Audit-identified avoidable factors in maternal and perinatal deaths in low resource settings: a systematic review

Studies from low and lower-middle income countries were considered, using World Bank criteria for stratification of country by gross national income (GNI) per capita [15]. Studies were eligible if they performed an audit of maternal or perinatal deaths using medical records, meetings of health workers, and/or interviews of health workers or patient families. Studies had to explicitly identify at least one avoidable factor in a maternal or perinatal death, and studies had to utilize definitions for maternal and perinatal deaths that were reasonably similar to those used by WHO. Maternal deaths are defined by WHO as “the death of a woman while pregnant or within 42 days of termination 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” [16]. The perinatal period, as defined by the WHO, “commences at 22 completed weeks (154 days) of gestation and ends seven completed days after birth” [17]. Perinatal mortality refers to stillbirths and newborn deaths in the first week of life [17]. For the purpose of this analysis, a factor in an “avoidable death” was defined as one that was assessed to be directly related to the death; in other words, if the factor had been avoided than the death would probably not have occurred [18]. Audits that did not report specific avoidable causes in maternal and/or perinatal deaths were excluded. Two investigators (HSM and JS) determined the eligibility of the articles independently and any discrepancy was resolved by a discussion between these two investigators.

We performed a systematic search of published literature from 1965 to November 2011 in Pubmed, Embase, CINAHL, POPLINE, LILACS and African Index Medicus. The search strategy included various combinations of exploded and focused MeSH headings and keyword searching using the terms Perinatal, Maternal, Mothers, Neonatal, Newborn, Infant, Mortality, Audit, Clinical Audit, Death, Fatal Outcome, Avoidable, Preventable, and Developing Countries. A previously described “snowball” search strategy was also performed in which papers were identified through bibliographies of key studies [19]. We also attempted to identify additional relevant studies, in particular non-published datasets, through queries to experts and international organizations concerned with maternal and perinatal health. Articles in all languages were considered and translated when necessary.

Each article that met inclusion criteria was assigned a quality score in order to impose a weighing scheme according to the quality of the study. The score, developed to analyze the quality of obstetrical care in low and middle income countries [20], utilizes a numerical scale ranging from zero (lowest quality) to nine (highest quality). Calculation of the quality score was conducted by assessing the following criteria: selection of audit cases (maximum 3 points), quality control during the audit process (maximum 4 points), and reliability of the audit (maximum 2 points) (See Nine-point clinical audit quality criteria). Two investigators (HSM and JS) independently scored each audit by assigning one point for each criterion met by the article. If there was a discrepancy between the scores of the two investigators, the two scores were averaged to arrive at a final score.

Nine-point clinical audit quality criteria adapted from Pirkle et al.[20].

Selection of cases

Description of study population with clear case definition

Description of sampling strategy

Consideration of missing cases

Data quality control

Criterion-based clinical audit pilot or pre-tested

Description of staff profile

Training of staff

Data entry validity checks

Reliability

Standardized data collection form

Inter-observer/inter-site variability assessed

A meta-analysis was performed for each avoidable factor identified. First, we calculated the percent of studies where the factor was determined to be a cause of a maternal or perinatal death. Next, for those studies where the factor was determined to be a cause of death, a “pooled” or combined estimate of the percentage of deaths in those studies due to the factor was calculated. The pooled estimate was calculated using a weighted average of the estimates from the individual studies via accepted meta-analysis techniques [21, 22], in which the study is treated as a random effect. We estimated the weighted averages, confidence intervals for the weighted averages, and the variance of the random study effect using an iteratively reweighted least squares approach [21]. The weights used in the iteratively reweighted least squares approach takes into account the sample size from each individual study, the variance of the random study effect, and the quality score of each study. This approach is semi-parametric in that is gives unbiased estimates regardless of the type of outcome variable (continuous, categorical or percentages). Finally, we determined the total number of attributable deaths for each avoidable factor. The study was conducted and reported according to PRISMA guidelines [23], (See Additional file 1). All analyses were performed using SAS/STAT 9.3 (SAS Institute, Cary, NC, 2011).

The search returned 3,775 results and an additional 16 papers were added from reference lists. There were 691 duplicate entries and 2,975 were subsequently removed after abstract screening. The full texts of the 125 remaining articles were reviewed and 39 met inclusion criteria (Figure 1). These 39 studies included a total of 44 datasets since five studies included multiple datasets; one study included datasets from two different time periods [24], one study included datasets from two different countries [25], and two studies included datasets from both maternal and perinatal populations [24, 26]. Two other studies analyzed deaths from different time periods [27, 28] but the time periods were contiguous in these studies and so they were considered as one dataset each. We were unable to identify unpublished audits of maternal and newborn deaths that met inclusion criteria. Therefore, only published data were used. We summarized the datasets and categorized them by type of measure examined [24‐62] (Table 1).

Forty-four datasets were included in the study, comprising a total of 6,205 audited deaths. The majority of the datasets, 82%, came from African studies. Two datasets (4.5%) were from Pakistan [48, 61], two (4.5%) from Indonesia [25, 30], and four (9.1%) from India [24]. Of note, the four datasets from India were obtained from the same investigation [24]. The oldest dataset was from 1967 [24] and the most recent datasets were from 2009 [47, 62]. Thirty-one of the 44 datasets (70.5%) examined maternal deaths while the other 13 datasets (29.5%) examined perinatal deaths. Study methods varied and there were an equal number of datasets that employed prospective methods (45.5%) as those that used retrospective methods (45.5%). Four of the datasets [27, 45, 48, 56] (9.1%) used a combined retrospective and prospective methodology. The majority of the datasets, 72.7%, described hospital populations, while the remaining 27.3% described both hospital and community populations. Many of the studies reported the maternal mortality rate and perinatal mortality rate from their audited populations, and these varied between studies. Just under half of the studies (47.7%) estimated the percentage of deaths that were thought to be avoidable in the sample audited, and this ranged from 1.5% to 88%. Quality scores for the audits ranged between 2–8 (See Table 1).

Overall there were 42 avoidable factors in maternal and perinatal deaths identified from the 44 datasets. These 42 factors fell into four general categories: health worker-oriented factors were the most common, accounting for 28 (66.7%) of the 42 factors; the next most common, in descending order of frequency, were patient-oriented factors (14.3%), administrative/supply factors (11.9%), and transport/referral factors (7.1%). All of the factors are summarized by category in Table 2 (within each category, factors are listed in order of relative number of attributable deaths).

The top 10 avoidable factors, listed in order of attributable deaths, are shown in Table 3. The number of deaths attributable to these factors ranged between 251 – 688 of the total sample of 6,205 audited deaths. Each of the top 10 avoidable factors was reported in a maternal or perinatal death in 10 – 23 (22.7% - 52.3%) of the datasets. The most common factors reported overall were patient and transport delays, each of which was a reported factor in 52.3% of the datasets. Health care worker-oriented factors were the most common category of factors among the top 10 factors, accounting for six out of ten factors. Substandard health worker practice was responsible for the most deaths (688 deaths), and we estimated that this factor contributed to 28.5% of the deaths in datasets where it was identified as an avoidable factor (See Additional file 2).