You are here

  • Home
  • Archive
  • Volume 13, Issue 11
  • Maternal near-miss prediction model development among pregnant women in Bahir Dar City administration, northwest Ethiopia: a study protocol

Article Text

Article menu
Health services research
Protocol
Maternal near-miss prediction model development among pregnant women in Bahir Dar City administration, northwest Ethiopia: a study protocol
  1. Yinager Workineh1,
  2. Getu Degu Alene2,
  3. Gedefaw Abeje Fekadu3
  1. 1Department of Nursing, Bahir Dar University, Bahir Dar, Ethiopia
  2. 2Department of Epidemiology and Biostatistics, Bahir Dar University, Bahir Dar, Ethiopia
  3. 3Department of Reproductive Health and Population Studies, Bahir Dar University, Bahir Dar, Ethiopia
  1. Correspondence to Yinager Workineh; workieyenie{at}gmail.com

Abstract

Introduction Maternal near-miss is a condition when a woman nearly died but survived from complications that happened during pregnancy, childbirth or within 42 days after delivery. Maternal near-miss is more prevalent among women in developing nations. Previous studies have identified the impact of different predictor variables on maternal near-miss but shared prognostic predictors are not adequately explored in Ethiopia. It is therefore necessary to build a clinical prediction model for maternal near-misses in Ethiopia. Hence, the aim of this study is to develop and validate a prognostic prediction model, and generate a risk score for maternal near-miss among pregnant women in Bahir Dar City Administration.

Methods and analysis A prospective follow-up study design will be employed among 2110 selected pregnant women in the Bahir Dar City administration from 1 May 2023 to 1 April 2024. At the initial antenatal visit, pregnant women will be systematically selected. Then, they will be followed until 42 days following birth. Data will be collected using structured questionnaires and data extraction sheet. The model will be created using Cox proportional hazard regression analysis. The performance of the model will be assessed based on its capacity for discrimination using c-index and calibration using calibration plot, intercept and slope. The model’s internal validity will be evaluated through the bootstrapping method. Ultimately, the model will be illustrated through a nomogram and decision tree, which will be made available to prospective users.

Ethics and dissemination Ethical approval has been obtained from the Institutional Review Board of the College of Medicine and Health Sciences, Bahir Dar University (protocol number 704/2023). Findings will be published in peer-reviewed journals and local and international seminars, conferences, symposiums and workshops. Manuscripts will be prepared and published in scientifically reputable journals. In addition, policy briefs will be prepared.

  • epidemiology
  • maternal medicine
  • prognosis
  • public health
http://creativecommons.org/licenses/by-nc/4.0/

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.

https://doi-org.ezproxy.u-pec.fr/10.1136/bmjopen-2023-074215

Statistics from Altmetric.com

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

    STRENGTHS AND LIMITATIONS OF THE STUDY

    • The data will be collected prospectively to minimise missing data.

    • The model will be developed based on easily and quickly identifiable individual level factors at the entrance of antenatal contact.

    • Recall bias related to last normal menstrual period may potentially affect the results.

    • Application of WHO maternal near-miss screening criteria may under estimate the detection of maternal near-miss.

    Introduction

    Maternal near-miss refers to a woman who almost lost her life but survived from a severe obstetric complication that occurred during pregnancy, childbirth or within 42 days following delivery.1 2 Severe acute maternal morbidity (SAMM) is a synonymous term for maternal near-miss.3

    The notion of maternal near-miss was devised by the WHO to pinpoint life-threatening situations during pregnancy, childbirth and postpartum.4 This concept allows for interventions focusing on the sequence of events that led to a woman with near to death or actual death.5 6 By using maternal near-miss statistics instead of maternal mortality rates (MMR), maternal healthcare system deficiencies and health priorities can be more swiftly identified.4 7 To apply this concept globally, WHO created diagnostic tools for maternal near-miss that encompass clinical, laboratory and management-based criteria.8 9 This tool was adapted and validated for sub-Saharan Africa (SSA) countries.10

    Maternal mortality and maternal near-miss are prominent health concerns on a global scale, especially in underprivileged countries. Roughly 303 000 women die annually due to complications during pregnancy and childbirth across the world.11 The 2017 United Nations Population Fund reported that every 2 min a woman died due to pregnancy or childbirth-related complications.12 Low-resource countries account for 99% of all maternal mortalities.13 According to the 2019 Mini-Ethiopian Demographic Health Survey (EDHS) report, Ethiopia’s MMR stood at 412 per 100 000 live births.14 This number is considerably higher than the average MMR worldwide (211 per 100 000 live births), but lower than the MMR in SSA (553 per 100 000 live births).15

    Maternal near-miss ranged from 0.80% to 8.23% based on disease-specific criteria, and 0.01%–2.99% based on management-related criteria.16 The maternal near-miss ratio was 18.57 per 1000 live-births globally.17 The smallest maternal near-miss ratio was found in Europe (3.10 per 1000 live birth),17 whereas the highest-burden of maternal near-miss was found in African and Asian countries.18 In SSA, the maternal near-miss ratio was 24.2 per 1000 live births.19 The prevalence of maternal near-miss in Ethiopia was 12.57% with the highest (26.5%) in the Amhara region.20

    The high burden of maternal near-miss is influenced by a multitude of complex risk factors. Delays in seeking care, reaching to care, receiving adequate and appropriate care21 are some of the risk factors for maternal near-miss. The delay in seeking care is linked to failure to recognise signs of complications, failure to perceive severity of illness, cost consideration, negative experience with health system, transportation difficulties and need of permission from family members.21 22 The reasons for the delay in accessing healthcare services include a considerable distance to the medical facility, poor road conditions and a shortage of transportation options.23 The uncompassionate demeanour of healthcare providers, inadequacy of supplies and essential equipments, unavailability or inadequate proficiency of medical staff and absence of urgency or comprehension of emergency situations are the reasons for delay in receiving adequate and appropriate care.21 A community’s delay to take responsibility can also contribute to maternal mortality as a result of the absence of a community-based and community-driven comprehensive approach to maternal health/well-being.24

    Several measures have been implemented to reduce the load, complications, causes and risk factors linked to maternal mortality and maternal near-miss. In 2012, WHO defined, conceptualised and evaluated severe maternal morbidity. The objective of this effort was to compile numerous definitions of maternal morbidity.25 26 In 2013, WHO established the Maternal Mortality Surveillance and Response unit, which concentrates on continuous evaluation of the reasons and factors that lead to maternal mortality.27

    In addition to the aforementioned global initiatives, individual risk assessment studies like clinical prediction models are important in enhancing maternal health. These prediction models can guide (1) clinical researchers to select appropriate study subjects; (2) patients to choose more beneficial steps for themselves; (3) doctors to make better clinical decisions and (4) health management departments to monitor and manage the quality of medical services and to allocate medical resources more rationally.28

    The effects of clinical prediction models can be nearly mirrored in any of the three-level prevention system of diseases including the primary (health promotion, prevention and control), secondary (early screening, early diagnosis and prompt treatment) and tertiary (rehabilitation programmes, preventing disease relapse, reducing mortality and disability and promoting functional recovery and quality of life). Prognostic predictive models can provide patients and doctors with a numerical risk value (probability) of identifying specific illness in the future based on current health condition.29

    The prediction model we are going to develop will be used in daily clinical practice and obstetric patients. It will help pregnant women to identify themselves as a risk group or not. In daily clinical practice, this model will provide clues for obstetricians to select high-risk obstetric patients for further screening, diagnosis and management. In general, the development of numerical and visual maternal near-miss model will assist obstetricians or other healthcare professionals, obstetric patients and their relatives to facilitate shared medical decision-making for diagnostic testing, initiating or discontinuing treatments or making lifestyle changes throughout the perinatal period.

    Different researchers investigated the burden and determinants of maternal near-miss in Ethiopia.20 30–37 These studies focused on the individual predictor effects on maternal near-miss. They did not identify shared characteristics of prognostic predictors as a whole, and did not directly indicate the risk stratification of obstetric patients. Researchers in other areas developed and validated the diagnostic prediction models for maternal death or severe obstetric morbidity among admitted patients.38–41 The diagnostic prediction model was developed for severe maternal outcome in SSA among admitted obstetric patients.42 But there are few prognostic prediction models for maternal near-miss. Hence, the development and validation of a prediction model for maternal near-miss among pregnant women is required in Ethiopia. Therefore, the aims of this study are the development and (internal) validation of a prognostic prediction model, as well as the generation of a risk score for maternal near-miss among pregnant women. The specific objectives are:

    1. To develop a prognostic model for maternal near-miss among pregnant women in Bahir Dar City administration, Northwest Ethiopia 2023–2024.

    2. To validate (internal) the prognostic model of maternal near-miss among pregnant women in Bahir Dar City administration, Northwest Ethiopia, 2023–2024.

    3. To create risk score for maternal near-miss among pregnant women in Bahir Dar City administration, Northwest Ethiopia, 2023–2024.

    Methods and analysis

    The components of this protocol were reported based on the transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) guideline or checklists43 (online supplemental file 1).

    Study design and period

    Prospective follow-up study will be conducted to develop prognostic prediction model of maternal near-miss. The focus of the model is to predict a future occurrence of maternal near-miss using individual level factors at the entrance of antenatal contact (ANC).

    Maternal near-miss in the follow-up period (O)=f (D1, D2, D3, …Dn)

    Where,

    O=occurrence of maternal near-miss

    D1…Dn=the predictors

    Individual-level variables will be used as prognostic predictors (Dn) to forecast the incidence of maternal near-miss (O) during the follow-up period. The occurrence of maternal near-miss among pregnant women as a function of individual level predictors is expressed as=f (age, height, weight…Dn).

    The study will be conducted from 1 May 2023 to 1 April 2024. The requirement of participants and data collection on prognostic predictors will be performed from 1 May 2023 to 5 August 2023 (base line period). The follow-up assessment of the outcome will be at any time after the enrolment of the participants. Hence, identification of the outcome will be carried out from 2 May 2023 to April 2024 (end line period).

    Study setting and participants

    The study will be carried out in Bahir Dar City administration in Northwest Ethiopia. Bahir Dar is 450 km away from Addis Ababa. Both urban and rural populations inhabit this city. There are three public hospitals, 11 healthcare centres, 15 health posts, 4 private hospitals, 56 private specialty clinics and 13 private medium clinics in the city.44

    First, health facilities in Bahir Dar City administration that provide ANC service will be identified. Second, the health facilities to be included in the study will be selected randomly. Third, the total sample size will be proportionally allocated to each health facility based on last year’s reports of first antenatal care visit. Fourth, pregnant women who come for their first ANC visit will be selected using a systematic sampling method. In this regard, in each data collection day, the woman who will arrive first in the ANC clinic will be selected as a starting participant. Then, every other ANC visitor will be selected. Finally, the selected pregnant women will be followed until they either experience the event or reach a state of censorship (figure 1). The event group will consist of women who experience maternal near-miss, while the censoring group will include women who withdraw from ANC visits, transfer out from the selected healthcare facility, are lost to follow-up or pass away during the follow-up period.

    Figure 1
    Figure 1

    Flow chart for study participants selection in Bahir Dar City administration, Northwest Ethiopia, 2023. ANC, antenatal contact.

    Eligibility criteria

    Participants who have no plans to relocate from the study area and are considered well enough to be interviewed by the interviewer will be included in the study. Pregnant women who experienced maternal near-miss at the beginning of the cohort and after 42 days of delivery, and who do not remember their last normal menstrual period will be excluded from the study participants.

    Predictors and their measurements

    Individual-level data on all predictors will be collected by trained midwives using an interview-administered questionnaire and an extraction sheet. These variables include sociodemographic characteristics such as age (in years), residence (coded as ‘0’ for rural and ‘1’ for urban), decision-making for healthcare (coded as ‘1’ for self, ‘2’ for spouse, ‘3’ for relatives and ‘4’ for jointly), height (in centimetres), weight (in kilograms) and mid-upper arm circumference (measured in centimetres). Additionally, obstetric factors such as parity (measured in number), plurality (coded as ‘0’ for single and ‘1’ for multiple), pregnancy intention (coded as ‘0’ for unplanned and ‘1’ for planned), gestational age (in weeks), inter-pregnancy interval (in months) and history of caesarean section (C/S) (coded as ‘0’ for no and ‘1’ for yes) will be recorded. Other factors to be considered include the history of pre-eclampsia (coded as ‘0’ for no and ‘1’ for yes), eclampsia (coded as ‘0’ for no and ‘1’ for yes), sepsis (coded as ‘0’ for no and ‘1’ for yes), haemorrhage (coded as ‘0’ for no and ‘1’ for yes), obstructed labour (coded as '0' for no and ‘1’ for yes), medical morbidity (coded as ‘0’ for no and ‘1’ for yes), history of stillbirth (coded as ‘0’ for no and ‘1’ for yes), history of abortion (coded as ‘0’ for no and ‘1’ for yes), distance from the health facility (measured in kilometres), timing of initial ANC (expressed in weeks) and birth preparedness and complication readiness (coded as ‘0’ for no and ‘1’ for yes). Additionally, baseline clinical indices such as blood pressure (measured in mm Hg) and haematocrit measurement (measured in percentage) will be recorded.

    Outcome

    Maternal near-miss will be diagnosed during the follow-up phase by trained health professionals using the WHO screening criteria.4Women who meet at least one of the clinical, laboratory or management-based criteria will be classified as the event group (maternal near-miss), while the remaining women will be classified as the censoring group. In addition to identifying maternal near-miss cases, pregnancy danger signs, referrals, withdrawals against medical advice, transfers out, deaths and any complications will be extracted from the maternal card during the follow-up period. The survival time will be measured in week (s) from the last date of normal menstrual period to the occurrence of maternal near-miss or censorship.

    Blinding assessment of predictors and outcome

    Blinding reduces the risk of bias that may be introduced in the model development. This will be done by measuring the predictors at the base line and outcome at the follow-up period. The data collectors for the baseline and end line surveys will be different.

    Sample size determination

    The sample was determined based on the minimum sample size calculation criteria for a time to event study. The minimum criteria for this calculation are as follows: (1) a minimum heuristic shrinkage factor, S, greater than 0.9 (targeting less than 10% overfitting), (2) a small difference between Nagelkerke’s R2app and R2adj (targeting less than 0.05 absolute difference) and (3) a small margin of error in the overall risk estimate (targeting less than 0.05 absolute error). These criteria, including the number of parameters (P), heuristic shrinkage factor (S), overall risk in the population and the model’s anticipated Cox-Snell R2 (or C-statistics) were reviewed from previous studies. Approximately 25 candidate parameters, 26.6% of maternal near-miss34 and C-statistics of 0.1142 were utilised to calculate the sample size. The sample size was then calculated using the Stata command ‘pmsampsize, type (b) rsquared (0.11) parameter25 prev (0.266)’. The resulting sample size was 1918 (table 1).

    View this table:
    Table 1

    Sample size calculation for prognostic prediction model development of maternal near-miss using minimum criteria of event-to-time method in Bahir Dar City, Ethiopia 2023–2024

    Finally, taking into account a non-response rate of 10%, the sample size for the development of a predictive model for maternal near-miss will be 2110.

    Data quality assurance

    Input from study participants and subject-matter experts will be used to construct and validate the questionnaire in terms of face and content validity. After tool validation, data collectors and supervisors will receive a 2 days of training to become familiar with the questionnaires, the data collection processes, the ethical considerations and a purpose of the study. Similarly, health professionals who work at antenatal, delivery and postnatal departments will be trained on the screening criteria of maternal near-miss. Then, a pretest will be conducted by four data collectors (Bachelor degree in Midwifery) and one supervisor (Epidemiologist). This pretest will be undertaken to ensure the accuracy of the data and to check for ambiguities in language after switching from English to Amharic. Data collection tools will be changed in light of the pretest’s results. Daily supervision of data collectors and daily verification of all collected data will be done by the supervisor. The data will be regularly checked for completeness, and any problems during data collection will be addressed appropriately.

    Statistical analysis methods

    Data processing

    The data from Epi-Collect5 will be downloaded and transferred to Microsoft Excel. The illogical values and steps will be checked during the design of the questionnaire/abstraction sheet in Epi-Collect5. The collected data will be examined for consistency and completeness. Then the data will be exported to R 4.2.2 software for analysis.

    Data cleaning will be performed to check the completeness of the data, remove or correct noise, outliers and missing values in order to prepare the data for the subsequent steps of model development and validation. The cleaned data will be further processed using feature selection and extraction algorithms, which will involve deriving new attributes and summarising the data. Data transformations, such as normalisation, will be applied to remove noise and correct inconsistencies in the data.

    For continuous variables, dichotomisation or categorization will be performed based on widely accepted clinical cut-off values. If there are no clinical cut-off points, restricted cubic splines or fractional polynomials will be applied to model non-linear relationships.

    The distribution of continuous variables will be examined graphically using a histogram and statistically using the Shapiro-Wilk test. If the data do not follow a normal distribution, data transformation and standardisation will be performed. Multicollinearity between each independent predictor will be checked using the variance inflation factor (VIF). If the VIF is greater than 10, there is no multicollinearity.

    Missing data handling

    Once the data are prepared for analysis, a thorough evaluation will be conducted to identify missing data. Missing data is a common problem that can impact the accuracy of classification and the models generated from data mining algorithms.

    The first step in dealing with missing data is to understand the patterns of missing values. The Hmisc library’s ‘naclus’ and ‘naplot’ functions, as well as the recursive partitioning library of Atkinson and Therneau, will be applied for this purpose. The ‘naclus’ function identifies variables that tend to be missing for the same participants and computes the proportion of missing values for each variable. The ‘rpart’ function builds a tree to predict which types of participants tend to have missing values.

    After understanding the patterns of missing values, statistical methods will be employed to handle the missing values. One common method is complete case analysis, which involves excluding all subjects with missing values for any potential predictor or outcome. Complete case analysis considers subjects with complete data for a specific predictor, even if they have missing values for other covariates not included in the specific model. This method discards information from subjects who have information on some predictors but not all, making it statistically inefficient. Therefore, methods that replace missing values with substituted values based on various criteria are preferred. These methods include: (1) replacing the missing value with a constant, (2) replacing the missing value with the mean of the field and (3) replacing the missing values with randomly generated values from the observed variable distribution.

    In the current model development, missing values will be handled by replacing them with randomly generated values from the observed variable distribution. Specifically, multiple imputation techniques will be used when the amount of missing data is less than 10%. The steps of multiple imputation involve: (1) replacing missing values with randomly selected values from specific distributions to create complete case data sets, (2) conducting the same analysis on each of these data sets and (3) pooling the results together. This imputation process will be repeated five times.

    Model building

    Predictor selection

    The prognostic predictors will be selected by considering existing knowledge of previously established predictors; applicability and costs of predictor measurement relevant to the targeted setting; and statistical power.

    Model estimation and specification

    Cox proportional hazard regression model will be used. The strength of the association will be measured in terms of HR at 95% confidence limits. The Cox proportional hazard model assumption will be checked using scaled Schoenfeld residual test and graphically with log–log Cox adjusted survival estimate. The model fitness will be checked using Cox-nell residuals test. Finally, the multivariable cox proportional hazard model will be declared at a p<0.05 with 95% CI as statistical significance.

    Model performance

    Calibration and discrimination are two crucial factors in assessing the prognostic accuracy of the model. At a group level, calibration assesses how accurately the absolute predicted risks align with the real hazards.45 Discrimination refers to the model’s capacity to differentiate between patients who experience the relevant event and those who do not.

    By utilising the concordance index, the model’s capacity for differentiation can be assessed (C-index). To compute the C-index, all possible pairs of patients with and without the outcome are examined. If the patient with the result has a higher expected risk than the patient without the result, the pair is considered to be in agreement. The C-index is equivalent to the area under the curve (AUC).46 The AUC values ranging from 0.9 to 1.0, 0.8 to 0.9, 0.7 to 0.8, 0.6 to 0.7 and 0.5 to 0.6 are classified as excellent (A), good (B), fair (C), poor (D) and fail (F), respectively, in order to evaluate the model’s ability to differentiate.

    A time-to-event C-statistic, such as Harrel’s C-statistic or Uno’s C-statistic, can be calculated for Cox proportional hazards prediction models.47 Two patients who both have the event can be matched together in these calculations, and if the patient who has the event first has the higher predicted risk, they are considered to be a concordant pair. Because this project is focused on survival, the time-to-event C-statistic method will be used to assess the model’s ability to discriminate.

    Calibration assesses if the predicted hazards are comparable to the actual hazards.47 Calibration can be evaluated through a calibration plot, intercept and slope. By computing a Cox proportional hazards model utilising the prognostic indicator, the calibration slope is determined. The calibration slope represents the coefficient of regression applied to the prognostic indicator.

    Internal validation

    Different techniques exist for internal verification of prognostic forecasting models, including cross-validation, bootstrapping and split-sample validation.48 The bootstrapping approach surpasses the split-sample and cross-validation methods in addressing optimism. Therefore, this study will utilise 10 000 random bootstrap samples with replacement on all predictors in the data set to validate the model. The C-index with a 95% CI, as well as sensitivity and specificity, will be employed in conjunction with the Hosmer-Lemeshow statistic.

    Model accuracy

    Using a Brier score, the accuracy of a set of probabilistic forecasts will be evaluated. The Brier score is the average of the squared disparities between the probabilistic predictions and the actual event outcomes, with the probabilistic forecasts being provided for those specific events.

    Clinical and public health impact assessment

    Decision curve analysis (DCA) is a commonly employed method to evaluate the efficacy of clinical prediction models. Traditional measures of diagnostic performance, such as sensitivity, specificity and area under the receiver operating characteristic curve do not consider the clinical value of a specific model. Instead, these measures only compare the diagnostic accuracy of different prediction models.49 Therefore, in this study, DCA will be utilised to assess the clinical and public health impacts of the maternal near-miss model.

    Model presentation

    Once the model is developed and validated, a risk score will be generated for maternal near-misses. This score will be based on predictors at the individual level and will be user-friendly. The coefficients of each predictor that are statistically significant in the multivariable cox proportional hazard regression model will be adjusted to calculate the risk scores. The cut-off point for the risk score and the development of a nomogram will be determined using the Youden index value (sensitivity+specificity−1) for each risk category. Additionally, a decision tree will be presented to potential users as part of the clinical prediction model.

    Patient and public involvement

    No patient or public has been involved while developing this study protocol.

    Ethics and dissemination

    The Institutional Review Board of the College of Medicine and Health Sciences, Bahir Dar University has granted ethical clearance (protocol number 704/2023) for this study. Results will be shared via scientific publications, conference presentations, community meetings, and policy briefs.

    Ethics statements

    Patient consent for publication

    Acknowledgments

    We thank Bahir Dar University for the facilities we have used while preparing this protocol.

    References

    1. ↵
      1. Filippi V,
      2. Chou D,
      3. Barreix M, et al
      . A new conceptual framework for maternal morbidity. Int J Gynecol Obstet 2018;141:4–9. doi:10.1002/ijgo.12463 Available: http://doi.wiley.com.ezproxy.u-pec.fr/10.1002/ijgo.2018.141.issue-S1
      OpenUrl
    2. ↵
      1. Bacci A,
      2. Lewis G,
      3. Baltag V, et al
      . The introduction of confidential enquiries into maternal deaths and near-Miss case reviews in the WHO European region. Reprod Health Matters 2007;15:145–52. doi:10.1016/S0968-8080(07)30334-0
      OpenUrlCrossRefPubMedWeb of Science
    3. ↵
      1. Souza JP,
      2. Cecatti JG,
      3. Faundes A, et al
      . Maternal near Miss and maternal death in the world health organization’s 2005 global survey on maternal and perinatal health. Bull World Health Organ 2010;88:113–9. doi:10.2471/BLT.08.057828
      OpenUrlCrossRefPubMedWeb of Science
    4. ↵
      1. World Health Organization
      . Evaluating the quality of care for severe pregnancy complications: the WHO near-Miss approach for maternal health. 2011.
    5. ↵
      1. Pattinson RC,
      2. Hall M
      . Near misses: a useful adjunct to maternal death enquiries. Br Med Bull 2003;67:231–43. doi:10.1093/bmb/ldg007
      OpenUrlCrossRefPubMedWeb of Science
    6. ↵
      1. Geller SE,
      2. Rosenberg D,
      3. Cox SM, et al
      . The continuum of maternal morbidity and mortality: factors associated with severity. Am J Obstet Gynecol 2004;191:939–44. doi:10.1016/j.ajog.2004.05.099
      OpenUrlCrossRefPubMedWeb of Science
    7. ↵
      1. Parmar NT,
      2. Parmar AG,
      3. Mazumdar VS
      . “Incidence of maternal "near-Miss" events in a tertiary care hospital of central Gujarat, India”. J Obstet Gynaecol India 2016;66:315–20. doi:10.1007/s13224-016-0901-3
      OpenUrlPubMed
    8. ↵
      1. Say L,
      2. Souza JP,
      3. Pattinson RC, et al
      . Maternal near Miss–towards a standard tool for monitoring quality of maternal health care. Best Pract Res Clin Obstet Gynaecol 2009;23:287–96. doi:10.1016/j.bpobgyn.2009.01.007
      OpenUrlCrossRefPubMed
    9. ↵
      1. Lam A,
      2. Ford S
      . Best practice & research clinical obstetrics and Gynaecology. Best Practice & Research Clinical Obstetrics and Gynaecology 2009;23:631–46.
      OpenUrl
    10. ↵
      1. Tura AK,
      2. Stekelenburg J,
      3. Scherjon SA, et al
      . Adaptation of the WHO maternal near Miss tool for use in sub–Saharan Africa: an international Delphi study. BMC Pregnancy Childbirth 2017;17:445. doi:10.1186/s12884-017-1640-x
    11. ↵
      1. Unicef
      . Trends in maternal mortality: 1990 to 2013. Geneva: World Health Organization, 2014.
    12. ↵
      1. United Nations Population Fund
      . Maternal health. 2019.
    13. ↵
      1. WHO
      . Maternal mortality fact sheet. World Health Organization, Division of Family Health Geneva, 2006.
    14. ↵
      Ethiopia mini demographic and health survey. 2019.
    15. ↵
      1. Srivastava N
      . n.d. Conditions of maternal mortality rate in India: unmet needs to take immediate action for increasing ratio’s.
    16. ↵
      1. Say L,
      2. Pattinson RC,
      3. Gülmezoglu AM
      . WHO systematic review of maternal morbidity and mortality: the prevalence of severe acute maternal morbidity (near Miss). Reprod Health 2004;1:1–5.:3. doi:10.1186/1742-4755-1-3
      OpenUrlCrossRefPubMed
    17. ↵
      1. Abdollahpour S,
      2. Heidarian Miri H,
      3. Khadivzadeh T
      . The global prevalence of maternal near Miss: a systematic review and meta-analysis. Health Promot Perspect 2019;9:255–62. doi:10.15171/hpp.2019.35
      OpenUrlPubMed
    18. ↵
      1. Tunçalp O,
      2. Hindin M,
      3. Souza J, et al
      . The prevalence of maternal near Miss: A systematic review. Obstetric Anesthesia Digest 2013;33:84. doi:10.1097/01.aoa.0000429114.02435.42
      OpenUrl
    19. ↵
      1. Tura AK,
      2. Trang TL,
      3. van den Akker T, et al
      . Applicability of the WHO maternal near Miss tool in sub-Saharan Africa: a systematic review. BMC Pregnancy Childbirth 2019;19:79. doi:10.1186/s12884-019-2225-7
    20. ↵
      1. Mengist B,
      2. Desta M,
      3. Tura AK, et al
      . Maternal near Miss in Ethiopia: protective role of Antenatal care and disparity in socioeconomic inequities: A systematic review and meta-analysis. Int J Africa Nurs Sci 2021;15:100332. doi:10.1016/j.ijans.2021.100332
      OpenUrl
    21. ↵
      1. Thaddeus S,
      2. Maine D
      . Too far to walk: maternal mortality in context. Social Science & Medicine 1994;38:1091–110. doi:10.1016/0277-9536(94)90226-7
      OpenUrl
    22. ↵
      1. Douthard RA,
      2. Martin IK,
      3. Chapple-McGruder T, et al
      . US maternal mortality within a global context: historical trends, current state, and future directions. J Womens Health (Larchmt) 2021;30:168–77. doi:10.1089/jwh.2020.8863
      OpenUrl
    23. ↵
      1. Barnes-Josiah D,
      2. Myntti C,
      3. Augustin A
      . “The “three delays” as a framework for examining maternal mortality in Haiti”. Social Science & Medicine 1998;46:981–93. doi:10.1016/S0277-9536(97)10018-1
      OpenUrl
    24. ↵
      1. MacDonald T,
      2. Jackson S,
      3. Charles M-C, et al
      . The fourth delay and community-driven solutions to reduce maternal mortality in rural Haiti: a community-based action research study. BMC Pregnancy Childbirth 2018;18:254. doi:10.1186/s12884-018-1881-3
    25. ↵
      1. Firoz T,
      2. Chou D,
      3. von Dadelszen P, et al
      . Measuring maternal health: focus on maternal morbidity. Bull World Health Organ 2013;91:794–6. doi:10.2471/BLT.13.117564
      OpenUrlCrossRefPubMed
    26. ↵
      1. Chou D,
      2. Tunçalp Ö,
      3. Firoz T, et al
      . Constructing maternal morbidity - towards a standard tool to measure and monitor maternal health beyond mortality. BMC Pregnancy Childbirth 2016;16:45. doi:10.1186/s12884-015-0789-4
    27. ↵
      1. World Health Organization
      . Maternal death surveillance and response: technical guidance information for action to prevent maternal death. 2013.
    28. ↵
      1. Chen L
      . Overview of clinical prediction models. Ann Transl Med 2020;8:71. doi:10.21037/atm.2019.11.121
    29. ↵
      1. Zhou Z-R,
      2. Wang W-W,
      3. Li Y, et al
      . In-depth mining of clinical data: the construction of clinical prediction model with R. Ann Transl Med 2019;7:796. doi:10.21037/atm.2019.08.63
    30. ↵
      1. Geze Tenaw S,
      2. Girma Fage S,
      3. Assefa N, et al
      . Determinants of maternal near-Miss in private hospitals in Eastern Ethiopia: A nested case–control study. Womens Health (Lond) 2021;17:17455065211061949. doi:10.1177/17455065211061949
    31. ↵
      1. Liyew EF,
      2. Yalew AW,
      3. Afework MF, et al
      . Distant and proximate factors associated with maternal near-Miss: a nested case-control study in selected public hospitals of Addis Ababa, Ethiopia. BMC Womens Health 2018;18:.:28. doi:10.1186/s12905-018-0519-y
    32. ↵
      1. Tenaw SG,
      2. Assefa N,
      3. Mulatu T, et al
      . Maternal near Miss among women admitted in major private hospitals in Eastern Ethiopia: a retrospective study. BMC Pregnancy Childbirth 2021;21:181. doi:10.1186/s12884-021-03677-w
    33. ↵
      1. Woldeyes WS,
      2. Asefa D,
      3. Muleta G
      . Incidence and determinants of severe maternal outcome in Jimma University teaching hospital, South-West Ethiopia: a prospective cross-sectional study. BMC Pregnancy Childbirth 2018;18:255. doi:10.1186/s12884-018-1879-x
    34. ↵
      1. Worke MD,
      2. Enyew HD,
      3. Dagnew MM
      . Magnitude of maternal near misses and the role of delays in Ethiopia: a hospital based cross-sectional study. BMC Res Notes 2019;12:585. doi:10.1186/s13104-019-4628-y
    35. ↵
      1. Abdulrazaq B,
      2. Getahun M,
      3. Mohammed A, et al
      . Determinant factors of maternal near Miss in selected health facilities of Berak Woreda, Oromia national regional state. Int J Sci Rep 2020;6:131. doi:10.18203/issn.2454-2156.IntJSciRep20201268
      OpenUrl
    36. ↵
      1. Gedefaw M,
      2. Gebrehana H,
      3. Gizachew A, et al
      . Assessment of maternal near Miss at Debre Markos referral hospital, Northwest Ethiopia: five years experience. OJEpi 2014;04:199–207. doi:10.4236/ojepi.2014.44026
      OpenUrl
    37. ↵
      1. Yemaneh Y,
      2. Tiruneh F
      . Proportion and associated factors of maternal near misses in selected public health institutions of Keffa, bench-Maji and Sheka zones of South nations nationalities and peoples regional state, South West Ethiopia, 2017.A Crossectional study. MEDICINE & PHARMACOLOGY [Preprint] 2017. doi:10.20944/preprints201804.0368.v1
    38. ↵
      1. Alam N,
      2. Hobbelink EL,
      3. van Tienhoven AJ, et al
      . The impact of the use of the early warning score (EWS) on patient outcomes: a systematic review. Resuscitation 2014;85:587–94. doi:10.1016/j.resuscitation.2014.01.013
      OpenUrlCrossRefPubMedWeb of Science
    39. ↵
      1. Morgan R,
      2. Williams F,
      3. Wright M
      . An early warning scoring system for detecting developing critical illness. Clin Intensive Care 1997;8:100.
      OpenUrl
    40. ↵
      1. Lappen JR,
      2. Keene M,
      3. Lore M, et al
      . Existing models fail to predict sepsis in an obstetric population with Intrauterine infection. Am J Obstet Gynecol 2010;203:573. doi:10.1016/j.ajog.2010.07.040
      OpenUrlPubMed
    41. ↵
      1. United Nations
      . Goal 3: sustainable development knowledge platform. 2018.
    42. ↵
      1. Umar A,
      2. Manu A,
      3. Mathai M, et al
      . Development and validation of an obstetric early warning system model for use in low resource settings. BMC Pregnancy Childbirth 2020;20:531. doi:10.1186/s12884-020-03215-0
    43. ↵
      1. Collins GS,
      2. Reitsma JB,
      3. Altman DG, et al
      . Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) the TRIPOD statement. Circulation 2015;131:211–9. doi:10.1161/CIRCULATIONAHA.114.014508
      OpenUrlAbstract/FREE Full Text
    44. ↵
      1. Central Statistical Agency
      . Population projection of Ethiopia for all regions at Wereda level from 2014–2017. J Ethnobiol Ethnomed 2013.
    45. ↵
      1. Nansubuga E,
      2. Ayiga N,
      3. Moyer CA
      . Prevalence of maternal near Miss and community-based risk factors in central Uganda. Int J Gynaecol Obstet 2016;135:214–20. doi:10.1016/j.ijgo.2016.05.009
      OpenUrl
    46. ↵
      1. Saucedo M,
      2. Esteves-Pereira AP,
      3. Pencolé L, et al
      . Understanding maternal mortality in women with obesity and the role of care they receive: a national case-control study. Int J Obes 2021;45:258–65. doi:10.1038/s41366-020-00691-4
      OpenUrl
    47. ↵
      1. Van Calster B,
      2. McLernon DJ,
      3. van Smeden M, et al
      . Calibration: the Achilles heel of predictive Analytics. BMC Med 2019;17:230.:230. doi:10.1186/s12916-019-1466-7
      OpenUrlCrossRefPubMed
    48. ↵
      1. Sun G-W,
      2. Shook TL,
      3. Kay GL
      . Inappropriate use of Bivariable analysis to screen risk factors for use in multivariable analysis. J Clin Epidemiol 1996;49:907–16. doi:10.1016/0895-4356(96)00025-x
      OpenUrlCrossRefPubMedWeb of Science
    49. ↵
      1. Fitzgerald M,
      2. Saville BR,
      3. Lewis RJ
      . Decision curve analysis. JAMA 2015;313:409–10. doi:10.1001/jama.2015.37
      OpenUrlCrossRefPubMed

    Supplementary materials

    • Supplementary Data

      This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

    Footnotes

    • Contributors YW, GDA and GAF contributed equally to the design of the study. YW drafted the manuscript and all the authors revised and approved the manuscript.

    • Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.

    • Competing interests None declared.

    • Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting or dissemination plans of this research.

    • Provenance and peer review Not commissioned; externally peer reviewed.

    • Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.