Predictor Variable Analysis to Determine Onset of Retinal Toxicity in Children on Vigabatrin Therapy

Abstract

Abstract: : Purpose: Infantile spasms (IS) is a rare form of childhood epilepsy affecting 2–5 children out of 10,000. The prognosis for children diagnosed with IS is very poor – seizure activity arrests neuro and psycho–motor development and is associated with mental retardation, cerebral palsy and a high mortality rate. Vigabatrin is an anti–epileptic drug that is very effective in the treatment of IS. It works by increasing the levels of GABA in the cerebral cortex. Vigabatrin also increases the levels of GABA in the retina and has been associated with bilateral concentric constrictions of the visual fields in 40% of the adult population being prescribed the drug. This is known as retinal toxicity. In the infant population, it is difficult to diagnose visual field constriction because the infants cannot undergo the same visual field tests that are done on adults. Therefore, in order to determine the onset of Vigabatrin toxicity in infants, other measures of visual aptitude must be performed. Previous research has shown that several responses of the electroretinogram change with Vigabatrin therapy. If visual changes can be discovered at an early stage then there is potential for minimizing or avoiding visual loss. It is therefore of prime importance to determine which variables can be used to predict early onset of toxicity. Methods: We are using Bayesian analysis to incorporate three groups of variables associated with toxicity into a predictive model. Those are: ERG variables that show change with VGB treatment, subject variables such as age and sex, and VGB variables such as cumulative dosage, days on drug, and mean daily dosages. The importance of each variable is conditional on what is known about the other variables and it’s direct relationship to toxicity. Data collected from 172 patients on Vigabatrin therapy at Sick Kids in Toronto, Canada was used for the analysis. Four models reflecting change over 6 months, 12 months, 18 months and 24 months were analyzed. Results: The models were able to correctly predict toxicity in 29 cases. This was done using cross–validation. The most accurate predictions were made with the 18 months model. Conclusions: Employing these predictive models we are able to determine which children are most likely to develop retinal toxicity. We will be implementing these predictive models into clinical practice so that children with higher toxicity predictions may be monitored more closely throughout their Vigabatrin therapy.

Keywords: clinical (human) or epidemiologic studies: biostatistics/epidemiology methodology • electroretinography: clinical • drug toxicity/drug effects