![The World Congress on Controversies in Obstetrics, Gynecology and Infertility (COGI). Click here to learn more.]({"type":"image" , "src" : "/sda/112382/JournalAd-COGI2024-Leaderboard.jpg"})
https://orcid.org/0000-0001-6873-2580
Abstract
This study analyses the contribution of system-related factors to maternal mortality in the low-resource setting of Mnazi Mmoja Hospital in Zanzibar, Tanzania. It is a retrospective cohort study including all maternal deaths (MD, n = 139) and maternal near-misses (MNM, n = 122) in Mnazi Mmoja Hospital with sufficient documentation during 2015 to 2018 (MD) and 2017 to 2018 (MNM). The number of admissions and surgical interventions per health care provider on the day of admission and the number of times vital signs were monitored per day were compared between MNM and MD cases using logistic regression. The mean number of times vital signs were monitored per day was associated with reduced odds of mortality (aOR 0.75, 95% CI 0.64–0.89), after adjustment for confounding factors such as severity of illness. The numbers of admissions or surgical procedures per health care provider were not associated with mortality. Concluding, the degree of monitoring of patients with life-threatening complications of pregnancy or childbirth is associated with the risk of mortality independent of the degree of severity. Preventing maternal mortality requires going beyond availability of essential interventions to tackle system-related factors that have a direct impact on the capacity to provide comprehensive care.
What is already known on this subject? Root cause analyses of maternal deaths have identified many system-related factors, such as availability of health care providers, adequate training, and motivation to sustain high intensity monitoring (Madzimbamuto et al. Citation2014; Mahmood et al. Citation2018).
What do the results of this study add? This is the first study to attempt to quantify the contribution of these system-related factors by comparing cases of maternal death with cases of maternal near-miss. We show that the degree of monitoring of patients with life-threatening complications is associated with the odds of mortality independent of the degree of severity. Even though this relation should not be regarded as causative, monitoring of vital signs can be seen as reflective of many system-related factors which hamper or facilitate comprehensive care.
What are the implications of these findings for clinical practice and/or further research? This study helps increase general understanding of the factors leading to progression from severe disease to death in a high-volume low-income setting.
Impact Statement
Introduction
Maternal mortality and morbidity rates remain high in low-income settings such as the archipelago of Zanzibar in Tanzania (Ministry of Health Citation2016). Although the number of facility deliveries and the number of women attending antenatal care (ANC) have increased (Ministry of Health Citation2016), this does not guarantee receiving skilled birth attendance, nor high quality of care (Fakih et al. Citation2016).
In low-resource settings, the focus has long been on increasing facility delivery rates and availability of essential interventions (Pearson and Shoo Citation2005; Mirkuzie et al., Citation2014; Mkoka et al. Citation2014; Binyaruka and Borghi Citation2017). However, the ability to provide comprehensive, high-quality care is largely influenced by characteristics of the health care system such as patient-to-care provider ratios, level of training, and the presence of a feedback-and-support structure. Factors such as these can be challenged in low-resource settings (Mills Citation2014). Audits of maternal deaths including root cause analyses have shown that system- and healthcare provider-related factors are more widespread than equipment- or supply problems (Madzimbamuto et al. Citation2014; Mahmood et al. Citation2018; Okonofua et al. Citation2018).
Although these factors are known to be prevalent in cases of maternal death, their true contribution to the mortality burden is unclear as they are most likely to be also present in women that did survive severe complications. If adequately selected, maternal near-miss cases are very similar to maternal mortality cases and can be used to quantify this contribution (Herklots et al. Citation2019).
This study explores the contribution of system-related factors on maternal mortality in Zanzibar’s Referral Hospital, Mnazi Mmoja Hospital by comparing their incidence in cases of maternal death to that of maternal near-misses.
Methods
Setting
Mnazi Mmoja Hospital (MMH) in Zanzibar, Tanzania, is a high-volume facility with over 12,000 deliveries per year, covering 30% of Zanzibar’s facility deliveries (Fakih et al. Citation2016; Nadkarni et al. Citation2019). It has a maternal death ratio of 401 maternal deaths per 100,000 deliveries (Herklots et al. Citation2017). Essential obstetrical interventions and an intensive care unit (ICU) are generally available and accessible, although not consistently of good quality.
The World Health Organisation’s (WHO) near-miss approach has been applied prospectively in MMH since April 2017 (Herklots et al. Citation2019). It has been shown to correctly identify a subgroup at very high risk of maternal mortality, with characteristics essentially comparable to maternal mortality cases.
Inclusion and exclusion
All maternal death (MD) cases and maternal near-miss (MNM) cases with documentation, from August 2015 to July 2018 for MD and from April 2017 to September 2018 for MNM occurring in Mnazi Mmoja Hospital were included.
Case identification was performed during the handover meetings of the Obstetrics & Gynaecology department. Basic characteristics of cases were recorded directly, and clinical files were earmarked for later collection after discharge. Maternal death was defined as ‘the death of a woman while pregnant or within 42 days of the end of pregnancy, irrespective of the duration and site of the pregnancy, from any cause related to or aggravated by the pregnancy or its management but not from accidental or incidental causes’ (WHO Citation2012). Maternal near-misses were identified using a setting-adjusted version of WHO’s near-miss criteria (see Appendix 1) (Herklots et al. Citation2017, Citation2019). Clinical files and maternal death reviews were analysed retrospectively. Clinical files that were incomplete or missing were excluded from the analysis and compared to included cases using baseline characteristics and the number of signs of organ dysfunction.
Outcome indicators and variables
We chose to assess three main health care system-related factors deemed to reflect the workload and the capacity of MMH’s staff to provide comprehensive care: (1) the number of admissions per health care provider (HCP) on the day of admission, (2) the number of surgical interventions per HCP on the day of admission, (3) the mean number of times vital signs were monitored per day from the first sign of organ dysfunction to death or recovery. Recovery was defined as discharge from the ICU or High Dependency Unit, or discharge from the general ward if the patient had not been sent to either.
We chose to focus on the day of admission in the first two outcome indicators. Adequate triage at admission is a prerequisite for providing high-quality care and therefore our first step in assessing system-related factors should start with this, instead of using, for example, the day of the first sign of organ dysfunction.
All diagnoses were categorised according to The WHO Application of ICD-10 to deaths during pregnancy, childbirth and the puerperium (ICD-MM) (WHO Citation2012). In cases with more than one diagnosis, only the diagnosis that was deemed most relevant to the maternal outcome after consultation with a consultant in obstetrics and gynaecology was included. In nine MD and twelve MNM, two diagnoses were deemed equally relevant to the maternal outcome and therefore both were included. The location of residence was categorised as urban or rural, reflecting a high or low density of health care facilities, respectively (Fakih et al. Citation2016). The number of surgical interventions was drawn from the anaesthesia records kept in the maternity theatre. The number of admissions was computed from monthly nurse reports. The number of doctors and nurses working at any given time is not recorded systematically, therefore, it was based on work rosters from November and December 2018.
Analysis
Statistical analysis was conducted in SPSS Statistics, version 25 (IBM, Armonk, NY, USA). Baseline characteristics (age, parity, mode of delivery, number of ANC visits, marital status, and level of education) were compared between the two groups with Pearson’s chi-square tests or Fisher’s exact test, when at least 20% of cells had expected counts <5. P values below 0.05 were considered statistically significant.
Logistic regression was applied to determine the relationship between system-related factors and maternal outcome. Backwards regression determined the relevant confounders and effect modifiers. Imputation was done in case of missing data, limited to creating a separate category for categorical variables and computing the mean for both outcome groups for numerical variables.
A sub-group analysis was performed excluding all MD occurring within 24 h of admission to avoid a spuriously lowered rate of monitoring per day in these cases.
Figures were generated in Jupyter notebook (jupyter.org) using the pandas and seaborn packages. For bootstrapping, the resample function of the Sklearn package was used, which applies random case resampling with replacement. This was performed 1000 times to estimate from the data the population variance in mortality (% MD).
Ethics
The Zanzibar Medical Research and Ethics Committee approved the study (ZAMREC, protocol reference numbers: ZAMREC/0001/August/005; ZAMREC/0001/Jan/17). Informed consent was deemed unnecessary due to anonymisation of the data collected from clinical files.
Results
A total of 364 patients were identified: 148 maternal deaths and 216 maternal near-misses, giving monthly averages of 4.1 MD and 12.5 MNM. The case ratio per 100,000 live births was 1:2.7 (426 MD:1150 MNM). Nine MD and 94 MNM were excluded from further analysis due to missing or incomplete files, and therefore missing data on the outcome indicators. Data on outcome indicators were complete for 139 MD and 122 MNM. Baseline characteristics and number of signs of organ dysfunction of included and excluded cases were comparable (Appendix 2).
Demographic characteristics are presented in . MD were significantly more likely to have had a spontaneous vaginal delivery and to display a higher number of signs of organ dysfunction.
Table 1. Demographic characteristics per maternal outcome.
In both groups, the most common diagnosis was obstetric haemorrhage, with postpartum haemorrhage (38.7% MD, 42.0% MNM) and ruptured uterus (12.9% MD, 27.5% MNM) as most prevalent within this category. Hypertensive disorders (eclampsia 38.2% MD, 50.0% MNM and (severe) pre-eclampsia 44.1% MD, 33.3% MNM) were the second most frequent diagnoses. Among MNM, the category ‘pregnancies with abortive outcome’ (ruptured ectopic pregnancy 25% MD, 100% MNM) and ‘other obstetric complications’ (postpartum cardiomyopathy (0 MD, 58.3% MNM), postpartum renal failure (0 MD, 16.7% MNM)) were more prevalent. By contrast there were significantly more ‘unknown/undetermined’ diagnoses among MD compared to MNM.
shows the outcome of descriptive analysis, followed by the outcomes of the logistic regression in . Backward regression determined the relevant confounders and effect modifiers: age, main diagnosis, number of signs of organ dysfunction, number of ANC visits, location of residence in relation to the hospital and calendar year. When adjusting for confounders, patients whose vital signs were measured more often, were less likely to die. Each additional vital sign measurement per day reduced the odds of mortality with a factor 0.75 (95% CI 0.64–0.89). The number of patients admitted and the number of surgeries performed per healthcare provider were not correlated to maternal mortality.
Table 2. Results descriptive analysis.
Table 3. Results univariate/multivariate logistic regression.
A sub-group analysis was performed, excluding all MD occurring within 24 h of admission, to avoid a spuriously lowered rate of monitoring of vital signs per day in these cases. The results were similar (MD n = 63, MNM n = 122, aOR = 0.69, 95% CI 0.53–0.91).
The results of the logistic regression suggest a strong correlation between mortality and the mean number of times vital signs are measured per day. To elucidate this relationship, we plotted the mean number of times vital signs were measured for both MD and MNM, broken down by disease severity (). As the severity of disease increased, the mean number of times vital signs were recorded increased accordingly in MNM cases. In MD, the increase is only significant in women with four or more signs of organ dysfunction but still falls behind the increase seen for MNM cases in terms of frequency and consistency. The higher number of women with a high number of signs of organ dysfunction in MD leads to an overall higher average number of times vital signs were monitored in that group, see unadjusted frequency in . However, when considering a subgroup of comparable severity, vital signs were measured more often in MNM than in MD cases, in concordance with the logistic regression ().
Figure 1. The number of vital signs measured per day as a function of the number of signs of organ dysfunction, split by outcome (MD, Maternal Death; MNM, Maternal Near-miss). The data are represented in violin plots, showing the distribution of the data points (individual patients). The median (white dot) and the interquartile range (black box, 25th–75th percentile) are indicated.
The relationship between the mean number of times vital signs were recorded and risk of mortality (%MD) is shown in . Except for the lowest risk cases i.e. one sign of organ dysfunction, all severity groups showed an increased survival associated with an increased frequency of monitoring. For women with two to three signs of organ dysfunction mortality rates are circa 50% lower with high-frequency monitoring (≥9 times/day) compared to low-frequency (1–3 times/day). This reduction levels off with increasing severity of disease.
Figure 2. Survival of patients (percentage Maternal Near-miss, y-axis) broken down by disease severity (individual panels) and by measured vital signs per day (x-axis). The line represents percent survival, whereas the shaded areas represent the standard deviation of 1000-fold bootstrapping. This assumes that the data are representative for the population. This requirement may be violated in groups with small numbers, for example in the panel with four signs of organ dysfunction, with 6–9 measured vital signs per day which contains just four patients, all MD.
Discussion
This study analysed the influence of three major health care system-related factors reflecting the workload of HCP in MMH on maternal mortality. The mean number of times vital signs were measured per day from the first sign of organ dysfunction to death or recovery was associated with significantly lowered odds of mortality, after correcting for disease severity and patient characteristics. An increase of one in the mean number of times the vital signs were measured per day, leads to 25% decrease in the odds of mortality. This study does not show a significant association between odds of mortality and the patient-to-HCP ratio or the number of surgical procedures per HCP.
Our study suggests that high frequency monitoring in severe maternal complications results in lower mortality up to a certain degree. This is in line with a secondary analysis of the WHO Multicountry Survey on Maternal and Newborn Health showing that availability of intensive care significantly reduces mortality (Soares et al. Citation2020). Our study also suggests, however, that availability is necessary but not sufficient. First, because at the extreme end of the spectrum (women with ≥5 signs of organ dysfunction), the currently provided care seems to have no real impact on mortality rates. Second, the lower level of monitoring in MD seems to indicate that not recognising severity of disease and therefore missing out on timely upscaling of care might contribute significantly to mortality. In other words, to have a significant impact on mortality, adequate and timely identification of severely ill patients and the capacity to treat them at intensive care level are both indispensable.
The retrospective nature of our study precludes strong conclusions about why care levels were not upscaled. System-related factors, such as workload or ability to recognise and divert care to the most severely ill are likely to play a role. We did not find a relation between workload indicators measured on the day of admission and mortality risk. However, this does not exclude a relationship of workload on other days during admission with mortality risk. Previous studies in Nigeria’s referral hospitals and mathematical predictions of mortality at MMH have shown a negative impact of patient-to-HCP ratio and workload on mortality (Okonofua et al. Citation2018; Nadkarni et al. Citation2019). In a setting where the workload is persistently high, it is possible that the differences on a day-to-day basis were too small to be able to find a significant correlation with chances of survival.
The lack of increase in monitoring frequency seen more often in MD cases than in MNM could reflect a lack of awareness of the severity of disease. In this case, systemic efforts should not only focus on providing intensive levels of care but also on factors that hamper adequate triage. For instance, by adequate training on early warning systems and spatial organisation that facilitates oversight of patients.
In both cases, including maternal near-miss cases based on a validated measure of disease severity, such as the number of signs of organ dysfunction used in our study, provides a clear framework for analysis of the level of care given. It helps to distinguish between possible shortcomings in terms of therapeutic capacities (level of severity too high compared to capacities such as seen in women with ≥5 signs of organ dysfunction) or inadequate care (low to intermediate level of severity but high mortality rates) during audits.
The main strength of the present study is its well-defined population. A clear set of internationally recognised criteria has been used to identify women with life-threatening complications with a high risk of mortality (Herklots et al. Citation2017). We have shown before that MNM and MD are very similar in the types of signs of organ dysfunction they experience. Moreover, we have validated the assumption that number of signs of organ dysfunction is related to the risk of mortality in this cohort (Herklots et al. Citation2019). This meant that we could adequately correct for the main confounder, the severity of the life-threatening conditions, by correcting for signs of organ dysfunction.
Despite the large size of the groups studied, there was a significant proportion (43.5%) of patient files of MNM that could not be used. Although MNM were identified prospectively, case analysis was performed later on and indicators were missing in many cases, mostly due to missing patient files. Baseline information was, however, available in all cases for comparison between included and excluded cases. The comparison did not show significant differences, which limits the risk of selection bias. The absolute number of MD with missing files was too low to perform this comparison but is therefore unlikely to affect the findings significantly.
Reporting bias could have occurred with more underreporting of vital signs at moments that workload was high. When vital signs are not documented, they cannot be consulted by other HCP in consequent shifts to evaluate changes in patients’ conditions. In other words, we consider poor documentation to be substandard care in itself.
Despite these shortcomings, this is the first time the concept of MNM is employed to deepen our understanding of the contribution of system-related factors leading to maternal mortality in low-resource settings. We recommend this to be an integral part of maternal morbidity and mortality case audits in any setting.
Conclusion
This study assessed the effect of system-related factors on maternal mortality. The frequency of vital signs monitoring was inversely related to risk of mortality after correcting for disease severity. Based on this study’s findings, investments should be made in attracting and keeping enough well-trained staff to monitor patients closely when needed, as this alone could significantly reduce mortality. Especially in a setting of limited resources this should be taken in consideration when setting priorities. Further audits combining MNM and MD could shed more light on system-related factors hampering the provision of adequate care.
Disclosure statement
The authors report no conflicts of interest. The authors alone are responsible for the content and writing of this article.
References
- Binyaruka P, Borghi J. 2017. Improving quality of care through payment for performance: examining effects on the availability and stock-out of essential medical commodities in Tanzania. Tropical Medicine & International Health 22:92–102.
- Fakih B, Nofly AAS, Ali AO, Mkopi A, Hassan A, Ali AM et al. 2016. The status of maternal and newborn health care services in Zanzibar. BMC Pregnancy and Childbirth 16:134.
- Herklots T, van Acht L, Khamis RS, Meguid T, Franx A, Jacod B. 2019. Validity of WHO's near-miss approach in a high maternal mortality setting. PLoS One 14:e0217135.
- Herklots T, van Acht L, Meguid T, Franx A, Jacod B. 2017. Severe maternal morbidity in Zanzibar's referral hospital: measuring the impact of in-hospital care. PLoS One 12:e0181470.
- Madzimbamuto FD, Ray SC, Mogobe KD, Ramogola-Masire D, Phillips R, Haverkamp M et al. 2014. A root-cause analysis of maternal deaths in Botswana: towards developing a culture of patient safety and quality improvement. BMC Pregnancy and Childbirth 14:231.
- Mahmood MA, Mufidah I, Scroggs S, Siddiqui AR, Raheel H, Wibdarminto K et al. 2018. Root-cause analysis of persistently high maternal mortality in a rural district of Indonesia: role of clinical care quality and health services organizational factors. Biomed Research International. 2018:1–11.
- Mills A. 2014. Health care systems in low- and middle-income countries. The New England Journal of Medicine 370:552–557.
- Ministry of Health, Zanzibar. 2016. The Zanzibar Annual Health Bulletin 2016. Department of Planning, P. A. R. (ed.). Zanzibar.
- Mirkuzie AH, Sisay MM, Reta AT, Bedane MM. 2014. Current evidence on basic emergency obstetric and newborn care services in Addis Ababa, Ethiopia; a cross sectional study. BMC Pregnancy and Childbirth 14:354.
- Mkoka DA, Goicolea I, Kiwara A, Mwangu M, Hurtig AK. 2014. Availability of drugs and medical supplies for emergency obstetric care: experience of health facility managers in a rural District of Tanzania. BMC Pregnancy and Childbirth 14:108.
- Nadkarni D, Minocha A, Harpaldas H, Kim G, Gopaluni A, Gravelyn S et al. 2019. Predicting resource-dependent maternal health outcomes at a referral hospital in Zanzibar using patient trajectories and mathematical modeling. PLoS One 14:e0212753.
- Okonofua F, Ntoimo L, Ogu R, Galadanci H, Abdus-Salam R, Gana M et al. 2018. Association of the client-provider ratio with the risk of maternal mortality in referral hospitals: a multi-site study in Nigeria. Reproductive Health 15:32.
- Pearson L, Shoo R. 2005. Availability and use of emergency obstetric services: Kenya, Rwanda, Southern Sudan, and Uganda. International Journal of Gynaecology and Obstetrics 88:208–215.
- Soares FM, Pacagnella RC, Tuncalp O, Cecatti JG, Vogel JP, Togoobaatar G, Souza JP. 2020. Provision of intensive care to severely ill pregnant women is associated with reduced mortality: Results from the WHO Multicountry Survey on Maternal and Newborn Health. International Journal of Gynaecology and Obstetrics 150:346–353.
- WHO. 2012. The WHO Application of ICD-10 to deaths during pregnancy, childbirth and the puerperium. ICD-MM.