CONTEXT: Developmental outcomes of very preterm (gestational age ≤32 weeks) or very low birth weight (<1500 g) children are commonly reported before age 3 years although the predictive validity for later outcomes are uncertain.
OBJECTIVE: To determine the validity of early developmental assessments in predicting school-age cognitive deficits.
DATA SOURCES: PubMed.
STUDY SELECTION: English-language studies reporting at least 2 serial developmental/cognitive assessments on the same population, 1 between ages 1 and 3 years and 1 at ≥5 years.
DATA EXTRACTION: For each study, we calculated the sensitivity, specificity, and positive and negative predictive values of early assessment for cognitive deficit (defined as test scores 1 SD below the population mean). Pooled meta-analytic sensitivity and specificity were estimated by using a hierarchical summary receiver operator characteristic curve.
RESULTS: We included 24 studies (n = 3133 children). Early assessments were conducted at 18 to 40 months and generally involved the Bayley Scales of Infant Development or the Griffiths Mental Development Scales; 11 different cognitive tests were used at school-age assessments at 5 to 18 years. Positive predictive values ranged from 20.0% to 88.9%, and negative predictive vales ranged from 47.8% to 95.5%. The pooled sensitivity (95% confidence interval) of early assessment for identifying school-age cognitive deficit was 55.0% (45.7%–63.9%) and specificity was 84.1% (77.5%–89.1%). Gestational age, birth weight, age at assessment, and time between assessments did not explain between-study heterogeneity.
LIMITATIONS: The accuracy of aggregated data could not be verified. Many assessment tools have been superseded by newer editions.
CONCLUSIONS: Early developmental assessment has poor sensitivity but good specificity and negative predictive value for school-age cognitive deficit.
- BSID —
- Bayley Scales of Infant Development
- BSID-II —
- Bayley Scales of Infant Development, second edition
- CI —
- confidence interval
- DOR —
- diagnostic odds ratio
- ESS —
- effective sample size
- HSROC —
- hierarchical summary receiver operator characteristics
- MDI —
- Mental Development Index
- NPV —
- negative predictive value
- PPV —
- positive predictive value
- QUADAS-2 —
- Quality of Diagnostic Accuracy Studies version 2
- VLBW —
- very low birth weight
The majority of outcome studies of preterm births report neurodevelopmental status at 18 or 24 months postterm age (corrected for prematurity). This practice stemmed from the emphasis of early outcome studies on measuring major disabilities, such as severe mental retardation, sensorineural hearing loss, blindness, and cerebral palsy. It has generally been felt that at 18 to 24 months of age, a meaningful assessment of neurodevelopment can be reliably conducted while achieving a good follow-up rate.
However, there are developmental and maturation changes that affect the diagnostic accuracy of findings in the first 2 years. Follow-up studies into school age and adolescence have regularly reported high rates of subtle disabilities that impact learning and social integration among seemingly “nondisabled” survivors of preterm birth.1–3 These “high-prevalence/low-severity dysfunctions” include low average IQ scores, learning disabilities, and attention and behavioral problems. Although assessment tools, such as the Bayley Scales of Infant Development (BSID), provide standardized mental (cognitive) scores from as early as 12 months of age, the correlation of the early mental scores with subsequent IQ at school age is unclear. Two studies reported moderate to substantial agreement between BSID, second edition (BSID-II) Mental Development Index (MDI) at age 2 years and full scale IQ at age 5 years among infants born at <30 weeks gestation or with very low birth weight (VLBW; birth weight <1500 g).4,5 Conversely, Hack et al6 described a considerable reduction in the proportions of extremely low birth weight infants (birth weight <1000 g) who were diagnosed with cognitive impairment (defined as standardized cognitive scores <70) from 39% at 20 months to 16% at 8 years of age when the children were tested sequentially. Applying the same diagnostic criteria, Roberts et al7 also found a reduction in the proportions of very preterm (gestational age <27 weeks) and extremely low birth weight infants with cognitive impairment, from 27.3% at age 2 years to 19.3% at age 8 years.
We aimed to perform a systematic search of the published literature and review the evidence for the predictive validity of early developmental assessment, conducted between the ages of 1 and 3 years, for school-age cognitive deficit in children born very preterm or with very low birth weight.
This study was registered in PROSPERO (international prospective register of systematic reviews) (identifier: CRD42012002168).
We conducted a systematic electronic search on PubMed to identify relevant English-language literature published since first January 1990. Studies published before 1990 were not included to focus the review on more contemporaneous preterm populations. The following search terms were used both as keywords and subject headings: (combinations of “preterm” or “premature” with “infant” or “neonate” or “children”) or (“low birth weight” or “extremely low birth weight”) and (“cogniti*” or “neurodevelopment*” or “mental retardation” or “disability” or “intelligence” or “IQ”). The electronic search was supplemented by a manual search of the reference lists of studies that met the inclusion criteria.
We sought to include all cohort and matched-control studies on study populations of infants born ≤32 weeks gestation or with VLBW, in which at least 2 serial assessments, consisting of a developmental assessment between 1 and 3 years of age and a cognitive assessment at ≥5 years of age, were conducted and reported using validated standardized psychometric assessments (eg, BSID, Wechsler Preschool and Primary Scale of Intelligence). We included a birth weight limit as many neonatal studies used a birth weight–based selection criterion. The titles and abstracts of studies retrieved from the electronic search were screened to identify relevant studies in the following 3 categories: (1) studies that reported both early developmental outcomes between ages 1 and 3 years as well as school-age cognitive outcomes at ≥5 years; (2) studies that only reported early developmental outcomes; and (3) studies that only reported school-age cognitive outcomes. To seek study populations that had received serial assessments in the relevant time periods, but with their results at different ages published in separate articles, we matched the authors and study location of articles in (2) and (3) to identify publications on the same population at different time points. We conducted full-text evaluation for studies that satisfied the initial screening process for final inclusion in the review. Studies that only reported outcomes in language or executive function (eg, memory) were excluded as they would not reflect the overall cognitive function of the study populations.
Data extracted included information on the study (study location, sampling method, eligibility criteria, sample size, attrition rates, and developmental and cognitive assessment tools employed) and the study population (years of birth of participants, mean gestational age and birth weight, ages at developmental and cognitive assessments, and mean test scores).
For this review, mild-moderate deficit was defined as developmental or cognitive test scores between 1 and 2 SD below the standardized or control group means. Severe deficit was defined as test scores >2 SD below the standardized or control group means. In studies where a control group of children born at full-term was recruited and assessed simultaneously, the mean and SD of the control group were used as the references for defining the presence of deficit. Data on the number of “true-positive”, “false-positive”, “false-negative,” and “true-negative” mild-moderate and severe cognitive deficits identified by early assessments were collated from each study. Unpublished data were sought from study authors through e-mail requests.
The quality of included studies was assessed using a checklist adapted from the Quality of Diagnostic Accuracy Studies, version 2 (QUADAS-2) appraisal tool.8 The QUADAS-2 tool uses “signaling questions” to judge bias in 4 domains: patient selection, index test, reference standard, and flow of participants through the study and timing of the index test. The applicability of the study to the review question in the first 3 domains was also assessed. In the context of this review, the index tests referred to the early developmental assessments and the reference standards were the school-age cognitive assessments. Table 1 lists the signaling questions and the quality standards set for this review. By appraising against the set standards, each study was given a rating of “low,” “high,” or “unclear” for risk of bias and concerns regarding applicability in each domain.
From each study, the estimated sensitivity, specificity, positive predictive values (PPV) and negative predictive values (NPV), and the corresponding 95% confidence interval (CI) of identifying any and severe school-age cognitive deficits by early developmental assessments were calculated. Because of the variation in impairment prevalence across studies, meta-analyses on PPV and NPV, which are dependent on prevalence rates, were not performed. Separate pooling of sensitivities and specificities from the studies, which ignore the correlation between the 2 measures, could lead to an underestimation of the diagnostic accuracy.9 Instead, a hierarchical summary receiver operator characteristics (HSROC) curve was used for meta-analysis.10 The HSROC model accounts for both within-study sampling variation and between-study heterogeneity using random effects. The output includes a summary operating point (pooled values for sensitivity and specificity) with 95% confidence region. Meta-regression was conducted by using bivariate models to test for the possible association between sensitivity and specificity and the following study-level variables: mean gestational age, mean birth weight, mean ages at assessments, time interval between assessments, and earliest year of birth of participants. Associations with sensitivity and specificity were tested separately and a likelihood ratio test was used to test both associations jointly. Because there were too many different types of assessment tools used to be categorized into reasonably homogenous groups, subgroup analysis on the association between the types of assessment tool and diagnostic validity was not robust and, therefore, not performed.
To investigate the possibility of publication and other sample size–related effects, a funnel plot of the log diagnostic odds ratio (DOR) against 1/(effective sample size [ESS])1/2 were tested for plot asymmetry using linear regression of the 2 variables, weighted by ESS.11 The DOR is a statistic measure defined as (true-positives × true-negatives)/(false-positives × false-negatives). The ESS is a function of the number of nondiseased (n1) and diseased (n2) participants, where ESS = (4n1 n2)/(n1 + n2).
All analyses were performed by using Stata statistical package, version 11.0 (Stata Corp, College Station, TX) and SAS 9.3 (SAS Institute Inc, Cary, NC).
The electronic literature search yielded 2844 unique citations; 2 additional studies were identified through a manual search. The flow of articles through the search and selection process is depicted in the PRISMA diagram in Fig 1. Fifty-four studies met the eligibility criteria. Data required for the review and meta-analysis were extractable directly from 6 articles. The authors of 18 of the remaining 48 studies contributed unpublished data. Therefore, 24 studies (37 articles)4–7,12–44 were included in this review, and their characteristics are detailed in Supplemental Table 3. The characteristics of eligible studies that were not included (year of publication, countries where the studies were conducted) were similar to those included in the review. For simplicity of referencing, studies that are represented by >1 article will be denoted by the first author and year of publication of the earliest article in tables and figures.
The study populations included 3133 children who were born at ≤32 weeks and/or had a birth weight <1500 g. The study populations in 4 studies12,17,27,29 consisted of participants who exceed the gestational age and birth weight limits set in this review. For these studies, only the subpopulation of participants who met our criteria was included in the analysis. The mean gestational ages at birth ranged from 25.0 to 33.1 weeks, and the mean birth weights were between 675 and 1298 g. A total of 37.0% (1159 children) of the included populations was born in the years 1972 to 1990, 49.6% (1555 children) in 1991 to 2000 and 13.4% (419 children) in 2000 to 2005. Of note, there were 20 participants from the study of Cohen17 that were born in early 1970s. We have not excluded studies on the basis of the time period the participants were born in because it allowed for analysis of the variability of diagnostic validity over time. Children with known genetic syndromes and congenital anomalies were excluded from the studies. Children with severe neurosensory (including blindness and deafness) and motor impairment were likely to be underrepresented in the cohort because 13 studies (contributing to 55% of the final sample) excluded children who were unable to complete the assessments as a result of their physical disabilities.5,12–15,17,19,21,29,33–35,37 The actual number of children excluded from the analysis for this reason is unknown because not all studies provided this information. Study participants were assessed between the ages of 18 and 40 months using the BSID in 13 studies, the Griffiths Mental Development Scales in 6 studies, the Stanford-Binet Intelligence Scale in 1 study, and the Brunet-Lezine Scale in 1 study. In 3 studies,16,23,29 >1 of these assessment tools were used. School-age cognitive assessments were conducted between the ages of 5 and 18 years, and 11 different tests were used.
The proportion of children diagnosed with developmental impairment (test scores >1 SD below the standardized or control group mean) varied widely among studies, ranging from 6.0%14 to 67.0%.6 The reported prevalence of school-age cognitive deficit was between 5.0%17 and 67.4%26 for mild-moderate (1–2 SD below mean) and 0.0%17,18 and 37.8%26 for severe impairment (>2 SD below mean). In 5 studies,7,21,27,28,34 the categorization of outcomes was based on the mean and SD of the scores achieved by concurrently recruited term-born controls. Wolke et al41 used cohort-specific cut-off points derived from a normative sample representative of the total population of infants in the Bavarian region to categorize impairments. It should be noted that the study population in Smith et al34 was from low to middle socioeconomic groups, and the mean test score achieved by the control group was about 0.5 SD below the normative mean. Using the results from the control group in this case could lead to an underestimation of the prevalence of impairment in this study. If the test standardized norm values were used, the prevalence of cognitive impairment diagnosed at 8 years of age would increase from 24.0% to 36.0% for mild-moderate impairment and from 6.0% to 6.6% for severe impairment.
Bias and Applicability of Included Studies
The proportions of studies considered to be at “low,” “high,” and “unclear” risk for bias and applicability concerns according to the QUADAS-2 appraisal are displayed in Fig 2. The quality of an individual study and the reasons for being considered at high risk for bias or of concern for applicability are detailed in Supplemental Table 4. The loss in follow-up of >30% of the eligible birth cohort was a main source of selection bias in the included studies. Although the overall risk of bias was low, the applicability of the results to our current population of preterm infants is concerning. This is because many of the included studies were conducted >20 years ago; the characteristics of the study populations would be different now, and some of the assessment tools used have been superseded by newer versions.
Predictive Validity of Early Developmental Assessment
The sensitivities, specificities, and PPV and NPV of early assessment for identifying any and severe cognitive deficit estimated from each study are presented in the forest plots in Fig 3. In studies where participants were examined at different time points within the 2 age ranges we studied, only the results from the assessment performed at the oldest age are presented. This gives a final sample size of 3060 children for the meta-analysis.
There was significant heterogeneity in the reported sensitivities and specificities among studies (P < .001 for both). The estimated sensitivities of diagnosing any impairment ranged from 17.0% to 90.5%. There appears to be a wider range and poorer precision (wider Confidence Intervals [CI]) in the estimated sensitivity than specificity across studies. This may reflect the presence of heterogeneity or may be due to estimates of sensitivity being based on smaller samples than estimates of specificity. The HSROC curves providing the pooled measures are presented in Fig 4. The summary points corresponded to a pooled sensitivity of 55.0% (95% CI, 45.7%–63.9%) and a pooled specificity of 84.1% (77.5%–89.1%) for the identification of any impairment. For the diagnosis of severe impairment, the pooled sensitivity was 39.2% (26.8%– 53.3%) and pooled specificity was 95.1% (92.3%–97.0%). Because the BSID-II was the most commonly used developmental test, a post-hoc meta-analysis of the subgroup of 11 studies that only used this tool for early developmental assessment showed a pooled sensitivity of 54.9% (39.5%–69.3%) for any impairment and 43.6% (23.5%–66.0%) for severe impairment; the corresponding specificities were 84.3% (70.1%–92.5%) and 96.4% (90.0%–98.8%). These values are similar to the results for the whole group.
None of the study-level variables examined (gestational age, birth weight, ages at assessments, time interval between assessments, and year of birth) were associated with sensitivity or specificity and therefore did not explain the heterogeneity present between studies (Table 2).
PPV estimates were most precise (narrower CIs) in studies in which the prevalence for any impairment was >40%, and ranged from 63.0% to 80.6%. For impairment prevalence >40%, PPV estimates for the prediction of any cognitive impairment were between 20.0% and 88.9%. In general, the NPV of early developmental assessments were high (range for “any impairment,” 47.8%–95.5%), particularly in predicting the absence of severe impairment (NPV range for “severe impairment,” 68.9%–100%).
Significance testing confirmed that asymmetry was not present in the funnel plot of the log DOR against the inverse of the square root of the ESS (Fig 5, P = .22), indicating the absence of sample size–related effects in the meta-analysis.
Through a systematic review of the literature, we found a substantial number of studies published in the past 20 years that have reported the early neurodevelopmental outcomes and later school-age cognitive abilities of children born very preterm or with VLBW. Although early assessments were generally accurate in predicting the absence of school-age cognitive deficits (high NPV), the identification and prediction of children who would have cognitive difficulties were weak. Meta-analysis of the data suggested that almost half of children who might experience cognitive difficulties at school-age were classified as having normal neurodevelopmental function at ages 1 to 3 years. Even for cases of severe cognitive deficit, the accuracy in early detection was low (meta-analytic sensitivity of 39.2%).
This review sought to answer a clinically relevant question that, for individual cohort studies, would involve lengthy follow-up and significant resources. One of the key strengths of the review is the systematic and comprehensive literature search that is highly sensitive in capturing all available data relevant to the research question in different settings. Because the sensitivity estimates from individual studies were based on small numbers of participants with cognitive impairment, the corresponding 95% CIs were very wide. The use of a meta-analytic approach increases the sample size and improves the precision of the pooled estimate.
However, we recognize weaknesses in our study. It is possible that the included studies represent a biased sample because a large number of eligible studies were not included because of nonresponse, refusal, or the data was no longer accessible. However, the nonincluded studies share similar study characteristics (year of publication, countries where the studies were conducted, inclusion criteria, assessment tools) to those in the review, and the funnel plot symmetry confirmed no publication bias by sample size. Hence, there is no reason to presume that different conclusions would be drawn. We used available aggregated data and were unable to verify data accuracy. Although we had attempted to focus the review on studies published since 1990, only 14 of the 24 included studies recruited participants born after 1990, and no participant was born in the last 10 years. The past couple of decades have seen an overall reduction in the proportions of survivors of very preterm birth with adverse neurodevelopmental outcomes at age 2 years,45–47 so we can expect the characteristics of the current preterm population to be different than those from past eras. Furthermore, the assessment tools used in the included studies, although validated and contemporary at the time of each study, have mostly been superseded by newer editions. For example, the BSID is now in its third edition.48 Recent studies have suggested that children achieve higher scores on the third edition of the Bayley Scales compared with the second edition when concurrently tested with both versions.49,50 Therefore, caution should be exercised when extrapolating from results based on earlier versions of the assessment tools. Although psychometric property differences exist, all the assessment tools provide comparative information of an individual’s development in reference to age-appropriate normative data on the same scale.
We investigated the source of heterogeneity between studies using meta-regression. This method has a few drawbacks. The statistical power to detect associations between the study estimates and the explanatory variables is related to the magnitude of the relationship between them, and is typically considered low in meta-regression.51 This was compounded by the narrow range of values available for each of the explanatory variables under evaluation. For example, the mean gestational age of the included studies ranged from 25.9 to 33.1 weeks. Hence, we cannot exclude the possibility of a type II error. More importantly, meta-regression is subjected to ecological fallacy (or aggregation bias). Therefore, to identify factors reliably that influence the validity of early developmental assessments, it would be necessary to use individual patient-level data.
Early intervention programs, initiated within the first 12 months of postterm life, are known to promote neurodevelopment among preterm infants.52 We do not have the necessary information on whether study participants were offered or received early intervention to evaluate its effect. A Cochrane review of 25 randomized controlled trials of early intervention programs reported that the cognitive benefit observed in infancy and at preschool age did not persist into school age.53 However, most of these programs terminate within the first year after birth, and little work has been done on cognitive rehabilitation or training programs that are sustained beyond toddlerhood. Neurodevelopmental assessment at 2 or 3 years of age is often used as the endpoint for postdischarge follow-up of very preterm or VLBW infants. Depending on the diagnosis at this stage, children are either referred for further intervention and support or discharged from follow-up. Reassuringly, we found the false-positive rate for early diagnosis of impairment to be low. It is likely that children with more severe impairments would be correctly identified at this stage. However, children with milder impairments, who are harder to diagnose, may miss out on the potential advantages of cognitive intervention programs.
Cognitive function in infancy is a poor predictor of later IQ in the general population.54 This may reflect real changes in cognitive function during childhood, unveiling of deficits in complex task performance that were nonessential in early childhood, or the increasing effect of social and environmental influences on cognitive outcomes over time. Other explanations may be the impact of behavior and attention during testing at different ages as well as the differences in the content and psychometric properties of early neurodevelopmental and later cognitive assessment tools. It has been reported that IQ scores from childhood to adulthood were more stable for very preterm/VLBW than for term-born individuals, particularly among those with severe cognitive impairment.55
In 2013, a meta-analysis similar to our study on the predictive value of the BSID on later very preterm and/or VLBW outcomes was published by Luttikhuizen dos Santos et al.56 They reported a strong positive correlation between BSID MDI in the first 3 years after birth and later cognitive scores (pooled correlation coefficient: 0.61; 95% CI, 0.57–0.64) that accounted for 37% of the variance in cognitive functioning. There are several important methodological differences between this meta-analysis and our study. Only studies using the BSID were included in the Dutch meta-analysis and studies published before 1990 were not excluded. The meta-analysis incorporated early neurodevelopmental data obtained before the age of 1 year and nearly half of the follow-up data were based on testing before school age. The convergent validity of MDI scores and cognitive scores may reflect the short interval between testing in this case. More crucially, the statistical measures used in our study (sensitivity and specificity) and the published meta-analysis (correlation coefficient) evaluate different test properties. Although sensitivity and specificity assessed the stability of diagnosis defined as a dichotomous variable, correlation coefficient measures the strength and direction of a linear relationship between 2 continuous variables. In a hypothetical scenario where the 1-year BSID MDI always fall 20 points below the IQ measured at 10 years, the measured correlation would be perfect, but the sensitivity would still be poor.
Early neurodevelopmental assessment has high specificity and NPV, but low sensitivity in identifying later school-age cognitive deficit. A significant number of older children and adolescents born very preterm or VLBW experience difficulties in school and are a group that might have benefitted from earlier support and intervention had their cognitive deficits been recognized. We would encourage future studies of the factors affecting the diagnostic and predictive accuracy of early neurodevelopmental assessments so as to identify follow-up schedules that have a maximal likelihood of detecting impairment.
Members of the Medicines for Neonates Investigator Group are Prof Deborah Ashby (Imperial College London, United Kingdom), Prof Peter Brocklehurst (University College London, United Kingdom), Prof Kate Costeloe (Queen Mary University of London, United Kingdom), Prof Elizabeth Draper (University of Leicester, United Kingdom), Prof Azeem Majeed (Imperial College London, United Kingdom), Prof Neena Modi (Imperial College London, United Kingdom) Prof Stavros Petrou (University of Warwick, United Kingdom), Prof Alys Young (University of Manchester, United Kingdom), Mrs Jane Abbott and Ms Zoe Chivers (Bliss Charity, London, United Kingdom), and Mrs Jacquie Kemp (London, United Kingdom).
We thank the following investigators who contributed unpublished and supplemental data for the review: Dr Haim Bassan (Tel Aviv University, Israel), Prof Arend Bos (Beatrix Children’s Hospital, Groningen, Netherlands), Dr Marie-Laure Charkaluk (Lille Catholic University, Paris Descartes University, France), Dr Sarale Cohen (retired), Prof Olaf Dammann (Tufts University, Boston, MA), Prof Linda de Vries (University Medical Centre, Utrecht, Netherlands), Prof Ermellina Fedrizzi (University of Padua, Italy), Prof Vineta Fellman (Lund University, Sweden), Prof Peter Gray (Mater Mothers’ Hospital, Brisbane, Australia), Prof Howard Kilbride (University of Missouri-Kansas City, Kansas City, MO), Prof Neil Marlow (University College London, United Kingdom), Prof Jennifer Pinto-Martin (University of Pennsylvania, Philadelphia, PA), Prof Jon Skranes (Norwegian University of Science & Technology, Trondheim, Norway), Prof Karen Smith (The University of Texas Medical Branch, Galveston, TX), Prof Mary Sullivan (University of Rhode Island, Kingston, RI), Prof Paul Swank (The University of Texas Health Science Center at Houston, Houston, TX ), Prof H. Gerry Taylor (Case Western Reserve University, Cleveland, OH), Dr Viena Tommiska (Helsinki University Central Hospital, Finland), Dr Norbert Veelken (Asklepios Klinik Nord, Hamburg, Germany), Prof Dieter Wolke (University of Warwick, United Kingdom), and Prof Lianne Woodward (Harvard Medical School, Boston, MA).
- Accepted May 24, 2016.
- Address correspondence to Neena Modi, MD, Section of Neonatal Medicine, Imperial College London, Chelsea & Westminster Hospital Campus, 369 Fulham Rd, London SW9 1NH, United Kingdom. E-mail:
This review has been registered in PROSPERO (international prospective register of systematic reviews; http://www.crd.york.ac.uk/PROSPERO/): identifier CRD42012002168.
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
FUNDING: This paper represents independent research funded by the National Institute for Health Research (NIHR) under its Programme Grants for Applied Research Programme (Reference RP-PG-0707-10010). The views expressed are those of the authors and not necessarily those of the National Health Service, the NIHR, or the Department of Health.
POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no potential conflicts of interest to disclose.
- Marlow N
- Doyle LW,
- Anderson PJ
- Munck P,
- Niemi P,
- Lapinleimu H,
- Lehtonen L,
- Haataja L; PIPARI Study Group
- Hack M,
- Taylor HG,
- Drotar D, et al
- Roberts G,
- Anderson PJ,
- Doyle LW; Victorian Infant Collaborative Study Group
- Deeks JJ
- Bruggink JL,
- Van Braeckel KN,
- Bos AF
- Gray PH,
- Burns YR,
- Mohay HA,
- O’Callaghan MJ,
- Tudehope DI
- Johnson S,
- Fawke J,
- Hennessy E, et al
- McGrath MM,
- Sullivan MC,
- Lester BM,
- Oh W
- Pinto-Martin J,
- Whitaker A,
- Feldman J, et al
- Smith KE,
- Landry SH,
- Swank PR
- Tommiska V,
- Heinonen K,
- Kero P, et al
- Mikkola K,
- Ritari N,
- Tommiska V, et al
- Gutbrod T,
- Wolke D,
- Soehne B,
- Ohrt B,
- Riegel K
- Saigal S,
- den Ouden L,
- Wolke D, et al
- Salt A,
- D’Amore A,
- Ahluwalia J, et al; East Anglian Very Low Birthweight Project Group
- Wilson-Costello D,
- Friedman H,
- Minich N, et al
- Doyle LW,
- Roberts G,
- Anderson PJ; Victorian Infant Collaborative Study Group
- Bayley N
- Bossuyt P,
- Davenport C,
- Deeks J,
- Hyde C,
- Leeflang M,
- Scholten R
- Spittle A,
- Orton J,
- Anderson PJ,
- Boyd R,
- Doyle LW
- Aylward GP
- Breeman LD,
- Jaekel J,
- Baumann N,
- Bartmann P,
- Wolke D
- Luttikhuizen dos Santos ES,
- de Kieviet JF,
- Königs M,
- van Elburg RM,
- Oosterlaan J
- Copyright © 2016 by the American Academy of Pediatrics