Predictive fat mass equations for spinal muscular atrophy type I children: Development and internal validation

Background: Body composition assessment is paramount for spinal muscular atrophy type I (SMA I) patients, as weight and BMI have proven to be misleading for these patients. Despite its importance, no disease-specific field method is currently available, and the assessment of body composition of SMA I patients requires reference methods available only in specialized settings. Objective: To develop predictive fat mass equations for SMA I children based on simple measurements, and compare existing equations to the new disease-specific equations. Design: Demographic, clinical and anthropometric data were examined as potential predictors of the best candidate response variable and non-linear relations were taken into account by transforming continuous predictors with restricted cubic splines. Alternative models were fitted including all the dimensions revealed by cluster analysis of the predictors. The best models were then internally validated, quantifying optimism of the obtained performance measures. The contribution of nusinersen treatment to the unexplained variability of the final models was also tested. Results: A total of 153 SMA I patients were included in the study, as part of a longitudinal observational study in SMA children conducted at the International Center for the Assessment of Nutritional Status (ICANS), University of Milan. The sample equally represented both sexes (56% females) and a wide age range (from 3 months to 12 years, median 1.2 years). Four alternative models performed equally in predicting fat mass fraction (fat mass/body weight). The most convenient was selected and further presented. The selected model uses as predictors sex, age, calf circumference and the sum of triceps, suprailiac and calf skinfold thicknesses. The model showed high predictive ability (optimism corrected coefficient of determination, R2 = 0.72) and internal validation indicated little optimism both in performance measures and model calibration. The addition of nusinersen as a predictor variable did not improve the prediction. The disease-specific equation was more accurate than the available fat mass equations. Conclusions: The developed prediction model allows the assessment of body composition in SMA I children with simple and widely available measures and with reasonable accuracy.