Background
Methods
Search strategy
Include | Exclude | |
Population | Pregnant women | |
Intervention / mediator | Behavioural risk factors (e.g. smoking, alcohol). Social risk factors (e.g. Environmental (housing, working)). Maternal health status (both mental and physical health) | Genetic risks for preterm birth |
Comparison across exposure | Comparison across socioeconomic strata (either individual or area-based) | |
Outcomes | Preterm birth and gestational age | Other birth outcomes (e.g. low birthweight) |
Publication characteristics: Inclusion / exclusion criteria | ||
Include | Exclude | |
Publication types | Primary studies from peer-reviewed literature, including those from reviews. Relevant secondary analyses (meta-analysis). Papers published or in-press. Working papers | Not primary research, e.g. letters, editorials, commentaries, conference proceedings, books and book chapters, meeting abstracts, lectures, and addresses. Previous reviews and meta-analyses, but relevant reviews were used to identify relevant primary studies |
Types of study | Analytical techniques that are relevant to research question: --Mediation --Attenuation Differential exposure | Other methods. Mediation or attenuation not specifically calculated within analysis |
Year of publication | 2000–2020 | |
Language | English language |
Selection
Data extraction
Effect Measure | Description |
---|---|
Total Effect | The overall effect of the exposure on an outcome: --For the difference method, this is the regression output for the exposure when not adjusted for the mediator. --For product of coefficients, this is the sum of direct and indirect effect |
Direct Effect | The effect of the exposure on an outcome when the intermediate variable is removed |
Indirect Effect | The effect of the exposure on an outcome through an intermediate variable |
Proportion Eliminated | How much of the total effect would be removed through action on the intermediate variable (setting the mediator to the same level for all pregnant women) [26]: --For the difference method, this is the difference between the total effect and regression output for the mediator-adjusted regression, divided by total effect (minus one if using exponentiated outputs) --For product of coefficients, this is the indirect effect divided by total effect [14] |
Quality-scoring
Integration
Results
Search results and description of included studies
Paper | Design | Country | Sample Size and Characteristics | Study Period | Mediation Analysis Approach | Measure of SES | Quality Score (/13) |
---|---|---|---|---|---|---|---|
Poulsen et al. (2019) [33] | Cohort | Denmark | 77,020 – National birth cohort (whole) | NS | Difference method using risk differences from linear regression | Maternal education: Short (≤ lower secondary) to long (degree; reference) | 10 |
Netherlands | 4,508 – Rotterdam birth cohort (whole) | NS | |||||
Norway | 78,267 – National birth cohort (whole) | NS | |||||
Ross et al. (2019) [34] | Cohort | United States (US) | 718,952 –Californian birth cohort (whole) | 2007–2012 | Product of coefficients/ Path analysis using Lavaan Package | Maternal education: At most high-school to more than high school (reference) | 9 |
Dolatian et al. (2014) [35] | Cohort | Iran | 500 – Random sample of pregnant women from stratified sample of four Tehran hospitals | 2011–2012 | Product of coefficients/ Path analysis using Lisrel Software | Income | 9 |
Clayborne et al. (2017) [36] | Cohort | Canada | 2,068 – Sample of pregnant women from Calgary and Edmonton Metropolitan Regions | 2008–2012 | Product of coefficients using PROCESS macro | Neighbourhood SES | 8 |
Dooley (2009) [37] [PhD thesis] | Cross-sectional | US | 28,793 – Hamilton County, Ohio, birth cohort (whole) | 2001–2003 | Product of coefficients/ Path analysis of multilevel modelling using Mplus | Neighbourhood concentrated disadvantage | 8 |
Mehra et al. (2019) [38] | Cohort | US | 138,494 – National convenience sample (retrospective) of births from all states using health insurance data | 2011 | Product of coefficients/ Path analysis of multilevel modelling using Mplus | Neighbourhood SES: most deprived quarter to least deprived (reference) | 8 |
Meng et al. (2013) [39] | Cross-sectional | Canada | 90,500—All births (including multiple) at three Ontario province public health units | 2000–2008 | Product of coefficients of multilevel modelling using both linear and logistic regression | Neighbourhood SES | 8 |
Mirabzadeh et al. (2013) [40] | Cohort | Iran | 500 – Random sample of pregnant women from stratified sample of four Tehran hospitals | 2012–2013 | Product of coefficients/ Path analysis using Lisrel Software | Composite comprising: maternal and spousal education, persons and cost/household area, car, computer | 8 |
Misra et al. (2001) a[41] | Cross-sectional | US | 735 – Urban university hospital sample of births to black mothers: drug users, women without prenatal care, and a systematic sample of the rest | 1995–1996 | Difference method using logistic regression | Lack of time and money | 8 |
Nkansah-Amankra et al. (2010) [42] | Cross-sectional | US | 8,064 – South Carolina state, stratified systematic sample of births | 2000–2003 | Difference method using multilevel logistic modelling | Neighbourhood SES: Proportion of residents in poverty | 8 |
Räisänen et al. (2013) [43] | Cross-sectionalb | Finland | 1,390,742 – National birth cohort (whole) | 1987–2010 | Difference method using logistic regression | Maternal occupation; blue collar relative to upper white collar (reference) | 8 |
Ahern et al. (2003) [44] | Case–Control | US | 1,496 cases + controls – A San Francisco hospital based sample of births: All preterm plus random selections of full-term, stratified by African American and White | 1980–1990 | Difference method using multilevel logistic modelling | Neighbourhood context | 7 |
Amegah et al. (2013) [45] | Cross-sectional | Ghana | 559 – Cape Coast’s four main healthcare facilities, random sample weighted by hospital or urban centre | 2011 | Difference method: Generalised linear model using Poisson Distribution and log link | Level of monthly income: low to upper middle and high (reference) | 7 |
van den Berg et al. (2012) [46] | Cohort | Netherlands | 3,821 – Amsterdam birth cohort (Dutch-only) (whole) | 2003–2004 | Difference method using logistic regression | Maternal education: years of education after primary school, low (< 6) to high (> 10; reference) | 7 |
Morgen et al. (2008) [47] | Cohort | Denmark | 38,131 primiparous & 37,849 multiparous – National birth cohort | 1996–2002 | Difference method using Cox regression | Maternal education; < 10 years to > 12 years (reference) | 7 |
Gisselmann and Hemström (2008) [48] | Cross-sectional | Sweden | 356,887 – National birth cohort (whole) | 1980–1985 | Difference method using logistic regression | Maternal occupation: Unskilled manufacturing manuals to middle non-manuals (reference) | 7 |
Niedhammer et al. (2012) [49] | Cohort | Republic of Ireland | 913 – Random sample of pregnant women (Irish-only) from two hospitals (urban and rural) | 2001–2003 | Difference method using Cox Regression | Maternal education: lower than to higher than secondary (reference) | 7 |
Jansen et al. (2009) [50] | Cohort | Netherlands | 3,830 – Rotterdam birth cohort (whole) | 2002–2006 | Difference method using logistic regression | Maternal education: low (< 4 years general secondary) to high (Master degree, PhD; reference) | 7 |
Quispel et al. (2014) a[51] | Cohort | Netherlands | 1,013 – Rotterdam, Apeldoorn, Breda: Random samples of pregnant women from primary, secondary, tertiary care | 2009–2011 | Difference method using logistic regression | Maternal education: low to moderate (reference) | 6 |
Gissler et al. (2003) [52] | Cross-sectional | Finland | 548,913 – National birth cohort (whole) | 1991–1999 | Difference method using logistic regression | Maternal occupation: blue collar to upper white collar (reference) | 6 |
Gray et al. (2008) [53] | Cohort | Scotland | 400,752 – National (hospital) birth cohort (whole) | 1994–2003 | Difference method using logistic regression | Neighbourhood SES: most deprived fifth to least deprived (area-based) (reference) | 6 |
de Oliveira et al. (2019) [54] | Case–Control | Brazil | 296 cases + 329 controls – Londrina sample of hospital births (including multiple) | 2006–2007 | Structural equation modelling | Socioeconomic vulnerability | 4 |
Quality assessment
Mediation approach
Association of SES and preterm birth
Paper | Effect of SES on PTB (95% confidence interval if available) | Mediator | Prevalence of mediator in sample | Proportion eliminated (95% confidence interval if available) |
---|---|---|---|---|
Poulsen et al. (2019) [33] Denmark | Total effect RD: 2.0 (1.4, 2.5) excess PTB/100 singleton deliveries | Smoking | 17% total; 39% short education, 8% long | 22% (11%, 31%)a |
Poulsen et al. (2019) [33] Netherlands | Total effect RD: 3.2 (0.8, 5.2) | 19% total; 41% short education, 7% long | 10% (-22%, 29%) | |
Poulsen et al. (2019) [33] Norway | Total effect RD: 2.0 (0.9, 3.0) | 9% total; 34% short education, 4% long | 19% (-1%, 29%)a | |
Ross et al. (2019) [34] | Direct coefficient: 0.072* Total effect coefficient 0.077 | Pre-eclampsia | 5% in black women, 3% in white women | 6.5%a |
Dolatian et al. (2014) [35] | Direct coefficient: 0.06* Total effect coefficient: 0.06126* | Perceived stress | Mean | 11.8%a |
Perceived social support through stress | Mean | Mediated effect in opposite directiona | ||
Combined | 2.1%a | |||
Clayborne et al. (2017) [36] | Total effect OR: 0.91 (0.64, 1.31) | Pre-pregnancy body mass index (BMI) | Mean | Cannot be estimated |
Gestational weight gain | Mean | Cannot be estimated | ||
Combined | Cannot be estimateda | |||
Dooley (2009) [37] (PhD thesis) | Direct effect: 43.29% increase in odds/standard deviation increase. Total effect**: 46.01%* | Medical risk | 13% | 2.9%a |
Smoking | 13% | 3.0%a | ||
Perceived neighbourhood support | Mean | No indirect effect | ||
Mehra et al. (2019) [38] | Direct effect coefficient: 0.036. Total effect coefficient: 0.059* | Hypertension | 10% | 22.0%a |
Infection | 28% | 16.9%a | ||
Meng et al. (2013) [39] | Total effect coefficient: 0.981 (0.626–1.337) | SES-related support | Composite measure | 11.7%a |
Psychosocial | Composite measure | 2.1%a | ||
Behavioural | Composite measure | 5.5%a | ||
Health | Composite measure | 6.4%a | ||
Mirabzadeh et al. (2013) [40] | Total effect coefficient: 0.1441a | Perceived social support through stress | Mean | 8.1%a |
Stress, depression, and anxiety | Mean | 22.5%a | ||
Combined | 30.6%a | |||
Misra et al. (2001) [41] | Total effect OR: 2.85 (1.85–4.40) | Psychosocial factors only | 26% severe stress | 44% |
Biomedical and psychosocial factors | 5% chronic disease | 64% | ||
Nkansah-Amankra et al. (2010) [42] | Total effect OR 1.34 (0.80–2.25) | Maternal stress (emotional, financial, spousal-related, traumatic) | 14% low poverty, 57% high poverty | No significant total effect |
Räisänen et al. (2013) [43] | Total effect OR: Extremely PTB 1.61 (1.38–1.89); Very PTB 1.48 (1.31–1.68); Moderately PTB 1.27 (1.22–1.32) | Smoking | 12% to 18% by gestational age category | 26% for extremely PTB 33% for very PTB 30% for moderately PTB |
Other factors and smoking | Composite measure | 39% for extremely PTB 50% for very PTB 41% for moderately PTB | ||
Ahern et al. (2003) [44] African-American | Total effect parameter estimate proportion unemployed: 44.4* | Cigarettes per day | Mean | 3% |
Ahern et al. (2003) [44] White | Total effect parameter estimate change in unemployed: -3.32 | No significant total effect | ||
Amegah et al. (2013) [45] | Total effect RR: 1.83 (1.31–2.56) | Malaria infection during pregnancy | 48% | No effect |
Pre-pregnancy BMI | 33% healthy weight | 17% | ||
Cooking fuel used | 18% LPG, 24% charcoal, 5% firewood | 22% | ||
Combined | 30% | |||
van den Berg et al. (2012) [46] | Total effect OR: 1.95 (1.19–3.20) | Smoking | 7% total, 33% in low educated, 2% in high educated | 43% |
Smoking and environmental tobacco exposure | 6% total, 27% in low educated, 1% in high educated | 39% | ||
Morgen et al. (2008) [47] | HR primiparous: 1.22 (1.04–1.42) HR multiparous: 1.56 (1.31–1.87) | Smoking | 26% to 35% by gestational age category | 5% in primiparous 23% in multiparous |
Alcohol | 40% to 45% by gestational age category | 5% in primiparous 4% in multiparous | ||
Binge drinking | 25% to 26% by gestational age category | 5% in primiparous no effect in multiparous | ||
Pre-pregnancy BMI | Mean | 9% in primiparous 2% in multiparous | ||
Gestational weight gain | Mean | 5% in primiparous 4% in multiparous | ||
Combined | 23% in primiparous 30% in multiparous | |||
Gisselmann and Hemström (2008) [48] | Total effect OR: 1.41* | Job control | Not stated | 44% |
Job hazards | Not stated | 5% | ||
Physical demands | Not stated | 22% | ||
All working conditions | Not stated | 46% | ||
Niedhammer et al. (2012) [49] | Total effect HR: 2.14 (1.05–4.38) | Rented home | 43% lower than secondary, 15% higher than secondary | 26% |
Crowded home | 18% lower than secondary, 5% higher than secondary | 13% | ||
Material factors | Composite | 33% | ||
Smoking | 46% lower than secondary, 16% higher than secondary | 2% | ||
Alcohol | 50% lower than secondary, 62% higher than secondary | 14% | ||
Behavioural | Composite | 10% | ||
Saturated fatty acids (nutritional factors) | 31% lower than secondary, 20% higher than secondary | 14% | ||
Material + behavioural | Composite Measure | 38% | ||
Material + behavioural + nutritional | Composite Measure | 42% | ||
Jansen et al. (2009) [50] | Total effect OR: 1.89 (1.28–2.80) | Mother’s age | Mean | 22% |
Mothers’ height | Mean | 22% | ||
Preeclampsia | 2% total, 1% high, 4% low education | 13% | ||
Intrauterine growth restriction (IUGR) | 1% total, 1% high, 2% low education | 12% | ||
Marital status (single) | 8% total, 3% high, 20% low education | 2% | ||
Pregnancy planning (unplanned) | 19% total, 10% high, 34% low education | No effect | ||
Financial concerns | 12% total, 5% high, 30% low education | 19% | ||
Long-lasting difficulties | Mean | 11% | ||
Psychopathology | Mean | 16% | ||
Working hours | Mean | No effect | ||
Smoking | 18% total, 5% high, 45% low education | 8% | ||
Alcohol consumption | 50% total, 68% high, 25% low education | 17% | ||
BMI | 67% total healthy weight, 75% high, 51% low education | 7% | ||
All except preeclampsia/IUGR/ working hours/pregnancy planning | Composite Measure | 69% | ||
All except working hours/pregnancy planning | Composite Measure | 89% | ||
Quispel et al. (2014) [51] | Total effect OR: 1.06 (1.02–1.10) | Depression score | 15% | No effect |
Gissler et al. (2003) [52] | Total effect OR: 1.35 (1.25–1.45) | Smoking | 15% total, 5% upper white collar, 26% blue collar workers | 42% |
Gray et al. (2008) [53] | Total effect OR: 1.49 (1.43–1.54) | Smoking | For 2 periods: 30% & 29% total, 15% for both periods in least deprived, 43% & 39% in most deprived | 45% |
de Oliveira et al. (2019) [54] | Direct standardised estimate: -0.083 | Inadequate prenatal care | Not stated | Cannot be estimateda |
Unwanted pregnancy | Cannot be estimateda |