Rationale: Many medications can cause cognitive side effects severe enough that a subset of patients discontinue therapy due to declines in quality of life. However, a full understanding of these deficits and the factors that confer vulnerability has yet to be reached. Here, we investigated whether a measure of speech content, lexical density (LD), is sensitive to the administration of two drugs known to cause cognitive deficits: topiramate (TPM), an anti-epileptic medication with additional indications for migraine and obesity, and lorazepam (LZP), a benzodiazipene used to treat anxiety and sleep disorders. We hypothesized that pre-treatment estimates of LD would be associated with the magnitude of drug-related changes in LD. Methods: We conducted a double-blind, randomized, placebo-controlled crossover study. After a baseline visit, subjects were randomly assigned to one of six possible sequences of three treatments: TPM (100, 150, or 200 mg), LZP (2 mg), and placebo (PBO) administered once each. At the next three visits, separated by two-week intervals, subjects received either a dose of TPM, LZP, or PBO according to their sequence. At baseline, and at .5, 2.5, and six hours after treatment, subjects completed spontaneous narrative (SN; Kirschbaum et al., 1993) and picture description (PD; Goodglass & Kaplan, 1983) tasks designed to elicit naturalistic speech samples. LD, ((adjectives+nouns+adverbs+verbs)/total words), was calculated separately for each task for each subject at each visit. To assess the magnitude of drug-related changes in LD, we calculated ‘relative change scores’ ((treatment-PBO)/PBO); enabling us to control for individual differences in unimpaired performance during the PBO visit. For each subject’s two treatment observations (TPM and LZP), we then constructed linear mixed effects models (LMEs) of relative change scores for each task; these models allowed us to determine whether each treatment differed significantly from PBO while adjusting for covariates such as treatment order, session number, drug level, sex, age, education, and estimated glomerular filtration rate. We also constructed regression models, adjusted for the same covariates, to investigate the relationship between baseline LD and the severity of LD declines observed after treatment. Results: Administration of TPM and LZP decreased LD on the SN task (TPM: ß = -.23, p < .0001; LZP: ß = -.18, p < .0001), and increased LD on the PD task (TPM: ß = .12, p < .005; LZP: ß = .15, p < .0005). Results of regression models showed no relationship between baseline LD and drug-related LD declines on the SN task (both p > .28), but robust relationships on PD (TPM: ß = -1.59, p = .017; LZP: ß = -1.97, p < .005): for both TPM and LZP, high LD on baseline PD was associated with larger LD declines after drug administration. Conclusions: We showed for the first time that pre-treatment measures of LD are associated with the severity of drug-related declines in LD. LD measures are quick and easy to acquire and may be clinically useful for predicting drug effects on language function. Funding: Please list any funding that was received in support of this abstract.: NIH NINDS RO1 NS076665; PI: Marino
