Confounds in Developmental Scores When Correcting for Prematurity Free

Definitive guidelines regarding correction for prematurity^1–3  are lacking. The American Academy of Pediatrics recommends correction to age 3,^4,5  but this is not implemented consistently. Correction becomes more critical with the decreasing age of viability.

Differences between corrected and uncorrected scores are greatest in infants born <28 weeks.^6,7  Rates of disabilities also increase at these gestational ages, questioning whether correction simply masks developmental deficits,⁸  causing them to be misinterpreted as temporary delays. Correction is applicable to all preterm infants; children with lower gestational ages at birth require extended correction.

The rapid developmental growth curve in infancy requires narrow normative age bands; childhood tests have several-month age bands. The assumption is that development is more rapid early on and the slope slows at later ages. On the Bayley-4,⁹  10 day age bands are found up to 5 months to 15 days, 30 day bands up to 36 months to 15 days, and 3 month intervals to 42 months to 30 days. For example, a child born at 26 weeks and tested at 12 months receives a corrected standard score of 100 but the uncorrected score is 85 (1 SD difference).

With early childhood testing, changes in the size of the age band norms are often overlooked. The Stanford-Binet-V for Early Childhood¹⁰  has 1 month age bands from 2 to 5 years and 3 month bands from 5 years onward. Therefore, if outcome is at 5 years, some children would fall into the 1-month band, whereas others would be compared with the 3 month age bands. The Differential Abilities Scales-II¹¹  has 3 month age bands up to 8 to 11. Despite consistent band widths, children born <28 weeks could span >1 age band. The Wechsler Preschool and Primary Scales of Intelligence IV¹²  has age bands ranging from 3 to 6 months, again adding variability.

With uncorrected scores, very or extremely preterm infants are at greater disadvantage if their age falls at the beginning of a 3 month age band because the child could be 2 to 3 months younger than one whose age falls in the upper end of the age band.¹³  Scores may also change when age correction is used at younger ages but discontinued at the later chronological ages of 3 to 4 years.

Norming groups are typically established on the basis of samples within discrete age bands that can vary widely. Discontinuities, evident in gaps or abrupt jumps in norms tables, may create a situation where the raw score yields the same scaled score for children within the age band but then jumps markedly for children in the next age band. Moreover, the test score distribution is only constant within the age band when it is very narrow and the growth curve is not too steep. Selecting the size of the age band for norms creation is equal parts psychometric science and clinical art.

The jumps between norms at successive age levels are exacerbated when correcting for prematurity. Raw test scores can have different interpretations with only a day or 2 difference in age. For example, on the Bayley-4, if the child’s age was 12 months, 15 days, a raw Cognitive score of 57 yields a standard score of 85 (16^th percentile); at 12 months, 16 days, the same raw score yields a score of 80 (ninth percentile),¹⁴  (0.33 SD difference). At some ages, the effect of a 1 day difference can be more substantial and could mean the difference between a moderate versus severe neurodevelopmental impairment classification.

One possible solution is to apply linear regression to predict average growth curves within and across age bands. Referred to as “continuous norming,” this approach may reduce the effect of wide window norms on correction.^15,16  This technique is particularly useful if the growth curve is steep, such as in the earliest months and years of development. The regression curve is plotted across the means by age; similarly, a curve is also fitted to the SDs across age. The fitted mean and fitted SD at any given day, week, or month of age can be used to produce norms for that age. The regression models also take into account the shape of the data distribution. The fitted mean and SD could be averaged to produce norms for different intervals (eg, weekly). Continuous norming does not assume development is static within an age band but conceptualizes age as continuous and models raw scores as a smooth, nonlinear function of age.

Advantages to regression-based continuous norming are that norms are more realistic and smaller sample sizes are needed.¹⁶  However, the number of tables necessary for the age interval selected require computerized scoring, which is a drawback. As a result, weekly or monthly norms should be considered for young and older children, respectively.

A major consideration is whether assessment is for clinical or research purposes. Corrected scores should be used into adolescence in research studies, but uncorrected scores are appropriate for clinical purposes in children aged >5. The child’s level of function is important: the likelihood of a major difference in corrected and uncorrected scores is minimal at ≥2 SDs below average. The effect of adjustment will be greater in children who are borderline to low average or above. In many venues, it is helpful to provide both corrected and uncorrected scores.^1,6 

The appropriateness of correction may depend on whether cognitive, motor, or language domains are involved.⁶  Age adjustment for prematurity was reported to overrate motor performance on the Bayley-III¹⁷  12 month assessment, but not cognitive performance. Other studies reported conflicting results, with correction being needed to bring motor milestones more in line with normative data.¹⁸  Language performance data are inconclusive.⁸  Using chronologic versus corrected age will affect categorical diagnoses and eligibility for intervention services.

In a theoretical model, Wilson-Ching et al²  used the Bayley-III normative data to identify cognitive raw scores that produced baseline standard scores of 70, 85, and 100, and compared these scores for 1, 2, 3, and 4 months of prematurity (assessing what the same raw score values are if the child was 1 to 4 months older). These comparisons were made at 6, 12, 24, and 36 months. Differences between scores at 1 and 2 months of prematurity washed out after 2 years; however, at 3 and 4 months of hypothetical prematurity, a 5-point difference in the cognitive composite score was still found at 36 months.

Aylward¹  employed a similar approach with Bayley-4 data, evaluating cognitive, motor, and language function at 6, 12, 24, and 36 months. Using a cutoff of 3 points difference in scores (0.20 SD) correction was needed at all ages over the first 3 years for language and motor function. With respect to the cognitive domain, correction was needed up through 24 months, but cognitive age adjustment was necessary at 36 months for those born 4 months early.

These data were obtained from two nationally representative, contemporary data sets of typically developing children without medical risk; however, the samples did not include preterm infants. Nonetheless, these data suggest a greater likelihood of correction addressing delay versus deficit.