The frequency of maternal morbidity: A systematic review of systematic reviews

Abstract Background Estimates of the burden of maternal morbidity are patchy. Objective To conduct a systematic review of systematic reviews of maternal conditions to: (1) make available the most up‐to‐date frequency estimates; (2) identify which conditions do not have reliable estimates; and (3) scrutinize the quality of the available reviews. Search strategy We searched Embase, MEDLINE, and CINAHL, combining terms for pregnancy, frequency (e.g. prevalence, incidence), publication type, and specific terms for each of 121 conditions. Selection criteria We included peer‐reviewed systematic reviews aiming to estimate the frequency of at least one of the conditions in WHO's list of maternal morbidities, with estimates from at least two countries. Data collection and analysis We present the frequency estimates with their uncertainty bounds by condition, region, and pregnancy/postpartum period. We also assess and present information on the quality of the systematic reviews. Main results Out of 11 930 found, 48 reviews were selected and one more was added. From 49 reviews we extracted 34 direct and 60 indirect frequency estimates covering 35 conditions. No review was available for 71% of the conditions on the WHO list. The extracted estimates show substantial maternal morbidity, spanning the time before and beyond childbirth. There were several gaps in the quality of the reviews. Notably, one‐third of the estimates were based only on facility‐based studies. Conclusions Good‐quality systematic reviews are needed for several conditions, as a research priority.

of morbidity resulted in 27 million morbid episodes in 2015. 2 These sources, however, underestimate the true burden of disease attributable to pregnancy-related conditions as they include only a few maternal conditions. 3 They ignore common conditions, such as postpartum depression, 3 and mild but prevalent conditions, such as urinary incontinence that affects over one-third of the pregnant population in Europe alone. 4 The WHO recently published a comprehensive list of maternal morbidities, comprising 121 direct and indirect conditions. 5 This list provides an important framework to understand what conditions constitute maternal morbidity, although the extent to which each of the listed morbidities contributes to the total burden remains unclear. 3 Addressing this gap in our knowledge is necessary to better prioritize conditions for intervention. Furthermore, identifying the conditions that we know the least about is also important so they can be included in the future research agenda. 2 The aim of this systematic review is to identify existing systematic reviews quantifying the burden of each of the conditions identified in WHO's list of maternal morbidities. Compiling this information will enable us to: (1) make available the most up-to-date frequency (e.g. prevalence, incidence) estimates on each maternal condition; (2) identify which conditions lack reliable estimates; and (3) discuss the quality of the systematic reviews and the reliability of the available estimates.

| Search strategy
We conducted a systematic search in Embase, MEDLINE, and the Cumulative Index to Nursing and Allied Health Literature (CINAHL) using a combination of free text terms and Medical Subject Headings (MeSH terms). We combined terms for the following domains: pregnancy (e.g. maternal, antenatal), frequency of the disease (e.g. prevalence, incidence), publication type (e.g. systematic review, metaanalysis), and specific terms for each of the 121 conditions described in the WHO maternal morbidity list by Chou et al. 5 The search strategy was prepared by AL, CC, and GG, with input from VF, SW, and an experienced librarian. The complete strategy is provided in Appendix S1. The search was restricted to humans and there were no language restrictions. The search was last run on July 23, 2016. In addition, we included further relevant systematic reviews known to the authors of this paper but not identified by the search, and we searched the reference lists of eligible studies. We used the MOOSE guidelines for conducting systematic reviews of observational studies to carry out and report on this review. 6

| Inclusion and exclusion criteria for selection of systematic reviews
We included peer-reviewed systematic reviews that aimed to estimate the frequency of at least one of the maternal conditions listed in Chou et al. 5 and which included estimates from at least two countries.
The latter was a way to ensure we included estimates representing a region rather than a specific country. We included systematic reviews that included at least one paper published in or after 2006, as an attempt to provide recent estimates.
We excluded papers that: (1) did not mention frequency of the outcome among pregnant women in the abstract; (2) only reviewed studies for certain subgroups (e.g. rural women, or women with a specific health condition, women giving birth to twins, or women with a previous cesarean delivery); (3) focused only on risk factors or consequences of a certain maternal condition; (4) were not systematic reviews; (5) primarily included interventions for or investigated the effect of a single individual characteristic of the relevant maternal conditions; (6) were not possible to access in full; and (7) provided insufficient information on their inclusion and exclusion criteria in the text and the authors did not provide this information after we had attempted to contact them twice.

| Data extraction
Data extraction from eligible reviews was carried out at two levels: (1) information on the overall paper; and (2) the specific frequency estimate. We did not extract information from the primary studies included in the eligible systematic reviews. For the study selection of the systematic reviews, one author (AL) screened titles and abstracts, and 10% of these were also screened by GG.
During the first level of data extraction, either AL, GG, or CC extracted data from eligible reviews such as the region reviewed, the databases searched, and the inclusion criteria applied for study selection. We also extracted detailed information on the quality of the systematic reviews. This was performed by AL with double extraction of 50% of the reviews (by GG). To assess the quality, we adapted the quality assessment tool for assessing systematic reviews proposed by Mann et al., 7 which is a modified version of the Overview Quality Assessment Questionnaire (OQAQ). Our adaptations included a question on whether authors clearly specified the source of the datawhether hospital, population, or unknown-and whether the search strategy was clearly laid out. The details on the modified OQAQ tool we used (which included 13 criteria) and the way we scored against it are available in Appendix S2. We did not provide numerical summary quality scores for each eligible review because these could mask the relative importance of the different quality indicators. Instead, we used a traffic light system, and we calculated the overall proportion of articles scoring a specific color (e.g. green) for each question.
For the second level of data extraction, either GG or AL extracted the frequency estimates together with information on whether these were population-or facility-based, the denominator for each, the countries they represented, the type of estimate (i.e. incidence or prevalence), and the diagnostic tools used for case ascertainment.
Estimates were classified as population based if: (1) authors clearly said they were population based, or (2) the sample was recruited from facilities in countries where virtually all deliveries happen in facilities.
For 50% of the papers included, we carried out double extraction of the frequency estimates and their details. Any discrepancies were resolved through discussion.
To select "the best" estimates to extract from systematic reviews where several frequency estimates were presented, we established the following rules that were applied in hierarchical order: 1. Select population-based estimates over (a) facility-based estimates, or (b) estimates combining both facility-and population-based estimates.

2.
When both pooled estimates (i.e. a weighted average) and the range of estimates from individual studies were provided, extract the pooled estimate.

3.
Select estimates covering the widest geographical area.

4.
Select the most recent estimates, in terms of data capture period.
For example, if a review reported a frequency estimate based on community-based studies and a separate estimate for facility-based studies, then we only extracted the estimate based on the communitybased studies. If a study reported weighted means based only on facilitybased studies separately for West Africa and for the whole of the African continent, then we selected only the estimate for Africa.
Estimates from systematic reviews where only a single study was identified are equivalent to reporting estimates from a single primary study; therefore, for the purposes of reporting estimates of maternal morbidity, we did not include estimates from such systematic reviews.
For any reviews that were eligible based on our inclusion and exclusion criteria, but from which we could not extract a frequency estimate, we have included the paper in the main description but we do not report an estimate from it.

| Analysis
We transformed all estimates into percentages for presentation and comparison. For a particular review and each condition reported on, we present the frequency and the type of estimate (prevalence or incidence), the uncertainty range, region, pregnancy period, diagnostic tools, and data source (facility-vs population-based). We report the region (or group of countries) for each estimate based on the countries covered by the primary studies included in the systematic review, which underpin each estimate. For those conditions reported by multiple eligible systematic reviews, we present estimates from each of those reviews. If those reviews reporting on the same condition included some of the same primary studies, we did not choose between the reviews because each review had distinct inclusion/exclusion criteria; for example, some reviews focused on certain countries or study designs. If a systematic review reported on multiple conditions of interest, we extracted estimates for each of these conditions.

| RESULTS
We identified 11 930 results from searches across Embase, MEDLINE, and CINAHL, of which 3481 were duplicates and 8302 were unrelated to the topic of interest after screening the title and abstract (Fig. 1). A total of 150 papers were selected for full-text review, of which three were added to the search results based on our previous knowledge.
Full-text review led to the exclusion of 102 papers for the reasons stated in Figure 1, including four articles that were excluded because they only reported composite outcomes, aggregating the frequency of multiple conditions in the WHO list. [8][9][10][11] We selected 48 eligible systematic reviews, and from searching their references we found one more. From these 49 eligible reviews, we extracted 34 direct and 60 indirect frequency estimates covering 35 conditions. The full list of included papers is provided in Appendix S3.

| Availability of systematic reviews
We found that for 71% of conditions in the WHO list by Chou et al. 5 there was no systematic review available (Appendix S3). The systematic reviews we found covered a substantial proportion (36%) of direct and coincidental maternal conditions, as well as several mental disorders (63%), and maternal infectious and parasitic diseases (46%).
Under the direct morbidity umbrella, our search did not yield any systematic review for three categories: (1) pregnancy-related infection, such as puerperal sepsis or mastitis; (2) cardiovascular obstetric complications such as peripartum cardiomyopathy; and (3) complications related to anesthesia.
In addition, we did not find any eligible systematic reviews for nine indirect morbidity categories listed by Chou et al., 5 as outlined in full in Appendix S3. For example, none were found in the category called "Other maternal diseases classifiable elsewhere but complicating pregnancy, childbirth and the puerperium," which includes anemia, and we also found none under "Diseases of the musculoskeletal system and connective tissue," including back pain. A systematic review on anemia was not considered eligible because it aimed to review primary studies investigating the risk factors for anemia, therefore excluding studies that investigated the frequency of anemia but did not report on effect-size estimates of risk factors. 12 Some conditions had multiple available systematic reviews. The highest number was identified for gestational diabetes (eight systematic reviews), 13-20 followed by infectious hepatitis, intimate partner violence, and postpartum depression, with four systematic reviews each. Although two systematic reviews were eligible and included, we have not reported frequency estimates for these because either the estimates were based on only one study per condition, 21 or they reviewed a variety of conditions and denominators that were difficult to combine to present summary estimates here 22 ; details are provided in Appendix S3.

| Characteristics of available systematic reviews
Details of the 94 frequency estimates extracted, including the denominator and the geographical area for each are presented by pregnancy period and by type of estimate in Table 1 (direct maternal morbidities) and Table 2 (indirect maternal morbidities). The systematic reviews used several types of prevalence (n=77) or incidence (n=17) estimates, including ranges, weighted means, crude means, and medians. The The systematic reviews covered different geographical and economic areas, e.g. the world, high-income countries (HICs), low-and middle-income countries (LMICs), or specific regions (Africa, Asia, Europe, etc.). Of the estimates we extracted, 17 (18%) were based on only two countries or it was not clear from the paper how many countries were included. Among the frequency estimates that included worldwide studies, the median number of countries contributing data was 10 (interquartile range, 7.5-20.5). Sub-Saharan Africa was the world region with the highest number of specifically dedicated systematic reviews (n=9).
Tables 1 and 2 describe the outcome assessment method behind each estimate. Information on assessment method at the estimate level was often scarce and poorly described. For direct morbidity estimates, the information on the assessment method underlying the estimates was unclear in five systematic reviews. 17,[23][24][25][26] In addition, some studies used assessment methods that are prone to bias. To take the example of gestational diabetes, Schneider et al., 18

| Quality of systematic reviews
There was much variation in the quality of the 49 systematic reviews, including some examples of excellent methodology and reporting. [27][28][29] Some aspects of quality were often found to be particularly poor, including insufficient reporting and methodological gaps (Fig. 2). For example, only 19 (39%) of the systematic reviews explicitly reported their language exclusions and the inclusion of grey literature, and only 21 (43%) provided a detailed description of the primary studies. Furthermore, for 16 (33%) of the reviews we did not have sufficient details on the data extraction process (e.g. use of independent extraction).
Information on data collection and sources was also lacking in many cases: for 19 (56%) of the direct morbidity estimates and 18 (30%) of the indirect estimates there was insufficient information to assess whether the data were from population-or facility-based sources. Overall, 32 (34%) of the estimates extracted included data from facility-based studies. Facility-based studies vary in terms of their representativeness. For example, when reviewing studies of the prevalence of malaria, Chico et al. 27

Author
Group of countries or region

| 31
Gon ET AL. at least once during pregnancy. Nevertheless, as the authors indicate, these estimates are only representative of those who attended antenatal care, and this paper includes studies from Africa going back at least two decades when antenatal care attendance was much lower.

Group of countries
One-third (n=16; 32%) of the systematic reviews did not explicitly report whether they performed a quality assessment of their primary studies. Even when a quality assessment was conducted, most studies did not use a standardized tool or did not report which tool they used or the results. Publication bias was assessed by only 15% of the systematic reviews.

| Frequency of maternal morbidity along the pregnancy-postpartum continuum
As shown in Table 1 and Figure 3, the toll of potentially life-threatening direct maternal morbidities is high, with postpartum hemorrhage being the most common, estimated at 6.2% based on the review by  (Table 1). Substantial direct maternal morbidity is also present throughout pregnancy with the prevalence of gestational diabetes mellitus estimated to be 5.1% in Africa 17 and 25.1% in the Western Pacific Region (Table 1, Fig. 1). 20 The frequency of indirect maternal morbidity is also high (see Table 2 and Fig. 4), particularly for mental health and infectious diseases. The prevalence of postpartum depression estimated for LMICs ranged from 1.0% to 26.3% according to Norhayati et al. 33 and from 4.9% to 50% according to Parsons et al. 34 In Africa, Sawyer et al. 35 estimated the prevalence of pregnancy-related depression at 18.3%.
Anxiety is another common health problem, with prevalence worldwide ranging between 4.4% and 39.0% 28 during pregnancy, and estimated to affect 8.5% of postpartum women on average. 28 The average prevalence of anxiety during pregnancy and the postpartum period in Africa has been estimated at 14% 35 (Table 2).
Regarding infectious diseases, 36 the estimated pooled HIV incidence rate in Sub-Saharan Africa is 4.7 per 100 person-years during pregnancy and 2.9 per 100 person-years during the postpartum period (Table 2). In Sub-Saharan Africa, based on one systematic review, 27 the reported prevalence of syphilis and chlamydia during pregnancy ranged between 2.5% and 2.9% and between 1.9% and 5.2%, respectively.
Estimates of these conditions across LMICs, as reported in another systematic review, range between 0.5% and 8.3% for syphilis and between 0.4% and 16.4% for chlamydia (Table 2). Across Sub-Saharan Africa, prevalence of malaria during pregnancy (peripheral parasitemia) ranges between 29.5% (in Eastern and Southern Africa) and 35.1% (in Western and Central Africa). 27 Estimates for hepatitis are high, with a median of 4.3% of pregnancies diagnosed with seroprevalence of hepatitis B serum antigen (HBsAg), and between 2.5% 37 and 3.0% 38 of pregnant women in Africa infected with hepatitis C (Table 2).
Many pregnancies are affected by non-life-threatening conditions.
Based on evidence predominately from HICs, nausea and vomiting have been reported to affect 69.4% of pregnant women 23 (Table 1). It was not clear whether the details for this matched the frequency estimate extracted.

Group of countries
c They define inclusion criteria for population as "studies on incontinence in population-based sample defined as from one or more district hospitals or from multiple clinics covering a defined geographic area." However, two countries contributing to the estimates were Turkey and Iran, for which hospital recruitment might not always be entirely appropriate.
Similarly, based on data from HICs, urinary incontinence has been variously estimated to affect 6.7% to 58.1% 4 or 26.0% to 75.0% 39 of women during pregnancy (Table 2). For some conditions, such as mental health disorders and infections, the timing of diagnosis may influence the frequency of the condition and thus explain differences in estimates between studies; this detail was not always reported. In Appendix S5, we summarize the case for postpartum depression, for which there are notable differences between the systematic reviews in how they summarized data from longitudinal studies reporting prevalence data for more than one time point.

| DISCUSSION
We conducted a systematic review of systematic reviews assessing the frequency of the 121 WHO maternal morbidities. 5 Women suffer F I G U R E 2 Quality assessment.  (1) that maternal morbidity is not a research priority; (2) that some conditions are challenging to define and measure; and (3) that others are very rare and hence unlikely to be covered in a systematic review.
Information on a wider range of maternal conditions and geographical areas should be gathered to produce better estimates.
The differences in prevalence reported for the same conditions, such as gestational diabetes and depression, may reflect actual differences between the populations and the widening inequalities between and within regions, 2 but they are also likely to be driven by methodological differences between the systematic reviews and the primary studies they included. Potential drivers include different assessment methods, varying definitions of the condition, and differences in the study populations and the timing of assessment. For example, the type of assessment method applied can double prevalence estimates for a condition such as gestational diabetes. 16 Generally, however, estimates for the same condition were relatively consistent; for example, estimates for postpartum hemorrhage varied between 6% and 11%, and obstetric fistula between 0% and 1.6% in LMICs.
Our results also highlight the existing gaps in the quality of methods and reporting used in systematic reviews on maternal conditions.
Crucially, for 56% of the direct and 30% of the indirect estimates, there was insufficient information to verify the population or data source.
Overall, 34% of the estimates extracted included facility-based studies. As discussed elsewhere, more reliable population-based estimates are needed, since mothers who access facilities are likely to be different to the ones who do not. 43  reported separately for studies that used population-based and facility-based data collection.

2.
Specify what assessment methods were used for each overall estimate presented. It is also good to report different summary estimates by diagnostic criteria. Try and avoid studies that include self-reported data except when this is an acceptable way of measuring the condition (e.g. nausea and vomiting). If self-reporting is included, discuss the primary studies assessing the validity of the self-report (sensitivity and specificity). Whether conditions arising during pregnancy should be quantified as incidence or prevalence heavily depends on the condition of interest, and the design and aims of a study. Yet many researchers use these terms interchangeably in the context of maternal morbidity; this is an issue that is beyond the scope of this study. However, we found that reviews of certain conditions for which incidence is of interest, such as postpartum depression, reported solely on prevalence. In systematic reviews, where several primary studies with a variety of designs are included, it can be difficult to choose the type of frequency to report. We call for future systematic reviews to clearly distinguish between incidence and prevalence estimates, to disaggregate these data, and to provide more discussion on this issue.
Our systematic review of systematic reviews is limited by the lack of grading based on diagnostic criteria. We chose not to perform such assessment because the primary studies in the included systematic reviews spanned across several conditions and decades, during which time the appropriateness of diagnostic criteria for different conditions changed. A further limitation is that we did not extract information directly from the primary studies identified by the systematic reviews-some systematic reviews included the same primary studies, and we did not always limit the time period for the publication of these primary studies-hence our reported frequencies represent a wide timescale. Overall, our review is limited by the quality of both the included systematic reviews and the primary studies they covered.
Finally, we only searched for systematic reviews rather than primary studies to assess the frequency of these conditions. We are aware of large-scale analyses of the frequency of important conditions such as anemia, 51 pregnancy-related infection, 52 and fistula, 53 which provide robust estimates for these conditions. We chose to focus, however, on systematic reviews that use standardized methods to aggregate existing data.
In conclusion, this review highlights both the existence of substantial maternal morbidity-spanning the time before and beyond childbirth-and major remaining gaps in the availability of systematic reviews for some maternal morbidities. Future systematic reviews should improve their quality standards, including the strict inclusion of population-based studies, and improvement of their review methods and their reporting, following available guidelines. With the changing burden of poor maternal health across the globe related to the obstetric transition, there is a pressing need to strengthen the evidence base for prioritizing action and further research. A central repository where results from new systematic reviews, using standardized terminology and metrics, can be stored and readily shared would be invaluable in tracking this shifting burden and in informing interventions to reduce the impact of maternal morbidities on women's lives.

AUTHOR CONTRIBUTIONS
GG, WJG, SW, and VF designed the research questions and methods.
AL, GG, CC, and SW conducted data extraction and analysis. GG prepared the manuscript. All authors (GG, AL, CC, WJG, SW, VF) provided feedback on the manuscript.

CONFLICTS OF INTEREST
The authors have no conflicts of interest to declare.