Table 4.

Regression model of AGS incidence and severity shows significant treatment effect in lovastatin versus simvastatin groups

Regression model	Genotype effect	Treatment effect	Interaction effect
Logistical regression, type 2 ANOVA (Muscas et al., 2019)	p = 6.22 × 10–12	p = 0.053	p = 0.263
Logistical regression (Muscas et al., 2019) + 100 mg/kg lovastatin from Osterweil et al. (2013; lovastatin groups separated)	p = 1.58 × 10–13	p = 0.00021	p = 0.4
Logistical regression (Muscas et al., 2019) + 100 mg/kg lovastatin from Osterweil et al. (2013; lovastatin groups collapsed)	p = 1.86 × 10–13	p = 9.22 × 10–5	p = 0.5
Logistical regression (Muscas et al., 2019) + all lovastatin groups from Osterweil et al. (2013)	p < 2.2 × 10–16	p = 8.08 × 10–9	p = 0.4
Multinominal regression (Muscas et al., 2019)	p = 8.62 × 10–12	p = 0.033	p = 0.34

Re-running the logistical regression comparing lovastatin and simvastatin treatments using a type 2 ANVOA shows a non-significant trend towards an effect of treatment. Adding data from the FVB 100 mg/kg lovastatin group originally published in Osterweil et al. (2013) shows a significant treatment effect either when kept separate or when collapsed into the existing lovastatin group. Adding data from additional lovastatin treatment groups from C57BL6 cohorts from Osterweil et al. (2013; 10, 30, and 100 mg/kg) further increases the significance of the treatment effect. As the interaction of genotype and treatment does not reach significance using this model, it may be that lovastatin corrects seizures in both WT and Fmr1^-/y mice equally; however, the low number of animals have seizures in the WT groups makes this difficult to assess. To compare lovastatin versus simvastatin treatment groups, a multinomial regression model of seizure severity scores with genotype and treatment effect was performed in R using the multinom function in the nnet package (see Extended Data Figure 1-3).