We aimed to test the effect of the derived genetic score for valproate response on valproate serum levels to confirm its validity to test the hypothesis whether valproate has an effect on ischemic stroke through serum level-dependent effects. Genetic variants that are associated with seizure response to valproate could be unrelated to the effect of valproate on ischemic stroke, if they influence a pathway that is not related to drug metabolism, but rather further downstream in its effect on seizure prevention. However, if the genetic variants predict valproate response through an effect on valproate serum levels below the threshold of impacting prescriber behavior, they are a proxy for genetically predicted drug exposure and thus can be used as an instrument for randomization in the test for an effect on ischemic stroke. In this special case of Mendelian randomization, this is assertion of the relevance assumption of the genetic variants.
Valproate serum levels were gathered from the primary care clinical data by using the Read codes ‘44W4.’ and ‘XE25d’. Values that were 0 (indicating off-valproate situations) and those higher than 200 μg/ml (potentially erroneous or not in μg/ml) were discarded. For each serum level value, the taken valproate dose at the time of measurement was approximated. First, the duration in days between the prescriptions before and after the date of the serum level measurement was calculated. Then, the quantity of tablets was multiplied by the dose of each tablet, divided by the duration of the prescription interval, yielding the average daily dose in mg per day. The association of valproate dose with valproate serum levels was tested in a linear regression model with valproate serum level as dependent variable and average daily valproate dose and the genetic score as independent variables. The model was additionally adjusted for age at the time of serum level measurement and sex. Levels for lamotrigine or levetiracetam were not available in the UK Biobank.
Valproate serum levels were gathered from the primary care clinical data by using the Read codes ‘44W4.’ and ‘XE25d’. Values that were 0 (indicating off-valproate situations) and those higher than 200 μg/ml (potentially erroneous or not in μg/ml) were discarded. For each serum level value, the taken valproate dose at the time of measurement was approximated. First, the duration in days between the prescriptions before and after the date of the serum level measurement was calculated. Then, the quantity of tablets was multiplied by the dose of each tablet, divided by the duration of the prescription interval, yielding the average daily dose in mg per day. The association of valproate dose with valproate serum levels was tested in a linear regression model with valproate serum level as dependent variable and average daily valproate dose and the genetic score as independent variables. The model was additionally adjusted for age at the time of serum level measurement and sex. Levels for lamotrigine or levetiracetam were not available in the UK Biobank.
