Introduction
Methods
Selection criteria
Type of study design
Type of outcomes
Type of population
Search strategy and process of study identification, selection and data extraction
Quality assessment
Study design (max = 12) | |
---|---|
Design | Cross-sectional vs longitudinal |
Sample size | Number of ecologic units included in the analysis as proportion of the total number of units, e.g. 100 countries of a total of 180 worldwide would be 55 %. |
Unbiased inclusion of units | Were the units included representative of the group for which inferences are being drawn? For example, for worldwide inferences, inclusion of only developed countries would be biased |
Level of data aggregation | Population to which the units refer to |
Level of inference | Use of the results of the analysis of the study’s sample data to draw inferences for individuals or groups (ecologic) |
Prespecification of ecologic units | Where the ecologic units selected to suit the hypothesis? (as opposed to selection motivated by convenience or necessity) |
Outcomes of interest included | Inclusion of all relevant outcomes (i.e. maternal and neonatal mortality and morbidity) or only of some outcomes. |
Source of data | Validity of the sources of data to represent the level that it refers to (e.g. the CS rate for one single hospital in one city would be an inadequate source of data to represent the national CS rate). |
Statistical methodology (max = 6) | |
Analytic methodology | All statistical methods are acceptable as long as they are used appropriately. A score was assigned based on the sophistication and flexibility of the method. |
Validity of regression | Did the adjustment have at least 10 units per covariate? |
Use of covariates | Did authors adjust analysis for desirable variables? Examples of socio-economic covariates: GDP or HDI. Examples of clinical covariates: proportion of women with diabetes or hypertensive disorders or obesity. |
Proper adjustment for covariates | Are the outcomes standardized or adjusted for certain factors before model adjustment? For standardized or adjusted outcomes, the standardized or adjusted factors should be included in the adjustment model. If standardized/adjusted outcomes are not used, this criterion is considered to have been met. |
Quality of reporting (max = 3) | |
Statement of study design | Did the authors present key elements of study design in the paper? |
Justification of study design | Did the authors justify the ecologic analysis, the rational and the specific objectives, including any prespecified hypotheses? |
Discussion of cross-level bias and limitations | Did the authors caution readers about the limitations of the ecologic design, the ecologic fallacy, the impossibility of extrapolating to a different level? |
Results
Study | Period, data sets and source | Outcomes | CS range | Design, Statistical method and adjustment factors | Quality scoring (max = 21) | Results and interpretation | Considerations for socio-economic factors |
---|---|---|---|---|---|---|---|
Althabe et al. 2006 [12] | 1991–2003 | ● Maternal mortality | 0.4–40 %) | Cross-sectional | 16 | The association between CS and MMR and NMR is different among countries. In medium- and high-income countries, there is no association between CS and MMR and NMR; in low-income countries, as CS rates increase, maternal and neonatal mortality decease. An arbitrarily selected 10 % CS rate threshold seems to have particular implications: a system with <10 % CS rate would be unlikely to cover the medical needs. | When adjusting for the considered factors (socio-economic), the observed association in low-income countries became non-significant for MMR. For NMR, the association remained but weakened. No adjustment was made for clinical factors. |
119 countries grouped as low-, medium- and high-income | (median 12.9 %) | Linear regression models | |||||
● Neonatal mortality (early) | Adjustment for: | ||||||
Main sources: DHS for developing countries, routine statistical surveillance systems or government reports for developed countries. | ● Gross National Income | ||||||
● Proportion of skilled birth attendant | |||||||
● Proportion of literate population | |||||||
Betrán et al. 2007 [2] | 1992–2003 | ● Maternal mortality | 0.4–40.5 % (weighted average 15 %) | Cross-sectional | 15 | In countries with high mortality, CS rate has a strong inverse association with MMR, NMR and IMR. This association weakens as mortality decreases. In low mortality countries the interpretation of the association is ambiguous. Data could support the suggestion that above a certain ceiling, higher CS rates may be associated with poorer outcomes. | No adjustment was made (neither for socio-economic or clinical factors). Authors acknowledged that most likely these factors are probably important confounders and that rising CS rates possibly mirrored a change in demographic or clinical risk profile in pregnant women. |
126 countries (89 % of global live births) | |||||||
LOWESS plots | |||||||
● Neonatal mortality | |||||||
Main sources: DHS for developing countries, routine statistical surveillance systems or government reports for developed countries. | |||||||
● Infant mortality | |||||||
Jurdi et al. 2004 [19] | 1995–2001 | ● Maternal mortality | 1.4–16 % | Cross-sectional | 15 | In this group of 18 countries there is a strong inverse association between CS and MMR and IMR. This is a heterogeneous group of countries with very diverse socio-economic and health indicators. Only 3 countries had CS rates above 15 % (Lebanon 15.1 %, Qatar 15.9 %, and Bahrain 16 %). | No adjustment was made (neither for socio-economic or clinical factors). But authors report, a significant positive association between CS and urban population, female literacy and Gross Domestic Product per capita. |
18 Arab countries | Spearman’s rank correlation (bivariate associations) | ||||||
Main sources: DHS or PAPCHILD surveys, UNFPA reports. | ● Infant mortality | ||||||
McClure et al. 2007 [20] | Not reported | ● Maternal mortality | Not available | Cross-sectional | 15 | In developing countries, as CS rates increased from 0 to 10–13 %, both MMR (0–10 %) and stillbirth (0–13 %) rates decreased sharply. Above 10 % CS rate, there was no significant association. In developed countries, no relationship was found. | Although this study stratifies by developed/developing countries, no further adjustment was attempted (neither for socio-economic or clinical factors). |
Piecewise regression models to explore if these relationships were consistent across the entire range of values; stepwise regression identified structural breaks in the regression lines. The sample was split at the breaks and least squares regression models were created for each of the sub-samples. Correlation and linear regression analyses were conducted. | |||||||
● Stillbirth | |||||||
188 countries grouped as developed (HDI > 0.80, n = 35) and developing n = 153) | |||||||
Main sources: World Health Report 2005 | |||||||
Silva et al. 2010 [21] | 1995–2007 (correlation for 2005) | ● Low birth weight | 22–54 %a | Cross-sectional | 15 | LBW rate was not correlated with CS rate. However, data suggested a non-linear trend: up to a CS rate = 30 %, LBW rates tended to decline as CS increased. For CS rates >30 %, LBW rates tended to increase with CS. Data support the hypothesis that increasing use of medical interventions in more developed settings may increase LBW rates. | No adjustment was made (neither for socio-economic or clinical factors). |
LOWESS regression and Spearman’s rank correlation (for testing) | |||||||
Brazil, 27 states | |||||||
Main sources: Government database | |||||||
Volpe et al. 2011 [13] | 2000–2009 | ● Maternal mortality | 0.4–41.9 % (median 13.8 %) | Cross-sectional | 15 | In countries with CS rates <15 %, higher CS rates were associated with lower MMR, NMR or IMR, and lower rates of LBW. In countries with CS rates >15 %, CS were not significantly associated with IMR or MMR (for MMR and CS, a marginally significant positive correlation was found). There was no evidence that CS > 15 % correlates to poorer, nor to better, maternal or child mortality rate outcomes. | No adjustment was made (neither for socio-economic or clinical factors). |
193 countries | |||||||
● Neonatal mortality | |||||||
Main sources: DHS for developing countries, routine statistical surveillance systems or government reports for developed countries. | ● Infant mortality | Non-linear exponential models were compared to quadratic models to regress IMR, NMR, MMR and LBWR rates to CS rate. The goodness-of-fit of models was compared using Akaike’s Information Criteria (AIC). | |||||
● Low birth weight | |||||||
Ye et al. 2014 [4] | 1980–2010 | ● Maternal mortality | CS range first year: 6.2–23 % | Longitudinal analysis | 18 | Most of the countries have experienced sharp increases in CS rates. Once CS rate reached 10 %, with adjustment for HDI and GDP, further increases in CS rate had no impact on MMR, NMR or IMR. Country-level CS rates above 10–15 % are hardly justified from the medical perspective. | Unadjusted analysis showed decline in mortality rates with increasing CS rates (up to 15 % for MMR and 20 % for NMR and IMR). After adjustment for HDI and GDP, the relationship disappeared and the curves become flat for CS rates above 10 %. The data points for CS rates <10 % were not sufficient to draw conclusions. No adjustment was made for clinical factors. |
19 developed countries | Two-level fractional | ||||||
Main sources: routine statistical surveillance systems or government reports. | ● Neonatal mortality | CS range last year: 14.3–32.2 % | polynomial model | ||||
Adjustment for: | |||||||
● Human Development Index (HDI) | |||||||
● Gross Domestic Product (GDP) | |||||||
● Infant mortality | |||||||
Zizza et al. 2011 [14] | 1994–2008 | ● Maternal mortality | 0.4–42.3 % | Cross-sectional | 15 | The analysis showed an inverse association between CS rates and MMR, and NMR for all geographical areas except for Europe. The piecewise regression provided the breakpoint beyond which an increased CS rate does not reflect an improvement in health care. The CS values for this breakpoint for NMR and MMR are 16 % and 9 %, respectively. For NMR, after 16 % there is a trend reversal; for MMR, after 9 %, it reaches a plateau. | No adjustment was made (neither for socio-economic or clinical factors). |
142 countries | ● Neonatal mortality | (weighted average 14.8 %) | Analysis of covariance (Ancova) and piecewise regressions | ||||
Main sources: DHS for developing countries, routine statistical surveillance systems or government reports for developed countries. |