Jonathan L. French (1), Amy Racine (2), Stef van Buuren (3), Jonas Haggstrom (4), and the Lifecycle, Auxology and Neurocognitive Development Surge Team (5) – representing the Healthy Birth, Growth and Development knowledge integration (HBGDki) community
(1) Metrum Research Group, Tariffville, CT, USA. (2) Novartis, Inc., Basel., Switzerland, (3) TNO, Utrecht, Netherlands, (4) Holy Diver Consulting, Stockholm, Sweden, (5) Sponsored by the Bill & Melinda Gates Foundation, Healthy Birth, Growth and Development Initiative.
Objectives: Assessment of neurocognitive development during the first 1000 days of life is important, particularly in children in low- and middle-income countries (LMIC). There are a variety of instruments used for these assessments, most of which are based on a defined set of tasks for the child to perform. Tasks are typically scored as a set of ordered categories. It has been proposed that the development score (D-score) is one way to integrate data collected using different scales and across different populations [1,2]. The objectives of this work were to evaluate the assumptions underlying the D-score using data from an LMIC population and to assess if the D-score can be used for between-population comparisons.
Methods: We connect a child’s D-score to the observed longitudinal outcomes through a Rasch model [1], a type of item-response theory model. Key assumptions of the Rasch model include invariance to the set of items used and common item-level difficulty across populations. Using data from a high income country (HIC) longitudinal study, the item-level difficulty values were previously estimated [1,2]. We evaluate the assumption of parameter invariance by comparing the estimated D-score based on the full set of items and matching items in the LMIC study instrument. We use discrimination plots to compare item difficulty and item discrimination across studies. Finally, we compare the longitudinal D-scores between the HIC and LMIC study populations.
Results: Comparison of the D-score based on the full-set and matched-set of items in the HIC study shows high correlations both overall and by age. Discrimination plots reveal that while many of the assumptions of the Rasch model hold, some difficulty parameters may differ across HIC and LMIC populations. These differences appear to be driven by items relating to language and motor skills. Compared to the HIC study, the D-scores in the LMIC study show average development at 6 months, higher than average scores at 15 months, and lower than average scores at 24 months.
Conclusions: The D-score shows promise for facilitating comparisons across populations. However, it has not been clearly validated for this purpose. While the D-score appears to be invariant to the choice of items, the item-level difficulties may depend on the population and/or instrument used to collect them. Additional work is needed to more fully evaluate the D-score, including comparisons using additional populations and neurocognitive development instruments.
References:
[1] Jacobusse G, Van Buuren S, and Verkerk PH. An interval scale for development of children aged 0-2 years. Stat Med 2006; 35(13): 2272-2283
[2] Van Buuren, S. Growth charts of human development. Stat Methods Med Res 2014; 23(4): 346-368
Reference: PAGE 25 () Abstr 5977 [www.page-meeting.org/?abstract=5977]
Poster: Methodology - Other topics