heart failure, cardiac arrest, hemorrhagic/septic shock, disseminated intravascular coagulation, intraventricular hemorrhage, hysterectomy, and maternal death. The study that generated the data used in this validation included women admitted in labor. Thus women with an absolute indication for c-section were excluded. While the c-section rate in this study was 27%, in 2012, the Mexican c-section rate was 55% (1). The dataset from this study can be used for applying the C-model and performing ROC analysis. However, given that this is a dataset with unique characteristics, results from this validation should only be extrapolated to the facility-based care for the population of women already presenting in labor in Mexico.

Mongolia
Maternity hospitals' data of the childbirth 2011 was used for the analysis. All deliveries took in 3 maternity hospitals in the capital city and The National Center for Maternal and Child Health were collected and managed by the Ministry of Health, Mongolia. It covered all births in the Ulaanbaatar plus all high risk deliveries referred to the tertiary level from all provinces.

Sri Lanka
For the C-model validation study, we selected the premier women's hospital with a higher average number of deliveries per year in Sri Lanka. Our aim was to have a random sample of women from this hospital to represent 5% of annual admissions. Based on the available records, around 9388 deliveries take place in this hospital annually. We randomly selected 496 deliveries (5.3% of total deliveries) taken place in the months of December 2013 and January 2014. Three MBBS qualified medical officers manually extracted all variables required for the validation from paper based patient records available in record room of the hospital. There were no maternal deaths or pregnant women with HIV infections admitted to the hospital during the study period.

Thailand
In Thailand, the C model was tested in a sample of 300 deliveries randomly selected from the medical record deliveries of Kalasin hospital during September-December 2013. Kalasin hospital is a regional hospital in the northeast region. The hospital provides 40 beds for obstetric care and has an average of 3700 deliveries annually. During the four months, caesarean section rate was 52.3 %. The data was kindly supported by Dr. Bunpode Suwannachart, the vice director of Kalasin regional hospital. His email address is bunpode@yahoo.com