Setting

Zanzibar is an East African archipelago, semi-autonomous to Tanzania and Mnazi Mmoja Hospital (MMH) is its referral hospital. The Obstetrics & Gynaecology (O&G) department of MMH accommodates 11,000 to 13,000 deliveries per year and on average one maternal death occurs per week. Health care services such as essential obstetrical interventions and an intensive care unit (ICU) are generally available and accessible, although not consistently of good quality. The in-hospital maternal mortality ratio (IH-MMR) is above 400 per 100,000 live births [11].

Study design

Data collection for this study was performed under two research projects (reference numbers ZAMREC/0001/AUGUST/005 and ZAMREC/001/JAN/17) approved by Zanzibar’s Medical Ethical Research Committee. Informed consent was waived because the study concerned only an analysis of clinical files with aggregated, anonymous outcomes.

All cases of maternal near miss and maternal mortality were prospectively identified amongst all women, pregnant or within 42 days after end of pregnancy, admitted to MMH between 1 April 2017 and 31 December 2018. Department-level data on the number of deliveries and live births were extracted from the hospital’s records. MNM cases were identified during the O&G department’s daily clinical staff meetings in which all severe cases were discussed. Simultaneously with this study, a maternal near-miss audit study was performed, contributing to improved case identification and awareness of maternal morbidity within the department. When a patient fulfilled a near-miss criterion, she was actively followed-up by one of the research assistants and the clinical staff. Throughout the entire study period, maternal death cases were identified soon after their occurrence. Maternal death reviews were conducted within 72 hours of the event by the local maternal death review committee consisting of medical doctors of the O&G department, nurse-midwives, an anaesthetist, health workers from the laboratory and blood bank and doctors and nurses from the ICU. Medical files of maternal near-misses and death, when retrieved, were saved, separate from the other files. Relevant information from the patient file, including socio-demographic characteristic, basic obstetric history, a summary of admission, and perinatal and maternal outcomes, was coded, anonymized, stored in a Microsoft Excel spreadsheet and backed-up, accessible to the study team members with a password.

Classification of MNM and mortality cases followed WHO’s near-miss approach. A patient in life-threatening condition was identified following the approach’s set of clinical, laboratory or management-based markers for organ dysfunction, adapted to local availability, as indicated in Table 1. The maternal outcome for these patients was either near-miss or death. The present markers were added to the individual case data in the beforementioned spreadsheet.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Markers for organ dysfunction, identifying cases in life-threatening condition.

https://doi.org/10.1371/journal.pone.0217135.t001

Guided by previous research experience in MMH’s O&G department, we anticipated issues of missing data and poor quality of data. Measures were taken to minimize the influence hereof, such as close collaboration of the researchers and research assistants with the department’s nurses, daily physical presence of the data collectors and weekly updates of data storage. Next to this, ongoing initiatives in the O&G department, in addition to the daily clinical staff meetings, also aimed at improvement of qualitative documentation in patient files. Amongst these projects were the full-time dedication of a registrar to department-level data and the beforementioned maternal near-miss audits. Nevertheless, poor documentation remained an issue.

Data collection was performed by junior investigators (TH, LA, RSK) and seven research assistants throughout the study period, none of whom were involved in patient care. Each research assistant worked on the study for a period of at least 12 weeks and was trained by their predecessor. Training and data collection were supervised by TH and TM. Data quality checks were performed weekly by TH in the spreadsheet and through discussion with the research assistants face to face or, in case of absence from the study site, over telephone. BJ performed additional checks on consistency and quality of the data.

Data analysis

Incidences and ratios of maternal outcomes were assessed with descriptive statistics. These outcome measures are defined in S1 Table. The distribution of dysfunction of the different organ systems was assessed with descriptive statistics in the group of women with severe maternal outcomes (SMO), including both MNM and MD, and separately for the MNM and MD groups. Relative risk ratios (RR) with a 95% confidence interval were calculated to compare the incidence of a specific type of organ dysfunction in MNM and MD cases.

To analyse the performance of the set of WHO near-miss criteria in this setting, its sensitivity, specificity, positive likelihood ratio and negative likelihood ratio were calculated, given an estimation of its accuracy. Performance in this is defined as the set of criteria’s ability to detect those patients with life-threatening, organ function-violating morbidity. The calculations were performed in the group of only the patients with severe complications.

To analyse the validity of the WHO near-miss approach in the study setting, the number of organ dysfunction markers has been plotted against the mortality risk and their relation has been evaluated using Pearson’s correlation coefficient.

Main findings

This study shows a strong correlation between the number of organ dysfunction markers and the risk of mortality, validating its use in low-resource, high-mortality settings. The WHO near-miss approach proves to be applicable in this setting with only 5.0% (4 out of 79) of maternal death cases without documentation of organ dysfunction markers.

Strengths and limitations

To our knowledge, this is the first study to analyse the correlation between markers of organ dysfunction and mortality risk since the seminal study in Brazil. As such, it is the first to validate WHO’s near-miss approach in a high mortality setting. Strengths of our study are its prospective character and the large cohort size enabling the results to even out random variations in the incidence of rare outcomes. The main limitations are the relatively poor quality of documentation in clinical files and the unavailability of a number of WHO near-miss criteria. This has surely caused underestimation of markers of organ dysfunction and might partly explain the difference in the slope of the correlation between number of organ dysfunctions and mortality risk between Zanzibar and Brazil.

Interpretation

This study demonstrates the validity of WHO’s near-miss approach in a high maternal mortality context. First, the criteria proved to have a high accuracy in identifying patients with life-threatening disorders at serious risk of dying. Second, the number of signs of organ dysfunction is strongly positively correlated to the mortality risk, comparable to that observed in the seminal study in Brazil [6]. Third, the number of women who died without documented signs of organ dysfunction is low, even in this setting where poor documentation is a major issue. Those cases were all four of women who arrived to MMH in near-death condition and there was no to little diagnostics and treatment, nor documentation, performed.

As expected in a high maternal mortality, low-resource setting, the mortality risk increases much more quickly with the number of organ dysfunction markers, compared to a moderate or low maternal mortality setting. A patient with one sign of organ dysfunction has a risk of mortality of 5%, one with two signs a risk of 10%. A comparable level of mortality risk is only reached in patients with three markers of organ dysfunction in Brazil. At least part of this distinction reflects the difference in quality of care between the Brazilian centres and MMH. This suggests that women in high-mortality settings do not die (solely) because their condition is more severe than elsewhere but because of sub-optimal quality of care.

In assessing cases of patients in life-threatening condition, underestimation likely has played a role as organ dysfunction markers are not measured comprehensively due to limited diagnostic and therapeutic possibilities in MMH. This is mostly reflected in the incidence of laboratory-based criteria, which is lower in high-mortality, low-resource settings, including this study’s, compared to moderate mortality settings, as shown in a large multi-country study [7]. Incomplete patient file documentation by health workers as well as a poor file-keeping infrastructure in MMH also contribute.

As expected, there is a difference in types of organ dysfunction between patients surviving and those dying in this cohort. The latter are more likely to have suffered from cardiovascular and/or respiratory collapse while the former are more likely to have experienced haematogical/coagulation dysfunction and uterine dysfunction. These findings are comparable to those found by others [1,7,8]. This distinction is insightful, showing that, despite identifying patients in life-threatening condition through the same set of criteria, the ones that survive are often on different pathways than the ones that die, and vice versa. This, firstly, encourages more research into the various characteristics within the group of severely ill patients, including sociodemographic, pregnancy-related, patient’s behaviour-related and received care-related characteristics. Secondly, it can inform policy-making for resource allocation, promoting the necessary diagnostic and management skills and tools, e.g. organ function laboratory test, CPR training and blood product availability.

Including severe morbidity next to mortality improves the assessment of a setting’s quality of maternal health care, going beyond the long-used indicator MMR. Doing this with a selected approach, that allows for local adjustment, enables comparability between settings and over time, which is relevant to making, implementing and reviewing policies for quality improvement. The mortality index allows to correct for the severity of the population studied because, irrespective of the background population case-mix, it puts the maternal death rate in perspective to the only group at significant risk of dying. As such, it provides a more accurate measure of the impact of clinical care than when using the maternal mortality rate alone, provided that near-miss cases are selected following well-defined, widely applicable, though locally-adjusted criteria.

The robustness of any conclusion on the quality of care is dependent on the quality of the data collected and challenges in this should be acknowledged and addressed in maternal morbidity registration. We recommend adhering to WHO’s near-miss criteria, adjusted to the specific setting as needed, because it enables meaningful comparison between similar reference populations. We show in our study that it is applicable and valid also in low-resource, high-mortality settings.