Effects of Growth Hormone Replacement Therapy on Bone Mineral Density in Growth Hormone Deficient Adults: A Meta-Analysis

Objectives. Growth hormone deficiency patients exhibited reduced bone mineral density compared with healthy controls, but previous researches demonstrated uncertainty about the effect of growth hormone replacement therapy on bone in growth hormone deficient adults. The aim of this study was to determine whether the growth hormone replacement therapy could elevate bone mineral density in growth hormone deficient adults. Methods. In this meta-analysis, searches of Medline, Embase, and The Cochrane Library were undertaken to identify studies in humans of the association between growth hormone treatment and bone mineral density in growth hormone deficient adults. Random effects model was used for this meta-analysis. Results. A total of 20 studies (including one outlier study) with 936 subjects were included in our research. We detected significant overall association of growth hormone treatment with increased bone mineral density of spine, femoral neck, and total body, but some results of subgroup analyses were not consistent with the overall analyses. Conclusions. Our meta-analysis suggested that growth hormone replacement therapy could have beneficial influence on bone mineral density in growth hormone deficient adults, but, in some subject populations, the influence was not evident.


Introduction
The major role of growth hormone (GH) during childhood is to promote bone growth and linear growth, but GH continues to have important metabolic actions throughout life. Besides growth, GH is known to affect body composition, bone mineralization, and lipid and glucose metabolism [1]. For instance, GH can accelerate bone turnover, which is supported by several lines of evidence. In vitro studies show that GH and its major effector, insulin-like growth factor-1 (IGF-1), are both mitogens for osteoblasts [2,3].
The condition of GH deficiency (GHD) has been accepted as a definite syndrome, and the clinical and biochemical abnormalities in GHD patients are also well known. They involve mainly the cardiovascular system, lipid metabolism, body composition, mineral metabolism, and quality of life [4,5]. For example, adult patients with childhood-onset or adult-onset GHD exhibit reduced bone mineral density (BMD) compared with healthy controls [6,7]. Moreover, clinical studies have shown that the prevalence of fractures is 2.7-3 times higher in GHD patients than in age-matched controls. Data from these studies suggest that the increased risk may be due to GHD rather than other pituitary hormone deficiencies [8,9].
Many studies have demonstrated that the abnormalities of GHD patients may be reversed by GH replacement therapy, but the evidence is not all conclusive. In particular, the effect of GH treatment on BMD is less clear, though it is well established that GH promotes longitudinal bone growth. Some studies suggest an improvement in BMD [10], some show no effect [11], and others suggest a decrease in BMD related to GH treatment [12]. Moreover, the association 2 International Journal of Endocrinology between GH treatment and BMD may be influenced by other factors such as gender, treatment time, GH dosage, or geographic location.
We, therefore, undertook a meta-analysis on the effects of GH replacement therapy on BMD based on available studies.

Search Strategy and Inclusion Criteria.
We systematically searched Medline, Embase, and Cochrane Library for studies written in English (from their commencements to December 2012). The search used the following terms: "growth hormone, " "GH, " "somatotropin, " "bone, " "bone mineral density" and "BMD. " The following three sites of BMD were included in this meta-analysis: spine, femoral neck (FN), and total body (TB).
Studies in humans of the effects of GH treatment on BMD, regardless of sample size, were included if they met the following criteria: (1) data were reported on at least one of the three sites (spine, FN, and TB) of BMD; (2) BMD was measured by dual-energy X-ray absorptiometry (DXA); (3) we only included studies in which mean BMD and standard deviations (SDs), or standard errors (SEs) were available; (4) adult subjects (>16 years old); (5) subjects were patients with GHD who received GH treatment. The excluded studies included reviews, editorials, comments, letters, and abstracts.

Data Extraction.
Two investigators independently reviewed the articles and selected eligible studies according to the inclusion criteria for eligible studies. Irrelevant studies were excluded. For studies with the same population resources or overlapping datasets, the most complete one was included. Study details and data were extracted independently and to a standardized electronic form by two investigators, and discrepancies were adjudicated by a third reviewer until consensus was achieved on every item. The following information was extracted from each study: last name of first author, year of publication, country, subject population, mean, BMD and SDs (or SEs) of subjects at baseline and after GH treatment.

Statistical Analysis.
For this meta-analysis, all data should be given as mean and SDs. In those studies, where values of SEs were originally reported, the values of SDs were calculated. When information was reported for more than one subpopulation in one study, each subpopulation was treated as a separate comparison in our meta-analysis. BMDs in the three sites (spine, FN, and TB) were continuous outcomes presented on different scales (g/cm 2 , -scores orscores), so we used a pooled standardized mean difference (SMD) with 95% confidence intervals (CI) calculated using the final follow-up values to analyze the effects of GH treatment on BMD. All data were initially analyzed with a fixed effects model. If heterogeneity was found, the analysis should be redone using a random effects model. A value of 0.05 was considered statistically significant.
Heterogeneity of the effect across studies was assessed by statistics, which is distributed as 2 statistics. 2 statistics were provided to quantify the percentage of total variation across studies that was attributable to heterogeneity rather than to chance. An 2 value >50% represented substantial variability, and heterogeneity was considered to be significant at < 0.10, a conservative standard for meta-analyses. In the presence of heterogeneity, sensitivity analyses were performed to identify the outlier studies. The influence of outliers was also assessed to evaluate the impact of their removal. Moreover, there might be effective modificationcaused by study-level characteristics including treatment time, GH dosage, manufacturer of DXA scanner, and geographic location. Thus, subgroup analyses were further conducted to detect the source of heterogeneity. Treatment time subgroups were defined as ≤2 years or >2 years. GH dosage subgroups were defined as fixed dosage and dosage depending on serum IGF-1 values. Manufacturer of DXA scanner subgroups was defined as GE-Lunar or Hologic Inc, since the studies using other manufacturers of DXA scanner (Norland) are too few. Geographic location subgroups were defined as Europe, America, or Oceania.
We performed a visual inspection of the funnel plot for publication bias. The funnel plot should be asymmetric when there is publication bias and symmetric in the case of no publication bias. We performed Egger and Begg tests to measure the funnel plot asymmetry using a significance level of < 0.05.
All statistical analyses were performed by using STATA 11.0 (Stata Corporation, College Station, TX, USA). The results of our research were reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines.
In all eligible studies, there were 3 studies separately providing the information on more than one subpopulation. Each subpopulation was treated as a separate comparison. A total of 936 subjects were included in this meta-analysis.

Association between GH Treatment and BMD of Spine.
We initially performedthe meta-analysis on all 18 studies (including 20 comparisons) with a fixed effects model. For the presence of significant heterogeneity ( 2 = 82.9%), the analysis wasredone using a random effects model. The results suggested significant association between GH treatment and increased BMD of spine (SMD = 0.540, 95% CI [0.272, 0.808], < 0.001; 2 = 82.9%, < 0.001 for test). Sensitivity analyses showed that there was an outlier study (study ID: Rota et al.). When the outlier study was omitted, 17 studies (including 19 comparisons) were included in the meta-analysis. The heterogeneity was decreased and the results also suggested significant association between GH treatment and increased BMD of spine (SMD = 0.429, 95% CI [0.263, 0.594], < 0.001; 2 = 50.0%, = 0.007 for test) (Figure 1(a)).
To further detect the source of heterogeneity, we performed subgroup analyses stratified by the characteristics (treatment time, GH dosage, manufacturer of DXA scanner, and geographic location) of the subjects. The results did not suggest significant association between GH treatment and BMD of spine in American subjects (SMD = 0.461, 95% CI [−0.049, 0.971], = 0.076; 2 = 76.3%, = 0.001 for test). But a significant association between GH treatment and increased BMD of spine in the other subgroups was found. Significant heterogeneity was removed or decreased in some subgroups. Table 3 summarizes the subgroup analyses results.

Association between GH Treatment and BMD of FN.
Similarly, we performed the meta-analysis on all 16 studies (including 18 comparisons) with arandom effects model. The results suggested significant association between GH treatment and increased BMD of FN (SMD = 0.476, 95% CI [0.190, 0.761], = 0.001; 2 = 83.0%, < 0.001 for test). Sensitivity analyses showed that there was also an outlier study (study ID: Rota et al.). When the outlier study was omitted, 15 studies (including 17 comparisons) were included in the meta-analysis. The heterogeneity was decreased and the results also suggested significant association between GH treatment and increased BMD of FN (SMD = 0.377, 95% CI [0.158, 0.595], = 0.001; 2 = 67.8%, < 0.001 for test) (Figure 2(a)).
We also performed subgroup analyses to further detect the source of heterogeneity. The results did not suggest significant association between GH treatment and BMD of FN in subjects treated by GH for ≤2 years (SMD = 0.289, 95% CI [−0.009, 0.587], = 0.057; 2 = 51.2%, = 0.045 for test) and American subjects (SMD = 0.501, 95% CI [−0.227, 1.229], = 0.177; 2 = 86.1%, < 0.001 for test). But significant association between GH treatment and increased BMD of FN in the other subgroups were found. Moreover, the significant heterogeneity was removed or decreased in some subgroups. Table 4 summarizes the subgroup analyses results.

Association between GH Treatment and BMD of TB.
Analogously, we performed the meta-analysis on all 11 studies (including 12 comparisons) with a random effects model. The results suggested significant association between GH treatment and increased BMD of TB (SMD = 0.242, 95% CI [0.019, 0.466], = 0.034; 2 = 69.6%, < 0.001 for test) (Figure 3(a)). Sensitivity analyses showed that there was no outlier study.
We also performed subgroup analyses to further detect the source of heterogeneity. The results did not suggest significant association between GH treatment and BMD of TB in subjects with treatment time 75.6%, = 0.043 for test), but as significant association between GH treatment and increased BMD of TB in the other subgroups was found. Moreover, the significant heterogeneity was removed or decreased in some subgroups. Table 5 summarizes the subgroup analyses results.

Heterogeneity and Publication
Bias. Significant heterogeneity was separately observed among the available studies on BMD of spine, FN, and TB. To detect the source of heterogeneity, we performed subgroup analyses stratified by the characteristics of the subjects. Significant heterogeneity was removed or decreased in some subgroups but still existed in other subgroups.  For the 17 studies (with an outlier study excluded) focusing on BMD of spine, both Egger's regression ( = 0.789) and Begg methods ( = 0.889) did not show publication bias (Figure 1(b)). For the 15 studies (with an outlier study excluded) focusing on BMD of FN, both Egger's regression ( = 0.285) and Begg methods ( = 0.303) did not show publication bias (Figure 2(b)). For the 11 studies focusing on BMD of TB, both Egger's regression ( = 0.309) and Begg methods ( = 0.631) did not show publication bias (Figure 3(b)).

Discussion
In our meta-analysis, we detected an outlier study (study ID: Rota 2008) through sensitivity analyses when we performed the meta-analysis on the association of GH treatment and BMD of spine and FN. In the study mentioned above, patients aged below 30 years and above 50 years were excluded, which might make it an outlier study.
We detected significant overall association between GH treatment and increased BMD of spine, FN, and TB. GH could exert both direct and indirect effects on bone. (1) For direct effects on bone, there was increasing evidence that the GH-IGF axis played a vital role in determining BMD and maintaining bone health and that perturbations in this axis might predispose to the development of osteoporosis. Although GH could act on cells directly through specific receptors [31,32], most of its anabolic actions were mediated through IGF-1 [33][34][35]. GH stimulated the secretion of IGF-1, largely from the liver, which then acted in an endocrine fashion. GH also stimulated IGF-1 locally in target tissues such as bone, where it might act in a paracrine or autocrine fashion [36,37]. Thus, the effect of GH on bone was mediated, at least in part by IGF-1, and bone mass was known to be linked to circulating levels of IGF-1 [38]. In vitro studies had shown that GH-IGF-1 bound to preosteoblasts or mature osteoblasts to induce differentiation and proliferation while also regulating osteoclastic differentiation and activity providing a mechanism to couple bone resorption and formation [39,40]. In addition, GH also increased biomarkers of bone turnover in normal subjects as well as adults and children with GHD [13,41]. In almost all of the included studies in our meta-analysis, the serum IGF1 levels of adult GHD patients were significantly increased by the GH treatment, which were listed in Table 2.
(2) For indirect effects on bone, It was known that GH had an anabolic effect on skeletal muscle, and it particularly seemed to increase muscle mass and isometric muscle strength when given in physiologically therapeutic doses to GHD patients. Klefter and Feldt-Rasmussen analyzed many trials measuring effects of GH on both muscle and bones [42], and then suggested that there could be a connection between increases in muscle mass and strength and changes in BMD in GHD patients treated with GH. This supported the present physiological concept that the mass and strength of bones were primarily determined by dynamic loads from the skeletal muscles [43,44].
Significant heterogeneity was found in our meta-analysis. Several study-level variables leading to heterogeneity were defined by subgroup analyses including treatment time, GH dosage, manufacturer of DXA scanner and geographic location. Some results of subgroup analyses were not consistent with the overall analyses.
Firstly, we did not detect significant association between GH treatment and BMD of FN and TB in subjects with treatment time ≤2 yr. GH-IGF1 stimulated bone remodeling which occurred as a biphasic process, dominated initially by bone resorption and only later by bone formation. This biphasic sequence might also explain the initial decrease in BMD reported in several clinical trials [45,46]. Thus, significant increases in BMD did not usually occur until 12-24 months of treatment, and clinical trials with duration of 24 months or less might not be expected to find significant increases in bone parameters.
Secondly, we did not detect significant association between GH treatment and BMD of TB in subjects received fixed GH dosage (weight-or surface-area-based dosing regimens). Early studies used weight-or surface-area-based dosing regimens that resulted in a higher GH dose than titrating GH dose to normalize the serum IGF-1 level in subsequent years [47]. And the use of dose titration means that it takes longer to establish the patient on a maintenance GH dose. Thus, known differences in the time until response of BMD to GH are recognized. In our meta-analysis, GH treatment time in the studies which determined the GH dosage depending on serum IGF-1 level is mostly longer than that isthe studies which used fixed GH dosage. So, the effect of the former dosing regimens on BMD might be more evident.
Thirdly, we did not detect significant association between GH treatment and BMD of TB in the subjects whose BMD 8 International Journal of Endocrinology    Fourthly, we did not detect significant association between GH treatment and BMD of spine, F, and TB in American subjects. In addition, we did not detect significant association between GH treatment and BMD of TB in European and Oceanian subjects but got a significant overall association between GH treatment and increased BMD of TB. Although GH was a major regulator of IGF-1 concentrations, other factors such as nutrition and insulin concentration were also important in its regulation. In different countries or geographic location, the nutrition or insulin concentration of GHD patients might also be different.
Heterogeneity was removed or decreased in some subgroups but still existed in other subgroups. Thus, in addition to treatment time, GH dosage, manufacturer of DXA scanner and geographic location, there might be other factors leading to heterogeneity. For instance, previous study indicated that gender might influence the association between GH treatment and BMD, and several studies suggested that men had a greater treatment response to GH replacement than women. The mechanisms underlying these gender differences were not fully understood, but sex hormones might play a role [49,50]. However, there were too few studies analyzing the results of GH treatment separately for males and females in the studies included in our meta-analysis, which made it impossible to undertake subgroup analyses stratified by gender. For the studies focus on BMD of spine, FN and TB, both Egger's regression and Begg methods did not show publication bias.
In our study, we mainly investigated the effects of GH treatment on BMD in adult GHD patients. Besides its beneficial effects on bone, GH treatment is also suggested to alleviate at least some of the aspects of the reduced physical and psycHological health associated with GHD in adult life [51,52]. Severe quality of life (QoL) impairment is evident in a significant proportion of adults with GHD, and the beneficial effects of physiological GH replacement on QoL in affected individuals are well documented [53][54][55]. However, the risk of adverse effects may increase in the GHD adults treated with GH, such as oedema, joint stiffness, and carpal tunnel syndrome [54]. Moreover, serum IGF1 levels increase in the GHD patients with GH treatment. In the general population, higher circulating IGF1 levels are associated with increased incidence of prostate, colorectal, and premenopausal breast cancer [56]. To date, however, there have been no published long-term studies in adults with GHD treated with GH with respect to the development of nonpituitary malignancies.
The present study has some limitations that should be considered. Firstly, because only studies that were indexed by the selected databases were included for data analysis, some relevant published studies or unpublished studies might be missed, which might have biased our results. Secondly, our meta-analysis only included adult subjects since it was inappropriate to pool data from studies in children, where growth had a major effect of GH therapy; with studies in adults, this was not the case.

Conclusions
Considered together, these studies seem to indicate that GH treatment had beneficial influence on BMD in GHD adults, but in some subject populations, the influence was not evident.