Derivation and validation of a mathematical model for predicting the response to exogenous recombinant human growth hormone (GH) in prepubertal children with idiopathic GH deficiency

Postmarkteting surveillance studies of recombinant human GH therapy, such as the Rabi Pharmacia International Growth Study (KIGS; Pharmacia & Upjohn, Inc,, international Growth Database), have accumulated extensive data concerning the characteristics and growth outcomes of children with various causes of short stature. These data provide an opportunity to analyze the factors that determine responsiveness to GH and allow the development of disease-specific growth prediction models. We undertook a multiple regression analysis of height velocity (centimeter per yr) with various patient parameters of potential relevance using data from a cohort of 593 prepubertal children with idiopathic GH deficiency (GHD) from the KIGS database. Our aim was to produce models that would have practical utility for predicting prepubertal growth during each of the first 4 yr of GH replacement therapy. These models were validated by a prospective comparison of predicted and observed growth outcomes in an additional 3 cohorts of prepubertal children with idiopathic GHD: 237 additional KIGS patients, 29 patients from the Australian OZGROW study, and 33 patients from Tubingen, Germany. The most influential variable for first year growth response was the natural log (ln) of the maximum GH response during provocation testing, which was inversely correlated with height velocity. The first year growth response was also inversely correlated with chronological age and height so score minus midparental height so score. First year growth was positively correlated with body weight SD score, weekly GH dose (ln), and birth weight SD score. Two first year models were developed using these parameters, 1 including and 1 excluding the maximum GH response to provocative testing. The former model explained 61% of the response variability, with a SD of 1.46 cm; the latter model explained 45% of the variability, with a SD of 1.72 cm. The two models gave similar predictions, although the model excluding the maximum GH response to testing tended to underpredict the growth response in patients with very low GH secretory capacity. For the second, third, and fourth year growth responses, 4 predictors were identified: height velocity during the previous year (positively correlated), body weight so score (positively correlnted) chronological age (negatively correlated. and weekly GH dose tin, positively correlated). The models far the second, third, and fourth year responses explained 40%, 37% and 30% of the variability, respectively, with SDS of 1.19, 1.05, and 0.95 cm, respectively. When the models were applied prospectively to the other cohorts, there were no significant differences between observed and predicted responses in any of the cohorts in any year of treatment. The fourth year response model gave accurate prospective growth predictions for the fifth to the eighth prepubertal years of GH: treatment in a subset of 48 KIGS patients. Analyses of Studentized residuals provided further validation of the models. The parameters used in our models do not explain all of the variability in growth response, but they have a high degree of precision (low error sos). Moreover, the parameters used are robust and easily accessible. These properties give the models' practical utility as growth prediction tools. The availability of longitudinal, disease-specific models will be helpful in the future for enabling growth-promoting therapy to be planned at the outset, optimized for efficacy and economy, and individualized to meet treatment goals based on realistic expectations.