ESPE2015 Poster Category 3 GH & IGF (68 abstracts)
aDepartment of Endocrinology and Metabolic Diseases, Polish Mothers Memorial Hospital Research Institute, Lodz, Poland; bDepartment of Automatics and Biomedical Engineering, AGH University of Science and Technology, Cracow, Poland
Background: Prediction of GH therapy effectiveness in children with short stature is an important issue in paediatric endocrinology.
Objective and hypotheses: The aim of the study was to create a linear regression model of GH therapy effectiveness, based on the data available before treatment.
Method: Retrospective analysis comprised the data of 150 short children (101 boys), diagnosed with isolated GH deficiency, who were treated with GH up to the attainment of final height (FH). The following parameters (input variables) were assessed before treatment for each patient: gender, chronological age (CA), bone age (BA), BA/CA ratio, height (expressed as hSDS), mothers and fathers height (expressed as mhSDS and fhSDS respectively), height velocity (HV), pubertal stage (labelled: prepubertal 0, pubertal i), GH peak after falling asleep and in two stimulation tests (all GH values log-transformed), IGF1 (expressed as S.D.s for age and sex), IGF1/IGFBP3 molar ratio, birth weight (expressed as S.D.s for gestational age). The output variable was FHSDS.
Results: The model was created on the data of 100 patients (learning group) and validated on the remaining 50 cases (testing group). The best model was described by the equation: FHSDS =0.683+0.529*hSDS0.286*IGF1 SDS0.152*HV+0.146*mhSDS+0.163*fhSDS The mean error (RMSE) of predicted FH S.D.s was 0.59 S.D. (3.5 cm) for learning group and 0.63 SD (3.8 cm) for testing group. The model explained 44% of variability of FH SDS in learning group and 36% in testing group.
Conclusion: Auxological indices and IGF1 secretion before treatment but not GH peak after failing asleep and in stimulation tests were significant predictors of GH therapy effectiveness in children with isolated GH deficiency. Relatively high amount of variability of FH S.D.s remains unexplained by the model, probably in part due to nonlinear dependencies between variables.