LONGITUDINAL ASSESSMENT OF HORMONAL AND PHYSICAL ALTERATIONS DURING NORMAL PUBERTY IN BOYS .4. PREDICTIONS OF ADULT HEIGHT BY THE BAYLEY-PINNEAU, ROCHE-WAINER-THISSEN, AND TANNER-WHITEHOUSE METHODS COMPARED

To determine how accurately the Roche-Wainer-Thisssen (RWT), Tanner-Wh itehouse (TW2), and Bayley-Pinneau (BP) prediction models estimated ad ult height, serial height predictions were made for 23 healthy boys (m ean initial age 10.4 +/- 1.1 years) every 8 months from 8-15 years of age. The RWT model was tested using Greulich-Pyle (RWT-GP) and Fels (R WT-Fels) bone ages. Stature was measured every 4 months until near fin al height was attained (growth rate <1cm . 8 mo(-1)). Mean age at near final height was 18.4 +/- 1.4 years. To assure that the predictions w ere as accurate and precise as possible, bone age assessments were mad e by experts in each method. To investigate the influence of maturatio n on the predictions, the boys were grouped by Fels bone ages: <11 yr, 11-13.99 yr, and 14-14.99 yr. Comparison of the prediction bias and o f the root mean square errors (RMSE) showed that the TW2 model gave th e most accurate results, followed by the RWT and BP models. The adult height was generally underpredicted by the TW2 model and overpredicted by the RWT and BP models. The RMSE was reduced for each of the models as the bone age approached maturity. The TW2 model had the smallest a verage RMSE in all bone age groups. In the <11 yr bone age group, the RWT-Fels, RWT-GP, and BP models produced RMSEs that were 16.4%, 18.4%, 62.1%, respectively, greater than the TW2 model. For the 11-13.99 yr group, RMSE by the RWT-Fels, RWT-GP, and BP models were 7.5%, 18.0%, a nd 15.2%, respectively, greater than the TW2 model. In the 14-14.99 yr group the RWT-GP model had a 45.5% greater RMSE than the TW2 model, w hereas the RWT-Fels model produced a RMSE only 15.2% greater than TW2. The RWT-Fels model produced a lower RMSE than the RWT-GP model for al l bone age groups. Although the data are probably as accurate and prec ise as presently possible, biologically significant error remains, esp ecially with overprediction of adult height in normally growing boys b y the BP and RWT models. It is recommended that regardless of the pred iction model implemented, caution be used when advising patients of th eir predicted adult height since all of the models tested had outlying predictions. (C) 1997 Wiley-Liss, Inc.