Table 1.
Study characteristics of multivariable prediction models for cognitive outcomes in childhood.
| Authors | Camargo-Figuera et al.20 | Camacho et al.21 | Eriksen et al.22 |
| Journal | BMC Pediatrics | Paediatric Perinatal Epidemiology | PLoS One |
| Continent | South America | Europe | Europe |
| Sample | Pelotas Birth Cohort | Millennium Cohort Study | Lifestyle During Pregnancy Study (sampled from the Danish National Birth Cohort) |
| Design | Prospective cohort | Prospective cohort | Prospective cohort |
| Sample size | 3312 | 9487 | 1782 |
| Year of recruitment | 2004 | 2000–2002 | 1997–2003a |
| Exclusion criteria | Conditions associated with very low IQ, e.g., severe mental retardation | Nil | Multiple pregnancies, language barrier, impaired hearing or vision, congenital disabilities implying mental retardation |
| Age at cognitive assessment | 6 | 3 | 5 |
| Cognitive assessment | Wechsler Primary and Preschool Scales of Intelligence -III | Bracken School Readiness Assessment | Wechsler Primary and Preschool Scales of Intelligence – Revised |
| Cognitive outcome variable | Binary | Binary | Continuous |
| Low IQ defined by a z-score <−1 | Not school ready defined by score <1 standard deviation below mean | ||
| Number of risk factors at outset | 32 | 29 | 27 |
| Rationale given for candidate variables | Yes—selected based on previous literature and availability | Yes—selected based on previous literature and availability | No–but broad range (>20) selected |
| Statistical model | Multivariable logistic regression | Multivariable logistic regression | Multivariable linear regression |
| Method of initial screening of candidate variables | Forward and backward stepwise selection | Forward and backward stepwise selection | Univariable association p ≥ 0.10 |
| Interaction terms fitted | No | No | No |
| Multicollinearity addressed/discussed | No | Yes | Yes |
| No. predictors in final model | 13 | 13 | 9 |
| Validation performed | Yes | Yes | No |
| Internal and external validation performed | Internal validation only | ||
| Predictive value measured | External validation | Internal validation | R squared 0.29 |
| Area under receiver operating curve (AUROC) 0.75 | AUROC 0.80 | ||
| Sensitivity 70.3% | Sensitivity 72% | ||
| Specificity 68% | Specificity 74% | ||
| Predictors in final model | Child—gender, height-for-age deficit; head circumference-for-age deficit | Child—gender, ethnicity, developmental milestones | Child—gender, birth weight, height, head circumference |
| Parental—breastfeeding, parental smoking, maternal perception of child’s health, skin colour | Parental—maternal age, maternal mental health, breastfeeding | Parental—maternal BMI, breastfeeding | |
| Socioenvironmental—parental employment status, maternal education, income, number of siblings, number of persons per room | Socioenvironmental—socioeconomic class, maternal education, income, number of children, employment status, housing type | Socioenvironmental—maternal IQ, parental education |
aIn 2003, a prospective follow-up of 1750 mother–child pairs was initiated, sampled on the basis of maternal alcohol drinking patterns from The Danish National Birth Cohort (DNBC).