The WHO criteria for pregnancy-related life-threatening conditions were found to be highly associated with maternal deaths. Survivors of the WHO pregnancy-related life-threatening conditions can be accurately classified as maternal near-miss cases. A severity score representing the total number of life-threatening conditions present in each case and a mathematical model describing the relationship between severity markers and maternal deaths have been developed.
Scoring systems have been used to evaluate severity and outcome of critically ill patients since many years. APACHE, SAPS, and SOFA systems are among the most used ones 
. Some of these systems are extensively used for intensive care benchmarking and quality of care assessment 
. Notwithstanding, these systems have been developed based on general critical care populations from developed countries. In these reference populations, obstetric patients were largely omitted or underrepresented, either because pregnant women were excluded, or because maternal deaths are rare events in the countries where these systems were developed. Another issue that should be acknowledged is that the physiological changes of pregnancy affect some of the markers used by general severity scoring systems leading to overestimation of severity 
. Also, diseases that are exclusive to this period of life (e.g. eclampsia, HELLP syndrome, acute fatty liver of pregnancy, amniotic fluid embolism), have peculiar characteristics that may not have been adequately addressed by scores designed for general populations. Owing to this, the performance of these systems in obstetric population is challenged, particularly in developing countries. Limitations are observed in the discriminatory power and the calibration of generic scoring systems when applied to obstetric population 
The maternal severity score and the MSI model, developed in this study, may contribute for a better assessment of severity of obstetric populations and enable a benchmark approach to quality of care of women experiencing severe complications related with pregnancy. As part of the strengths of this analysis, the MSI model was developed in a large multicenter study, which had an appropriate sample both in terms of number of critically ill obstetric patients and maternal deaths. The study population was large enough to allow adequate model development and testing in different subsets of the study population as methodologically recommended 
. In the end, the MSI model was found to be robust, presenting good performance and discriminatory power. There are also two additional potential advantages that should be noted: the MSI model was developed based on the WHO criteria for pregnancy-related life-threatening conditions and the study database reflects the standard of care provided in obstetric referral hospitals from a developing country.
The WHO criteria for pregnancy-related life-threatening conditions are part of a strategy promoted by WHO for assessing and improving quality of maternal health care 
. These criteria are used in the identification of maternal near-miss cases in clinical audits and other near-miss studies. Together with routine implementation, there are several near-miss research projects being currently conducted around the world using these criteria. In addition to external validation, these initiatives may favor further dissemination of the maternal near miss concept and enable the use of the MSI model as a benchmark tool.
As in other severity models, the MSI model reflects the characteristics and standards of care received by the population that provided data for its development. Brazil is the largest country in Latin America and the world’s fifth largest country (both by geographical area and population). It is an upper-middle income country which overall MMR is estimated as 56 maternal deaths per 100,000 live births in 2010 by WHO 
. This study includes a substantial proportion of all maternal deaths that are estimated to have occurred in the country during the data-collection period (8%, range of uncertainty 5–13%). The fact that the MSI model has been developed in an obstetric population is relevant for the applicability of this tool in other populations. This is particularly important in the context of other scoring systems (e.g. the APACHE family) that were developed in generic populations of developed countries. We believe that the application of the MSI to obstetric populations of other developing countries is more direct as compared to extrapolating results from non-obstetric populations from developed countries. However, it should be noted that the standard of care provided by the participating facilities is being used as the reference for the MSI estimates. One would expect that there may be still some limitations and constraints in the quality of care provided by the participating facilities. Thus, the underlying aim of using the MSI as a benchmark is to assess the health service performance against a standard and, through interventions to improve quality of care, achieve a superior performance.
This study has some limitations that are worth noting. First, the non-random nature of the facility sampling process may have introduced some level of selection bias, potentially impairing the country representativeness of this study. On the other hand, the convenience sampling approach was realistic and made this study feasible. Precautions have been taken to maximize country representativeness. An analysis based on the intra-cluster correlation coefficients provided some evidence supporting the success of these precaution measures 
. Second, this study is largely based on information obtained from medical records. In order to reduce the chances of recording bias, information from medical records was complemented with information obtained directly from the assisting staff (if relevant information was missing and in case of doubt). In addition, several procedures to optimize quality of data have been put in place. Third, the study population is essentially provided by referral hospitals which tend to concentrate the more severe cases: the MMR observed in this study is about three times the overall MMR estimated for the country. Another aspect that deserves noting is the relationship between the various covariates within the MSI model. The maternal severity score (i.e. the total number of severity markers present in each case) is positively correlated with maternal mortality and as the number of life-threatening conditions increase, the death probability increases. If a life-threatening condition is identified at hospital arrival or within the first 24 hours of hospital stay, there is an increase in the risk of death, possibly denoting the fact that the woman has arrived in the hospital already in a very severe condition. Cancer and a cardiovascular or respiratory failure substantially increase the death risk. Two covariates (i.e. severe pre-eclampsia and hysterectomy) have negative coefficients denoting a “protective” association within the model. At the first glance this may seem counterintuitive, but these negative coefficients have to be considered in the context of severe maternal morbidity. Our interpretation to the negative coefficient associated to pre-eclampsia is that women presenting a severe health condition due to pre-eclampsia have the potential of a better outcome as compared to women in the same level of severity having other complications. This can be due to the fact that severe pre-eclampsia tend to be a transient complication and effective strategies to manage women with pre-eclampsia exist (e.g. fetal delivery, magnesium sulfate and anti-hypertensive drugs) and may have been used in the population that provided data to this model, resulting in reduced death risks. Similarly, hysterectomy plays a game-changing role in the outcome of women with uterine-related haemorrhage and infection.
There are several potential applications to both Maternal Severity Score and MSI. A primary application is determining the level of complexity and severity of a certain obstetric population. For example, a district hospital treating a population with an average maternal severity score
0.5 is expected to require much more material and human resources than another district hospital treating a population with a maternal severity score
0.1. These two hypothetical district hospitals receive two different case-mix and the maternal severity score can be used to put the health service in context and support decision making for resource allocation. Another primary application is the health impact evaluation, as part of quality of care assessment. The average MSI can provide an estimation of the expected number of maternal deaths for a selected population. For example, a hypothetical obstetric population being treated in an intensive care unit or in a high dependency facility in a tertiary hospital has an average MSI of 10%. It means that in a group of 100 women treated in this facility it would be expected the occurrence of 10 maternal deaths. If 20 maternal deaths have taken place in this population, one could conclude that there may be some opportunities being missed in this facility and a strategy to improve care is needed. The MSI allows also inter-hospital and over-time comparisons. Another possible application is in research. In a randomized controlled trial, for instance, it is worthwhile determining if both trial arms are comparable in terms of severity. Severity can functions as a major confounder: trial results can differ because of unbalances in the severity of the study populations (e.g. populations containing more severe cases tend to present worse health outcomes in comparison to populations with less severe cases). The maternal severity score and the MSI can be used for adjusting for the case-mix, through stratification or as a covariate in statistical modeling.
In summary, the maternal severity score provide an estimation of the overall severity associated with a specific women or a selected population. Similarly, the MSI provides an estimate of the death risk. To maximize usability, the Appendix S1
contains a maternal severity score and maternal severity index calculator. As a final remark, the estimates derived from the MSI model are to be used with caution. A MSI with 95% of death risk means that among 100 women with similar conditions, 95 women may die. However, the model is not able to differentiate if the specific woman is among the 5 that will survive. Thus, MSI estimates should not directly guide the management of critically ill patients.
The identification of maternal near-miss cases using the WHO list of pregnancy-related life-threatening conditions is valid, as these conditions are accurately associated with maternal deaths. The MSI model adequately describes the relationship between severity markers and maternal deaths. The MSI model can be used as a tool for benchmarking, population severity assessment and case-mix adjustment. The use of the MSI model within a maternal near-miss approach has the potential of contributing to the assessment and improvement of maternal health care, particularly that required by women experiencing severe maternal morbidity. Further studies assessing the performance of the MSI model in other populations are welcome.