Children born very preterm (VPT) are at high risk of cognitive impairment that impacts their educational and social opportunities. This study examined the predictive accuracy of assessments at 2, 4, 6, and 9 years in identifying preterm children with cognitive impairment by 12 years.
We prospectively studied a regional cohort of 103 children born VPT (≤32 weeks’ gestation) and 109 children born term from birth to corrected age 12 years. Cognitive functioning was assessed by using age-appropriate, standardized measures: Bayley Scales of Infant Development, Second Edition (age 2); Wechsler Preschool and Primary Scale of Intelligence (ages 4 and 6); and Wechsler Intelligence Scale for Children, Fourth Edition (ages 9 and 12).
By 12 years, children born VPT were more likely to have severe (odds ratio 3.9; 95% confidence interval 1.1–13.5) or any (odds ratio 3.2; 95% confidence interval 1.8–5.6) cognitive impairment compared with children born term. Adopting a severe cognitive impairment criterion at age 2 under-identified 44% of children born VPT with later severe impairment, whereas a more inclusive earlier criterion identified all severely affected children at 12 years. Prediction improved with age, with any delay at age 6 having the highest sensitivity (85%) and positive predictive value (66%) relative to earlier age assessments. Inclusion of family-social circumstances further improved diagnostic accuracy.
Cognitive risk prediction improves with age, with assessments at 6 years offering optimal diagnostic accuracy. Intervention for children with early mild delay may be beneficial, especially for those raised in socially disadvantaged family contexts.
Approximately half of children born very preterm have mild to moderate cognitive impairments by school age. Early identification is challenging because cognition is difficult to assess at younger ages. It is unclear when school-aged cognitive risk is best predicted.
Cognitive risk prediction improves with age during early childhood, with assessments at age 6 years offering the best diagnostic accuracy. Both mild and severe early delay place children at risk later, with risk being further exacerbated by family-social disadvantage.
Children born very preterm (VPT) remain at high risk for neurodevelopmental impairments despite advances in neonatal care.1–3 These impairments span multiple domains, with cognitive difficulties affecting approximately half of surviving children.1,4–7 Between 40% and 50% of children born VPT meet criteria for either mild or severe cognitive or intellectual impairment, which is defined as an IQ >1 SD below the normative mean. These cognitive difficulties are in turn associated with high rates of special education service use,8,9 longer-term educational underachievement,10 social1,5,6 and mental health11 difficulties, as well as reduced earning and employment potential in adulthood.12,13 This suggests that even milder cognitive impairments may have significant impacts on functioning over the life course.
Early neurodevelopmental intervention is therefore critical to mitigate these adverse long-term effects. Not only is the brain undergoing rapid development during early childhood but it is also characterized by a high degree of neural plasticity and sensitivity to positive and negative environmental influences.14,15 Yet, a major challenge for early identification of cognitive impairment is that some deficits do not manifest until older ages, when the demands of the environment exceed the developmental capabilities of the child. This raises important questions regarding the optimal duration of developmental monitoring to ensure accurate and timely identification and intervention for children born VPT with clinically significant cognitive and learning needs.16,17
The Bayley Scales of Infant Development18 represent the most commonly used measure of cognitive ability before age 3. After this time, cognition is typically assessed by using standardized intelligence measures.19,20 Multiple studies have followed children born VPT or extremely preterm prospectively to examine the relations between early ability and middle childhood cognitive functioning, with variable results.21–31 Some studies report good concordance between Bayley assessments and later cognitive function in children,21,23,25,30 whereas others suggest poor correlation with preschool-26,31 and school-aged22,24,27 cognitive outcomes. One study of extremely preterm survivors examined the accuracy of the Bayley scales in predicting cognitive function at school age using multiple age assessments.23 Results suggested that early assessments were relatively good predictors of later cognitive function, with accuracy of risk prediction improving with age. The generalizability of these observations to VPT survivors remains uncertain.
An additional and important consideration is the social context in which children are raised given that family socioeconomic factors also impact cognitive development.4,5,32–34 Mangin et al35 found that family-social adversity contributed additively to preterm children’s cognitive risk in middle childhood. Other data suggest that over time, environmental factors may play an increasingly important role in shaping VPT children’s cognitive development than earlier perinatal exposures.33 This highlights the importance of considering not only the timing of earlier assessments but also the extent of family-social disadvantage when assessing need for ongoing monitoring and intervention for children born VPT.
Therefore, our aims in this study were as follows:
Examine the extent of severe and any (mild or severe) cognitive impairment in children born VPT compared with those born term at corrected ages 2, 4, 6, 9, and 12 years. For consistency with other studies, children were classified as showing severe impairment if their IQ score was >2 SDs below the term group mean and any impairment (mild or severe) if their IQ was >1 SD below the term mean at each assessment point.1,6
Examine the predictive accuracy of standardized cognitive measures at ages 2, 4, 6, and 9 in identifying cognitive impairment at age 12 in children born VPT.
Assess whether cognitive risk prediction for children born VPT could be further improved by considering the family-social context.
Methods
Sample
Two groups of children were included. The VPT group comprised 103 children born at ≤32 weeks’ gestational age (GA) consecutively admitted into a level III NICU at Christchurch Women’s Hospital in New Zealand from 1998 to 2000 (92% recruitment) and followed through age 12 years. Exclusion criteria included congenital abnormalities and non–English-speaking parents. Recruited infants did not differ from nonrecruited infants on clinical or family-social factors. Excluding post–NICU-discharge deaths (n = 3), retention at ages 4, 6, 9, and 12 was 98%, 97%, 96%, and 97%, respectively.
The term group comprised 109 children born at 37 to 41 weeks’ gestation identified from hospital birth records, recruited at age 2, and followed through age 12 years. Children were matched for sex and pregnancy due date, with 62% (n = 113) of regionally representative eligible infants being included. Nonparticipation reasons included primarily family circumstances.6 Retention rates at ages 4, 6, 9, and 12 were 96%, 96%, 97%, and 96%, respectively.
Procedures
Procedures and measures were approved by the Canterbury Regional Ethics Committee with written informed consent obtained from all parents and/or guardians. All children underwent comprehensive neurodevelopmental assessments close to their second, fourth, sixth, ninth, and 12th birthdays (corrected for GA). Cognitive assessments were completed by a blinded registered psychologist. Although GA correction through age 12 is not routinely performed clinically, this approach was adopted to reduce the likelihood of overestimation of cognitive impartment among children born VPT.36 At ages 2, 4, and 6, families were surveyed about their children’s participation in early intervention and/or special education services. Teachers were also questioned regarding remedial support at age 6, and clinical agencies were contacted to confirm the nature and duration of services children received.
Measures
At corrected age 2 years, cognitive function was estimated by using the Bayley Scales of Infant Development, Second Edition.18 At 4 and 6 years’ corrected age, the short form of the Wechsler Preschool and Primary Scale of Intelligence, Revised19 was administered, consisting of 2 verbal (Comprehension and Arithmetic) and 2 performance (Picture Completion and Block Design) subtests that correlate highly with full-scale IQ (r = 0.89–0.92).37 One preterm child at age 4 and 2 children at age 6 were assigned the minimum IQ score of 40 because of severe disability. One term child was excluded at age 4 (incomplete assessment), and one preterm child was excluded at age 6 (administration error).
At corrected ages 9 and 12 years, a short form of the Wechsler Intelligence Scale for Children, Fourth Edition was administered,20 consisting of 5 subtests that encompass the indices of the full form: Verbal Comprehension, Perceptual Reasoning, Working Memory, and Processing Speed. Three preterm children were assigned a score of 40, and 1 preterm child had an IQ estimated from 2 subtests (Vocabulary and Matrix Reasoning) at 9 years because of inability to complete the assessment. At 12 years, 2 preterm children were assigned an IQ of 40 (severe disability).
Across all assessments, the average 1-SD cutoff point ranged from 91.5 (age 2) to 95.3 (age 6), with the average 2-SD cutoff score ranging from 70.6 (age 2) to 83.7 (age 6).
Family-Social Risk
We collected 5 measures of family-social risk during the first 2 years of life: maternal minority ethnicity (non–New Zealand European), maternal age at child birth (<21 years), maternal education (did not graduate high school), single-parent family, and family socioeconomic status (SES) (semiskilled, unskilled, or unemployed). Each variable was coded as either present or absent and then summed to form a composite family-social risk index. Because few children were exposed to >2 risk factors, the index was operationalized to reflect 0, 1, or ≥2 family-social risk factors.
Statistical Analysis
We calculated means, SDs, and percentages of relevant demographic and clinical characteristics and compared them by term status using the t test and χ2 test, respectively. The magnitude of between-group differences in cognitive performance by age was assessed by using Cohen’s d or χ2 and/or Fisher’s exact tests. For any (IQ >1 SD) cognitive impairment measures, the odds ratios (ORs) and 95% confidence intervals (CIs) were assessed by using logistic regression models, both unadjusted and adjusted for family-social risk. The classification accuracy of identifying cognitive impairment at age 12 by using severe (IQ >2 SDs) and any (IQ >1 SD) criteria at ages 2, 4, 6, and 9 years was evaluated on the basis of sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and receiver operating curve (ROC). Effect modification by family-social risk was examined by comparing the classification accuracy of the best cognitive impairment predictor by family-social risk strata. All tests of statistical significance were 2 tailed with α < .05. Analyses were conducted by using SAS 9.4 (SAS Institute, Inc, Cary, NC).
Results
Sample Characteristics
The clinical and family characteristics of the 2 study groups are shown in Table 1. Notably, infants in the VPT group were more likely to be of multiple gestation and small for GA; they were also more likely to be born to less educated women of lower SES. Reflecting these increased levels of social disadvantage, children born VPT were 2.6 times more likely to be raised in families with ≥2 family-social risk factors (33% VPT; 13% term). With regard to support services, 63 children (61%) born VPT received intervention by 4 years: 48 received 1 service, 12 received 2 services, and 3 received 3 services. In addition, more children born VPT received remedial support at 6 years (43% VPT; 20% term).
Study Sample: Clinical and Demographic Characteristics (N = 212)
| VPT (N = 103) | Term (N = 109) | Test Statistica | P | |
|---|---|---|---|---|
| Child clinical characteristics | ||||
| GA, mean (SD) | 28 (2) | 39 (1) | −45.1 | <.001b |
| Birth wt, mean (SD) | 1061 (314) | 3585 (411) | −50.4 | <.001b |
| Male sex, n (%) | 52 (50) | 60 (55) | 0.4 | .51 |
| Twin birth, n (%) | 34 (33) | 4 (4) | 31.0 | <.001b |
| Small for GA, n (%) | 11 (11) | 1 (1) | 9.5 | .002b |
| Culture-proven sepsis, n (%) | 30 (29) | 0 (0) | na | — |
| Oxygen use at 36 wk, n (%) | 36 (35) | 0 (0) | na | — |
| ROP stage 3 or 4, n (%) | 4 (4) | 0 (0) | na | — |
| IVH grade 3 or 4, n (%) | 6 (6) | 0 (0) | na | — |
| Moderate to severe white matter abnormality, n (%) | 19 (18) | 0 (0) | na | — |
| Postnatal dexamethasone, n (%) | 6 (6) | 0 (0) | na | — |
| Early intervention and learning support | ||||
| Early intervention services by 4 y, n (%) | 63 (61) | 1 (1) | — | <.001b |
| Remedial support at 6 y, n (%) | 44 (43) | 22 (20) | — | <.001b |
| Family and social characteristics | ||||
| Maternal age, mean (SD) | 31 (5) | 31 (4) | −0.4 | .69 |
| Ethnic minority, n (%) | 14 (14) | 13 (12) | 0.1 | .74 |
| Single-parent family, n (%) | 20 (19) | 13 (12) | 2.2 | .14 |
| Mother did not complete high school, n (%) | 42 (41) | 20 (18) | 12.6 | <.001b |
| Low family SES, n (%) | 31 (30) | 11 (10) | 13.1 | <.001b |
| Family-social risk index | ||||
| None, n (%) | 38 (37) | 66 (61) | 11.9 | <.001b |
| Low, 1 family-social risk factor, n (%) | 31 (30) | 28 (26) | 0.5 | .47 |
| High, 2 or more family-social risk factors, n (%) | 34 (33) | 14 (13) | 12.3 | <.001b |
| VPT (N = 103) | Term (N = 109) | Test Statistica | P | |
|---|---|---|---|---|
| Child clinical characteristics | ||||
| GA, mean (SD) | 28 (2) | 39 (1) | −45.1 | <.001b |
| Birth wt, mean (SD) | 1061 (314) | 3585 (411) | −50.4 | <.001b |
| Male sex, n (%) | 52 (50) | 60 (55) | 0.4 | .51 |
| Twin birth, n (%) | 34 (33) | 4 (4) | 31.0 | <.001b |
| Small for GA, n (%) | 11 (11) | 1 (1) | 9.5 | .002b |
| Culture-proven sepsis, n (%) | 30 (29) | 0 (0) | na | — |
| Oxygen use at 36 wk, n (%) | 36 (35) | 0 (0) | na | — |
| ROP stage 3 or 4, n (%) | 4 (4) | 0 (0) | na | — |
| IVH grade 3 or 4, n (%) | 6 (6) | 0 (0) | na | — |
| Moderate to severe white matter abnormality, n (%) | 19 (18) | 0 (0) | na | — |
| Postnatal dexamethasone, n (%) | 6 (6) | 0 (0) | na | — |
| Early intervention and learning support | ||||
| Early intervention services by 4 y, n (%) | 63 (61) | 1 (1) | — | <.001b |
| Remedial support at 6 y, n (%) | 44 (43) | 22 (20) | — | <.001b |
| Family and social characteristics | ||||
| Maternal age, mean (SD) | 31 (5) | 31 (4) | −0.4 | .69 |
| Ethnic minority, n (%) | 14 (14) | 13 (12) | 0.1 | .74 |
| Single-parent family, n (%) | 20 (19) | 13 (12) | 2.2 | .14 |
| Mother did not complete high school, n (%) | 42 (41) | 20 (18) | 12.6 | <.001b |
| Low family SES, n (%) | 31 (30) | 11 (10) | 13.1 | <.001b |
| Family-social risk index | ||||
| None, n (%) | 38 (37) | 66 (61) | 11.9 | <.001b |
| Low, 1 family-social risk factor, n (%) | 31 (30) | 28 (26) | 0.5 | .47 |
| High, 2 or more family-social risk factors, n (%) | 34 (33) | 14 (13) | 12.3 | <.001b |
IVH, intraventricular hemorrhage; na, not applicable; ROP, retinopathy of prematurity; —, not available.
Continuous distributions (normally distributed) compared by group status by using a t test and binary variables compared by using a χ2 test.
Comparison statistically significant at α < .05 based on a 2-sided test.
Extent of Cognitive Impairment
Examination of scores at each age indicated that children born VPT performed significantly worse than those born term across all cognitive measures (Fig 1). Correspondingly, significantly more children born VPT were subject to severe (10%–16%) or any (34%–49%) cognitive impairment compared with children born term.
Comparison of the distributions of cognitive performance by group status across ages of assessment: 2, 4, 6, 9, and 12 years. The smoothed distribution of scores among children born VPT and term are represented on kernel density estimate curves. The histograms represent the distribution of IQ scores by group status. The kernel density estimate curves represent nonparametric estimates of the probability functions for IQ by VPT status. For the kernel density estimate curves, the parameter that determines the degree of smoothness in the estimated density function (ie, the bandwidth) is based on an approximation of the mean integrated square error calculated as the sum of the integrated squared bias and the variance. IQ_4, IQ at age 4; IQ_6, IQ at age 6; IQ_9, IQ at age 9; IQ_12, IQ at age 12.
Comparison of the distributions of cognitive performance by group status across ages of assessment: 2, 4, 6, 9, and 12 years. The smoothed distribution of scores among children born VPT and term are represented on kernel density estimate curves. The histograms represent the distribution of IQ scores by group status. The kernel density estimate curves represent nonparametric estimates of the probability functions for IQ by VPT status. For the kernel density estimate curves, the parameter that determines the degree of smoothness in the estimated density function (ie, the bandwidth) is based on an approximation of the mean integrated square error calculated as the sum of the integrated squared bias and the variance. IQ_4, IQ at age 4; IQ_6, IQ at age 6; IQ_9, IQ at age 9; IQ_12, IQ at age 12.
Table 2 shows the extent of severe and any cognitive impairment among all children, with children in the VPT group obtaining significantly lower cognitive and/or IQ scores than children born term across all assessments (P < .05). By age 12, 14 children (VPT, n = 11 [11%]; term, n = 3 [3%]; P = .03) met criteria for severe impairment, and 52 (VPT, n = 39 [38%]; term, n = 13 [12%]; P < .001) met criteria for any cognitive impairment.
Rates of Severe (IQ >2 SDs) and Any (Mild or Severe: IQ >1 SD) Cognitive Impairment at Each Time Point by Group Status
| Age of Assessment | Cognitive Impairment Threshold | VPT (N = 103), % | Term (N = 109), % | Effect Size, d | Comparisons by Group | |||||
|---|---|---|---|---|---|---|---|---|---|---|
| Test Statistica | Pb | Unadjusted ORc | 95% CI | Adjusted ORc | 95% CI | |||||
| 2 | −0.5 | |||||||||
| Severe | 13 | 5 | — | .04b | 2.7 | (1.0–7.2) | 2.3 | (0.8–6.5) | ||
| Any | 40 | 15 | 15.3 | <.001b | 2.7 | (1.6–4.5) | 2.5 | (1.4––4.2) | ||
| 4 | −0.7 | |||||||||
| Severe | 10 | 2 | — | .02b | 5.1 | (1.1–22.7) | 3.4 | (0.7–15.7) | ||
| Any | 35 | 12 | 14.7 | <.001b | 2.8 | (1.6–5.0) | 2.4 | (1.3–4.4) | ||
| 6 | −0.8 | |||||||||
| Severe | 16 | 4 | — | .004b | 4.2 | (1.4–12.0) | 3.4 | (1.1–10.2) | ||
| Any | 49 | 16 | 25.7 | <.001b | 3.1 | (1.9–4.9) | 2.8 | (1.7–4.7) | ||
| 9 | −0.6 | |||||||||
| Severe | 13 | 3 | — | .008b | 4.5 | (1.3–15.5) | 3.2 | (0.9–11.4) | ||
| Any | 34 | 12 | 14.5 | <.001b | 2.8 | (1.6–5.0) | 2.4 | (1.3–4.3) | ||
| 12 | −0.6 | |||||||||
| Severe | 11 | 3 | — | .03b | 3.9 | (1.1–13.5) | 2.8 | (0.8–10.2) | ||
| Any | 38 | 12 | 19.2 | <.001b | 3.2 | (1.8–5.6) | 2.8 | (1.5–5.1) | ||
| Age of Assessment | Cognitive Impairment Threshold | VPT (N = 103), % | Term (N = 109), % | Effect Size, d | Comparisons by Group | |||||
|---|---|---|---|---|---|---|---|---|---|---|
| Test Statistica | Pb | Unadjusted ORc | 95% CI | Adjusted ORc | 95% CI | |||||
| 2 | −0.5 | |||||||||
| Severe | 13 | 5 | — | .04b | 2.7 | (1.0–7.2) | 2.3 | (0.8–6.5) | ||
| Any | 40 | 15 | 15.3 | <.001b | 2.7 | (1.6–4.5) | 2.5 | (1.4––4.2) | ||
| 4 | −0.7 | |||||||||
| Severe | 10 | 2 | — | .02b | 5.1 | (1.1–22.7) | 3.4 | (0.7–15.7) | ||
| Any | 35 | 12 | 14.7 | <.001b | 2.8 | (1.6–5.0) | 2.4 | (1.3–4.4) | ||
| 6 | −0.8 | |||||||||
| Severe | 16 | 4 | — | .004b | 4.2 | (1.4–12.0) | 3.4 | (1.1–10.2) | ||
| Any | 49 | 16 | 25.7 | <.001b | 3.1 | (1.9–4.9) | 2.8 | (1.7–4.7) | ||
| 9 | −0.6 | |||||||||
| Severe | 13 | 3 | — | .008b | 4.5 | (1.3–15.5) | 3.2 | (0.9–11.4) | ||
| Any | 34 | 12 | 14.5 | <.001b | 2.8 | (1.6–5.0) | 2.4 | (1.3–4.3) | ||
| 12 | −0.6 | |||||||||
| Severe | 11 | 3 | — | .03b | 3.9 | (1.1–13.5) | 2.8 | (0.8–10.2) | ||
| Any | 38 | 12 | 19.2 | <.001b | 3.2 | (1.8–5.6) | 2.8 | (1.5–5.1) | ||
—, not available.
Proportions by group status were compared by using the χ2 test and Fisher’s exact test (when cell sizes were small [<5]).
α < .05 based on a 2-sided test.
Odds of any (>1 SD) cognitive impairment among children (VPT versus term); adjusted analyses for family-social risk status (0, 1, or 2).
Given the higher rates of family disadvantage among the VPT group (ORs; Table 2), we examined to what extent between-group differences may partly reflect the effects of social risk. Accounting for family-social risk in regression models (adjusted ORs; Table 2) attenuated effect estimates by roughly 10%. Nonetheless, the odds of cognitive impairment remained higher for children born VPT even after adjustment for family-social risk.
Predictive Utility of Earlier Assessments in Identifying 12-Year Cognitive Risk
Figure 2 examines the predictive accuracy of cognitive impairment classification at 2, 4, 6, and 9 years in identifying VPT children at risk for cognitive impairment at 12 years. Results show that adopting a severe impairment predictor criterion had poorer sensitivity in identifying cases of severe cognitive impairment at age 12 than any impairment criterion. Specifically, the severe cognitive impairment predictor criterion at ages 2, 4, and 6 missed between 18% and 44% of severe cases (Fig 2A), whereas the more inclusive any impairment criterion successfully predicted all severe impairment cases at 12 years (Fig 2B).
Classification accuracy of cognitive impairment at age 12 (severe: IQ >2 SD; any: IQ >1 SD) among VPT children using measures of cognitive impairment at 2, 4, 6, and 9 years. Prediction of severe cognitive impairment at 12 years using earlier (A) severe and (B) any impairment. Prediction of any cognitive impairment at 12 years using earlier (C) severe and (D) any impairment criteria.
Classification accuracy of cognitive impairment at age 12 (severe: IQ >2 SD; any: IQ >1 SD) among VPT children using measures of cognitive impairment at 2, 4, 6, and 9 years. Prediction of severe cognitive impairment at 12 years using earlier (A) severe and (B) any impairment. Prediction of any cognitive impairment at 12 years using earlier (C) severe and (D) any impairment criteria.
Further examination of diagnostic accuracy showed that the diagnostic precision of early severe impairment in identifying any cognitive impairments at age 12 was modest (Fig 2C). In contrast, employing a criterion of early Mental Developmental Index (MDI) and/or IQ >1 SD to identify any cognitive impairment cases at age 12 years had generally good diagnostic properties (Fig 2D). That is, 100% of the severe cases and 43% (age 4) to 79% (age 6) of the any impairment cases at 12 years were identified by using the earlier any impairment predictors. PPV for the any impairment predictors was moderate (56% [age 2] to 66% [age 6]), whereas the specificity (73% [ages 2 and 6] to 81% [age 9]) and NPV (76% [age 4] to 89% [age 6]) were high. Results suggest that any delay at age 6 was particularly sensitive in detecting risk of any cognitive impairment at age 12 years among children born VPT.
Risk prediction was further assessed by plotting the age-specific ROCs for the prediction of any cognitive impairment at age 12 using any impairment criterion at earlier ages (Fig 3). Results from this analysis confirmed the above observation that any cognitive impairment at age 6 offers the best prediction of any cognitive impairment at 12 years.
ROCs: diagnostic accuracy in predicting any cognitive impairment (IQ >1 SD) at age 12 by using any impairment at ages 2, 4, 6, and 9 years. Any cognitive impairment cutoffs are marked with black circles. Among earlier assessments, the model with any IQ delay at age 6 has optimal discriminatory ability. AUC, area under the curve.
ROCs: diagnostic accuracy in predicting any cognitive impairment (IQ >1 SD) at age 12 by using any impairment at ages 2, 4, 6, and 9 years. Any cognitive impairment cutoffs are marked with black circles. Among earlier assessments, the model with any IQ delay at age 6 has optimal discriminatory ability. AUC, area under the curve.
Effect Modification by Family-Social Risk
Figure 4 shows the extent to which additional consideration of a child’s family-social circumstances (2+ family-social risk factors) improved the diagnostic accuracy of any cognitive impairment at age 6 predicting any later cognitive impairment during middle childhood. Diagnostic accuracy was better among children in the high versus low family-social risk group with superior PPV (79% vs 58%), sensitivity (88% vs 82%), and specificity (75% vs 72%).
ROCs: effect modification by family-social risk of the diagnostic accuracy of any cognitive impairment at age 6 predicting any cognitive impairment at 12 years. Any cognitive impairment cutoffs are marked with black circles. Any IQ delay at 6 years in high–social-risk strata has better discriminatory ability. AUC, area under the curve.
ROCs: effect modification by family-social risk of the diagnostic accuracy of any cognitive impairment at age 6 predicting any cognitive impairment at 12 years. Any cognitive impairment cutoffs are marked with black circles. Any IQ delay at 6 years in high–social-risk strata has better discriminatory ability. AUC, area under the curve.
Discussion
This study is the first to evaluate the predictive accuracy of early standardized assessments in identifying VPT children at risk for cognitive delay through middle school. Study strengths included the recruitment of representative cohorts of children, high sample retention, and availability of cognitive assessments throughout early and middle childhood using well-validated measures. Study findings and implications are discussed below.
Similar to others,4,25–27 we found that children born VPT had higher rates of cognitive impairment relative to their term-born peers, with odds of any impairment ranging from 3.7 (95% CI 1.9–7.4) at age 2 to 5.0 (95% CI 2.6–9.6) at age 6. Notably, most children born VPT experience mild (27.2%) rather than severe (10.7%) cognitive impairment at 12 years. This finding has potential clinical implications because children with mild impairment often do not qualify for school assistance programs.
In keeping with other longitudinal studies,21,23,25 we found that early severe cognitive impairment was a relatively good indicator of continued problems into middle school. However, a high proportion (38%) of children born VPT were subject to either mild or severe cognitive impairment at 12 years. Importantly, 44% of these children would have been missed at their 2-year assessment if only those with severe impairment were deemed eligible for ongoing monitoring. Examination of the diagnostic accuracy of early childhood assessments further suggested that adopting a more inclusive (MDI and/or IQ >1 SD) criterion offered improved prediction for both severe and any cognitive impairment at 12 years in children born VPT. Given the long-term ramifications that even milder cognitive impairments may have on functional outcomes, adopting a more inclusive approach when identifying at-risk children may be warranted.
Consistent with an earlier study of children born extremely preterm,23 we found that the predictive accuracy of cognitive measures improved with age. This was further confirmed by ROC analyses, which indicated that IQ delay at 12 years was optimally predicted by any IQ delay at age 6. Notably, IQ performance at age 9 was also a good predictor of any cognitive delay at age 12 given its more proximal assessment (Fig 3). However, such a late assessment is likely of limited use if the goal is early detection and proactive intervention for the child and family. Our finding that 2- and 4-year assessments were not as good predictors of school-aged functioning as the 6- and 9-year assessments is consistent with other reports22,26,27 and may reflect the limitations of early evaluations that are designed to gauge a child’s general developmental level rather than accurately predict higher-order cognitive functions.
We recognize that rates of any cognitive impairment were particularly elevated at age 6 in both groups. This finding may represent a measurement artifact. Alternatively, it could reflect the increased cognitive expectations experienced by children at age 6 relative to their earlier preschool years. All study children started elementary school around their fifth birthday, with this transition and its accompanying increased cognitive demands and expectations potentially resulting in milder deficits becoming more apparent. Furthermore, the high rates of cognitive difficulties in the VPT cohort at 6 years might also reflect the additive effect of family-social risk. Findings suggest that monitoring cognitive functioning of children born VPT until age 6 might be beneficial to create a safety net for this high-risk population during the challenging transition to school.
Risk-stratification findings suggest that children born VPT who experienced ≥2 family-social risk factors were at additional risk of persistent cognitive impairment above and beyond the risk conferred by earlier delay, with prediction of impairment at 12 years being superior in the high– versus low–social-risk subgroup. This reaffirms the additive effects of prematurity and social disadvantage on cognitive function and emphasizes the importance of not only considering children’s early developmental functioning but also the extent of social adversity when determining a child’s eligibility for developmental monitoring and/or intervention.
Limitations
Our data indicate that having any cognitive impairment at age 6 appears to be the best predictor of cognitive impairment at age 12 from the models evaluated. Given the relatively high false-positive rate (34%), this model has its limitations; however, this can be improved by using risk stratification.
Furthermore, it is likely that in addition to family-social risk, other factors associated with the child-rearing environment may contribute to children’s cognitive functioning and risk of delay. Intervention support services may have also impacted later cognitive functioning; yet, taking this into account is challenging given that the children who received support are those identified with early impairment.
The current study focuses on cognitive risk prediction from early childhood measures often employed by developmental monitoring programs. Our additional measure of early family-social risk was a relatively simple composite of factors extracted from clinical data. However, we acknowledge that other factors, including medical risk, parental mental health, family stability, and parenting, are likely to play a role in shaping cognitive outcomes. Future research is important to better understand the developmental pathways that modify cognitive risk for children born VPT and assess whether the inclusion of additional factors in risk-prediction models improves the diagnostic accuracy of early assessments. In addition, examination of the predictive accuracy of these approaches for other neurodevelopmental outcomes is warranted.
Implications
Our findings highlight the potential benefit of monitoring children at high risk with early delay until elementary school. We acknowledge that this would result in a higher number of referrals and potentially increased short-term costs. Developmental follow-up is costly,38 yet early developmental services are valuable and positively impact preterm children’s cognitive39 and preacademic skills.40 Future work should examine which specific strategies and interventions have the greatest potential to positively impact cognitive outcomes in children born VPT.
Conclusions
Cognitive impairment in middle school is poorly predicted by early severe delay. Monitoring children born VPT until age 6, intervening for children with early mild cognitive delay, and assisting families with social disadvantages are factors that warrant consideration in supporting preterm children in achieving their best potential long-term.
Acknowledgments
We give special thanks to the study families for their time and support of this project.
Study enrollment and patient monitoring and follow-up occurred at Christchurch Women’s Hospital in New Zealand.
Dr Erdei conceptualized the research questions of this study, performed the analysis and interpretation of the data in conjunction with the coauthors, drafted the initial manuscript, and edited the manuscript; Dr Cherkerzian contributed to the statistical methods and analysis and critically reviewed and helped revise the present manuscript; Ms Morris conducted the initial analysis and critically reviewed the present manuscript; Prof Woodward and Associate Prof Austin conceptualized the design of this study in addition to the larger study on which this analysis is based, coordinated and supervised data collection, contributed to the data analysis plan, and critically reviewed and helped revise the present manuscript; and all authors approved the final manuscript as submitted and agree to be accountable for all aspects of the work.
FUNDING: Funded by the Neurological Foundation, Lottery Grants Board, Canterbury Medical Research Foundation, and Health Research Council of New Zealand.
Comments