Context: Growth analyses have traditionally been done by non-structural descriptive statistics or by fitting models. Assuming reciprocity of weight and height on each other, we utilize ML to predict ages 7-18y height based on height and weight data up to age 6y.
Methods: Primary and secondary features of 1596 subjects(798 boys) 0-19y from the longitudinal GrowUp 1974 Gothenburg cohort with emphasis on infancy were utilized to train multiple regressors. The accuracy of the model was evaluated 5-fold by cross validation, and by testing 600 additional subjects of the same study. The system also was validated with 180 subjects of the Edinburgh Longitudinal Growth Study.
Results: Random Forest Regressor using gender and heights data at age 3.3-6.0y produced the most accurate predictions. The prediction accuracy against actual final height (R-square) increase from 0.580 at age 7 and 0.592 at age 13 to 0.837 at age 18. Actual heights were 173.8±9.2(SD) and predicted heights 173.3±8.0cm. Verification of prediction for 600 additional GrowUp children showed prediction/actual R2 of 0.76, when for the Edinburgh cohort - 0.38.
Conclusions: 1.ML and AI was used successfully for the first time to predict adult height based on early (<=6y) height and weight. 2.Prediction accuracy at age 0-6y for age 18y height is better than the bone-age based TW3 method. 3.The best features for prediction are sex and heights at age 3.3-6.0y, when childhood growth velocity has stabilized. 4.The success of ML strongly depends on the structure of data sampling and cannot be easily inferred between dissimilar cohorts.
