Statistical modelling to estimate patients’ weight in Types 1?3 SMA


Topic:

Clinical Trials

Poster Number: Virtual

Author(s):

Richard Houghton, F. Hoffmann-La Roche Ltd, Matt Kent, Genesis Research, C. Simone Sutherland, F. Hoffmann-La Roche Ltd, Isaac Gravestock, F. Hoffmann-La Roche Ltd, Eleni Gaki, Roche Products Ltd, Tammy McIver, Roche Products Ltd, Laurent Servais, I-Motion Institute, Hôpital Armand Trousseau

Background
Individuals with spinal muscular atrophy (SMA) may have atypical weight with possible correlation to other disease morbidity, such as feeding difficulties and reduced mobility. Statistical models that describe weight deviations in patients with SMA compared with the general population are limited.

Methods
Using baseline data from the FIREFISH (Type 1 SMA; NCT02913482), SUNFISH (Types 2 and 3 SMA; NCT02908685), JEWELFISH (Types 1?3 SMA; NCT03032172) and NatHis-SMA (Types 2 and 3 SMA; NCT02391831) studies, percent deviation of patient weight from World Health Organization median references was calculated. Reference weights were based on age, sex and height standards up to age 19 years. Percent weight deviation was linearly regressed on 10 a priori predictors. Four statistical models were developed using 75% of the data, and model fits tested using the remaining 25%. Final model selection was based on goodness-of-fit and statistical parsimony.

Results
This analysis included 526 patients with median age of 10 (range 0?61) years, and 50% were female. Non-sitting motor status (n=145; 28%) and prior exposure to SMA disease-modifying therapies (DMTs; n=160; 30%) were associated with below-reference weights, whereas sitting and more-advanced motor statuses; SMA type; age; sex; region; survival of motor neuron 2 gene copy number; feeding support; history of scoliosis; gastroesophageal reflux disease; and pneumonia were not. An alternative model without prior treatment and a null model had comparable ability to estimate reference weight in the 25% hold-out set. The intercept term was significant in all models, indicating that, on average, non-sitting patients with SMA weigh less than the general population reference weights according to age, sex and height.

Conclusion
This study provided a validated algorithm for estimating the weight of patients with SMA based on few input parameters. Variability in weight is primarily accounted for by age, sex and height. A simple yet accurate prediction of patient weight could be useful in clinical practice or for resource planning, as dose regimens for some DMTs are weight based. Associations implied by our models are not necessarily indicative of causal relationships.