Figure 2 shows a harvest plot demonstrating significance and direction of interaction by outcomes, country, age group, sample size and study quality. Overall, 22 estimates were included from the 18 studies, as some studies included multiple measures and multiple age groups. Cognitive outcomes were the most examined outcome, and the youngest age groups were the most frequently studied (eight preschool and seven primary school). One study was in the harvest plot as it covered multiple age groups.45 Outcomes that do not fit into the categories have not been included in this plot. Table 3 shows the outcome, sample size and effects for each study.

Interaction for each outcome by age group, sample size, quality score, effect of socioeconomic status (SES) and effect of preterm birth or gestational age (GA) for included studies

Eleven studies examined cognitive outcomes, with 15 interaction estimates (one study used three age groups). The first study by Beauregard et al33 used evidence from a UK birth cohort study to estimate the effect of household poverty (income less than 60% of the median) and GA at birth categories (early/moderate PTB (24–33 weeks), late preterm (34–36), early term (37–38) and term (39–40)) on various educational outcomes at ages 3 (school readiness and naming vocabulary), 5 (naming vocabulary, picture similarity and pattern construction), and 7 years (word reading, pattern construction and number skills). Household poverty was significantly associated with lower scores for all outcomes in all age groups in interaction models. PTB categories were significantly associated with lower scores for school readiness (not early term) and naming vocabulary (only early/moderate PTB) at age 3, pattern construction at ages 5 and 7, and number skills at age 7 (for early/moderate PTB only). The interaction between early/moderate PTB and poverty for word reading at age 7 was superadditive. The interaction between early term and poverty for pattern construction at age 7 had a subadditive effect. No other significant interactions were found, suggesting no overall clear pattern was noted; the effect of the two exposures was not modified by the other.

The second study by Beauregard et al34 used evidence from a USA population-based birth cohort study to estimate the effect of a composite measure of SES (parental education, occupation and income) and lower GA on cognitive ability at age 2, and reading and mathematics at age 6 (Kindergarten). Both decreasing SES and earlier GA groups had significant proportional negative effects on cognitive ability (at age 2) and both reading and mathematics assessment. No significant interactions between GA and SES were found, and direction of estimated interaction effect was not noted.

Mallinson et al35 used data on all births in Wisconsin to estimate the effects of maternal Medicaid coverage (government-provided health insurance), maternal education and GA (as a continuous variable) on standardised testing and passing literacy benchmark at kindergarten age. Lower maternal education and Medicaid coverage were both significantly associated with lower scores and not passing benchmark as was decreasing GA. Interaction between Medicaid or education and GA was significant (interaction scale not stated) and suggested lower SES exacerbated effects of lower GA on outcomes when harmonised.

Richards et al36 used data from a retrospective birth cohort in Georgia, USA to estimate the effects of neighbourhood deprivation and PTB on mathematics failure during first grade. PTB and increasing deprivation were both associated with significantly increased risk of failure. Significant superadditive and submultiplicative interactions were seen, while the relative excess risk due to interaction (RERI) was not significant. This suggests two potential scenarios, exposure to both PTB and low SES exacerbate the negative effects of each and the RERI calculation was underpowered, or the superadditive interaction is a spurious finding.

Ekeus et al37 used Swedish register data to estimate the effects of a composite measure of SES and PTB on intelligence performance in 18-year-old to 19-year-old young men at conscription. Lower SES and PTB were both associated with significantly lower performance, while a significant supermultiplicative interaction was found for moderate PTB (33–36 weeks) only.

Lindström et al38 used Swedish register data to estimate the effects of household occupation and PTB on postsecondary education in 23–29 years old. Lower GA groups had significantly lower rates of postsecondary education and a significant positive interaction (type not stated) with household occupation was found; low SES exacerbated the effects of PTB.

Bilsteen et al41 used Danish register data to estimate the effects of maternal education and GA on educational attainment at age 28. Lower GA was associated with significantly lower odds of tertiary or secondary education and no significant interaction was seen. The effect of maternal education was not stated.

Gisselmann et al40 used Swedish register data to estimate the effects of parental education and PTB on school performance at ages 15–16. Both lower education and PTB were associated with lower achievement, and significant interaction demonstrated that low SES exacerbated the effect of PTB (interaction scale not stated).

ElHassan et al44 used data on births in Arkansas, USA, to estimate the effect of low SES (using maternal education and maternal insurance coverage measured at birth) and GA groups on maths and literacy attainment between the third and eighth grades (8–14 years old). Both low SES and extremely low GA group were associated with significantly lower scores in both subjects, but no significant multiplicative interaction was seen (the effects of SES are the same in all GA groups, they are multiplicative).

Doyle et al48 conducted an Australian study using data on consecutive extremely preterm or extremely low birth weight births born during the study period who survived, that looked at the effects of education or social class and PTB on reading, spelling and maths at ages 8 and 18. Lower GA and low SES were associated with worse scores at age 8, and only low SES at age 18. No significant interaction was found.

Peacock et al49 used data from a UK-based cohort study to estimate the effect of social class and GA on Key Stage 1 (KS1) scores (ages 5–7). Children born preterm were less likely to attain KS1 and specific areas (reading, writing and mathematics). No significant interaction was found.

Four studies examined the interaction between SES and GA on developmental outcomes. Potijk et al42 used a prospective cohort from the Netherlands to estimate the effects of a composite measure of SES and GA on developmental delay at age 4. Decreasing SES and GA were both significant for increasing/worsening communication delay, and submultiplicative interaction was seen. This means that the relative effect of having both low SES and PTB on communication delay is lower than if the risks of both were multiplied. For other developmental outcomes (total score, fine motor, gross motor, problem-solving, personal–social), interaction was multiplicative (the effect of both risks is the same as each single risk combined). Notably, none of these outcomes had both risks significant independently. Shown on figure 2 as no clear effect.

Ene et al46 used a record-based cohort to examine the effects of area-based deprivation and PTB on speech and language concerns at age 2.5. Lower GA was associated with higher levels of concern, while SES was not associated. No significant additive interaction was found.

de Laat et al47 used data from a Dutch population-based cohort to estimate effects of PTB and two measures of SES (maternal education and perceived income inadequacy) on total difficulties at ages 5–6 assessed by mothers and teachers. Difficulties were more common in preterm children on maternal assessment. No significant multiplicative interaction was found.

Dall’oglio et al50 used data from a very small Italian cohort study to examine the effects PTB categories and parental education on Griffiths Mental Developmental Scales at age 4. Both PTB and low SES were significantly associated with worse scores. No significant interaction was found.

Three studies examined interaction between GA and SES on mental health outcomes. Lindström et al39 used Swedish register data to estimate the effects on hospital admissions for a psychiatric diagnosis by age 23–29. The effect of SES was not stated; however, lower GA was significantly associated with higher odds of admission. A significant interaction was found; lower SES exacerbated the GA impacts (scale for interaction not stated).

Potijk et al43 use a Dutch population-based cohort to estimate the effects on total behavioural problems, internalising problems and externalising problems. Low SES and PTB were both associated with more total and internalising problems, while externalising problems were only associated with PTB. No significant interaction was found; all combinations were multiplicative.

Lindström et al45 used Swedish register data to estimate the effects of SES and PTB category on ADHD medication prescription from age 6 through to 19. A significant interaction was found; the effect of moderate PTB was exacerbated by low SES.

Lindström et al38 estimated the effects of household occupation and PTB on disability and unemployment in 23–29 years old. No significant interaction was found. Bilsteen et al41 estimated the effects of maternal education and GA on income tertile and source of income. Lower GA was significantly associated with lower odds of being in the highest income tertile, and higher odds of getting income from cash benefits and disability pension. There was no significant interaction found.

We split the sample sizes into three broad groups; less than 10 000 (eight studies), between 10 000 and 30 000 (four studies), and over 100 000 (seven studies). One study provided estimates for two sample size categories.33 Of the studies that found a significant interaction between SES and GA, one had a sample size in the less than 10 000 range (for developmental outcomes),42 one was between 10 000 and 30 000 (for cognitive outcomes)40 and the other six studies were all over 100 000.35–39 45 Two studies with a sample over 100 000 found no significant interaction for assorted outcomes post school.38 41 The study with less than 10 000 participants only found a significant interaction for one out of five outcomes.42

The most frequently adjusted confounder was sex of the child (15/18). Other variables included child age at outcome or year of birth (9), number of siblings or maternal parity (8), maternal age at birth (7), ethnicity or race (6), and number of parents or parental/maternal marital status (6). Birth weight, multiple gestation, Apagar score (measured at birth), smoking during pregnancy and gestational diabetes were sporadically included.

Baranowska-Rataj et al32 used Swedish register data (1 087 750) to determine the effect of PTB and parental education/occupation/income quintile separately on grades in the final year of compulsory schooling, compared with non-preterm siblings. Worse grade outcomes were observed for very and extremely preterm groups. When combining the two exposures (dummy variables which combine PTB category and SES), there were no clear differences in the effects.

Kroll et al30 used routine UK data to estimate the effects of area-based deprivation and GA groups on infant mortality. For children born at less than 32 weeks gestation, there were no socioeconomic inequalities in all-cause mortality. For those born between 32 and 37 weeks, low SES was associated with higher all-cause mortality.

Ray et al31 used a Canadian population-based cohort study to estimate the effects on neonatal mortality. PTB significantly increased risk and stratification by income quintile and did not change the pattern.

Searches were rapidly updated, and single screened, in Scopus and advanced Google up until July 2023. A small UK study found that GA and area-based SES had no interaction but GA and family-based measures of SES had some significant interactive effects on brain structure measures.51 A large Danish register-based study found a superadditive interaction between GA and SES on the risk of not completing schooling at 15–16 years old.52 A small study from USA found no significant interaction between PTB and social adversity on executive function.53 These studies are in keeping with our current findings.