Late Neuroimaging Predicts Neurodevelopmental Impairment at 18-22 Months in Preterm Infants Born at 24-27 Completed Weeks’ Gestation


Hintz SR, Barnes PD, Bulas D, et al. Neuroimaging and neurodevelopmental outcome in extremely preterm infants. Pediatrics 2015;135:e32-42. PMID: 25554820.


John Flibotte, MD
Assistant Professor of Clinical Pediatrics
Children’s Hospital of Philadelphia &Perelman School of Medicine at the University of Pennsylvania




In a population of infants born at gestational age 24-27 completed weeks, how accurately do early cranial ultrasound (CUS), late CUS and near term MRI findings predict neurodevelopmental impairment (NDI) outcomes at 18-22 months? Furthermore does an MRI at near term enhance predictive ability of NDI at 18-22 months over late CUS findings alone?


      • Design: Nested Prospective Cohort, where cohort was nested within SUPPORT Randomized Controlled Trial1,2
      • Allocation: Observational, does not apply.
      • Blinding: Two central readers, who were masked to patient outcomes, reported all neuroimaging results. Certified examiners who determined the primary outcome were unaware of treatment allocations or clinical course of the infants.
      • Follow-up period: 18-22 months’ corrected age
      • Setting: 16 Neonatal Research Network Centers between 2005-2009
      • Patients:
        • Nested cohort of the SUPPORT Trial (NCT00233324)1,2
          • 24 to 27 completed gestational weeks
          • Enrolled in SUPPORT
          • MRI was obtained at 35 to 42 weeks’ gestational age, and only infants for whom the MRI was obtained within 2 weeks of the CUS were included for analysis
      • Intervention: The study question centered on the ability of two different modalities of neuroimaging, CUS and MRI, to predict outcomes (see below). To evaluate the importance of timing of the CUS, investigators considered both “early” CUS (4-14 days of life) and “late” CUS (35-42 weeks postmenstrual age). MRI studies were obtained typically at 35-42 weeks postmenstrual age and ideally within 7 days of the late CUS. MRI results were only considered if obtained within 2 weeks of the late CUS. It is not clear from the methods section of the paper whether imaging was obtained per protocol or based on clinical grounds.Imaging results were analyzed based on presence or absence of major findings. Major findings were defined for each imaging modality as follows:
        • Early CUS major findings: ICH Grade III/IV; cystic PVL
        • Late CUS major findings: cystic PVL; porencephalic cysts; moderate (Ventricular: brain ratio of 1:3-2:3) to severe (>2:3) ventricular enlargement on either or both sides; ventricular shunt
        • MRI major findings: moderate or severe white matter abnormality (WMA), as previously defined3; significant cerebellar lesions.
        • All images were read centrally by 2 observers with high inter-rater reliability (US Kappa 0.75-0.88; and MRI 0.96).
      • Outcomes: This was a non-interventional observational cohort study. There were two primary outcomes:
        • Neurodevelopmental impairment (NDI) or death
        • NDI defined as: Bayley III score <70; moderate-to-severe CP; GMFCS > 2, severe hearing impairment, or bilateral severe visual impairment
        • Significant gross motor impairment or death
        • Gross motor impairment defined as: moderate-to-severe CP or GMFCS >2 regardless of formal diagnosis of CP
        • Secondary analyses included association of imaging findings with the following additional outcomes: cognitive score; cognitive score <85; cognitive score <70; any CP; moderate to severe CP; NDI; significant gross motor impairment; unimpaired/ mildly impaired; NDI or death.
      • Analysis and Sample Size:
        • Sample size was not calculated, with empirical accrual of all infants in the SUPPORT trial that met the entry criteria. 480 infants met inclusion criteria of having late CUS and brain MRI within 2 weeks of each other.
        • Unadjusted analysis for association between imaging findings and outcomes were performed using standard methods: chi-squared, Fisher’s exact test or ANOVA.
        • Stepwise modeling was used to construct receiver operator characteristic (ROC) curves to establish associations with the 2 binary outcomes established a-priori (i.e. NDI or death & significant motor impairment or death). The following 4 factors were added in stepwise fashion with calculation of resulting areas under the curve (AUC): 1) perinatal/ neonatal factors; 2) Early CUS composite adverse findings; 3) Late CUS composite adverse findings; 4) MRI adverse findings.
        • Of note, the perinatal factors considered in constructing ROC curves for each outcome of interest were selected from a predefined list of 12 total factors if they had proven association with the outcome when explicitly tested. For the outcome of NDI or death, the perinatal factors included were: race, late sepsis, BPD and postnatal steroids. For the outcome of significant gross motor impairment or death, the perinatal factors included were: race, multiple gestation, maternal insurance, late sepsis, BPD and postnatal steroids.
      • Patient follow-up: 480 infants were enrolled and had complete neuroimaging data including a late CUS and brain MRI within 2 weeks of that ultrasound. 15 infants died before 18 months’ correct age and no follow-up was obtained for them but they were included in the composite outcomes. 20 infants were lost to follow-up; a BSID III was obtained for 441 children and neurosensory exam for 445. Therefore, follow-up was 456/480 (95%) for the outcome of NDI or death and 460/480 (96%) for the outcome of gross motor impairment or death.


Independent associations of neuroimaging findings and outcomes are presented in Table 6 and tested within full and limited models. These models were defined by which imaging results were included (see below). The notable findings included:

In a full model, where all types of imaging were included:

  1. Significant association of late CUS adverse findings with both NDI or death and significant gross motor impairment or death (OR 9.8 and 10.9 respectively)
  2. Significant association of cerebellar lesions on MRI with both NDI or death and significant gross motor impairment or death (OR 3 and 5.2 respectively)

In limited models:

  1. Excluding late CUS, MRI with moderate or severe WMA & MRI with significant cerebellar lesions were significantly associated with NDI or death (OR 2.4 and 2.7, respectively) and significant gross motor impairment or death (OR 2.8 and 4.5, respectively)
  2. Excluding MRI, late CUS was significantly associated with NDI or death (OR 11.9) and significant gross motor impairment or death (OR 13.2)

The authors presented their results in Table 7, as follows:

Outcome Model Variables AUC 95% CI
NDI or death Perinatal/neonatal 0.743 0.67-0.82
Perinatal/neonatal+ Early CUS 0.773 0.70-0.84
Perinatal/neonatal+ Early + Late CUS 0.800 0.73-0.87
Perinatal/neonatal+ Early CUS+ MRI 0.809 0.75-0.87
Perinatal/neonatal+ Early+ Late CUS+ MRI 0.825 0.76-0.89
Significant gross motor impairment or death Perinatal/neonatal 0.833 0.75-0.92
Perinatal/neonatal+ Early CUS 0.859 0.79-0.93
Perinatal/neonatal+ Early + Late CUS 0.885 0.82-0.95
Perinatal/neonatal+ Early CUS+ MRI 0.892 0.83-0.96
Perinatal/neonatal+ Early+ Late CUS+ MRI 0.908 0.85-0.97

The authors did not present data for traditional measures of diagnostic testing, including sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV).


Results of early neuroimaging were not predictive of future neurodevelopmental outcomes in any of the models, highlighting the importance of near term neuroimaging for formulating prognosis. Importantly, near term MRI did enhance predictive ability over late cranial ultrasound, but only minimally. In environments where MRI is not easily available, late cranial ultrasound alone may be sufficient.


Prognosis is an important part of neonatal practice and many care decisions are made in the context of predictions of neurodevelopmental outcomes.4 Traditionally, information obtained from bedside ultrasound has been used as the basis of these projections, largely based on early scans. However, the expanding availability and use of MRI has allowed for potentially improved ability to predict.3,5-7

Prior studies had suggested that late US close to discharge was a better marker for accurate prediction, and indeed it was used as part of the outcome cluster in randomized controlled trials of various therapies.8 Now in a large study, Hintz et al. present data from a cohort of preterm infants in the post-surfactant era that demonstrates that late imaging is more reliable than early imaging at predicting neurodevelopmental outcomes. Importantly, they provide longitudinal comparisons within the same cohort of the diagnostic performance of early ultrasound, late (term equivalent) ultrasound, and term equivalent MRI. Interestingly, they demonstrate that the ability to predict outcomes is only mildly enhanced with the addition of MRI to late ultrasound results.

Several models are presented in a stepwise fashion and the ability to predict outcomes is presented as calculated areas under the curve and adjusted odds ratios. However, they do not present estimates of sensitivity, specificity, positive/ negative predictive values or likelihood ratios. Clinicians are not always used to thinking of odds ratios, and particularly where a diagnostic pathway can be used, it may help to use an alternative strategy. Based on the data in Tables 2-5, the following measures can be calculated for the compound outcome of neurodevelopmental outcome or death:

Modality and Findings Sens Spec PPV NPV +LR -LR
Early CUS: Major Findings vs No Major Findings 0.27 0.92 0.31 0.91 3.5 0.79
Late CUS: Major Findings vs No Major Findings 0.28 0.97 0.54 0.91 8.8 0.74
MRI with Any WMA vs None 0.92 0.23 0.13 0.96 1.2 0.32

In a clinical context, the ability of testing to alter decision-making is most often summarized by the above measures. Higher positive or lower negative likelihood ratios associated with a test have greater influence on changing pre-test probability of a condition when the test results are present or absent.9,10 These values enable the clinician to use the local point prevalence of disorders to then find on a nomogram the likely post-test probability. From the above table, it is apparent that late CUS with major findings provides the greatest positive likelihood ratio, in excess of MRI. In keeping with its known higher sensitivity, MRI provides the most substantial negative likelihood ratio.

Ultrasound has long been the mainstay for predicting future outcomes and has shown good specificity when major findings are evaluated. For the outcome of cerebral palsy (CP), De Vries et al11 demonstrated that serial ultrasound can predict future CP outcome with specificity in excess of 90% and sensitivity of 70-80% depending on the risk profile of the population in which it is used. However, ex preterm neonates presenting with later developmental impairments in spite of normal ultrasounds in the NICU12 have led some to advocate for MRI prior to discharge to more accurately predict risk profile, particularly for subtle neurodevelopmental impairments.3,6 The sensitivity and specificity of MRI in Hintz et al. is similar to that reported by Woodward et al. (92% vs 84% and 23% vs 34%, respectively).3

The strengths of this study are the large population size, the blinded evaluations of imaging, and the required temporal proximity of late CUS and MRI. It is unclear from the methods, whether the imaging was obtained according to protocol. However, if the imaging was obtained based on clinical grounds, there is inherent bias introduced. The authors themselves point out limitations including the overall low rates of adverse outcomes of infants in this study. If the rates of adverse outcomes were higher, the positive predictive value of all modalities might be improved. The authors also provided extensive modeling that included consideration of factors that are known to be important in predicting outcomes, including perinatal history and postnatal co-morbidities13, together with imaging. Such analysis demonstrates the potential clinical value of this work in the full context within which it would be applied rather than analyzing imaging findings in isolation. One potential drawback to this approach is the possibility of over-analysis, given the number of co-factors considered. Finally, they demonstrate that conventional MRI only marginally improves predictive ability compared to late ultrasound, highlighting the importance of term equivalent ultrasound for all preterm neonates rather than reliance on early ultrasound findings alone.

Regardless of what imaging is chosen, it is clear that no imaging is perfect in formulating predictions. The authors themselves make this point, as does an accompanying editorial that was published in the same issue of Pediatrics and explicitly highlighted the “perils of prediction”.14 In the absence of certainty, the parent perspective and the impact of earlier diagnosis on outcomes are additional factors that warrant further investigation.15,16


  1. SUPPORT Study Group of the Eunice Kennedy Shriver NICHD Neonatal Research Network, Finer NN, Carlo WA, et al. Early CPAP versus surfactant in extremely preterm infants. N Engl J Med. 2010;362(21):1970–1979. doi:10.1056/NEJMoa0911783.
  2. SUPPORT Study Group of the Eunice Kennedy Shriver NICHD Neonatal Research Network, Carlo WA, Finer NN, et al. Target ranges of oxygen saturation in extremely preterm infants. N Engl J Med. 2010;362(21):1959–1969. doi:10.1056/NEJMoa0911781.
  3. Woodward LJ, Anderson PJ, Austin NC, Howard K, Inder TE. Neonatal MRI to Predict Neurodevelopmental Outcomes in Preterm Infants. N Engl J Med. 2006;355(7):685–694. doi:10.1056/NEJMoa053792.
  4. de Vries LS, van Haastert IC, Benders MJNL, Groenendaal F. Myth: Cerebral palsy cannot be predicted by neonatal brain imaging. Seminars in Fetal and Neonatal Medicine. 2011;16(5):279–287. doi:10.1016/j.siny.2011.04.004.
  5. de Vries LS, Benders MJNL, Groenendaal F. Imaging the premature brain: ultrasound or MRI? Neuroradiology. 2013;55(S2):13–22. doi:10.1007/s00234-013-1233-y.
  6. Mirmiran M, Barnes PD, Keller K, et al. Neonatal brain magnetic resonance imaging before discharge is better than serial cranial ultrasound in predicting cerebral palsy in very low birth weight preterm infants. Pediatrics. 2004;114(4):992–998. doi:10.1542/peds.2003-0772-L.
  7. Whyte HEA, Blaser S. Limitations of routine neuroimaging in predicting outcomes of preterm infants. Neuroradiology. 2013;55 Suppl 2:3–11. doi:10.1007/s00234-013-1238-6.
  8. Schmidt B, Davis P, Moddemann D, et al. Long-term effects of indomethacin prophylaxis in extremely-low-birth-weight infants. N Engl J Med. 2001;344(26):1966-72.
  9. Fagan TJ. Letter: nomogram for Bayes theorem. N Engl J Med. 1975;(293):257.
  10. Caraguel CGB, Vanderstichel R. The two-step Fagan’s nomogram: ad hoc interpretation of a diagnostic test result without calculation. Evidence-Based Medicine. 2013;18(4):125–128. doi:10.1136/eb-2013-101243.
  11. de Vries LS, Van Haastert I-LC, Rademaker KJ, Koopman C, Groenendaal F. Ultrasound abnormalities preceding cerebral palsy in high-risk preterm infants. The Journal of Pediatrics. 2004;144(6):815–820.
  12. Laptook AR. Adverse Neurodevelopmental Outcomes Among Extremely Low Birth Weight Infants With a Normal Head Ultrasound: Prevalence and Antecedents. Pediatrics. 2005;115(3):673–680. doi:10.1542/peds.2004-0667.
  13. Schmidt B. Impact of Bronchopulmonary Dysplasia, Brain Injury, and Severe Retinopathy on the Outcome of Extremely Low-Birth-Weight Infants at 18 Months: Results From the Trial of Indomethacin Prophylaxis in Preterms. JAMA: The Journal of the American Medical Association. 2003;289(9):1124–1129. doi:10.1001/jama.289.9.1124.
  14. Eichenwald EC. Neuroimaging of Extremely Preterm Infants: Perils of Prediction. Pediatrics. 2015;135(1):e176-177. doi: 10.1542/peds.2014-2025.
  15. Pearce R, Baardsnes J. Term MRI for small preterm babies: do parents really want to know and why has nobody asked them? Acta Paediatrica. 2012;101(10):1013–1015. doi:10.1111/j.1651-2227.2012.02767.x.
  16. McIntyre S, Morgan C, Walker K, Novak I. Cerebral palsy–don’t delay. Dev Disabil Res Rev. 2011;17(2):114–129. doi:10.1002/ddrr.1106.

1 Comment

  1. It is surprising that the paper did not include estimations of the diagnostic value of MRI and US (presented as sensitivity, specifity, predictive values etc).

    @JohnFlibotte, thanks for a thorough commentary and for including those measurement of diagnostic value.

Comments are closed.