Share this article on:

A Prediction Score for Maternal Mortality in Senegal and Mali

Huchon, Cyrille MD, PhD; Dumont, Alexandre MD, PhD; Traoré, Mamadou MD; Abrahamowicz, Michal PhD; Fauconnier, Arnaud MD, PhD; Fraser, William MD, MSc; Fournier, Pierre MD, MSc

doi: 10.1097/AOG.0b013e31828b33a4
Original Research

OBJECTIVE: To develop and validate a maternal mortality score to identify patients at risk of in-hospital death in developing countries.

METHODS: We performed a prospective observational study in 46 referral hospitals in Senegal and Mali, starting October 1, 2007. Derivation of a maternal mortality score was performed, using generalized estimating equation, on patients included during the first 6 months of the study (301 deaths out of 43,624 deliveries) and validated on patients included during the next 6 months (345 deaths out of 46,328 deliveries).

RESULTS: Nine criteria were independently associated with maternal death: severe anemia in pregnancy, malaria diagnosed during pregnancy, parity greater than 4, fewer than three antenatal visits, referral from another health facility, antepartum or postpartum hemorrhage, preeclampsia or eclampsia, uterine rupture, and genital infection or sepsis. The maternal mortality score, ranging from 0 to 100, occupies an area under the receiver operating characteristics curve of 0.89 (95% confidence interval [CI] 0.87–0.91). The low-risk group for maternal mortality, based on a score less than 10, has a negative predictive value of 99.9% (95% CI 99.8–99.9) and a negative likelihood ratio of 0.18, ruling out maternal mortality with a probability of 0.13% (95% CI 0.09–0.17). Sensitivity of the score to identify patients at risk of in-hospital death was 85.0% (95% CI 80.5–88.8). Validation of the score yielded a sensitivity of 87.8% (95% CI 83.9–91.1), a negative predictive value of 99.9% (95% CI 99.8–99.9), and a probability of maternal death of 0.12% (95% CI 0.08–0.17) in the low-risk group.

CONCLUSION: The maternal mortality score could help health care professionals to identify patients at risk of maternal mortality who need careful management.


A maternal mortality score, derived and validated in populations from Senegal and Mali, can be used to detect patients at risk for in-hospital maternal mortality.

International Health Unit, Centre de Recherche du Centre Hospitalier de l'Université de Montréal, the Centre de Recherche du CHU Sainte Justine, and the Department of Gynecology and Obstetrics, CHU Sainte Justine, Université de Montréal, and the Department of Epidemiology and Biostatistics, McGill University, Montreal, Quebec, Canada; the Institut de Recherche pour le Développement, Paris, France; Faculté de Pharmacie, Université Paris Descartes, Sorbonne Paris Cité, France; URFOSAME, Referral Health Center of Commune V, Bamako, Mali; and EA 7285 Risk and Safety in Clinical Medicine for Women and Perinatal Health, University of Versailles St-Quentin, Versailles, France.

Corresponding author: Cyrille Huchon, MD, PhD, International Health Unit, Centre de Recherche du Centre Hospitalier de l'Université de Montréal/Université de Montréal, Canada; e-mail:

Funded by the Canadian Institutes of Health Research (CIHR), grant number 200602MCT-157547-RFA-CFCF-100169. Dr Huchon is supported by the Strategic Training Program in Global Health Research, a partnership of the CIHR and the Québec Population Health Research Network and a mobility grant from the French National College of Gynecologists and Obstetricians. Dr Dumont is supported by grants from CIHR and Fonds de Recherche en Santé du Québec. Dr Fraser is supported by CIHR through a Canada Research Chair. The ALARM International Program (trial intervention) was coordinated by the Society of Obstetricians and Gynecologists of Canada in collaboration with the Ministry of Health, Faculty of Medicine and professional associations in each country. The funders played no role in the study design, data collection and analysis, the decision to publish, or the preparation of the article.

Financial Disclosure The authors did not report any potential conflicts of interest.

Presented at the 36th Meeting of the French National College of Gynecologists and Obstetricians, December 5–7, 2012, Paris-La Défense, France.

The authors thank the medical and administrative staffs at the participating centers for help with data collection and The Centre de Recherche du Centre Hospitalier de l'Université de Montréal for help with language editing.

In sub-Saharan Africa, maternal mortality and morbidity are major problems. Reducing these rates is the aim of the fifth Millennium Development Goal, which is very unlikely to be attained in this region of the world.1

In-hospital maternal mortality in sub-Saharan African countries is high, although it varies markedly between studies (0.1–5%).2–10 The leading causes of maternal death are hemorrhage, preeclampsia or eclampsia, prolonged obstructed labor, uterine rupture, puerperal sepsis, and abortion complications. In some contexts, indirect causes of maternal death such as human immunodeficiency virus, acquired immunodeficiency syndrome and malaria make a significant contribution.11 Several studies have identified factors independently associated with in-hospital maternal mortality.2–5,8,9,12 However, independent prognostic factors are quite variable between studies, probably as a result of differences in patient populations, settings, variables collected, and statistical methods. Thus, it remains difficult to provide clinicians working in maternity wards in poor-resource settings with simple and reliable criteria for identifying patients at risk of in-hospital death to help them decide whether the patient should be managed on a high-priority basis and in which setting.

Therefore, a large multicenter study was conducted in Mali and Senegal during the preintervention period of a cluster randomized trial to measure in-hospital maternal mortality before implementation of a quality care improvement program13 and assess predictors of in-hospital mortality in patients delivering at referral hospitals included in the trial. We sought to develop and validate a score that would be useful in clinical practice to identify patients at risk of maternal mortality in developing countries.

Back to Top | Article Outline


The trial was approved by the ethics committee at Sainte-Justine Hospital in Montreal, Canada, which manages the operating funds, and by the national ethics committees in Senegal and Mali. The QUARITE (quality of care, risk management and technology in obstetrics) trial is registered on the Current Controlled Trials web site under number ISRCTN46950658 (

The study protocol of the QUARITE trial and patient report form has already been reported elsewhere.13 The present prospective observational study was conducted at 46 referral hospitals (24 in Senegal and 22 in Mali). A hospital was eligible for the trial if it had functional operating rooms and performed more than 800 deliveries annually. The 46 hospitals included in the trial are representative of the existing health system in Senegal and Mali, taking into account the variety of contexts (urban compared with rural) and levels of care (primary compared with secondary referral health facilities). The centers were asked to include prospectively all consecutive patients delivering at the maternity unit from October 1, 2007, to September 30, 2008, in Senegal and from November 1, 2007, to October 31, 2008, in Mali. Patients were followed up in the hospital until they were discharged. Because this study was carried out during the preintervention phase of the trial, there were no constraints or guidelines covering investigations, admission and discharge decisions, or treatments.

All patients who delivered in one of the participating facilities during the study period were eligible for enrollment, but women who delivered at home or in another center were not eligible. Women who died before delivery or were transferred to another hospital after delivery were excluded from the statistical analyses presented in this article.

A comprehensive operations manual describing study procedures, data collection requirements, and variable definitions was provided to each participating institution. In each center, a midwife and a physician were in charge of data collection and quality control. Any procedural or definitional questions from the data collectors were referred to the study coordinator, who visited each hospital every 3 months. Data were collected on standardized patient forms and then entered using double-entry, out-of-range, logical-error checking and data compare procedure of Epi-Info 2000 software. The quality control of clinical data were carried out in three stages. The first stage corresponded to the quarterly visits of the national coordinator. During these visits, the coordinator verified that the data collection was exhaustive by comparing the number of eligible patients on the hospital's birth register with the number of patient forms collected. A complementary procedure was carried out to monitor the thoroughness of data on maternal deaths, which are generally underregistered in the maternity ward, by identifying the eligible maternal deaths among all the female deaths that occurred in the facility using the various registries available: admissions, hospitalizations (maternity and other services), operating rooms, and morgue. On a random sample of patient forms, the coordinator checked the quality of data collected. The completion rate was estimated as the proportion of patient forms that contained 100% of the following information: date of entry, patient identification, date of discharge, and vital status of the mother on date of discharge. The concordance rate was estimated as the proportion of patient forms whose information was concordant with the hospital registers and medical records. Both the completion rate and the concordance rate were expected to be above 75%. If the completion or concordance rate was between 50% and 75%, the coordinator checked the data quality on a new random sample of patient forms. A completion or concordance rate of less than 50% triggered the verification of all patient forms. A second control of missing or abnormal data was carried out at the national coordination centers before data entry. If necessary, the missing information was obtained from the collectors of each facility by telephone or fax. The third step was carried out by the data manager after the data were entered and transmitted to the trial coordination center at the Université de Montréal. An audit report of the database, including lists of duplications and missing or abnormal data, was sent quarterly to the national coordination centers, which were responsible for correcting any errors.

Data collected included patient demographic information, history, antenatal care attendance, pathologies diagnosed during pregnancy, type of admission, mode of delivery, obstetric complications, treatments, and vital status of the mother and newborn(s) at hospital discharge. The set of variables was chosen for clinical reasons and should be easily measured by health care staff with various competencies at admission time and during labor and delivery.

In-hospital maternal mortality was measured with the vital status of the mother (dead) before hospital discharge. To develop the prediction score for maternal mortality, available patients from the first 6 months of the study in each country were selected to constitute the developmental data set, whereas the remaining patients became the validation data set.

Each of the possible explanatory variables collected for each patient in the trial was independently evaluated for its association with maternal mortality using Pearson's χ2 for qualitative variables and Student’s t test for quantitative variables. Continuous variables that yielded P values <.1 in the univariable analysis were dichotomized based on the area under the receiver operating characteristics curve for use in a predictive model.14

Primary analyses relied on a generalized estimating equation extension of multivariable logistic regression with exchangeable covariance structure to account for the potential clustering of the deaths within hospitals.15 Initial generalized estimating equation model included all variables with P values <.10 in the univariable analyses. Then backward elimination (with P>.05 criterion for elimination) was used to select the best combination of variables that were independent statistically significant predictors of in-hospital maternal mortality.

Finally, the maternal mortality score was based on the items found to be statistically significant (two-tailed P<.05) in the final multivariable generalized estimating equation model. The number of maternal mortality score points contributed by each score item was obtained by rounding up the coefficients of the multivariable generalized estimating equation model to generate a simple scale. The area under the receiver operating characteristics curve of the maternal mortality score was calculated with 95% confidence interval (CI). The probability of maternal mortality, sensitivity, specificity, positive likelihood ratio, negative likelihood ratio, and positive and negative predictive values were calculated for different maternal mortality score values. When a score for a given item was missing for a patient, the item was considered absent and the corresponding item score was assigned a value of 0.

Risk groups for maternal mortality were then constructed by cutoffs for the maternal mortality score, selected using the receiver-operating characteristics curve, to maximize classification rates and usefulness of the score. Probability of maternal mortality was calculated with 95% CI in each risk group. Sensitivity of the score was calculated by the number of patients who died, not classified in the low-risk group, on the total of maternal deaths. The predictive ability of the maternal mortality score was then tested by using the validation data set to confirm the classification accuracy of the score. Analyses were carried out using Stata 11.0.

Back to Top | Article Outline


Of the 90,284 enrolled patients, 332 were excluded (177 women who died before delivery and 155 women transferred to another health facility), leaving 89,952 for the development and validation of the maternal mortality score. The first 6 months of the study included 43,624 patients with an in-hospital maternal mortality rate of 0.69% (301 deaths out of 43,624 deliveries) for derivation of the score, whereas the second period of study used for the validation of the score included 46,328 patients with an in-hospital maternal mortality rate of 0.74% (345 deaths out of 46,328 deliveries). In-hospital maternal mortality rates were not significantly different between the two periods (P=.33). Table 1 presents, for each country and type of hospital, the number of patients, in-hospital maternal mortality rate, age, type of admission, proportion of patients with obstetric complications, cesarean delivery, and length of stay in the hospital. The in-hospital maternal mortality rates ranged from 0.15% in the hospitals in Bamako, the capital of Mali, to 1.31% in the regional hospitals of Senegal, but it is important to note that these are crude mortality rates. The proportion of women referred from other health care facilities or who presented at least one obstetric complication and cesarean delivery rates were higher in regional or district hospitals than in the hospital located in the capital cities (Dakar or Bamako). In all the hospital strata, the length of stay in the hospital was higher for cesarean deliveries than for vaginal deliveries.

Table 1

Table 1

Univariable analysis of potential predictors of maternal mortality is presented in Table 2. Definitions of the variables are presented in Box 1. Of the 15 potential variables selected for the model, nine were significant in the generalized estimating equation and were included in the final prediction score. The proportion of patients with missing values for these variables ranged from 0.04% (parity) to 1.08% (number of antenatal visits). Table 3 presents the coefficients of the multivariable generalized estimating equation model for maternal deaths in the developmental sample.

Table 2

Table 2

Back to Top | Article Outline

Box 1 Definitions of Variables Cited Here...

Parity: Number of full-term pregnancies for the patient.

Number of antenatal visits: Antenatal visits during current pregnancy.

Referral from other health facility: Patient who was admitted to another health care facility and secondarily referred to the hospital, whatever the means of transportation.

Genital infection or sepsis: Chorioamnionitis, fever greater than 38.5°C, purulent vaginal discharge, bacteriemia, or septicemia.

Preeclampsia or eclampsia: Hypertension greater than140/90 mm Hg with proteinuria greater than 1+ on dipstick, convulsions.

Antepartum or postpartum hemorrhage: Vaginal bleeding before delivery or excessive blood loss during the third stage of labor with or without transfusion.

Malaria: Diagnosed in the current pregnancy.

Severe anemia: Hemoglobin less than7 g/dL during pregnancy before delivery.

Uterine rupture: Clinical symptoms (pain, fetal distress, acute loss of contractions, hemorrhage) or intrauterine fetal death that led to laparotomy, at which the diagnosis of uterine rupture is confirmed; or laparotomy for uterine rupture after vaginal birth.

Prolonged or obstructed labor: Occurring before delivery.

Premature rupture of membranes: Large gush of clear vaginal fluid before the beginning of labor.

Prolonged pregnancy: Pregnancy greater than 42 weeks of amenorrhea.

Preterm labor or preterm birth: Before 37 weeks of amenorrhea

HIV or AIDS: Positive HIV serology or AIDS during the current pregnancy.

Gestational diabetes mellitus: Diagnosed during the current pregnancy.

Suspected macrosomia: Excessive symphysis-fundal height for gestational age (single pregnancy).

Intrauterine growth restriction: Insufficient symphysis-fundal height or biometry (echography) for gestational age.

Pathologic pelvis: Contracted pelvis with a diminution of the transversal diameter of the pelvis or an asymmetric pelvis.

HIV, human immunodeficiency virus; AIDS, acquired immunodeficiency syndrome.

Table 3

Table 3

The score is made up of the following variables: parity, malaria diagnosed during pregnancy, severe anemia diagnosed in pregnancy, number of prenatal visits, type of admission (women referred urgently from other health facilities or coming by themselves), and four major obstetric complications (preeclampsia or eclampsia, antepartum or postpartum hemorrhage, uterine rupture, genital infection, or sepsis).

The results of the data quality analyses confirmed that the variables in the prediction score demonstrated good reliability. The rate of agreement between the patients' forms and clinical records for selected variables was 83.7%. For the four variables related to demographic characteristics, medical history, and current pregnancy, information was either collected by interviewing the patient and her relatives or by reviewing the patient's prenatal record. For the five variables related to obstetric complications, the diagnosis during the first 24 hours in the hospital is taken into account.

The points assigned to each of the nine predictors in the score are presented in Box 2. Points assigned for each variable range from three (for parity greater than four) up to 19 (for severe anemia diagnosed during pregnancy). Total value of the score for a patient was obtained by adding the points for each item present in a patient.

Back to Top | Article Outline

Box 2 Predictors of Maternal Mortality and Points Assigned Cited Here...

Parity greater than four: 3 points

Fewer than three antenatal visits: 5 points

Referral from other health facility: 8 points

Malaria: 10 points

Uterine rupture: 13 points

Genital infection or sepsis: 13 points

Preeclampsia or eclampsia: 14 points

Antepartum or postpartum hemorrhage: 15 points

Severe anemia: 19 points

Total maternal mortality score: out of 100 points

The area under the receiver operating characteristics curve (Fig. 1) of the score is 0.89 (95% CI 0.87–0.91). Using the appropriate logistic transformation, probability of maternal mortality can be calculated using the equation: probability of maternal death=1[Fraction Slash][1+exp (0.125×score value−6.51)]. To avoid this calculation, risk groups referring to the score values are presented in Table 4.

Fig. 1

Fig. 1

Table 4

Table 4

The low-risk group comprised patients with maternal mortality scores less than 10 (Table 4). For those patients who had a total maternal mortality score less than 10 points in the developmental population, the probability of maternal mortality was 0.13% (45 of 35,258, 95% CI 0.09–0.17). This cutoff value of 10 has a negative predictive value of 99.9% (95% CI 99.8–99.9) and a negative likelihood ratio of 0.18. A total maternal mortality score greater than 10 provided a sensitivity of 85.0% (95% CI 80.5–88.8) to detect in-hospital maternal mortality.

The intermediate-risk group comprised patients with maternal mortality scores ranging from 10 to 25 points. In the developmental population, the probability of maternal mortality was 1.79% (95% CI 1.49–2.15) in this intermediate-risk group.

The high-risk group comprised patients with maternal mortality scores greater than 25 points. In this high-risk group, the probability of maternal death (positive predictive value) in the developmental population was 7.71% (138 of 1,791, 95% CI 6.51–9.04) (Table 4). This cutoff value of 25 points for the mortality score produced high specificity (96.2%, 95% CI 96.0–96.4) and positive likelihood ratio (12.1) values.

Probabilities of maternal death in the validation data set, provided by the score, were in the expected ranges in each risk group. Observed mortality was 0.12% (42 of 36,202, 95% CI 0.08–0.17%) in the low-risk group for maternal death, 1.58% (123 of 7,761, 95% CI 1.32–1.89) in the intermediate-risk group, and 7.61% (180 of 2,365, 95% CI 6.57–8.75) in the high-risk group. Application of the score to those patients yielded an area under the receiver operating characteristics curve of 0.90 (95% CI 0.89–0.92). In the validation data set, a cutoff value of 10 provided a sensitivity of 87.8% (95% CI 83.9–91.1) and an negative predictive value of 99.9% (95% CI 99.8–99.9). A cutoff value of 25 yielded a specificity of 95.2% (95% CI 95.0–95.4) in the validation data set.

Back to Top | Article Outline


The score developed in this study provides useful knowledge to health care professionals in referral hospitals in Mali and Senegal and can serve as a reference in their decision to manage patients delivering in their health facilities.

In-hospital maternal mortality rates were highly variable between facilities, ranging from 0.15% to 1.31% of all deliveries. As reported for other hospitals in African countries, hemorrhage and preeclampsia or eclampsia are major causes of maternal mortality.16 As expected, the following obstetric complications were associated with a high case-fatality rate: uterine rupture, genital infection or sepsis, hemorrhage, and hypertensive disorders.16,17 Furthermore, women referred from other health facilities were at higher risk of in-hospital maternal mortality, as shown by other authors.17

Indirect causes of in-hospital maternal mortality revealed in our score have already been highlighted in other studies.11 Anemia has been investigated as the cause of 4% of maternal deaths in Africa in a systematic review.16 Patients presenting severe anemia are more likely to die from hemorrhage, even moderate. Malaria is also known to be a cause of in-hospital maternal mortality in endemic countries.11,17 Inclusion of the diagnosis of malaria in our score confirms the need to take indirect causes of maternal mortality into account when dealing with obstetric complications, as suggested by other authors.11

We used data collected in most referral hospitals of Senegal and Mali. All deliveries were prospectively recorded and data quality was regularly controlled.13 The 46 hospitals included in the trial were representative of the existing health systems in both countries. Our findings may be generalized to other countries with similar health care systems and other referral hospitals with similar populations. However, our study has some limitations. First, we had information about mortality and morbidities only until the women's discharge from hospital. Therefore, the frequency of some maternal outcomes may have been underestimated, especially for women delivering vaginally, who are usually discharged earlier than women who have cesarean deliveries. Second, diagnoses during current pregnancy (eg, severe anemia) may have been reported differently among the participating hospitals. The resulting misclassifications most likely lead to categorizing some “exposed” women (eg, those with severe anemia) as “nonexposed.” These misclassifications may have underestimated the strength of the corresponding associations. Third, our survey included only hospitals with functional operating rooms and 800 or more deliveries per year. The results therefore cannot be generalized to primary health care centers. For the same reasons, some of our results, especially crude in-hospital maternal mortality rates, should not be regarded as representative of entire countries or regions. Finally, our analyses must account for the dependence of outcomes of individual patients who delivered in the same hospital. To take the resulting clustering effect into account, our score was derived based on the variables found to be statistically significant in the generalized estimating equations extension of multivariable logistic regression.

The predictors identified in this study confirm that patients with uterine rupture, hemorrhage, preeclampsia or eclampsia, or genital infection or sepsis should be managed with high priority by qualified health professionals in comprehensive emergency obstetric care departments, especially if the patient is grand multiparous, referred from another health facility, had severe anemia or malaria diagnosed during pregnancy, or had an insufficient number of antenatal care visits. Given the human resources crisis in Mali and Senegal, the availability of qualified personnel (midwives and doctors) is problematic and many tasks are delegated to less qualified health personnel (students, matrons, nursing assistants). These professionals may play a crucial role in improving maternal outcomes in referral hospitals if they are involved in appropriate tasks and are adequately trained.

In clinical practice, this score could be implemented and used on the frontline by nonqualified health personnel by recording simple items from the score: parity, number of antenatal visits, referral from another health facility, malaria or severe anemia during pregnancy as recorded on the antenatal visit card, convulsions (eclampsia), purulent vaginal discharge, and hyperthermia (sepsis). If the total value of the maternal mortality score is more than 10, even without all the criteria, they should immediately alert qualified professionals. Education of nonqualified health personnel by the doctor or midwife to detect signs of uterine rupture (acute pain with loss of contraction) and preeclampsia (blood pressure greater than 140/90 mmHg, proteinuria greater than 1+) may be easy to fill in the entire score. Our results indicate that they should be trained to detect these complications. Qualified professionals could also finish filling in the form with the score as a backup system. Early detection of patients at risk of maternal mortality by using the mortality score and immediate management by midwives or doctors should improve maternal outcomes.18

Our score to detect patients at risk of in-hospital maternal mortality in Africa uses previously known risk factors for in-hospital maternal mortality. Using a multivariable analysis, each independent item in the score has its own weight to predict in-hospital maternal mortality. Identification of patients at risk for in-hospital maternal mortality by the score should lead to appropriate management through recourse to suitable health resources and professionals. Appropriate management of patients in clinical practice could reduce maternal mortality. This potential reduction of mortality could be measured by its implementation in routine clinical practice in Africa.

Back to Top | Article Outline


1. Salama P, Lawn J, Bryce J, Bustreo F, Fauveau V, Starrs A, et al.. Making the Countdown count. Lancet 2008;371:1219–21.
2. Chukudebelu WO, Ozumba BC. Maternal mortality at the University of Nigeria Teaching Hospital, Enugu: a 10-year survey. Trop J Obstet Gynaecol 1988;1:23–6.
3. Dumont A, Gaye A, de Bernis L, Chaillet N, Landry A, Delage J, et al.. Facility-based maternal death reviews: effects on maternal mortality in a district hospital in Senegal. Bull World Health Organ 2006;84:218–24.
4. Igberase GO, Ebeigbe PN. Maternal mortality in a rural referral hospital in the Niger Delta, Nigeria. J Obstet Gynaecol 2007;27:275–8.
5. Kampikaho A, Irwig LM. Risk factors for maternal mortality in five Kampala hospitals, 1980–1986. Int J Epidemiol 1990;19:1116–8.
6. Lankoande J, Sondo B, Ouedraogo C, Ouedraogo A, Kone B. [Maternal mortality in the National Hospital Center of Ouagadougou (Burkina Faso). Report of 123 cases collected in 1995] [in French]. Rev Epidemiol Sante Publique 1997;45:174–6.
7. Malle D, Ross DA, Campbell OM, Huttly SR. Institutional maternal mortality in Mali. Int J Gynaecol Obstet 1994;46:19–26.
8. Tebeu PM, Ngassa P, Kouam L, Major AL, Fomulu JN. Maternal mortality in Maroua Provincial Hospital, Cameroon (2003–2005). West Indian Med J 2007;56:502–7.
9. Teguete I, Traore Y, Dennis N, Mounkoro N, Traore M, Dolo A. A 19-year retrospective investigation of maternal mortality at Point G National Hospital, Bamako, Mali. Int J Gynaecol Obstet 2010;108:194–8.
10. Thonneau PF, Matsudai T, Alihonou E, De Souza J, Faye O, Moreau JC, et al.. Distribution of causes of maternal mortality during delivery and post-partum: results of an African multicentre hospital-based study. Eur J Obstet Gynecol Reprod Biol 2004;114:150–4.
11. Cross S, Bell JS, Graham WJ. What you count is what you target: the implications of maternal death classification for tracking progress towards reducing maternal mortality in developing countries. Bull World Health Organ 2010;88:147–53.
12. Souza JP, Gulmezoglu A, Lumbiganon P, Laopaiboon M, Carroli G, Fawole B, et al.. Caesarean section without medical indications is associated with an increased risk of adverse short-term maternal outcomes: the 2004–2008 WHO Global Survey on Maternal and Perinatal Health. BMC Med 2010;8:71.
13. Dumont A, Fournier P, Fraser W, Haddad S, Traore M, Diop I, et al.. QUARITE (quality of care, risk management and technology in obstetrics): a cluster-randomized trial of a multifaceted intervention to improve emergency obstetric care in Senegal and Mali. Trials 2009;10:85.
14. Tsuruta H, Bax L. Polychotomization of continuous variables in regression models based on the overall C index. BMC Med Inform Decis Mak 2006;6:41.
15. Zeger SL, Liang KY. Longitudinal data analysis for discrete and continuous outcomes. Biometrics 1986;42:121–30.
16. Khan KS, Wojdyla D, Say L, Gulmezoglu AM, Van Look PF. WHO analysis of causes of maternal death: a systematic review. Lancet 2006;367:1066–74.
17. Gumanga SK, Kolbila DZ, Gandau BB, Munkaila A, Malechi H, Kyei-Aboagye K. Trends in maternal mortality in Tamale Teaching Hospital, Ghana. Ghana Med J 2011;45:105–10.
18. Pacagnella RC, Cecatti JG, Osis MJ, Souza JP. The role of delays in severe maternal morbidity and mortality: expanding the conceptual framework. Reprod Health Matters 2012;20:155–63.
© 2013 by The American College of Obstetricians and Gynecologists. Published by Wolters Kluwer Health, Inc. All rights reserved.