Maximizing Knowledge Extraction From Patient-Reported Outcome Data Using Exposure-Outcome Item Response Modeling...

This investigation was undertaken to maximally extract hidden knowledge from an efavirenz-based trial data set using an item response theory-based approach to exposure-outcome analysis. The aim was to understand the influence of efavirenz exposure on the underlying neuropsychiatric impairment in HIV/AIDS patients. Data from 196 individuals with 4136 neuropsychiatric impairment symptom observations at baseline and 2 and 12 weeks of 600-mg efavirenz-based therapy was analyzed. The 7 symptoms were categorized as sleep disorders (3), hallucinations (3), and cognitive impairment (1). A longitudinal item response theory model incorporating 3 latent variables based on the symptom categories and a linear disease progression model with a symptomatic drug effect was developed in NONMEM 7.4.1. The model adequately characterized the observed symptoms and revealed the hidden knowledge on the informativeness of symptoms in characterizing the underlying neuropsychiatric impairment. Informativeness, which was affected by underlying impairment severity and efavirenz therapy duration, varied among symptoms. Sleep disorders were the most efavirenz-sensitive symptom category. Vivid dreams and auditory hallucinations were most informative in their respective symptom categories. Mini-Mental State Examination score cutoff levels used to classify the severity of cognitive impairment did not distinctively correspond with neuropsychiatric impairment severity. Efavirenz treatment effect on the severity of neuropsychiatric impairment was not more than 15%. Simulation of individual symptoms at the 400-mg dose only reduced the prevalence of sleep disorder symptoms. Use of the exposure-item response theory modeling approach maximally extracted hidden knowledge about efavirenz-induced neuropsychiatric impairment and appropriately characterized the impact of dose reduction on specific neuropsychiatric impairment symptoms.