[Prediction of growth from an early age: curve matching with the TNO Growth Predictor]. / Predictie van groei vanaf jonge leeftijd: 'curve matching' met de TNO Groeivoorspeller.

Curve matching is a new big data technique to predict an outcome given earlier measurements. Here we apply curve matching to predict the future growth of a specific child, the target child. The method searches in large datasets of longitudinal growth data for other children who are similar to the target child in terms of factors that influence growth. The observed growth curves of these matched children provide valuable insights into the future growth of the target child. The TNO Groeivoorspeller (TNO Growth Predictor) plots the expected growth of the target child, as well as the uncertainty of the prediction. Curve matching is a general technique that can also be used for other health measures. The key requirement is the availability of relevant longitudinal data on the outcome and its determinants.

An accurate and efficient identification of children with psychosocial problems by means of computerized adaptive testing.

BACKGROUND: Questionnaires used by health services to identify children with psychosocial problems are often rather short. The psychometric properties of such short questionnaires are mostly less than needed for an accurate distinction between children with and without problems. We aimed to assess whether a short Computerized Adaptive Test (CAT) can overcome the weaknesses of short written questionnaires when identifying children with psychosocial problems. METHOD: We used a Dutch national data set obtained from parents of children invited for a routine health examination by Preventive Child Healthcare with 205 items on behavioral and emotional problems (n = 2,041, response 84%). In a random subsample we determined which items met the requirements of an Item Response Theory (IRT) model to a sufficient degree. Using those items, item parameters necessary for a CAT were calculated and a cut-off point was defined. In the remaining subsample we determined the validity and efficiency of a Computerized Adaptive Test using simulation techniques, with current treatment status and a clinical score on the Total Problem Scale (TPS) of the Child Behavior Checklist as criteria. RESULTS: Out of 205 items available 190 sufficiently met the criteria of the underlying IRT model. For 90% of the children a score above or below cut-off point could be determined with 95% accuracy. The mean number of items needed to achieve this was 12. Sensitivity and specificity with the TPS as a criterion were 0.89 and 0.91, respectively. CONCLUSION: An IRT-based CAT is a very promising option for the identification of psychosocial problems in children, as it can lead to an efficient, yet high-quality identification. The results of our simulation study need to be replicated in a real-life administration of this CAT.