Estimation of standing height in spina bifida: model development and validation

Abstract

Objective

Childhood standing height has been estimated from arm span-related (height_AS) models. The authors aimed to develop and cross-validate a height_AS model in individuals with spina bifida (SB) and examine the accuracy of existing height_AS models.

Methods

Participants were individuals with sacral and low-lumbar SB (n = 14) and non-SB (n = 83), 7–16 years old. Arm span, age, sex, and group (SB vs. non-SB) were candidate height predictors. Sequential regression and leave-one-out cross-validation approaches were used for the model development (M1) and cross-validation (M1–M5). Existing models were: an SB-specific model from Polfuss et al. (M2) and non-SB specific models from Gauld et al. (M3), Mulu et al. (M4), and Zverev et al. (M5) studies.

Results

Arm span and group explained 95 % of the variance in height (R² = 0.95; p < 0.001; SEE = 3.666 cm) and were included in the M1. Mean differences between actual and estimated height were 0.0 cm (M1), 0.4 cm (M2), and 0.5 cm (M5), all not significant (p > 0.05). However, Bland-Altman analysis revealed some variability in the predictability of the models across participants with limits of agreement ranging from 7.4 to 10.9 cm. Considerable errors were observed with M3 (mean diff: −5.58 cm, 95 % CI: −1.6, −20.2 cm), and M4 (mean diff: 10.5 cm, 95 % CI: −13.8, −27.3 cm).

Conclusions

Models (M1, M2 and M5) may accurately estimate standing height in groups of children with SB. However, due to the wide limits of agreement, caution is recommended when applying these models for individual height estimations.