Long term results of growth hormone treatment in France in children of short stature: population, register based study

Subjects

The prescription of growth hormone has been regulated in France since 1973 under the medical control of Association France Hypophyse (the organisation responsible for prescribing growth hormone in France). Human extracted growth hormone—and from 1986-7 onwards growth hormone produced by recombinant DNA technology—was used. All deliveries of growth hormone to hospitals' pharmacies were registered, and relevant data on treated children were recorded on standardised charts. All children who started treatment before 31 December 1989 were identified. Only patients with Turner's syndrome (n=88), whose outcome has been reported elsewhere,14 were excluded, giving a study group of 3233 children.

Data collected

At baseline (start of treatment) and follow up visits (every 3 months until the treatment ended) the following data were recorded by individual paediatricians and communicated to Association France Hypophyse: height, weight, chronological age, bone age,15 stage of puberty,16 17 dose, frequency of injection, type of growth hormone (human or recombinant), and associated treatments. Parents' height, aetiology of short stature, and the results of two stimulation tests were provided. Growth hormone assays were performed by individual centres.

Growth hormone deficiency was diagnosed on the basis of auxometric data and growth hormone stimulation tests and was coded according to a standardised classification. The decision to continue or end the treatment was revised annually by Association France Hypophyse. The criterion for ending treatment was a growth velocity <20 mm in the previous 6 months with a bone age >=13 years for girls and >=14.5 years for boys.

Data from follow up visits were available until 31 December 1993. At that date, 1700 children had met the criterion for treatment to be ended, 923 had stopped treatment before meeting the criterion, and 610 who had not yet met the criterion were continuing treatment.

Analysis of growth and statistical methods

Growth was analysed as proposed by Cole.18 Standard deviation (SD) scores of height and weight for age and sex were calculated by using standards from a cohort of French individuals who had reached adult height in the late 1970s.19 Target height SD score was calculated as the mean of parental SD scores, obtained from standards of their (approximate) generation.20 The formula proposed by Ranke21 was used to account for the correlation between the heights of the parents.

The difference between the baseline height SD score and the score at later time points was calculated for 1, 2, 3, and 4 years of growth hormone treatment and at the end of treatment. Mean heights at the end of treatment corresponded to bone age 15.3 years (male) and 13.8 years (female) and were above 97.3% (male) and 97.8% (female) of adult height.22

A model for predicting height at the end of treatment (final height) was constructed in several stages.18 Firstly, the model had to predict outcome from demographic and auxometric information about the child at baseline (chronological and bone ages, stage of puberty, weight SD score, and target height SD score). Secondly, information about aetiology of short stature (diagnosis, peak concentration of growth hormone during stimulation tests) and about growth hormone treatment (dose, frequency of injection, type of growth hormone) and associated treatments were added to significant predictors of the preceding stage. The final stage was to test interactions between the treatment variables (dose, frequency of injection, type of growth hormone) and between these variables and weight and time interval.18 To account for the effect of regression towards the mean,23 all models included height SD score at baseline, time interval between baseline and final measurements, and the interaction between these two variables.18 As the main outcome response—final height SD score—was approximately normally distributed, no transformation was needed. Two potential predictors (mean dose and maximum stimulated peak concentration of growth hormone) were log transformed to yield more closely normally distributed variables.

The children who were treated after puberty had started were analysed separately and divided according to sex (age at puberty differs for girls and boys). The computations were performed with the SAS statistical package.24

Characteristics of subjects and treatments

Evolution of outcomes

The mean annual change in height SD score was 0.3 during the first year of treatment, decreasing steadily thereafter (fig 1). Overall, the difference between the baseline height SD score and the final height score (after a mean of 4.3 years of treatment) was 1.0 (table 1). Thus the mean final height SD score was −2.0 (that is, 1.61 m for boys and 1.49 m for girls); the deficit in height SD score was −1.1 with respect to the genetic target.

Fig 1

Mean difference between height SD score at baseline and score at later time points, plotted against years of treatment for all children in study. Numbers refer to numbers of children in each category. Standard errors of mean SD scores are about 0.05

• Download figure
• Open in new tab
• Download powerpoint

Altogether, 35% (923/2623) of children stopped receiving growth hormone treatment before the scheduled end of treatment (38% (668/1776) of children with idiopathic deficiency v 27% (215/784) of those with organic deficiency, P<0.0001; other baseline characteristics were similar for the continued or interrupted treatment groups (data not shown)). Interruption within the first 2 years of treatment was motivated by insufficient response (insufficient growth) in 98 children (fig 1) and by tiredness in 46.

Children whose treatment started after puberty responded less well to treatment from the first year of treatment onwards. In the pubertal group, the difference between the final height SD score and the baseline score was 0.8 for boys (treated for a mean of 2.8 years) and 0.5 for girls (2.7 years).

Prediction of final height

Model construction for final height SD score led to the three models shown in table 3. These models are similar and all include chronological age (with a second order polynomial decreasing effect) and target height. These two variables are the strongest predictors, which, together with effect variables in regression to the mean, explain 42% to 48% of outcome variance. Aetiology of short stature also had a predictive value for outcome. Craniospinal irradiation was always associated with a lower final height. Prepubertal children with idiopathic growth hormone deficiency responded less well to treatment than those with organic deficiency, whereas the opposite pattern was observed in pubertal children. Maximum stimulated peak concentration of growth hormone had an independent predictive value only in children treated before puberty. Variables in growth hormone treatment had much lower predictive value. Only type of growth hormone was related to the outcome: children who received human hormone had a mean height SD score 0.3 lower than children who received recombinant hormone; adjusting for mean dose of growth hormone and mean number of injections did not modify these associations. However, type of growth hormone explained only 1% of outcome variance. Puberty blockade was associated with a poorer outcome in children treated after puberty. The other study variables—namely, sex, weight SD score, pubertal stage, year that treatment started, mean dose of growth hormone, and mean number of injections—were not associated with outcome. Finally, we examined the correlation between the differences in height SD score at the start of treatment compared with the scores both after 1 year of treatment and at later time points: these correlations were low and inconsistent between groups.

Growth hormone deficiency v no deficiency

Sixty eight children of short stature but without growth hormone deficiency were treated on a compassionate basis. These children, mainly male (42) and prepubertal (56), had especially severe short stature (mean height SD score −3.4) and short parents (mean target height SD score −1.6). Annual height gains were low (fig 2), and 40 children stopped receiving treatment before the scheduled end of treatment. For those who were treated until the scheduled end, the final height SD score (mean duration of treatment 3.8 years) was −2.4. Analysis of the children with growth hormone deficiency provided results nearly identical to those obtained for the entire group (data not shown).

Fig 2

Mean difference between height SD score at baseline and score at later time points, plotted against years of treatment for short normal children. Numbers refer to numbers of children in each category. Standard errors of mean SD scores are about 0.10

• Download figure
• Open in new tab
• Download powerpoint

Factors for predicting final height

This study identifies several factors that may be of predictive value for final height. In particular, chronological age, rather than bone age, is a strong predictor. The role of age, independent of treatment duration, was observed in both prepubertal and pubertal children. This relation possibly explains in part why the results were poorer for the children treated after the onset of puberty. This agrees with recent studies that identified age as the best prognostic feature.9 10 11 12 13 There are several possible explanations. Growth hormone deficiency causes a progressive and cumulative growth deficit, which is likely to be better compensated by early rather than late treatment.9 Also, as children age, growth is less directly influenced by growth hormone, and the influence of other factors such as sex steroids increases. Thirdly, children who are short at a given moment can be so because their growth velocity (known to vary substantially with time) had previously been low. The earlier this moment, the higher would be the expected subsequent growth velocity. This is consistent with a study by Greco et al, which reported that only one in three normal children of the 1958 birth cohort in Great Britain who were short at age 7 became a short adult.27

Target height was also highly predictive of final height. This result, also observed in most previously published studies,9 10 11 12 13 underlines the strong genetic influence on growth outcomes. This dependence should be carefully considered when interpreting outcomes in treated children who have a low target height—in particular, in those with idiopathic deficiency.9 12

Aetiology of short stature also predicted final height. As expected, craniospinal irradiation was associated with the worst outcome (as radiation has a large deleterious effect on vertebral growth). Results for other types of growth hormone deficiency were apparently conflicting. Outcome was better for prepubertal children with organic deficiency than with idiopathic deficiency. The opposite was found for patients treated after the onset of puberty. This again suggests that pubertal growth may be less influenced by growth hormone than by other factors—for example, sex steroids—so that compensation of growth hormone deficiency would be less effective than before puberty. Besides, the outcome for short children without deficiency was especially poor, a result which supports recent studies suggesting that growth hormone treatment has little value for such children.28

The only treatment variable that predicted height at the end of treatment was type of growth hormone (human or recombinant). Recombinant growth hormone was superior to the human type, a reassuring result as comparative trials, stopped because of Creutzfeldt-Jakob disease, were inconclusive. Unlike some short term studies,9 10 11 12 we did not find any relation between final height and mean (log) dose of growth hormone or frequency of injections. Puberty inhibitor was associated with poorer outcome in adolescent children, whereas most did not have precocious puberty. This fits with a recent report indicating that puberty blockade does not improve height gain in children with normal puberty.29 The independent predictive values of weight and maximum stimulated peak concentration of growth hormone remain unclear. We did not find consistent association between final height and these two variables, in agreement with some13 but not other studies.9 12

Despite the significant proportion of children who discontinued growth hormone treatment before the scheduled end of treatment, the evidence of a weak or no correlation between the difference in height SD score at the start of treatment compared with the scores both after 1 year of treatment and at later time points does not support the concept of responsiveness to growth hormone in growth hormone deficiency. This has implications for clinical practice: the final outcome cannot be predicted by the initial response.

Methodological considerations

Several methodological aspects of this study should be considered. Firstly, 65 % of treated children received only, or partly, human growth hormone. Thus the outcome described may be less favourable than the current outcome with generalised treatment with recombinant growth hormone. However, the scarcity of growth hormone treatment before 1987 is likely to have led to a better selection of short children being treated. Secondly, doses and injection frequencies of growth hormone were 30% to 50% lower than currently recommended in several countries. However, we did not find any relation between dose or frequency and height gain (perhaps partly because of the relatively narrow ranges of treatment methods). Thirdly, adult heights of treated children were not available, and a surrogate had to be used—height at the scheduled end of treatment, which corresponds at least to 97.3% (male) and 97.8% (female) of adult height in normal individuals.22 However, it seems unlikely that these children would gain much, in terms of the height SD score, after the discontinuation criteria had been met and treatment stopped. Fourthly, all data were obtained from routine examination in daily practice, and their reliability may therefore be questioned. In particular, the use of different growth hormone tests and assays may have resulted in substantial variability of reported peak concentrations of growth hormone and thus have limited the power to show a predictive value for this variable. Fifthly, analysis of data from observation cohorts such as ours should not be confused with data from controlled studies. Little information is available on the spontaneous adult height of the children with growth hormone deficiency, a highly heterogeneous group. However, a thorough analysis of our results allows the design of rational trials to reassess the clinical utility of growth hormone treatment in, for example, idiopathic forms of the deficiency. Finally, all analyses were adjusted on height at baseline and time between measurement to allow for regression to the mean.18 We also used height SD score rather than height itself since the standardised value eliminates the large dependency of height on age and sex.

Conclusion

This large, register based cohort study suggests that the outcome among children treated with growth hormone in the years 1973-93 was not as good as expected. Growth hormone treatment did not restore the genetic growth pattern and thus should not be considered as a replacement treatment, and even less as a “panacea” for children of short stature.30 Our findings, together with recent evidence on the limitations of growth hormone deficiency diagnostic tests,31 32 raise questions about indications for and methods of growth hormone treatment. Further research is needed to determine optimal use of these treatments. The large costs (Fr80 000; £10 000 a year in France for the cost of growth hormone alone) and highly demanding characteristics of growth hormone treatment require sensible prescription.