Predicting non-response to intraglandular botulinum neurotoxin A injections for drooling in children with neurodevelopmental disabilities.

Abstract

To develop robust multivariable prediction models for non-response to (1) submandibular botulinum neurotoxin A (BoNT-A) injections and (2) concurrent submandibular and parotid (four-gland) injections, to guide treatment decisions for drooling in children with neurodevelopmental disabilities, including cerebral palsy. This was a retrospective cohort study including 262 children (155 males/107 females, median age 7 years 11 months [IQR 5 years 1 month], range 4 years 0 months - 17 years 11 months) receiving submandibular injections and 74 children (52 males/22 females, median age 7 years 7 months [IQR 4 years 3 months], range 4 years 9 months - 18 years 8 months) receiving four-gland injections. Multivariable logistic regression analyses were used to estimate associations between candidate predictors and non-response 8 weeks after injection. Ninety-six children (37%) were non-responders to submandibular injections, for which developmental age was the strongest predictor (adjusted odds ratio [aOR] 2.13; 95% confidence interval [CI] 1.02-4.45 for developmental age <4 years or 4-6 years with IQ <70). Other characteristics that showed a trend towards an increased risk of non-response were diagnosis, sex, and head position. Thirty-four children (46%) were non-responders to four-gland injections, for which tongue protrusion (aOR 3.10; 95% CI 1.14-8.43) seemed most predictive, whereas multiple preceding submandibular injections (aOR 0.34; 95% CI 0.10-1.16) showed a trend towards being protective. Predictors were, however, unstable across different definitions of non-response and both models (i.e. submandibular and four-gland) had insufficient discriminative ability. Potential predictors of non-response to BoNT-A injections were identified. Nevertheless, the developed prediction models seemed inadequate for guidance of treatment decisions. Developmental age seemed most predictive of non-response to submandibular botulinum neurotoxin A injections. Non-response to concurrent submandibular and parotid injections was best predicted by tongue protrusion and number of previous injections. Multivariable prediction models including these clinical characteristics were unable to discriminate well. Predictors differed when non-response was defined using alternative outcome measures.