Subjects and study protocol
The study was a 2-year prospective, randomized, open-label, multicentre trial in short prepubertal children with isolated GHD or ISS [
22] naïve to GH treatment. Individual GH responsiveness was estimated by our growth response prediction model for children with GHD and ISS [
23], and the patients were randomized in 1:2 proportions to receive either a standard or an individualized GH dose during 2 years of catch-up growth to a preset growth target, mid-parental height SDS (MPH
SDS). Randomization variables included gender, weight
SDS at birth, height
SDS at age 1 year, GH
maxAITT (during an arginine–insulin tolerance test), GH
max24h profile (during 24h spontaneous GH sampling), age and height
SDS at start, the child’s height
SDS difference to its MPH
SDS (diff MPH
SDS) at start, and predicted 1
st year Δ height
SDS[
23]. No patients with syndromes, chronic diseases or complete GH insensitivity were included in the study population.
Upon inclusion, all patients had a height
SDS below −2.0 [
24] and a growth velocity below −1.0 SDS. 128 children (38 girls, 90 boys) followed the protocol [
22]. Both the results from the GH
maxAITT and the GH
max24h profile were used to separate children with GHD from those with ISS. Classic GHD was defined based on a GH
max below 32 mU/L (using polyclonal assay, WHO IRP 80–505) corresponding to 24 mU/L (using monoclonal assay) and equivalent to the ‘old cut-off of 10 μg/L’ [
18]. According to this definition, 39 children had isolated GHD and 89 had ISS. However, when using GH
max results exclusively from the AITT, 90 children were assigned the diagnosis of GHD and 38 the diagnosis of ISS. Additionally, all of our patients fulfilled the criteria for GHD according to a growth velocity lesser than −1.0 SDS, IGF-I below −1 SD of sex and age specific references, bone age retardation of more than 1,5 years [
18,
22].
To address the question concerning GH thresholds and to study the possible dose-dependency, the analysis focused on the individualized treatment group (n=87). We used data from prepubertal children included in a trial, randomized to individual GH doses in the range of 17–100 μg/kg/d according to their growth-related GH responsiveness; with dose-adjustment for the estimated difference to the preset height target (i.e. MPH
SDS) at 2 years on GH in order to make it possible for each child to reach its MPH
SDS within a 2 year period. [
22]. Children with a higher predicted growth response, calculated before start of treatment, received a lower individual GH dose, and those with a lower predicted growth response received a higher GH dose. The GH doses used in the individualized-dose group were 17 μg/kg/day (n=3), 33 μg/kg/day (n=27), 40 μg/kg/day (n=10), 50 μg/kg/day (n=26), 66 μg/kg/day (n=14), and 100 μg/kg/day (n=7): the mean GH dose in this group was 49 μg/kg/day [
22].
The fixed GH dose group was used as a control group for estimating a metabolic effect of the dose selection procedure in the trial by regressing the metabolic variables on the intended dose of the fixed GH dose group. The intended dose is the GH dose that the patients would have been given if they had been randomized to the group treated with individualized dose. Since this group was randomised to receive the fixed GH dose, any metabolic effect of the intended dose implicates an effect of the dose selection procedure independent of the dose given. None of the six metabolic effects appeared to be related to the dose selection procedure in the trial. Δ ALP, Δ LSTSDS, Δ insulin and Δ IGF-ISDS showed no significant effect, whereas Δ LVDd and to a lesser extend Δ heightSDS were even somewhat negative correlated to the intended dose of the fixed GH dose group (data not shown).
Body composition
Body composition was measured by dual-energy X-ray absorptiometry (DXA), using only one DPX-L scanner (Lunar Co., Madison, WI) at each study centre. Regular harmonization between the centres was performed following GCP/GMP. DXA assessment results in a three-compartment model of the body consisting of fat mass, lean soft tissue (LST) mass and bone mineral content (BMC). All analyses were conducted using the extended analysis program for total body analysis including bone mineral density (BMD).
Statistics
For all analyses, the assumptions of normality were assessed by analysis of skewness, kurtosis and frequency histograms. A p-value of < 0.05 was considered to be statistically significant.
Delta values (Δ-values) for the metabolic variables were calculated in order to quantify changes at 2 year of treatment compared with baseline. S-shaped piecewise linear regression models were fitted with GH dose as the predictor variable and the Δ-value of the metabolic variables as response variables. It consists of 3 pieces: a horizontal head and tail and a linear piece in the middle and was plotted based on the GH-dose effect on the dependent metabolic variables. Fitting this regression comes down to fitting a bounded linear model with four parameters; the usual slope and intercept of a linear regression and an upper and lower bound for the fitted model values. The GH dose effect was given by the maximum range of the fitted piecewise function, which is equal to the difference between the fitted upper and lower bound. Half of the GH dose effect on the respective variables was calculated based on the value halfway between the lowest and highest level of the piecewise function (50% Δ effect). The number of cases within each dose group was taken into account by weighting. Corresponding 90% confidence bands were calculated.
The fitted upper and lower bounds minus the fitted intercept, divided by the fitted slope provide the two breakpoints of the predictor variable separating the middle linear part from the horizontal head and tail. The ED 50% required to achieve (50% Δ effect) is located halfway the two breakpoints of the predictor variable and a corresponding 90% confidence interval is computed. We considered the ED 50% values of pairs of metabolic variables as significantly different if their ED50% values were mutually outside each other's 90% confidence interval, each providing a 5% significance level one-sided. This ED 50% is halfway the two breakpoints of the predictor variable separating the middle linear part from the horizontal head and tail.
A one-way analysis of variance (ANOVA) with GH dose as a bounded continuous predictor was performed to test the piecewise linear GH effect. A non-parametric comparison of group means (robust test of equality of means – Welch test and Brown–Forsythe test) was conducted when variances of dependent variables were not equal across groups. To examine the influence from the 17 μg/kg/day dose group consisting of only three children, analysis were repeated with these children excluded. No significant differences resulted compared to the the complete study population. Only data consistently significant were reported.
Statistical analyses were performed with SPSS 17.0 (SPSS Inc., Chicago, USA) and with Matlab version 7.13.0 (R2011b, The Mathworks, Natick, MA, USA).