Prediction of long-term response to recombinant human growth hormone in Turner syndrome: Development and validation of mathematical models

It has become common practice to apply GH treatment in short Turner syndrom e patients with the objective of promoting growth. The variability in respo nse and the high costs of this treatment demand the individualization and o ptimization of therapy. Based on 686 prepubertal Turner patients from the K abi International Growth Study (KIGS; Pharmacia & Upjohn, Inc. Internationa l Growth Database), we undertook a multiple regression analysis of height v elocity (centimeters per yr) by using various parameters of potential relev ance. Derived prediction models for the first 4 yr of GH treatment were val idated with 76 additional KIGS patients and 81 patients from Tuebingen, Ger many. Among the 6 predictors identified, the most influential variable for first year growth response was the natural log (1n) of the weekly GH dose. The first year growth response was also correlated with age and distance be tween height and target height (SD score; both negative) and body weight SD , number of GH injections per week, and oxandrolone treatment given additio nally (positive). The first year model explains 46% of the variability, wit h 1 SD of 1.26 cm. For the second to fourth years, 5 predictors were identi fied: height velocity during previous years, weekly GH dose (1n), weight so , oxandrolone therapy (all positive), and age (negative). These models expl ained 32%, 29%, and 30% of the variability, respectively, with SD scores of 1.1, 1.0, and 1.0 cm, respectively. When the models were applied to the ot her cohorts, no significant difference was noted between observed and predi cted responses. Although the parameters used in our models do not entirely explain the variability in the growth response in Turner syndrome, the para meters themselves were clinically relevant to our present understanding and proved to be of high precision. Some of the tested markers, such as karyot ype, do not contribute to the growth response. These variables make the mod els practical and suitable for planning beneficial and cost-effective thera py.