Study design
A cross-sectional study including 109 reported and reviewed maternal deaths from two regions in Tanzania that occurred between 1st January and 31st December 2018 was conducted. Routine MDSR categorization of cause of deaths and care delays was compared with those assigned by an independent expert panel of obstetricians with additional information from VA. To compute the completeness of maternal deaths reported by the MDSR we used the number of infants that received Bacillus Calmette-Guerin (BCG) vaccine, as a proxy for live births as previously recommended, (22) to calculate the MMR for the two regions in 2018.
Study setting
The study was conducted in Lindi and Mtwara regions of Southern Tanzania with a total population of about 2 million (23). The two regions have two regional referral hospitals, 12 district hospitals, three private/mission hospitals, 40 health centres and 399 dispensaries. The MMR in Lindi and Mtwara was 456 and 579 per 100,000 live births in 2013 (24). The fertility rate is one of the lowest (3.8) in Tanzania. Most women deliver in facilities, 80.8% and 81.3% in Lindi and Mtwara, respectively. Caesarean section rate is 6.0% in Lindi and 10.3% in Mtwara (5).
The MDSR system in Tanzania and categorization of cause of death
Each health facility that provides delivery services in Tanzania has a standard MDSR committee as stipulated in the guideline (21). In regional and district hospitals, where most deaths occur, MDSR committee is composed of a multidisciplinary team of clinical and non-clinical staff such as obstetrician (if available), medical doctors, clinical officers, nurses and midwives from maternity ward, facility management, laboratory personnel and other supporting staff . The committee meets within seven days after a suspected maternal death has occurred. Before the meeting, a designated person prepares a narrative summary using information from medical files, interviews of health care providers and relatives who cared for the patient. There is no clear guide on how and which relatives should be interviewed. During the meeting the summary is presented and discussed and when necessary more information is obtained from medical files or health care providers who cared for the patient. Findings from the meeting are summarised in a maternal death reporting form which includes demographic characteristics, medical information, underlying medical cause of deaths, description of contributing non-medical factors along the three phases of delays and a plan of action (21). The MDSR guideline recommends the underlying medical cause of death to be categorized following ICD 10 rules but the training and the guideline does not provide a formal training on this. The reporting form in MDSR guideline has a short list of example of causes and ICD 10 codes to be used during reporting. (See Annex 1)
Outcomes
Our main outcome was the underlying medical cause of death defined as disease condition that started chain of events that led to death e.g. Postpartum Haemorrhage (PPH) (11). Underlying causes of deaths are grouped into nine groups that are mutually exclusive, totally inclusive and descriptive of all underlying causes of maternal deaths (Box 1)
Type | Group number and name |
Direct maternal deaths | 1. Pregnancy with abortive outcome |
| 2. Hypertensive disorders in pregnancy, childbirth and the puerperium |
| 3. Obstetric Haemorrhage |
| 4. Pregnancy related infection |
| 5. Other obstetric complications |
| 6. Unanticipated complications of management |
Indirect maternal deaths | 7. Non-obstetric complications |
Unspecified | 8. Unknown/undetermined |
Death during pregnancy child birth and puerperium | 9. Coincidental causes |
Box 1: ICD MM groups of causes
As stipulated in Tanzania MDSR guideline, delays in health seeking or provision of care deemed to have contributed to the maternal deaths were grouped using the three delays model, stipulating delays 1) to deciding to seek care 2) to reach care including transport and 3) to receive appropriate care in facilities (21). Several delays may have contributed to a death. Phase one delays are delays at household and personal level that lead to late or lack of seeking care. It includes the time from onset of disease at home until decision to seek care is made by the patient, family or both. Phase two delays are concerned with access to health care such as availability of health facility, roads and transport issues, and constitute time from when decision to seek care is made until arrival at a proper health facility. Phase three delays occurred in health facilities and are more concerned with time, equipment and supplies, structure, management errors, human resource and referral system, and constitutes time until adequate treatment or care begins.
Data sources and measurements
Data collection followed three steps: i) abstracting information from MDSR documents ii) performing VA, and iii) the independent obstetrician panel review.
The first author AS, in close collaboration with regional Reproductive and Child Health Coordinators, abstracted information using a pre-defined checklist from maternal deaths narrative summaries, death review report forms and district monthly death report summaries (date of death, age, facility, village and cause of death)..
The field team (AS and VA interviewers) then traced families using demographic information such as names of the deceased, place of death, district and date of death, home address, name of village/street leader, name of husband/partner and other information, for VA interviews.
VA interviews were conducted using the translated standard questionnaire provided by WHO (25). The questionnaire was piloted and the translation were reviewed and corrected accordingly. In addition to the standard inquiries, questions relating to the three phases of delays were added.
The field team commenced the process of finding families for VA interview by visiting and enquiring at facility where death occurred or where the deceased attended her antenatal clinic. They were then taken to the family through local government leaders. At the family`s home, after being introduced they explained in detail the purpose of VA. Then one of the interviewers identified person (s) that was (were) present during illness and death and conducted VA with them.
Using the VA questionnaire as well as copies of available medical files a group of experts, consisting of two experienced obstetricians in MDSR from Muhimbili University of Health and Allied Sciences and one from Mtwara regional hospital reviewed all maternal deaths. The author, AS, was among the panel members and had previously been trained on ICD-MM. The panel members neither conducted the VA interviews nor documented any information used by the reviews.
First, the expert panel went through VA questionnaire and determined the underlying cause from the information by consensus. Second, the panel went through the medical files and reviewed all available information. Based on these two sources, the panel determined the i) underlying cause of death including the ICD coding, ii) contributing medical causes and iii) three phases of delays by consensus (11).
Quantitative variables
Data were processed using MS Excel and then transferred to SPSS computer program version 25.Proportions of each underlying medical cause categorized by MDSR system and the expert panel of obstetricians were computed. The underlying medical cause and differences between the routine MDSR system and obstetricians panel were tabulated. As the routine MDSR system used a shortlist of ICD codes while the expert panel used the full number of ICD-MM the comparison had to use a pragmatic approach. For example, when the obstetrician panel categorized a death to be caused by PPH due to atony, coagulopathy or retained placenta, then this was considered to be in agreement if MDSR system categorized the same death as PPH (Non traumatic). Also PPH (Traumatic) for MDSR system was decided to be in agreement if obstetrician panel categorized the same case as PPH (vaginal tear, cervical tear, uterine tear during caesarean section excluding uterine rupture).
Statistical methods
Cohen`s K statistic was used to determine level of agreement in categorizing the underlying causes. We defined < 0 as no agreement, 0-0.2 as slight agreement, 0.21-0.4 as fair, 0.41-0.6 as moderate, 0.61-0.8 as substantial and 0.81-1 as almost perfect agreement (26) . Agreement between the obstetricians’ panel and the MDSR system on the three delays was calculated by determining percentage difference between the two groups.