68-3

29/130

Figure 3　Difference between final and predicted height (Model 1)R-square = 0.56, P < 0.0001Figure 4　Correlation between final and predicted height (Model 3) R-square = 0.49, P < 0.0001Figure 5　Correlation between final and predicted height (Model 4) R-square = 0.61, P < 0.0001Finally, we added the height of the parents as a parameter and as a result, it showed higher accu-racy than model 2 (model 4). The equation was final height = 0.32156081 × actual measured height − 4.6652063 × calendar age + 0.41903909 × father’s height + 0.34952508 × mother’s height − 0.740469 × bone age (TW2) + 62.1007751. The R-square was 0.61 (P < 0.0001) (Figure 5).In this study, we assessed the accuracy of the two conventional models of final height prediction for Japanese young people, in a group of football players. One model uses bone age, calendar age, and actual measured height as variables and the other uses the parents’ height. We demonstrated that the model using parents’ height was more accurate (R-square was 0.52 compared with the bone age model developed by Matsuoka et al.). In the bone age model developed by Matsuoka et al., accuracy was good for a final height of between 170 to 175 cm, but less accurate outside this range. The discrepancy was distributed linearly, and we therefore optimized the coefficients with the same four variables. The R-square became 0.49, but the accuracy was still lower than using the parents’ heights (R-square = 0.52). In a new model with bone age added as well as parental height, the R-square became 0.61. As a result, a new predic-tive model by changing the coefficients of the bone age model developed by Matsuoka et al. and adding the height of the parents as variables demonstrated better accuracy. According to our data, the predic-tion model that did not use bone age was less accu-rate than our newly developed model. However, we think that the parental height model developed by Ogata et al. has sufficient accuracy and could be used in athletes without radiation exposure. This study aimed to develop prediction models at the 6th grade of primary school. Although several models for predicting final height using bone age have been reported14-16), the accuracy varies. Models with low prediction accuracy uses bone age at a single point, while those with high accuracy use changes in height and bone age over a period. As individuals’ growth speed differs17, 18), taking into account the growth process may improve accuracy. However, previous reports showed that considering growth curves did not have good accu-225Discussion