Table 3 Assessment of bias in studies.
From: Socioeconomic disparities in the postnatal growth of preterm infants: a systematic review
| Â | Selection bias | Confounding bias | Missing data, attrition bias | Detection bias |
|---|---|---|---|---|
Excluded “socially vulnerable” premature infants, including those born to unmarried mother 101. | Infants of unemployed mothers were significantly heavier and longer at birth than infants of employed mothers (179 g and 1.1 cm difference between means). Birthweight and birth length were not adjusted for when analysing the effect of maternal employment status on growth variation. Other variables also shown to be associated with growth variation (e.g. problem at birth, problems at discharge and sex) were also not adjusted for in this analysis. | The number of infants for whom anthropometric measurements were taken at 6 months corrected age is not reported. When direct measurement was not possible at any time point, corresponding growth data were collected from the records. | Weight, height and head circumference measurements were taken at the same time of day according to standard guidelines. Measurements were taken twice to ensure consistency. | |
Risk of bias (Ahn) 16: | High | High | Unknown | Low |
Bocca-Tjeertes 17 | • Children were recruited from 13 community healthcare centres, covering approximately 25% of all children in the Netherlands, at the 4 year well-child visit attended by 97% of children. • Mothers in the non-response group were more likely to be of non-Dutch ethnicity and have a lower level of education. | Differences in baseline characteristics between infants of mothers with different levels of education, or infants in families with different levels of income, were not reported. Maternal height and small for gestational age status were significantly associated with head circumference growth restraint at 1 year, but not adjusted for in the multivariable logistic regression analysis of the association between maternal education and head circumference growth restraint. | 20% of measured growth data was missing for children at each time point. | The fourth Dutch growth survey was used as a comparator group to determine growth restraint. Children with birthweight less than 2500 g, those with non-Dutch parents, those with diagnosed growth disorders or on medications known to affect growth were excluded in this survey 102. |
Risk of bias (Bocca-Tjeertes) 17 | High | Moderate | High | Low-Moderate |
Sammy 18 | All infants discharged from the newborn unit between April and June 2014 appear to have been screened. Only 3 out of 115 infants screened were not enrolled in the study. | Differences in baseline characteristics between infants of mothers with different levels of education were not reported. Gestational age, shown to be associated with the early growth of preterm infants, is not adjusted for in the analysis of the association between maternal education level and growth. | 6 of 112 infants (5%) did not have their growth assessed at 2 weeks post discharge from the neonatal unit – there were 2 neonatal deaths, and 4 infants were lost to follow-up. | Infants were assessed at 2 weeks post discharge from the neonatal unit for the presence or absence of growth restraint. The timing of follow-up was not standardised. The outcomes “optimal growth” and “growth deficit” are not defined. |
Risk of bias (Sammy) 18 | Low | High | Low | High |
Teranishi 19 | Data from nationwide a cohort study of white, singleton children born in 1958 with perinatal and 23 year data available were used. | Differences in baseline characteristics between preterm infants in social classes I & II, and IV and V, were not reported. The associations of other variables with height deficit – the growth outcome reported by parental socioeconomic status – are not investigated. | Height data were available for: • 83% of subjects at 7 years. • 84% of subjects at 11 years. • 75% of subjects at 16 years. • 95% of subjects at 23 years. • The number and characteristics of preterm-born subjects for whom height data were available at each time point is unknown. | Heights were measured at ages 7, 11 and 16. Heights were self-reported at age 23. |
Risk of bias (Teranishi) 19 | Moderate | Unknown | Unknown | High (due to self-reported height at age 23). |
Holmqvist 20 | Only “low risk” infants without any maternal or foetal complications were included. The characteristics of infants lost to follow-up from the original cohort of 61 preterm infants103 are not reported. | Infants of well-educated mothers had a higher gestational age than those of less well-educated mothers (34 versus 33 weeks); no statistical comparison is reported. No confounding factors were adjusted for in analyses of the association between maternal education level and growth outcomes, but the presence of a neurodevelopmental disorder was associated with weight growth increments from 0 – 3 months and 3 – 7 months. | The number of infants for whom growth data was available at each time point, and their characteristics, are not reported. | The use of chronological versus corrected age to determine timing of follow-up is not specified. Age ranges of infants at follow-up appointments are not reported. Student’s t tests were used to compare weekly growth within discrete time periods instead of statistical methods which allow comparison of growth trends over the entire study period. |
Risk of bias (Holmqvist 20) | High | High | Unknown | High |
Ghods 21 | 65 of 238 eligible infants (27%) were excluded due to loss to follow-up (11%), incomplete medical records (14%), and inability to intrapolate the head circumference at 1 year (3%). It is unclear whether infants excluded due to loss to follow-up were lost before 1 year or up to 66 months corrected age. | Differences in baseline characteristics between infants in different socioeconomic strata were not reported. Other factors associated with head circumference (HC) catch up (e.g. birthweight z-score, birth length z-score, birth HC z-score, gestational age, small for gestational age status, the number of hospital days, neonatal comorbidities, breastfeeding characteristics) were not adjusted for in the analysis of the association between maternal education and HC catch up. | HC measurements for 9 infants (5%) were interpolated from adjoining data. The characteristics of these infants are not reported. | The presence of HC catch up is determined by the measurement at 1 year corrected age, but the age range of infants at the 1 year follow-up is not reported. Figure 1 of the study shows that infants’ head circumference z-score varies notably over time. The method used to determine families’ Home Facilities and Financial Situation was interviewer-dependent. The language used during the psychologist interview to measure parental socioeconomic status is not specified; over 40% of infants were from immigrant families. |
Risk of bias (Ghods 21) | Moderate | High | Low | High |
Ni 22 | The cohort was recruited from a birth cohort study, and participants without data at 19 years (the last follow-up) were excluded. Participants with data at 19 years were more likely to have mothers with higher education and parents with higher occupational socioeconomic status compared to the original cohort. | Differences in baseline characteristics between infants of mothers with different educational attainment, or between infants of parents with different occupational socioeconomic status were not reported. Change in weight z-score from 2.5 to 6 years, shown in a regression analysis to be significantly associated with BMI at 19 years, was not adjusted for when reporting BMI by parental socioeconomic status. | There was no missing growth data at 19 years as this was specified as an exclusion criterion. | Only BMI, a calculated product of weight and height, is presented according to parental socioeconomic status. No other growth outcomes are reported. |
Risk of bias (Ni 22) | High | High | Low | High |
Liang 25 | Inclusion criteria detailed that parents should have had “basic reading and comprehension skills”. The exclusion criteria excluded infants with severe illness, e.g. those with major comorbidities, infants who received invasive respiratory therapy, and infants who required surgery. | Differences in baseline characteristics according to primary caregiver education, and monthly household income, were not reported. No other variables shown to be associated with growth at 12 months were adjusted for when growth outcomes were reported according to monthly household income. The multiple regression analyses of growth outcomes at 12 months included other covariates significant in univariate analyses in addition to monthly household income. Ambiguity relating to covariates included in regression analyses (e.g. lack of reference time point for “feeding mode” variable) entails a risk of residual confounding. | Table 2 in the study appears to indicate that all 115 infants in the intervention group, and all 100 infants in the control group were followed-up at all time points until 18 months of age. However, it is not specified whether lack of attendance at a follow-up visit was a criterion for exclusion. The authors report a “20% loss of follow-up rate” in the Methods – it is unclear whether this was an estimated or actual rate. | The time range for which measurements were taken at the 12 month time point is not reported. It is not specified whether the time points refer to corrected or postnatal ages. |
Risk of bias (Liang 25) | High | High for raw growth outcomes reported according to monthly household income. Moderate for multiple regression analyses. | Unknown | Low-Moderate |
Fu 24 | The characteristics of those excluded due to lack of data for assessment of childhood overweight/obesity, or missing data on study variables, is not reported. As the sample were recruited from cohort enrolled at birth, lack of attendance at the 4 – 7 year follow-up visit is likely to be associated with greater socioeconomic deprivation 35,36. | Differences in baseline characteristics according to maternal occupation and maternal education status are not reported. The categorisation of the maternal occupation variable (which includes the category “Others”) entails a risk of residual confounding. Other variables shown to be significantly associated with childhood overweight/obesity in univariate analyses (e.g. birthweight, birth length, formula-feeding, trajectory of BMI z-score during the first year of corrected age) were not adjusted for when reporting the incidence of childhood overweight/obesity by parental socioeconomic status. | There was no attrition as lack of data for the detection of childhood overweight/obesity at 4–7 years was specified as an exclusion criterion. | No other growth outcomes apart from childhood overweight/obesity are reported according to parental socioeconomic status. |
Risk of bias (Fu 24) | Moderate | High | Low | High |
Sices 26 | The inclusion and exclusion criteria are reasonable to allow for the measurement of the specified growth outcomes, however, infants without at least two out of three consecutive growth measurements were excluded. The characteristics of the excluded infants are not reported. | Differences in baseline characteristics according to maternal education are not reported. Other variables shown to be associated with growth failure (e.g. chronic lung disease, cerebral palsy) are not accounted for when reporting the prevalence of growth failure during the different time periods by maternal education level. | All infants were assessed at 40 weeks, 93% of infants were seen at 4 months, 90% were seen at 8 months and 98% were seen at 20 months. 82% of infants attended all 3 visits after 40 weeks. The characteristics of those missing data at follow-ups is not reported. | Growth outcomes relating to length and head circumference are not reported. The comparisons between infants with growth failure in any study period and no growth failure is not reported. For the 5 infants (3%) with growth failure in more than one time period, only the first instance of growth failure is considered in the analysis; the parental socioeconomic status of these infants is not reported. |
Risk of bias (Sices 26) | Moderate | High | Low-Moderate | Low-Moderate |
Peterson 27 | Loss to follow-up in the prospectively recruited cohort of very low birthweight infants, and the characteristics of those lost to follow-up, is not reported. As over 60% of the initial sample were recruited from the cohort enrolled at birth, lack of attendance at the school-age follow-up visit is likely to be associated with greater socioeconomic deprivation 35,36. | Differences in baseline characteristics according to maternal education status are not reported. Other variables shown to be associated with subnormal head circumference (HC) at school age (e.g. birthweight, small for gestational age status, neonatal risk score, cerebral palsy and overall neurosensory impairment at school age) were not adjusted for when reporting the incidence of subnormal HC by parental socioeconomic status. Authors report provided “health data” at the school-age assessment, but only report the rates of cerebral palsy and overall neurosensory impairment between children with and without subnormal HC – data for other comorbidities are not reported. | All children had school-age head circumference measurements as lack of this measurement was specified as an exclusion criterion. | Growth outcomes for weight and height are not reported by parental socioeconomic status; 37% of infants with subnormal HC also had subnormal weight and height (less than 2 SD below the mean). The school-age measurements took place between 5.9 and 9 years, but the analysis of subnormal HC by parental socioeconomic status does not account for the time at which the measurements were taken. Blinding of assessors to infant birthweight status (particularly as authors hypothesised that those with birthweight < 750 g would be more likely to have subnormal HC) when measuring HC is not specified. |
Risk of bias (Peterson 27) | Moderate | High | Low | High |
Kelleher 23 | Infants were excluded if they were lost to follow-up before 30 months, if their mothers could not communicate adequately in English, or if their mothers reported drug or alcohol abuse or psychiatric hospitalisation. These criteria disproportionately exclude participants who are socioeconomically deprived33,34,35,36. Infants requiring intensive medical intervention (e.g. hospitalisation > 60 days after 40 weeks corrected age, or oxygen support for > 90 days) and those with severe neurodevelopmental abnormalities were also excluded. | Differences in baseline characteristics according to maternal education status or family income are not reported. The association between maternal education and the incidence of failure to thrive (FTT) is not adjusted for comorbidities even though children with FTT appear to be overrepresented among those with cerebral palsy, bronchopulmonary dysplasia and congenital heart disease. | The percentages of infants who attended each follow-up visit for growth outcome assessment, and the characteristics of these infants, are not reported. | The blinding of assessors to the intervention group in the IHDP trial is not specified. The stepwise variable selection process used for the multiple logistic regression analysis may have failed to highlight family income as a significant independent variable due to collinearity with maternal education. The choice of reference group for the maternal education variable (some college education without a college degree) in the multivariable regression model does not allow for comparison of infants at the two extremes of maternal education ( < High school versus ≥ college education). |
Risk of bias (Kelleher 23) | High | High | Unknown | Moderate |
